guest_gpu: fix private AGC shader register regression

2026-10-28 08:13:05 +00:00 · 2026-09-01 04:08:41 +04:00
431 changed files with 36630 additions and 88873 deletions
@@ -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
@@ -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 controller_settings_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|controller_settings|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 controller_settings_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|controller_settings|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 controller_settings_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|controller_settings|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
@@ -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
@@ -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"
+    }
 }
+
@@ -2,31 +2,10 @@ set(BUILD_SHARED_LIBS OFF CACHE BOOL "" FORCE)

 add_subdirectory(cpuinfo)

+# The Windows guest red-zone patcher needs a decoder/encoder and a small
+# trampoline assembler. Pin the same revisions used by shadPS4's implementation.
 include(FetchContent)

-# 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 guest instruction patcher needs a decoder/encoder and a small
-# trampoline assembler. Pin the same revisions used by shadPS4's implementation.
 set(XBYAK_BUILD_TESTS OFF CACHE BOOL "" FORCE)
 FetchContent_Declare(
 	xbyak
@@ -48,57 +27,9 @@ FetchContent_Declare(
 )

 FetchContent_MakeAvailable(xbyak zydis)
-target_compile_definitions(xbyak INTERFACE XBYAK_NO_EXCEPTION)
-
-# 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)
@@ -155,9 +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)
@@ -189,7 +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)
@@ -14,7 +14,7 @@
 #include <cpuinfo/internal-api.h>
 #include <cpuinfo/log.h>

-/* Polyfill recent CPUFAMILY_ARM_* values for older platform headers */
+/* Polyfill recent CPUFAMILY_ARM_* values for older SDKs */
 #ifndef CPUFAMILY_ARM_MONSOON_MISTRAL
 	#define CPUFAMILY_ARM_MONSOON_MISTRAL   0xE81E7EF6
 #endif
@@ -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
@@ -1,7 +1,7 @@
-cmake_minimum_required(VERSION 3.19...4.3)
+cmake_minimum_required(VERSION 3.12...4.3)

 if(CMAKE_VERSION VERSION_GREATER_EQUAL 4.0)
-	set(CMAKE_POLICY_VERSION_MINIMUM 3.10 CACHE STRING "Minimum policy version for bundled dependencies")
+	set(CMAKE_POLICY_VERSION_MINIMUM 3.5 CACHE STRING "Minimum policy version for bundled dependencies")
 endif()

 get_property(KYTY_GENERATOR_IS_MULTI_CONFIG GLOBAL PROPERTY GENERATOR_IS_MULTI_CONFIG)
@@ -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")
@@ -142,7 +126,7 @@ if (KYTY_LINKER STREQUAL LD AND NOT LINUX)
 	set(KYTY_LD_OPTIONS "-Wl,--image-base=0x100000000000")
 endif()

-project(Kyty${KYTY_PROJECT_NAME}${CMAKE_BUILD_TYPE}${KYTY_COMPILER} VERSION 0.3.0)
+project(Kyty${KYTY_PROJECT_NAME}${CMAKE_BUILD_TYPE}${KYTY_COMPILER} VERSION 0.2.2)

 include_directories(
 	${KYTY_THIRD_PARTY_DIR}/gtest/include
@@ -244,7 +228,7 @@ file(GLOB kyty_emulator_src CONFIGURE_DEPENDS
 	"${KYTY_SOURCE_DIR}/loader/*.h"
 )

-find_program(KYTY_GLSLANG_VALIDATOR NAMES glslangValidator glslang REQUIRED)
+find_program(KYTY_GLSLANG_VALIDATOR glslangValidator REQUIRED)
 set(gpu_tiler_shader_dir "${KYTY_SOURCE_DIR}/graphics/host_gpu/shaders")
 set(gpu_tiler_generated_dir "${PROJECT_BINARY_DIR}/gpu_tiler_shaders")
 set(gpu_tiler_shader_names
@@ -259,7 +243,7 @@ set(gpu_tiler_shader_names
 	depth
 	promote_d16
 	demote_d16
-	color_transform
+	swap_bgra16
 )
 file(GLOB gpu_tiler_shader_includes CONFIGURE_DEPENDS "${gpu_tiler_shader_dir}/gpu_tiler_*.inc")
 foreach(shader_name IN LISTS gpu_tiler_shader_names)
@@ -301,7 +285,6 @@ set(gpu_blit_generated_dir "${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
 	${KYTY_THIRD_PARTY_DIR}/cpuinfo/include
@@ -393,10 +390,8 @@ add_kyty_full_emulator_test(shader_cfg_tests "${KYTY_TESTS_DIR}/shaderCfgTests.c

 add_executable(scalar_provenance_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/ScalarProvenanceTests.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/gpu_format.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ReadLaneElimination.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/SrtWalker.cpp"
 )
 target_link_libraries(scalar_provenance_tests fmt::fmt)
@@ -409,35 +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 ${inc_headers})
-
-add_executable(trophy_system_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/TrophySystemTests.cpp"
-)
-target_link_libraries(trophy_system_tests common)
-target_include_directories(trophy_system_tests PRIVATE ${inc_headers})
-
-add_executable(avplayer_file_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/AvPlayerFileTests.cpp"
-	"${KYTY_SOURCE_DIR}/loader/timer.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"
@@ -460,15 +431,16 @@ add_executable(shader_vertex_metadata_tests EXCLUDE_FROM_ALL
 )
 target_include_directories(shader_vertex_metadata_tests PRIVATE ${inc_headers})

-add_executable(resource_materialization_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/ResourceMaterializationTests.cpp"
+add_executable(shader_stage_runtime_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/ShaderStageRuntimeTests.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/gpu_format.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/shaderStageRuntime.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/SrtWalker.cpp"
 )
-target_link_libraries(resource_materialization_tests fmt::fmt)
-target_include_directories(resource_materialization_tests PRIVATE ${inc_headers})
+target_link_libraries(shader_stage_runtime_tests fmt::fmt)
+target_include_directories(shader_stage_runtime_tests PRIVATE ${inc_headers})

 add_executable(resource_tracking_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/ResourceTrackingTests.cpp"
@@ -479,6 +451,7 @@ add_executable(resource_tracking_tests EXCLUDE_FROM_ALL
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ShaderInfoCollection.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/BindingLayout.cpp"
 )
 target_link_libraries(resource_tracking_tests fmt::fmt)
 target_include_directories(resource_tracking_tests PRIVATE ${inc_headers})
@@ -491,53 +464,6 @@ 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 ${inc_headers})
-
-add_executable(controller_settings_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/ControllerSettingsTests.cpp"
-)
-target_link_libraries(controller_settings_tests common fmt::fmt SDL3::SDL3-static)
-target_include_directories(controller_settings_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"
@@ -548,7 +474,6 @@ 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"
@@ -615,34 +540,17 @@ 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 trophy_system COMMAND $<TARGET_FILE:trophy_system_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 shader_stage_runtime COMMAND $<TARGET_FILE:shader_stage_runtime_tests>)
 	add_test(NAME resource_tracking COMMAND $<TARGET_FILE:resource_tracking_tests>)
 	add_test(NAME event_queue_lifetime COMMAND $<TARGET_FILE:event_queue_lifetime_tests>)
 	add_test(NAME 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 controller_settings COMMAND $<TARGET_FILE:controller_settings_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
@@ -660,27 +568,24 @@ if(BUILD_TESTING)

 	add_test(NAME texture_cache_layered_image
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --layered-image-only)
-	add_test(NAME host_image_allocation
-		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --host-image-allocation-only)
 	add_test(NAME texture_cache_image_views
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --image-view-cache-only)
 	add_test(NAME texture_cache_storage_sampled
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --storage-sampled-only)
 	add_test(NAME texture_cache_depth_readback
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --depth-readback-only)
-	add_test(NAME compute_meta_clear_classification
-		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)

-	add_test(NAME texture_cache_image_overlap
-		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --image-overlap-only)
-	add_test(NAME texture_cache_htile_clear
-		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --htile-clear-only)
-	add_test(NAME buffer_cache_ranges
-		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --buffer-cache-range-only)
+	if(WIN32)
+		# These tests still depend on the Windows multisample-depth path.
+		add_test(NAME texture_cache_image_overlap
+			COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --image-overlap-only)
+		add_test(NAME texture_cache_htile_clear
+			COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --htile-clear-only)
+		add_test(NAME buffer_cache_ranges
+			COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --buffer-cache-range-only)
+	endif()

 	add_custom_target(kyty_tests DEPENDS
 		ime_dialog_tests
@@ -689,13 +594,9 @@ if(BUILD_TESTING)
 		image_page_table_tests
 		memory_tracker_tests
 		page_manager_tests
-		archive_file_tests
-		trophy_system_tests
-		avplayer_file_tests
 		bit_array_tests
-		lru_cache_tests
 		shader_vertex_metadata_tests
-		resource_materialization_tests
+		shader_stage_runtime_tests
 		resource_tracking_tests
 		event_queue_lifetime_tests
 		kernel_file_system_tests
@@ -703,47 +604,13 @@ 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
-		controller_settings_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"
-		"${KYTY_SOURCE_DIR}/loader/x64InstructionEmulator.cpp"
-		"${KYTY_SOURCE_DIR}/loader/guestInstructionPatcher.cpp"
-	)
-	target_link_libraries(macos_sse4a_tests common fmt::fmt Zydis::Zydis xbyak::xbyak)
-	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)

-set(KYTY_TROPHY_UNLOCK_SOUND "${CMAKE_CURRENT_SOURCE_DIR}/assets/sounds/trophy-unlock.wav")
-add_custom_target(kyty_trophy_sound
-	COMMAND ${CMAKE_COMMAND} -E make_directory "$<TARGET_FILE_DIR:kyty_emulator>/assets/sounds"
-	COMMAND ${CMAKE_COMMAND} -E copy_if_different "${KYTY_TROPHY_UNLOCK_SOUND}"
-		"$<TARGET_FILE_DIR:kyty_emulator>/assets/sounds/trophy-unlock.wav"
-	VERBATIM)
-add_dependencies(kyty_emulator kyty_trophy_sound)
-install(FILES "${KYTY_TROPHY_UNLOCK_SOUND}" DESTINATION assets/sounds)
-
 target_link_libraries(kyty_emulator ${kyty_emulator_link_libraries})
 if (WIN32)
 	target_link_libraries(kyty_emulator opengl32)
@@ -779,16 +646,7 @@ 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)
@@ -806,10 +664,6 @@ 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 .)
@@ -1,20 +1,16 @@
 # KytyPS5

-[![Build KytyPS5 (Windows)](https://img.shields.io/github/check-runs/KytyPS5/KytyPS5/main?nameFilter=Build%20KytyPS5%20%28Windows%29&label=Build%20KytyPS5%20%28Windows%29)](https://github.com/KytyPS5/KytyPS5/actions/workflows/build.yml)
-[![Build KytyPS5 (Linux)](https://img.shields.io/github/check-runs/KytyPS5/KytyPS5/main?nameFilter=Build%20KytyPS5%20%28Linux%29&label=Build%20KytyPS5%20%28Linux%29)](https://github.com/KytyPS5/KytyPS5/actions/workflows/build.yml)
-[![Build KytyPS5 (macOS)](https://img.shields.io/github/check-runs/KytyPS5/KytyPS5/main?nameFilter=Build%20KytyPS5%20%28macOS%29&label=Build%20KytyPS5%20%28macOS%29)](https://github.com/KytyPS5/KytyPS5/actions/workflows/build.yml)
+[![Build KytyPS5 (Windows)](https://img.shields.io/github/actions/workflow/status/KytyPS5/KytyPS5/build.yml?branch=main&event=push&label=Build%20KytyPS5%20%28Windows%29)](https://github.com/KytyPS5/KytyPS5/actions/workflows/build.yml)
+[![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
@@ -40,20 +36,20 @@ 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>Saros</strong><br>
-      <img src="docs/screenshots/ps5-03.png" width="300" alt="Saros running in KytyPS5">
+      <strong>Dreaming Sarah</strong><br>
+      <img src="docs/screenshots/ps5-03.png" width="300" alt="Dreaming Sarah running in KytyPS5">
    </td>
  </tr>
  <tr>
@@ -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 6</strong><br>
-      <img src="docs/screenshots/ps5-06.jpg" width="300" alt="UFC 6 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
 ```

@@ -136,13 +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) (`glslang` or `glslangValidator`) on `PATH`
- Python 3 on `PATH`; the bundled SPIRV-Tools runs it at configure time

 The Microsoft C++ compiler (`cl.exe`) is not supported; use `clang-cl`.

@@ -171,19 +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 python3 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
@@ -198,50 +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
@@ -252,7 +203,7 @@ Requirements:

 - An Apple Silicon Mac with Rosetta 2 installed (`softwareupdate --install-rosetta`)
 - Xcode (or the Command Line Tools)
- Homebrew packages: `brew install cmake ninja glslang python`
+- Homebrew packages: `brew install cmake ninja glslang`
 - Qt 6 (Concurrent, Network, Widgets) with x86-64 support. The official Qt installation is
  universal and works; Homebrew's Qt is arm64-only and will not link

@@ -334,21 +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
@@ -392,6 +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.
@@ -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
-}
@@ -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; };
-        });
-    };
-}
@@ -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
-  '';
-}
@@ -1,4 +0,0 @@
-let
-  pkgs = import <nixpkgs> { };
-in
-import ./nix/devshell.nix { inherit pkgs; }
@@ -26,8 +26,6 @@ target_link_libraries(common
 		cpuinfo
 		kyty_git_version
 		spdlog::spdlog
-		ZArchive::zarchive
-		nlohmann_json::nlohmann_json
 )

 get_property(inc_headers TARGET common PROPERTY INCLUDE_DIRECTORIES)
@@ -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
@@ -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
@@ -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_ */
@@ -1,5 +1,6 @@
 #include "common/assert.h"

+#include "common/debug.h"
 #include "common/logging/log.h"
 #include "common/subsystems.h"
 #include "kytyGitVersion.h"
@@ -11,10 +12,25 @@

 namespace Common {

+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS && KYTY_BUILD == KYTY_BUILD_DEBUG &&                    \
+    KYTY_COMPILER == KYTY_COMPILER_CLANG
+constexpr int PRINT_STACK_FROM = 4;
+#else
+constexpr int PRINT_STACK_FROM = 2;
+#endif
+
 static std::string BuildFatalReport(const char* title, std::string_view text, const char* file,
                                    int line) {
-	return fmt::format("--- Build ---\n{}\n{}\n{} in {}:{}\n", KYTY_BUILD_LABEL, title, text, file,
-	                   line);
+	DebugStack stack;
+	DebugStack::Trace(&stack);
+
+	std::string report = "--- Build ---\n" KYTY_BUILD_LABEL "\n--- Stack Trace ---\n";
+	for (int i = PRINT_STACK_FROM; i < stack.depth; i++) {
+		report += fmt::format("[{}] {:016x}\n", i - PRINT_STACK_FROM,
+		                      static_cast<uint64_t>(stack.GetAddr(i)));
+	}
+	report += fmt::format("{}\n{} in {}:{}\n", title, text, file, line);
+	return report;
 }

 static int DbgReport(const char* title, std::string_view text, const char* file, int line) {
@@ -43,7 +59,6 @@ int DbgExitHandler(const char* file, int line, fmt::text_style style, std::strin
 }

 void DbgExit(int status) {
-	Subsystems::EmergencyShutdownActive();
 	std::fflush(nullptr);
 	std::_Exit(status);
 }
@@ -29,7 +29,11 @@ void DbgExit(int status);

 } // namespace Common

+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS || KYTY_PLATFORM == KYTY_PLATFORM_LINUX
 #define EXIT_HALT() (Common::DbgExit(321), 1)
+#else
+#define EXIT_HALT() (std::_Exit(321), 1)
+#endif

 #ifndef KYTY_FINAL
 #define EXIT_IF(x)                                                                                 \
@@ -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_ */
@@ -1,7 +1,20 @@
 #include "common/debug.h"

+#include "common/platform/sysDbg.h"
+
 namespace Common {

+void DebugStack::Print(int from, [[maybe_unused]] bool with_name) const {
+	for (int i = from; i < depth; i++) {
+		printf("[%d] %016" PRIx64 "\n", i - from, static_cast<uint64_t>(GetAddr(i)));
+	}
+}
+
+void DebugStack::Trace(DebugStack* stack) {
+	stack->depth = DEBUG_MAX_STACK_DEPTH;
+	SysStackWalk(stack->stack.data(), &stack->depth);
+}
+
 std::string Debug::GetCompiler() {
 #if KYTY_COMPILER == KYTY_COMPILER_CLANG
 	return "clang";
@@ -24,4 +37,8 @@ std::string Debug::GetLinker() {
 #endif
 }

+std::string Debug::GetBitness() {
+	return "64";
+}
+
 } // namespace Common
@@ -2,16 +2,42 @@
 #define KYTY_COMMON_DEBUG_H_

 #include "common/common.h"
+#include "common/stringUtils.h"

-#include <string>
+#include <array>
+#include <cstring>

 namespace Common {

+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS && KYTY_BUILD == KYTY_BUILD_DEBUG &&                    \
+    KYTY_COMPILER == KYTY_COMPILER_CLANG
+constexpr int DEBUG_MAX_STACK_DEPTH = 20;
+#else
+constexpr int DEBUG_MAX_STACK_DEPTH = 15;
+#endif
+
 namespace Debug {
 std::string GetCompiler();
 std::string GetLinker();
+std::string GetBitness();
 } // namespace Debug

+struct DebugStack {
+	int                                      depth {0};
+	std::array<void*, DEBUG_MAX_STACK_DEPTH> stack {};
+
+	[[nodiscard]] uintptr_t GetAddr(int i) const { return reinterpret_cast<uintptr_t>(stack[i]); }
+
+	void Print(int from, bool with_name = true) const;
+
+	void CopyTo(DebugStack* s) const {
+		std::memcpy(s->stack.data(), stack.data(), sizeof(void*) * depth);
+		s->depth = depth;
+	}
+
+	static void Trace(DebugStack* stack);
+};
+
 } // namespace Common

 #endif /* KYTY_COMMON_DEBUG_H_ */
@@ -23,7 +23,6 @@ void Load(const ConfigOptions& cfg) {
 	EXIT_IF(g_config == nullptr);
 	EXIT_IF(cfg.user_name.empty() || cfg.user_name.size() > MAX_USER_NAME_LENGTH);
 	EXIT_IF(!IsConfiguredUserIdValid(cfg.user_id));
-	EXIT_IF(cfg.controller_speaker_volume > 100 || cfg.controller_vibration_intensity > 100);

 	*g_config = cfg;
 }
@@ -44,22 +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;
-}
-
-uint32_t GetControllerSpeakerVolume() {
-	return g_config->controller_speaker_volume;
-}
-
-uint32_t GetControllerVibrationIntensity() {
-	return g_config->controller_vibration_intensity;
-}
-
 PresentMode GetPresentMode() {
 	return g_config->present_mode;
 }
@@ -72,18 +55,6 @@ bool FullscreenEnabled() {
 	return g_config->fullscreen_enabled;
 }

-bool HideCursorEnabled() {
-	return g_config->hide_cursor_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);
 }
@@ -104,7 +75,7 @@ ShaderOptimizationType GetShaderOptimizationType() {
 	return g_config->shader_optimization_type;
 }

-LogDirection GetShaderLogDirection() {
+ShaderLogDirection GetShaderLogDirection() {
 	return g_config->shader_log_direction;
 }

@@ -124,7 +95,7 @@ bool GraphicsDebugDumpEnabled() {
 	return g_config->graphics_debug_dump_enabled;
 }

-LogDirection GetPrintfDirection() {
+OutputDirection GetPrintfDirection() {
 	return g_config->printf_direction;
 }

@@ -132,8 +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() {
@@ -152,26 +123,10 @@ bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }

-bool SyncRawImageBuffersEnabled() {
-	return g_config->sync_raw_image_buffers;
-}
-
-bool TessellationEnabled() {
-	return g_config->tessellation_enabled;
-}
-
-bool TrophyEnabled() {
-	return g_config->trophy_enabled;
-}
-
 bool PlayGoHackEnabled() {
 	return g_config->playgo_hack_enabled;
 }

-bool SkipNoticeScreen() {
-	return g_config->skip_notice_screen;
-}
-
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 bool RedZoneProtectionEnabled() {
 	return g_config->red_zone_protection_enabled;
@@ -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;
@@ -46,38 +47,27 @@ 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;
-	uint32_t               controller_speaker_volume      = 50;
-	uint32_t               controller_vibration_intensity = 100;
-	PresentMode            present_mode                = PresentMode::Mailbox;
+	PresentMode            present_mode                = PresentMode::Fifo;
 	int32_t                gpu_index                   = -1;
 	bool                   fullscreen_enabled          = false;
-	bool                   hide_cursor_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                   sync_raw_image_buffers      = false;
-	bool                   tessellation_enabled        = false;
-	bool                   trophy_enabled              = true;
 	bool                   playgo_hack_enabled         = false;
-	bool                   skip_notice_screen          = false;
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	bool red_zone_protection_enabled = false;
 #endif
@@ -90,23 +80,16 @@ uint32_t GetScreenWidth();
 uint32_t GetScreenHeight();
 const std::string& GetUserName();
 int32_t  GetUserId();
-const std::string& GetAudioInputDevice();
-const std::optional<ControllerColor>& GetControllerColor();
-uint32_t GetControllerSpeakerVolume();
-uint32_t GetControllerVibrationIntensity();
 PresentMode GetPresentMode();
 int32_t GetGpuIndex();
 bool     FullscreenEnabled();
-bool     HideCursorEnabled();
-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();
@@ -114,10 +97,10 @@ std::filesystem::path GetCommandBufferDumpFolder();

 bool GraphicsDebugDumpEnabled();

-LogDirection          GetPrintfDirection();
+OutputDirection       GetPrintfDirection();
 std::filesystem::path GetPrintfOutputFile();

-bool ProfilerEnabled();
+ProfilerDirection GetProfilerDirection();

 bool SpirvDebugPrintfEnabled();

@@ -125,11 +108,7 @@ bool GpuAssistedValidationEnabled();

 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
-bool SyncRawImageBuffersEnabled();
-bool TessellationEnabled();
-bool TrophyEnabled();
 bool PlayGoHackEnabled();
-bool SkipNoticeScreen();
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 bool RedZoneProtectionEnabled();
 #endif
@@ -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>()) {
@@ -71,7 +69,7 @@ File::File(): m_p(std::make_unique<FilePrivate>()) {
 }

 File::~File() {
-	Close();
+	EXIT_IF(m_p->f != nullptr);
 }

 File::File(const std::filesystem::path& name): 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);
 }

@@ -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);
@@ -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_ */
@@ -5,17 +5,14 @@

 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #include <windows.h> // IWYU pragma: keep
-#else
-#include <algorithm>
+#elif defined(__APPLE__)
 #include <csignal>
-#include <cstdlib>
-#include <initializer_list>
-#include <unistd.h>
-#if defined(__APPLE__)
 #include <sys/ucontext.h>
 #else
+#include <csignal>
+#include <initializer_list>
 #include <ucontext.h> // IWYU pragma: keep
-#endif
+#include <unistd.h>
 #endif

 // IWYU pragma: no_include <errhandlingapi.h>
@@ -35,51 +32,6 @@ static_assert(decltype(g_handler)::is_always_lock_free);
 static_assert(decltype(g_install_state)::is_always_lock_free);
 #endif

-#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX
-
-// macOS uses the same POSIX platform setting and needs a signal stack too.
-class ThreadSignalStack {
-public:
-	ThreadSignalStack() {
-		const auto page_size = static_cast<size_t>(::getpagesize());
-		const auto stack_size =
-		    (std::max<size_t>(64 * 1024, MINSIGSTKSZ) + page_size - 1) & ~(page_size - 1);
-		if (::posix_memalign(&m_memory, page_size, stack_size) != 0) {
-			return;
-		}
-
-		stack_t stack {};
-		stack.ss_sp   = m_memory;
-		stack.ss_size = stack_size;
-		if (::sigaltstack(&stack, &m_previous) != 0) {
-			std::free(m_memory);
-			m_memory = nullptr;
-		}
-	}
-
-	~ThreadSignalStack() {
-		if (m_memory != nullptr && ::sigaltstack(&m_previous, nullptr) == 0) {
-			std::free(m_memory);
-		}
-	}
-
-	[[nodiscard]] bool IsInitialized() const { return m_memory != nullptr; }
-
-	KYTY_CLASS_NO_COPY(ThreadSignalStack)
-
-private:
-	void*   m_memory = nullptr;
-	stack_t m_previous {};
-};
-
-bool InitializeThreadSignalStack() {
-	// Keep fault handling off guest stacks, which GPU tracking can make read-only.
-	thread_local ThreadSignalStack signal_stack;
-	return signal_stack.IsInitialized();
-}
-
-#endif
-
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS

 static LONG WINAPI ExceptionFilter(PEXCEPTION_POINTERS exception) noexcept {
@@ -301,7 +253,7 @@ bool InstallHandler(Handler handler) {
 #elif defined(__APPLE__)
 	struct sigaction sa {};
 	sa.sa_sigaction = SignalHandler;
-	sa.sa_flags     = SA_SIGINFO | SA_ONSTACK;
+	sa.sa_flags     = SA_SIGINFO;
 	sigemptyset(&sa.sa_mask);
 	// The guest signal-dispatch path (KernelRaiseException) interrupts threads with
 	// SIGUSR1; block it while a fault is being resolved so a stop-the-world request
@@ -322,7 +274,8 @@ bool InstallHandler(Handler handler) {
 	struct sigaction action {};
 	action.sa_sigaction = SignalHandler;
 	sigemptyset(&action.sa_mask);
-	action.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
+	// Fault resolution needs the normal thread stack.
+	action.sa_flags = SA_SIGINFO | SA_RESTART;

 	for (const int signal_number: {SIGSEGV, SIGBUS, SIGILL}) {
 		if (::sigaction(signal_number, &action, nullptr) != 0) {
@@ -39,10 +39,6 @@ using Handler = bool (*)(const ExceptionInfo&);

 bool InstallHandler(Handler handler);

-#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX
-bool InitializeThreadSignalStack();
-#endif
-
 } // namespace Common::HostException

 #endif /* KYTY_COMMON_HOST_EXCEPTION_H_ */
@@ -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 {});
@@ -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();
@@ -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_ */
@@ -30,6 +30,8 @@ struct sys_dbg_stack_info_t {
 #endif
 };

-void SysStackUsage(sys_dbg_stack_info_t& s); // NOLINT(google-runtime-references)
+void SysStackWalk(void** stack, int* depth);
+void SysStackUsage(sys_dbg_stack_info_t& s);          // NOLINT(google-runtime-references)
+void SysStackUsagePrint(sys_dbg_stack_info_t& stack); // NOLINT(google-runtime-references)

 #endif /* KYTY_COMMON_PLATFORM_SYSDBG_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);
@@ -8,6 +8,7 @@

 #include <cstdlib>
 #include <cstring>
+#include <execinfo.h>
 #include <pthread.h>
 #include <sys/param.h>
 #include <sys/types.h>
@@ -16,6 +17,56 @@
 #include <libgen.h> // POSIX basename() lives here on macOS, not in <cstring>
 #endif

+// Avoid unwinding a guest-owned stack.
+static bool OnOwnStack() {
+	const char* probe = reinterpret_cast<const char*>(&probe);
+
+	pthread_attr_t attr {};
+#if defined(__APPLE__)
+	const auto* top  = static_cast<const char*>(pthread_get_stackaddr_np(pthread_self()));
+	const auto  size = pthread_get_stacksize_np(pthread_self());
+	(void)attr;
+	return top != nullptr && size != 0 && probe < top && probe >= top - size;
+#else
+	if (pthread_getattr_np(pthread_self(), &attr) != 0) {
+		return false;
+	}
+	void*      base = nullptr;
+	size_t     size = 0;
+	const bool ok = pthread_attr_getstack(&attr, &base, &size) == 0 && base != nullptr && size != 0;
+	pthread_attr_destroy(&attr);
+	if (!ok) {
+		return false;
+	}
+	const auto* low = static_cast<const char*>(base);
+	return probe >= low && probe < low + size;
+#endif
+}
+
+void SysStackWalk(void** stack, int* depth) {
+	if (stack == nullptr || depth == nullptr || *depth <= 0) {
+		if (depth != nullptr) {
+			*depth = 0;
+		}
+		return;
+	}
+
+	if (!OnOwnStack()) {
+		*depth = 0;
+		return;
+	}
+
+	const int n = ::backtrace(stack, *depth);
+	*depth      = (n < 0 ? 0 : n);
+}
+
+void SysStackUsagePrint(sys_dbg_stack_info_t& stack) {
+	printf("stack: (0x%" PRIx64 ", %" PRIu64 ")\n", static_cast<uint64_t>(stack.commited_addr),
+	       static_cast<uint64_t>(stack.commited_size));
+	printf("code: (0x%" PRIx64 ", %" PRIu64 ")\n", static_cast<uint64_t>(stack.code_addr),
+	       static_cast<uint64_t>(stack.code_size));
+}
+
 void SysStackUsage(sys_dbg_stack_info_t& s) {
 	pid_t pid = getpid();

@@ -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);
@@ -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
@@ -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_ */
@@ -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.
@@ -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_ */
@@ -15,6 +15,18 @@
 #include "common/assert.h"
 #include "common/platform/sysDbg.h"

+void SysStackWalk(void** stack, int* depth) {
+	*depth = static_cast<int>(CaptureStackBackTrace(0, static_cast<DWORD>(*depth), stack, nullptr));
+}
+
+void SysStackUsagePrint(sys_dbg_stack_info_t& stack) {
+	printf("stack: (0x%" PRIx64 ", %" PRIu64 ") + (0x%" PRIx64 ", %" PRIu64 ") + (0x%" PRIx64
+	       ", %" PRIu64 ")\n",
+	       static_cast<uint64_t>(stack.reserved_addr), static_cast<uint64_t>(stack.reserved_size),
+	       static_cast<uint64_t>(stack.guard_addr), static_cast<uint64_t>(stack.guard_size),
+	       static_cast<uint64_t>(stack.commited_addr), static_cast<uint64_t>(stack.commited_size));
+}
+
 void SysStackUsage(sys_dbg_stack_info_t& s) {
 	MEMORY_BASIC_INFORMATION mbi {};
 	[[maybe_unused]] size_t  ss = VirtualQuery(&mbi, &mbi, sizeof(mbi));
@@ -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);

@@ -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
@@ -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;
 	}
 }

@@ -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)

@@ -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, '\\', '/');
 }
@@ -157,18 +257,86 @@ 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 std::string Utf16ToUtf8(std::u16string_view utf16) {
-	if (utf16.empty()) {
-		return {};
+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(const char16_t* utf16) {
+	std::u16string                                                    input(utf16);
 	std::wstring_convert<std::codecvt_utf8_utf16<char16_t>, char16_t> convert;
-	return convert.to_bytes(utf16.data(), utf16.data() + utf16.size());
+	return convert.to_bytes(input);
+}
+
+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
@@ -1,12 +1,15 @@
 #include "common/threads.h"

 #include "common/assert.h"
+#include "common/debug.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 +28,8 @@
 #ifdef KYTY_WIN_CS
 #include <windows.h> // IWYU pragma: keep
 // IWYU pragma: no_include <winbase.h>
-constexpr DWORD KYTY_CS_SPIN_COUNT = 4000;
+constexpr DWORD    KYTY_CS_SPIN_COUNT          = 4000;
+constexpr uint64_t KYTY_SLEEP_SPIN_LIMIT_100NS = 500; // 50 us

 #ifndef CREATE_WAITABLE_TIMER_HIGH_RESOLUTION
 #define CREATE_WAITABLE_TIMER_HIGH_RESOLUTION 0x00000002
@@ -36,6 +40,30 @@ 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 +133,14 @@ 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 +152,17 @@ 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 +198,37 @@ 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 +285,10 @@ 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 +366,7 @@ 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 +407,7 @@ void CondVar::Wait(Mutex* mutex) {
 	}
 	cpp_lock.release();
 #endif
+	UnregisterCondWaiter(m_cond_var.get());
 }

 void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
@@ -340,6 +416,7 @@ 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 +431,7 @@ 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 +446,16 @@ 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() {
@@ -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);
@@ -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);
@@ -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();
@@ -1,453 +0,0 @@
-#include "common/trophies.h"
-
-#include "common/file.h"
-
-#include <algorithm>
-#include <array>
-#include <charconv>
-#include <fmt/format.h>
-#include <nlohmann/json.hpp>
-#include <span>
-
-namespace Common::Trophies {
-namespace {
-
-using Json  = nlohmann::json;
-using Files = std::map<std::string_view, std::span<const std::byte>>;
-
-constexpr uint64_t MaxUnlocksSize = uint64_t {1} << 20u;
-constexpr uint64_t MaxPackageSize = uint64_t {1} << 30u;
-
-uint64_t ReadBigEndian(const std::byte* data, size_t size) {
-	uint64_t value = 0;
-	for (size_t i = 0; i < size; ++i) {
-		value = (value << 8u) | std::to_integer<uint8_t>(data[i]);
-	}
-	return value;
-}
-
-int ReadId(const Json& value) {
-	if (value.is_number_integer()) {
-		const auto number = value.get<int64_t>();
-		return number >= 0 && number <= INT32_MAX ? static_cast<int>(number) : -1;
-	}
-	if (!value.is_string()) {
-		return -1;
-	}
-	const auto& text        = value.get_ref<const std::string&>();
-	int         id          = -1;
-	const auto [end, error] = std::from_chars(text.data(), text.data() + text.size(), id);
-	return error == std::errc {} && end == text.data() + text.size() && id >= 0 ? id : -1;
-}
-
-std::optional<uint64_t> ReadUnsigned(const Json& value) {
-	if (value.is_number_unsigned()) {
-		return value.get<uint64_t>();
-	}
-	if (value.is_number_integer()) {
-		const auto number = value.get<int64_t>();
-		return number >= 0 ? std::optional<uint64_t>(static_cast<uint64_t>(number)) : std::nullopt;
-	}
-	if (!value.is_string()) {
-		return std::nullopt;
-	}
-	const auto& text        = value.get_ref<const std::string&>();
-	uint64_t    number      = 0;
-	const auto [end, error] = std::from_chars(text.data(), text.data() + text.size(), number);
-	return error == std::errc {} && end == text.data() + text.size()
-	           ? std::optional<uint64_t>(number)
-	           : std::nullopt;
-}
-
-Json ReadJson(const Files& files, std::string_view name) {
-	const auto entry = files.find(name);
-	if (entry == files.end()) {
-		return {};
-	}
-	return Json::parse(entry->second.begin(), entry->second.end(), nullptr, false);
-}
-
-bool ReadFiles(std::span<const std::byte> data, Files& files, uint64_t& size) {
-	if (data.size() < 0x40 || data.size() > MaxPackageSize ||
-	    ReadBigEndian(data.data(), 4) != 0xb228c60a || ReadBigEndian(data.data() + 4, 4) != 1) {
-		return false;
-	}
-	size             = ReadBigEndian(data.data() + 8, 8);
-	const auto count = ReadBigEndian(data.data() + 0x10, 4);
-	const auto toc   = ReadBigEndian(data.data() + 0x14, 4);
-	if (size < 0x40 || size > data.size() || count > 4096 || toc > size ||
-	    0x20 + count * 0x40 > size - toc) {
-		return false;
-	}
-	for (uint64_t i = 0; i < count; ++i) {
-		const auto*            entry = data.data() + toc + 0x20 + i * 0x40;
-		const auto*            name  = reinterpret_cast<const char*>(entry);
-		const std::string_view filename(name, std::find(name, name + 0x20, '\0') - name);
-		const auto             offset = ReadBigEndian(entry + 0x20, 8);
-		const auto             length = ReadBigEndian(entry + 0x28, 8);
-		if (offset > size || length > size - offset) {
-			return false;
-		}
-		files.emplace(filename, data.subspan(offset, length));
-	}
-	return true;
-}
-
-Package ParsePackage(std::span<const std::byte> data, int console_language) {
-	Files    files;
-	uint64_t size = 0;
-	if (!ReadFiles(data, files, size)) {
-		return {};
-	}
-	const auto conf = ReadJson(files, "tropconf.json");
-	if (!conf.is_object() || !conf.contains("trophies") || !conf["trophies"].is_array() ||
-	    conf["trophies"].size() > 1000 || !conf.contains("defaultLanguage") ||
-	    !conf["defaultLanguage"].is_string()) {
-		return {};
-	}
-	static constexpr std::array<std::string_view, 30> locales = {
-	    "ja-JP", "en-US", "fr-FR",   "es-ES",   "de-DE",  "it-IT", "nl-NL", "pt-PT",
-	    "ru-RU", "ko-KR", "zh-Hant", "zh-Hans", "fi-FI",  "sv-SE", "da-DK", "no-NO",
-	    "pl-PL", "pt-BR", "en-GB",   "tr-TR",   "es-419", "ar-AE", "fr-CA", "cs-CZ",
-	    "hu-HU", "el-GR", "ro-RO",   "th-TH",   "vi-VN",  "id-ID"};
-	const auto locale =
-	    locales[console_language >= 0 && console_language < locales.size() ? console_language : 1];
-	auto meta = ReadJson(files, fmt::format("tropmeta_{}.json", locale));
-	if (meta.is_null()) {
-		meta = ReadJson(files, fmt::format("tropmeta_{}.json",
-		                                   conf["defaultLanguage"].get_ref<const std::string&>()));
-	}
-	if (!meta.is_object() || !meta.contains("metadata") || !meta["metadata"].is_object()) {
-		return {};
-	}
-	const auto& texts = meta["metadata"];
-	if (!texts.contains("titleMetadata") || !texts["titleMetadata"].is_object() ||
-	    !texts["titleMetadata"].contains("name") || !texts["titleMetadata"]["name"].is_string() ||
-	    !texts.contains("trophyMetadata") || !texts["trophyMetadata"].is_array()) {
-		return {};
-	}
-	Package package;
-	package.title = texts["titleMetadata"]["name"].get<std::string>();
-	package.groups.emplace(-1, package.title);
-	uint64_t icon_bytes = 0;
-	for (const auto& definition: conf["trophies"]) {
-		if (!definition.is_object() || !definition.contains("id") ||
-		    !definition.contains("grade") || !definition["grade"].is_string()) {
-			return {};
-		}
-		Trophy trophy;
-		trophy.id              = ReadId(definition["id"]);
-		const auto grade       = definition["grade"].get<std::string>();
-		const auto grade_index = std::string_view("PGSB").find(grade);
-		if (trophy.id < 0 || grade.size() != 1 || grade_index == std::string_view::npos) {
-			return {};
-		}
-		trophy.grade = static_cast<int>(grade_index) + 1;
-		if (definition.contains("groupId")) {
-			trophy.group_id = ReadId(definition["groupId"]);
-			if (trophy.group_id < 0) {
-				return {};
-			}
-		}
-		if (definition.contains("platinumTrophyId")) {
-			trophy.platinum_id = ReadId(definition["platinumTrophyId"]);
-		}
-		trophy.hidden     = definition.contains("hidden") && definition["hidden"] == true;
-		trophy.has_reward = definition.contains("hasReward") && definition["hasReward"] == true;
-		if (definition.contains("unlockCondition")) {
-			const auto& condition = definition["unlockCondition"];
-			if (condition.is_object()) {
-				trophy.progressive =
-				    condition.contains("progressive") && condition["progressive"] == true;
-				if (condition.contains("udsStatId")) {
-					trophy.uds_stat_id = ReadUnsigned(condition["udsStatId"]);
-					if (!trophy.uds_stat_id) {
-						return {};
-					}
-				}
-				if (condition.contains("targetValue")) {
-					trophy.target = ReadUnsigned(condition["targetValue"]);
-				}
-				const auto comparison = condition.find("comparator");
-				if (comparison != condition.end()) {
-					if (*comparison == "ge")
-						trophy.comparison = Comparison::GreaterEqual;
-					else if (*comparison == "gt")
-						trophy.comparison = Comparison::Greater;
-					else if (*comparison == "le")
-						trophy.comparison = Comparison::LessEqual;
-					else if (*comparison == "lt")
-						trophy.comparison = Comparison::Less;
-				}
-			}
-		}
-		const auto icon = files.find(fmt::format("trop{:04}.png", trophy.id));
-		if (icon != files.end()) {
-			if (icon->second.size() > size - icon_bytes) {
-				return {};
-			}
-			trophy.icon_png.assign(icon->second.begin(), icon->second.end());
-			icon_bytes += icon->second.size();
-		}
-		package.groups.try_emplace(trophy.group_id);
-		if (package.groups.size() > 50 ||
-		    !package.trophies.emplace(trophy.id, std::move(trophy)).second) {
-			return {};
-		}
-	}
-	if (texts.contains("groupMetadata") && texts["groupMetadata"].is_array()) {
-		for (const auto& group: texts["groupMetadata"]) {
-			if (group.is_object() && group.contains("id") && group.contains("name") &&
-			    group["name"].is_string()) {
-				const auto found = package.groups.find(ReadId(group["id"]));
-				if (found != package.groups.end()) {
-					found->second = group["name"].get<std::string>();
-				}
-			}
-		}
-	}
-	for (const auto& text: texts["trophyMetadata"]) {
-		if (!text.is_object() || !text.contains("id")) {
-			continue;
-		}
-		const auto found = package.trophies.find(ReadId(text["id"]));
-		if (found == package.trophies.end()) {
-			continue;
-		}
-		for (const auto& [key, field]: {std::pair {"name", &found->second.name},
-		                                {"detail", &found->second.description},
-		                                {"reward", &found->second.reward}}) {
-			if (text.contains(key) && text[key].is_string()) {
-				*field = text[key].get<std::string>();
-			}
-		}
-	}
-	return package;
-}
-
-} // namespace
-
-Package LoadPackage(const std::filesystem::path& path, int console_language) {
-	File file(path, File::Mode::Read);
-	if (file.IsInvalid() || file.Size() > MaxPackageSize) {
-		return {};
-	}
-	return ParsePackage(file.ReadWholeBuffer(), console_language);
-}
-
-std::filesystem::path PackagePath(uint32_t service_label) {
-	return std::filesystem::path(PackageDirectory) / fmt::format("trophy{:02}.ucp", service_label);
-}
-
-std::filesystem::path UnlocksPath(const std::filesystem::path& root, std::string_view title_id,
-                                  int user_id, uint32_t service_label) {
-	if (title_id.empty() || !std::all_of(title_id.begin(), title_id.end(), [](char c) {
-		    return (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || (c >= '0' && c <= '9') ||
-		           c == '_' || c == '-';
-	    })) {
-		return {};
-	}
-	return root / "_Trophies" / title_id /
-	       fmt::format("trophies_{}_{}.json", user_id, service_label);
-}
-
-UnlockData LoadUnlockData(const std::filesystem::path& path) {
-	File file(path, File::Mode::Read);
-	if (file.IsInvalid() || file.Size() > MaxUnlocksSize) {
-		return {};
-	}
-	const auto bytes = file.ReadWholeBuffer();
-	const auto json  = Json::parse(bytes.begin(), bytes.end(), nullptr, false);
-	if (!json.is_object() || !json.contains("unlockedTrophies") ||
-	    !json["unlockedTrophies"].is_array()) {
-		return {};
-	}
-	UnlockData unlocks;
-	for (const auto& value: json["unlockedTrophies"]) {
-		const auto id = ReadId(value);
-		if (id >= 0) {
-			unlocks.unlocked.insert(id);
-		}
-	}
-	if (json.contains("unlockedAt") && json["unlockedAt"].is_object()) {
-		for (auto it = json["unlockedAt"].begin(); it != json["unlockedAt"].end(); ++it) {
-			const auto id = ReadId(Json(it.key()));
-			if (id >= 0 && unlocks.unlocked.contains(id)) {
-				const auto tick = ReadUnsigned(it.value());
-				if (tick && *tick >= UnixEpochTick && *tick <= 315537897599999999ULL) {
-					unlocks.timestamps.emplace(id, *tick);
-				}
-			}
-		}
-	}
-	return unlocks;
-}
-
-bool SaveUnlockData(const std::filesystem::path& path, const UnlockData& unlocks) {
-	if (path.empty() || !File::CreateDirectories(path.parent_path())) {
-		return false;
-	}
-	Json timestamps = Json::object();
-	for (const auto& [id, tick]: unlocks.timestamps) {
-		if (unlocks.unlocked.contains(id)) {
-			timestamps[std::to_string(id)] = tick;
-		}
-	}
-	const auto text =
-	    Json {{"unlockedTrophies", unlocks.unlocked}, {"unlockedAt", timestamps}}.dump();
-	auto temporary = path;
-	temporary += ".tmp";
-	File file;
-	if (text.size() > MaxUnlocksSize || !file.Create(temporary)) {
-		return false;
-	}
-	uint32_t written = 0;
-	file.Write(text.data(), static_cast<uint32_t>(text.size()), &written);
-	const bool complete = written == text.size() && file.Flush();
-	file.Close();
-	std::error_code error;
-	if (complete) {
-		std::filesystem::rename(temporary, path, error);
-		if (!error) {
-			return true;
-		}
-	}
-	std::filesystem::remove(temporary, error);
-	return false;
-}
-
-UdsRules LoadUdsRules(const std::filesystem::path& path) {
-	File file(path, File::Mode::Read);
-	if (file.IsInvalid() || file.Size() > MaxPackageSize) {
-		return {};
-	}
-	const auto bytes = file.ReadWholeBuffer();
-	Files      files;
-	uint64_t   size = 0;
-	if (!ReadFiles(bytes, files, size)) {
-		return {};
-	}
-	const auto definitions = ReadJson(files, "stats_definition.json");
-	const auto extraction  = ReadJson(files, "stats_extraction.json");
-	if (!definitions.is_object() || !definitions.contains("statDefinitionArray") ||
-	    !definitions["statDefinitionArray"].is_array() || !extraction.is_object() ||
-	    !extraction.contains("statsExtractionRuleArray") ||
-	    !extraction["statsExtractionRuleArray"].is_array()) {
-		return {};
-	}
-	// Only direct unsigned latest-value stats are evaluated here. Aggregated stats
-	// and filtered/nested extraction require UDS state and event types we do not implement.
-	std::map<uint64_t, bool> stats;
-	for (const auto& definition: definitions["statDefinitionArray"]) {
-		if (!definition.is_object() || !definition.contains("statId") ||
-		    !definition.contains("aggregation") || definition["aggregation"] != "latest" ||
-		    !definition.contains("dataType")) {
-			continue;
-		}
-		const auto  id   = ReadUnsigned(definition["statId"]);
-		const auto& type = definition["dataType"];
-		if (!id || (type != "uint64" && type != "uint32")) {
-			continue;
-		}
-		const uint64_t maximum = type == "uint32" ? UINT32_MAX : UINT64_MAX;
-		if ((definition.contains("minValue") && ReadUnsigned(definition["minValue"]) != 0) ||
-		    (definition.contains("maxValue") && ReadUnsigned(definition["maxValue"]) != maximum)) {
-			continue;
-		}
-		stats.emplace(*id, type == "uint32");
-	}
-	UdsRules events;
-	for (const auto& rule: extraction["statsExtractionRuleArray"]) {
-		if (!rule.is_object() || !rule.contains("condition") || !rule.contains("action")) {
-			continue;
-		}
-		const auto& condition = rule["condition"];
-		const auto& action    = rule["action"];
-		if (!condition.is_object() || condition.size() != 1 || !condition.contains("eventName") ||
-		    !condition["eventName"].is_string() || !action.is_object() ||
-		    !action.contains("input") || !action["input"].is_string() ||
-		    !action.contains("output") || !action["output"].is_object() ||
-		    !action["output"].contains("statId")) {
-			continue;
-		}
-		const auto  id    = ReadUnsigned(action["output"]["statId"]);
-		const auto& input = action["input"].get_ref<const std::string&>();
-		if (!id || !stats.contains(*id) || !input.starts_with("$.") || input.size() == 2 ||
-		    input.find_first_not_of(
-		        "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789_", 2) !=
-		        std::string::npos) {
-			continue;
-		}
-		events[condition["eventName"].get<std::string>()].push_back(
-		    {*id, input.substr(2), stats.at(*id)});
-	}
-	return events;
-}
-
-std::filesystem::path UdsPackagePath(uint32_t service_label) {
-	return std::filesystem::path("sce_sys/uds") / fmt::format("uds{:02}.ucp", service_label);
-}
-
-std::vector<int> FindUdsTrophies(const Package& package, std::string_view event_name,
-                                 const std::map<std::string, UdsInteger>& properties) {
-	const auto rules = package.event_rules.find(std::string(event_name));
-	if (rules == package.event_rules.end()) {
-		return {};
-	}
-	std::vector<int> ids;
-	for (const auto& [id, trophy]: package.trophies) {
-		if (trophy.grade == 1 || !trophy.uds_stat_id || !trophy.target) {
-			continue;
-		}
-		for (const auto& rule: rules->second) {
-			const auto value = properties.find(rule.input);
-			if (rule.stat_id != *trophy.uds_stat_id || value == properties.end()) {
-				continue;
-			}
-			std::optional<uint64_t> number;
-			if (rule.is_uint32) {
-				if (const auto* input = std::get_if<uint32_t>(&value->second)) number = *input;
-			} else if (const auto* input = std::get_if<uint64_t>(&value->second)) {
-				number = *input;
-			}
-			if (!number) continue;
-			bool matched = false;
-			switch (trophy.comparison) {
-				case Comparison::GreaterEqual: matched = *number >= *trophy.target; break;
-				case Comparison::Greater: matched = *number > *trophy.target; break;
-				case Comparison::LessEqual: matched = *number <= *trophy.target; break;
-				case Comparison::Less: matched = *number < *trophy.target; break;
-				case Comparison::None: break;
-			}
-			if (matched) {
-				ids.push_back(id);
-				break;
-			}
-		}
-	}
-	return ids;
-}
-
-uint32_t Progress::Percentage() const {
-	const auto points        = total_grade[2] * 6 + total_grade[3] * 2 + total_grade[4];
-	const auto earned_points = earned_grade[2] * 6 + earned_grade[3] * 2 + earned_grade[4];
-	return points == 0 ? 0 : earned_points * 100 / points;
-}
-
-Progress GetProgress(const Package& package, const UnlockData& unlocks, std::optional<int> group) {
-	Progress progress;
-	for (const auto& [id, trophy]: package.trophies) {
-		if (group && trophy.group_id != *group) {
-			continue;
-		}
-		++progress.total;
-		++progress.total_grade[trophy.grade];
-		if (unlocks.unlocked.contains(id)) {
-			++progress.earned;
-			++progress.earned_grade[trophy.grade];
-		}
-	}
-	return progress;
-}
-
-} // namespace Common::Trophies
@@ -1,87 +0,0 @@
-#ifndef KYTY_COMMON_TROPHIES_H_
-#define KYTY_COMMON_TROPHIES_H_
-
-#include <array>
-#include <cstddef>
-#include <cstdint>
-#include <filesystem>
-#include <map>
-#include <optional>
-#include <set>
-#include <string>
-#include <string_view>
-#include <variant>
-#include <vector>
-
-namespace Common::Trophies {
-
-inline constexpr char     PackageDirectory[] = "sce_sys/trophy2";
-inline constexpr uint64_t UnixEpochTick      = 62135596800000000ULL;
-
-enum class Comparison { None, GreaterEqual, Greater, LessEqual, Less };
-
-struct Trophy {
-	int                     id          = 0;
-	int                     group_id    = -1;
-	int                     platinum_id = -1;
-	int                     grade       = 0;
-	std::optional<uint64_t> target;
-	std::optional<uint64_t> uds_stat_id;
-	Comparison              comparison  = Comparison::None;
-	bool                    progressive = false;
-	std::string             name;
-	std::string             description;
-	std::string             reward;
-	std::vector<std::byte>  icon_png;
-	bool                    hidden     = false;
-	bool                    has_reward = false;
-};
-
-struct UdsRule {
-	uint64_t    stat_id;
-	std::string input;
-	bool        is_uint32;
-};
-
-using UdsRules   = std::map<std::string, std::vector<UdsRule>>;
-using UdsInteger = std::variant<int32_t, uint32_t, uint64_t>;
-
-struct Package {
-	std::string                title;
-	std::map<int, std::string> groups;
-	std::map<int, Trophy>      trophies;
-	UdsRules                   event_rules;
-};
-
-struct UnlockData {
-	std::set<int>           unlocked;
-	std::map<int, uint64_t> timestamps;
-};
-
-struct Progress {
-	std::array<uint32_t, 5> total_grade {};
-	std::array<uint32_t, 5> earned_grade {};
-	uint32_t                total  = 0;
-	uint32_t                earned = 0;
-
-	[[nodiscard]] uint32_t Percentage() const;
-};
-
-[[nodiscard]] Progress GetProgress(const Package& package, const UnlockData& unlocks,
-                                   std::optional<int> group = std::nullopt);
-
-[[nodiscard]] Package LoadPackage(const std::filesystem::path& path, int console_language);
-[[nodiscard]] std::filesystem::path PackagePath(uint32_t service_label);
-[[nodiscard]] std::filesystem::path UdsPackagePath(uint32_t service_label);
-[[nodiscard]] UdsRules              LoadUdsRules(const std::filesystem::path& path);
-[[nodiscard]] std::vector<int> FindUdsTrophies(const Package& package, std::string_view event_name,
-                                               const std::map<std::string, UdsInteger>& properties);
-[[nodiscard]] std::filesystem::path UnlocksPath(const std::filesystem::path& root,
-                                                std::string_view title_id, int user_id,
-                                                uint32_t service_label);
-[[nodiscard]] UnlockData            LoadUnlockData(const std::filesystem::path& path);
-[[nodiscard]] bool SaveUnlockData(const std::filesystem::path& path, const UnlockData& unlocks);
-
-} // namespace Common::Trophies
-
-#endif // KYTY_COMMON_TROPHIES_H_
@@ -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
@@ -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
@@ -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);
@@ -23,12 +23,6 @@ enum class ContextStateOperation : uint32_t {
 	PushClear = 3,
 };

-struct IndirectDrawRegisters {
-	uint32_t vertex_offset;
-	uint32_t instance_offset;
-	uint32_t index_offset;
-};
-
 class Pm4Execution {
 public:
 	[[nodiscard]] bool MadeProgress() const noexcept { return m_made_progress; }
@@ -38,12 +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;
 };
@@ -68,6 +61,7 @@ public:

 	void            BufferInit();
 	void            BufferFlush();
+	void            BufferFlushAndWait();
 	void            BufferWait();
 	HW::Context&    GetCtx() { return m_ctx; }
 	HW::UserConfig& GetUcfg() { return m_ucfg; }
@@ -78,18 +72,18 @@ 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);
-	void DrawIndexAuto(DrawAutoArgs args);
-	void DrawIndirect(uint32_t data_offset, IndirectDrawRegisters registers,
-	                  uint32_t draw_initiator, bool indexed);
+	void DrawIndex(uint32_t index_count, const void* index_addr, uint32_t flags, uint32_t type,
+	               uint32_t instance_count = 0, const void* object_ids = nullptr,
+	               uint32_t render_target_slice_offset = 0, int32_t vertex_offset_add = 0,
+	               uint32_t first_instance = 0);
+	void DrawIndexOffset(uint32_t index_offset, uint32_t index_count, uint32_t flags);
+	void DrawIndexAuto(uint32_t index_count, uint32_t flags,
+	                   uint32_t render_target_slice_offset = 0);
+	void DrawIndirect(uint32_t data_offset, uint32_t draw_initiator, bool indexed);
 	void DrawIndirectMulti(uint32_t data_offset, uint32_t max_count_or_count,
 	                       const volatile uint32_t* count_addr, uint32_t stride_in_bytes,
-	                       IndirectDrawRegisters registers, uint32_t draw_initiator, bool indexed);
+	                       uint32_t draw_initiator, bool indexed);
 	void WriteAtEndOfPipe32(uint32_t cache_policy, uint32_t event_write_dest,
 	                        uint32_t eop_event_type, uint32_t cache_action, uint32_t event_index,
 	                        uint32_t event_write_source, void* dst_gpu_addr, uint32_t value,
@@ -108,7 +102,7 @@ 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);

@@ -139,7 +133,7 @@ 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; }

@@ -153,10 +147,16 @@ 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();
+	void SubmitNonIndexedDraw(uint32_t vertex_count, uint32_t flags,
+	                          uint32_t render_target_slice_offset, uint32_t first_vertex,
+	                          uint32_t first_instance);
+
 	CommandScheduler&   GetScheduler() const { return m_renderer.GetCommandScheduler(); }
 	CommandBuffer&      CurrentBuffer() { return GetScheduler().Current(); }
+	void                CheckBuffer() const { GetScheduler().CheckActive(); }
+	GpuResourceManager& GetGpuResources() const { return m_renderer.GetGpuResources(); }

 	RenderContext&   m_renderer;
 	HW::Context      m_ctx;
@@ -24,6 +24,7 @@ constexpr auto MakeContextDispatchTable() {
 	g_hw_ctx_func[Pm4::DB_DEPTH_CLEAR]               = HwCtxSetDepthClear;
 	g_hw_ctx_func[Pm4::PA_SC_SCREEN_SCISSOR_TL]      = HwCtxSetScreenScissor;
 	g_hw_ctx_func[Pm4::DB_STENCIL_INFO]              = HwCtxSetStencilInfo;
+	g_hw_ctx_func[Pm4::DB_SHADING_RATE_ENCODING]     = HwCtxSetDepthShadingRateEncoding;
 	g_hw_ctx_func[Pm4::PA_SU_HARDWARE_SCREEN_OFFSET] = HwCtxSetHardwareScreenOffset;
 	g_hw_ctx_func[Pm4::PA_SC_WINDOW_OFFSET]          = HwCtxSetWindowOffset;
 	g_hw_ctx_func[Pm4::PA_SC_WINDOW_SCISSOR_TL]      = HwCtxSetWindowScissor;
@@ -139,15 +140,13 @@ constexpr auto MakeShaderDispatchTable() {
 	g_hw_sh_func[Pm4::COMPUTE_TMPRING_SIZE]            = HwShSetCsRegisters;
 	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;
-	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_HS]  = HwShSetRegisters;
-	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS] = HwShSetRegisters;
-	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS] = HwShSetRegisters;
+	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_PS]  = HwShIgnoreRegisters;
+	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS]  = HwShIgnoreRegisters;
+	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS] = HwShIgnoreRegisters;
+	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_GS] = HwShIgnoreRegisters;
+	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_HS]  = HwShIgnoreRegisters;
+	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS] = HwShIgnoreRegisters;
+	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS] = HwShIgnoreRegisters;

 	return g_hw_sh_func;
 }
@@ -45,6 +45,8 @@ uint32_t HwCtxSetStencilClear(CommandProcessor&, uint32_t, uint32_t, const uint3
 uint32_t HwCtxSetDepthClear(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetScreenScissor(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetStencilInfo(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
+uint32_t HwCtxSetDepthShadingRateEncoding(CommandProcessor&, uint32_t, uint32_t,
+                                          const uint32_t*, uint32_t);
 uint32_t HwCtxSetHardwareScreenOffset(CommandProcessor&, uint32_t, uint32_t, const uint32_t*,
                                      uint32_t);
 uint32_t HwCtxSetWindowOffset(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
@@ -98,7 +100,7 @@ uint32_t HwShSetCsUserSgpr(CommandProcessor&, uint32_t, uint32_t, const uint32_t
 uint32_t HwShIgnoreUserAccumulator(CommandProcessor&, uint32_t, uint32_t, const uint32_t*,
                                   uint32_t);
 uint32_t HwShSetCsRegisters(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
-uint32_t HwShSetRegisters(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
+uint32_t HwShIgnoreRegisters(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwUcSetPrimitiveType(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwUcSetIndexType(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwUcSetObjectId(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
@@ -136,6 +136,7 @@ static HW::ColorControl DecodeColorControl(uint32_t value) {

 static HW::DepthControl DecodeDepthControl(uint32_t value) {
 	HW::DepthControl r {};
+	EXIT_NOT_IMPLEMENTED((value & 0xc0000000u) != 0);

 	r.stencil_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCIL_ENABLE) != 0;
 	r.z_enable            = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_ENABLE) != 0;
@@ -184,6 +185,7 @@ static HW::RenderControl DecodeRenderControl(uint32_t value) {
 	    KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_STENCIL_TO_COLOR) != 0;
 	r.copy_centroid          = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_CENTROID) != 0;
 	r.copy_sample            = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_SAMPLE);
+	EXIT_NOT_IMPLEMENTED(r.copy_depth_to_color || r.copy_stencil_to_color);

 	return r;
 }
@@ -314,17 +316,8 @@ static void HwCtxSetDepthBoundsRegister(CommandProcessor& cp, uint32_t cmd_offse
 	}
 }

-static void HwCtxSetDepthMetadataRegister(CommandProcessor& cp, uint32_t cmd_offset,
-                                         uint32_t value) {
-	if (cmd_offset == Pm4::DB_RENDER_OVERRIDE) {
-		HW::DepthRenderOverride control;
-		control.force_z_valid       = (value & 0x20000000u) != 0;
-		control.force_z_dirty       = (value & 0x08000000u) != 0;
-		control.force_stencil_valid = (value & 0x40000000u) != 0;
-		control.force_stencil_dirty = (value & 0x10000000u) != 0;
-		cp.GetCtx().SetDepthRenderOverride(control);
-	}
-}
+static void HwCtxIgnoreDepthMetadataRegister([[maybe_unused]] uint32_t cmd_offset,
+                                             [[maybe_unused]] uint32_t value) {}

 static void HwCtxIgnoreFovWindow([[maybe_unused]] uint32_t value) {}

@@ -397,7 +390,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetDepthMetadataRegisters) {
 	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;

 	for (uint32_t i = 0; i < num_values; i++) {
-		HwCtxSetDepthMetadataRegister(cp, cmd_offset + i, buffer[i]);
+		HwCtxIgnoreDepthMetadataRegister(cmd_offset + i, buffer[i]);
 	}

 	return num_values;
@@ -607,6 +600,17 @@ static void HwCtxSetDepthShadingRateEncodingRegister(CommandProcessor& cp, uint3
 	ctx.SetDepthRenderTarget(target);
 }

+KYTY_HW_CTX_PARSER(HwCtxSetDepthShadingRateEncoding) {
+	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
+
+	EXIT_NOT_IMPLEMENTED(num_values != 1);
+	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_SHADING_RATE_ENCODING);
+
+	HwCtxSetDepthShadingRateEncodingRegister(cp, buffer[0]);
+
+	return num_values;
+}
+
 static HW::EqaaControl ParseEqaaControl(uint32_t value) {
 	HW::EqaaControl r;

@@ -988,7 +992,7 @@ static void HwShIgnoreComputeRegister([[maybe_unused]] uint32_t cmd_offset,
 static void HwShIgnoreShaderRegister([[maybe_unused]] uint32_t cmd_offset,
                                     [[maybe_unused]] uint32_t value) {}

-KYTY_HW_SH_PARSER(HwShSetRegisters) {
+KYTY_HW_SH_PARSER(HwShIgnoreRegisters) {
 	const auto reg_num = (cmd_id >> 16u) & 0x3fffu;
 	EXIT_NOT_IMPLEMENTED(reg_num == 0);

@@ -1330,13 +1334,26 @@ 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;
 }
@@ -1481,10 +1498,10 @@ 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;
@@ -1602,15 +1619,15 @@ KYTY_CP_OP_PARSER(CpOpDrawIndex) {
 		auto*    index_addr =
 		    reinterpret_cast<void*>(buffer[1] | (static_cast<uint64_t>(buffer[2]) << 32u));
 		uint32_t max_instance_count = buffer[3];
-		uint32_t instance_count     = buffer[6];
-		uint32_t flags              = buffer[7];
+		auto*    object_ids =
+		    reinterpret_cast<void*>(buffer[4] | (static_cast<uint64_t>(buffer[5]) << 32u));
+		uint32_t instance_count = buffer[6];
+		uint32_t flags          = buffer[7];

 		EXIT_NOT_IMPLEMENTED(instance_count > max_instance_count);
 		EXIT_NOT_IMPLEMENTED((flags & ~0xa0u) != 0);

-		cp.DrawIndex({.index_count    = index_count,
-		              .index_addr     = index_addr,
-		              .instance_count = instance_count});
+		cp.DrawIndex(index_count, index_addr, 0, 1, instance_count, object_ids);

 		return 8;
 	}
@@ -1625,7 +1642,7 @@ KYTY_CP_OP_PARSER(CpOpDrawIndex) {
 		EXIT_NOT_IMPLEMENTED(index_count > max_index_count);
 		EXIT_NOT_IMPLEMENTED((flags & ~0x20u) != 0);

-		cp.DrawIndex({.index_count = index_count, .index_addr = index_addr});
+		cp.DrawIndex(index_count, index_addr, 0, 1);

 		return 5;
 	}
@@ -1641,12 +1658,8 @@ KYTY_CP_OP_PARSER(CpOpDrawIndirect) {
 	const auto data_offset    = buffer[0];
 	const auto draw_initiator = buffer[3];
 	const bool indexed        = (cmd_id == 0xc0032500);
-	// The native indexed packet enables the high-half first-index destination in word 3.
-	const IndirectDrawRegisters registers {
-	    buffer[1] & 0xffffu, buffer[2] & 0xffffu,
-	    indexed && (buffer[2] & (1u << 28u)) != 0 ? buffer[1] >> 16u : Pm4::SH_NOP};

-	cp.DrawIndirect(data_offset, registers, draw_initiator, indexed);
+	cp.DrawIndirect(data_offset, draw_initiator, indexed);

 	return 4;
 }
@@ -1664,17 +1677,13 @@ KYTY_CP_OP_PARSER(CpOpDrawIndirectMulti) {
 	const auto stride_in_bytes = buffer[7];
 	const auto draw_initiator  = buffer[8];
 	const bool indexed         = (cmd_id == 0xc0083800);
-	// Multi-draw moves the first-index enable to word 4; the destinations stay in 2/3.
-	const IndirectDrawRegisters registers {
-	    buffer[1] & 0xffffu, buffer[2] & 0xffffu,
-	    indexed && (buffer[3] & (1u << 28u)) != 0 ? buffer[1] >> 16u : Pm4::SH_NOP};

 	if (count_indirect == 0) {
 		count_addr = nullptr;
 	}

 	cp.DrawIndirectMulti(data_offset, max_count_or_count, count_addr, stride_in_bytes,
-	                     registers, draw_initiator, indexed);
+	                     draw_initiator, indexed);

 	return 9;
 }
@@ -1692,7 +1701,7 @@ KYTY_CP_OP_PARSER(CpOpDrawIndexOffset) {
 	EXIT_NOT_IMPLEMENTED(index_count > max_index_size);
 	EXIT_NOT_IMPLEMENTED((flags & ~0x20u) != 0);

-	cp.DrawIndexOffset(index_offset, index_count);
+	cp.DrawIndexOffset(index_offset, index_count, 0);

 	return 4;
 }
@@ -1707,7 +1716,7 @@ KYTY_CP_OP_PARSER(CpOpDrawIndexAuto) {

 	EXIT_NOT_IMPLEMENTED((flags & ~0x22u) != 0);

-	cp.DrawIndexAuto({.vertex_count = index_count});
+	cp.DrawIndexAuto(index_count, 0);

 	return 2;
 }
@@ -1953,7 +1962,7 @@ 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;
 }
@@ -2011,14 +2020,6 @@ 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);
 		}

@@ -2273,6 +2274,7 @@ 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;
@@ -2280,10 +2282,13 @@ 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) {
@@ -2321,9 +2326,7 @@ 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;
 	}
@@ -3012,28 +3015,28 @@ void GraphicsInitJmpTablesCxIndirect() {
 		HwCtxIgnoreCbDccControl(value);
 	};
 	g_hw_ctx_indirect_func[Pm4::DB_COUNT_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
+		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 	};
 	for (auto cmd_offset = Pm4::DB_SRESULTS_COMPARE_STATE0;
 	     cmd_offset <= Pm4::DB_SRESULTS_COMPARE_STATE1; cmd_offset++) {
 		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-			HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
+			HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 		};
 	}
 	g_hw_ctx_indirect_func[Pm4::DB_RENDER_OVERRIDE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
+		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::DB_RENDER_OVERRIDE2] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
+		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::DB_DFSM_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
+		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::DB_RMI_L2_CACHE_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
+		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::CB_RMI_GL2_CACHE_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
+		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::TA_BC_BASE_ADDR] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 		HwCtxIgnoreBorderColorTableAddr(cmd_offset, value);
@@ -3629,39 +3632,18 @@ 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);
 	};
-	for (uint32_t offset = Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS;
-	     offset <= Pm4::SPI_SHADER_USER_DATA_ADDR_HI_GS; offset++) {
-		g_hw_sh_indirect_func[offset] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-			cp.GetShCtx().SetGsUserDataAddress(cmd_offset - Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS,
-			                                   value);
-		};
-	}
+	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+		HwShIgnoreShaderRegister(cmd_offset, value);
+	};
+	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+		HwShIgnoreShaderRegister(cmd_offset, value);
+	};
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_CHKSUM_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
@@ -3671,13 +3653,12 @@ 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;
@@ -3836,25 +3817,6 @@ 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;
@@ -3862,11 +3824,6 @@ void GraphicsInitJmpTablesUcIndirect() {
 		r.vertex_group_size    = KYTY_PM4_GET(value, GE_CNTL, VERT_GRP_SIZE);
 		cp.GetUcfg().SetGeControl(r);
 	};
-	g_hw_uc_indirect_func[Pm4::UC_PARAMETER_OVERSUBSCRIPTION] = [](KYTY_HW_UC_INDIRECT_ARGS) {
-		(void)cp;
-		(void)cmd_offset;
-		(void)value;
-	};
 	g_hw_uc_indirect_func[Pm4::GE_USER_VGPR_EN] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		HW::GeUserVgprEn r;
 		r.vgpr1 = KYTY_PM4_GET(value, GE_USER_VGPR_EN, EN_USER_VGPR1) != 0;
@@ -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,
@@ -15,62 +15,58 @@ struct FormatInfo {
 	uint32_t           render_target_bytes_per_element;
 	bool               sampled_texture;
 	bool               uint_texture;
-	bool               sint_texture = false;
 };

 constexpr FormatInfo kFormatInfo[] = {
 	{BufferFormat::k8UNorm, 1, 0, 1, true, false},
-	{BufferFormat::k8UScaled, 1, 0, 0, true, false},
-	{BufferFormat::k8SNorm, 1, 0, 1, false, false},
+	{BufferFormat::k8SNorm, 0, 0, 1, false, false},
 	{BufferFormat::k8UInt, 1, 0, 1, true, true},
 	{BufferFormat::k16UNorm, 2, 0, 2, true, false},
 	{BufferFormat::k16SNorm, 2, 0, 2, true, false},
 	{BufferFormat::k16UInt, 2, 0, 2, true, true},
-	{BufferFormat::k16SInt, 2, 0, 2, true, false, true},
+	{BufferFormat::k16SInt, 2, 0, 2, false, false},
 	{BufferFormat::k16Float, 2, 0, 2, true, false},
 	{BufferFormat::k8_8UNorm, 2, 0, 2, true, false},
-	{BufferFormat::k8_8UScaled, 2, 0, 0, true, false},
 	{BufferFormat::k8_8SNorm, 2, 0, 2, true, false},
 	{BufferFormat::k8_8UInt, 2, 0, 2, true, true},
-	{BufferFormat::k8_8SInt, 2, 0, 2, true, false, true},
+	{BufferFormat::k8_8SInt, 2, 0, 2, false, false},
 	{BufferFormat::k32UInt, 4, 0, 4, true, true},
-	{BufferFormat::k32SInt, 4, 0, 4, true, false, true},
+	{BufferFormat::k32SInt, 4, 0, 4, false, false},
 	{BufferFormat::k32Float, 4, 0, 4, true, false},
 	{BufferFormat::k16_16UNorm, 4, 0, 4, true, false},
 	{BufferFormat::k16_16SNorm, 4, 0, 4, true, false},
 	{BufferFormat::k16_16UInt, 4, 0, 4, true, true},
-	{BufferFormat::k16_16SInt, 4, 0, 4, true, false, true},
+	{BufferFormat::k16_16SInt, 4, 0, 4, false, false},
 	{BufferFormat::k16_16Float, 4, 0, 4, true, false},
 	{BufferFormat::k11_11_10UInt, 4, 0, 4, true, true},
 	{BufferFormat::k11_11_10Float, 4, 0, 4, true, false},
-	{BufferFormat::k10_11_11Float, 4, 0, 4, true, false},
 	{BufferFormat::k10_10_10_2UNorm, 4, 0, 4, true, false},
 	{BufferFormat::k10_10_10_2UInt, 4, 0, 4, true, true},
 	{BufferFormat::k8_8_8_8UNorm, 4, 0, 4, true, false},
 	{BufferFormat::k8_8_8_8SNorm, 4, 0, 4, true, false},
 	{BufferFormat::k8_8_8_8UInt, 4, 0, 4, true, true},
-	{BufferFormat::k8_8_8_8SInt, 4, 0, 4, true, false, true},
+	{BufferFormat::k8_8_8_8SInt, 4, 0, 4, false, false},
 	{BufferFormat::k32_32UInt, 8, 0, 8, true, true},
-	{BufferFormat::k32_32SInt, 8, 0, 8, true, false, true},
+	{BufferFormat::k32_32SInt, 8, 0, 8, false, false},
 	{BufferFormat::k32_32Float, 8, 0, 8, true, false},
 	{BufferFormat::k16_16_16_16UNorm, 8, 0, 8, true, false},
 	{BufferFormat::k16_16_16_16SNorm, 8, 0, 8, true, false},
 	{BufferFormat::k16_16_16_16UInt, 8, 0, 8, true, true},
-	{BufferFormat::k16_16_16_16SInt, 8, 0, 8, true, false, true},
+	{BufferFormat::k16_16_16_16SInt, 8, 0, 8, false, false},
 	{BufferFormat::k16_16_16_16Float, 8, 0, 8, true, false},
 	{BufferFormat::k32_32_32UInt, 12, 0, 12, true, true},
-	{BufferFormat::k32_32_32SInt, 12, 0, 12, true, false, true},
+	{BufferFormat::k32_32_32SInt, 12, 0, 12, false, false},
 	{BufferFormat::k32_32_32Float, 12, 0, 12, true, false},
 	{BufferFormat::k32_32_32_32UInt, 16, 0, 16, true, true},
-	{BufferFormat::k32_32_32_32SInt, 16, 0, 16, true, false, true},
+	{BufferFormat::k32_32_32_32SInt, 16, 0, 16, false, false},
 	{BufferFormat::k32_32_32_32Float, 16, 0, 16, true, false},
-	{BufferFormat::k8Srgb, 1, 0, 1, true, false},
-	{BufferFormat::k8_8Srgb, 2, 0, 2, true, false},
+	{BufferFormat::k8Srgb, 1, 0, 0, true, false},
+	{BufferFormat::k8_8Srgb, 2, 0, 0, true, false},
 	{BufferFormat::k8_8_8_8Srgb, 4, 0, 4, true, false},
 	{BufferFormat::k9_9_9_5Float, 4, 0, 0, true, false},
 	{BufferFormat::k5_6_5UNorm, 2, 0, 2, true, false},
 	{BufferFormat::k5_5_5_1UNorm, 2, 0, 2, true, false},
-	{BufferFormat::k1_5_5_5UNorm, 2, 0, 2, false, false},
+	{BufferFormat::k1_5_5_5UNorm, 0, 0, 2, false, false},
 	{BufferFormat::k4_4_4_4UNorm, 2, 0, 2, true, false},
 	{BufferFormat::kFmask8_S2_F1, 1, 0, 1, true, false},
 	{BufferFormat::kFmask8_S4_F1, 1, 0, 1, true, false},
@@ -222,24 +218,18 @@ bool IsFmaskTextureFormat(BufferFormat format) {
 	return format >= BufferFormat::kFmask8_S2_F1 && format <= BufferFormat::kFmask64_S16_F8;
 }

-TextureNumericClass SampledTextureNumericClass(BufferFormat format) {
+bool IsSampledTextureFormat(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
-	if (info == nullptr || !info->sampled_texture) {
-		return TextureNumericClass::Unsupported;
-	}
-	if (info->sint_texture) {
-		return TextureNumericClass::Sint;
-	}
-	return info->uint_texture ? TextureNumericClass::Uint : TextureNumericClass::Float;
+	return info != nullptr && info->sampled_texture;
+}
+
+bool IsUintTextureFormat(BufferFormat format) {
+	const auto* info = FindFormatInfo(format);
+	return info != nullptr && info->uint_texture;
 }

 BufferFormat RemapTextureFormat(BufferFormat format) {
-	switch (format) {
-		case BufferFormat::k8UScaled: return BufferFormat::k8UNorm;
-		case BufferFormat::k8_8UScaled: return BufferFormat::k8_8UNorm;
-		case BufferFormat::k11_11_10UInt: return BufferFormat::k32UInt;
-		default: return format;
-	}
+	return format == BufferFormat::k11_11_10UInt ? BufferFormat::k32UInt : format;
 }

 } // namespace Libs::Graphics::Prospero
@@ -37,14 +37,13 @@ struct RenderTargetFormatEncoding {
 	}
 };

-enum class TextureNumericClass { Unsupported, Float, Uint, Sint };
-
 RenderTargetFormatEncoding ResolveRenderTargetFormat(ChannelLayout layout, ChannelType type);
 uint32_t                   NumBytesPerElement(BufferFormat format);
 uint32_t                   BlockCompressedBytesPerBlock(BufferFormat format);
 uint32_t                   RenderTargetBytesPerElement(BufferFormat format);
 bool                       IsFmaskTextureFormat(BufferFormat format);
-TextureNumericClass        SampledTextureNumericClass(BufferFormat format);
+bool                       IsSampledTextureFormat(BufferFormat format);
+bool                       IsUintTextureFormat(BufferFormat format);
 BufferFormat               RemapTextureFormat(BufferFormat format);

 } // namespace Libs::Graphics::Prospero
@@ -13,6 +13,7 @@
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/sync.h"
+#include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window.h"
 #include "graphics/shader/shader.h"
@@ -35,34 +36,40 @@ 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;

-struct DrawIndirectArgs {
-	uint32_t vertex_count_per_instance;
-	uint32_t instance_count;
-	uint32_t start_vertex_location;
-	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;
+	}

-struct DrawIndexedIndirectArgs {
-	uint32_t index_count_per_instance;
-	uint32_t instance_count;
-	uint32_t start_index_location;
-	uint32_t base_vertex_location;
-	uint32_t start_instance_location;
+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 +146,20 @@ 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 +172,22 @@ 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 +215,6 @@ 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 +254,10 @@ void CommandProcessor::BufferFlush() {
 	GetScheduler().Flush();
 }

+void CommandProcessor::BufferFlushAndWait() {
+	GetScheduler().FlushAndWait();
+}
+
 void CommandProcessor::BufferWait() {
 	BufferInit();
 	GetScheduler().Finish();
@@ -359,12 +372,8 @@ 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 +413,7 @@ 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 +441,6 @@ 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 +546,15 @@ void GuestGpu::ThreadRun(void* data) {

 bool GuestGpu::Process(Submission& submission) {
 	const bool first_slice = !submission.started;
-	auto&      cp          = GetProcessor(submission.queue_id);
+	if (first_slice && RenderDocCaptureRequested()) {
+		Common::LockGuard render_lock(m_renderer.GetMutex());
+		RenderDocStartCapture();
+	}
+	auto& cp = GetProcessor(submission.queue_id);
+
+	if (first_slice && submission.reset_processor) {
+		cp.Reset();
+	}

 	if (first_slice) {
 		submission.started = true;
@@ -577,11 +593,11 @@ 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 +619,18 @@ 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 +662,20 @@ 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 +683,21 @@ 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;
@@ -689,7 +710,7 @@ void CommandProcessor::ProcessPm4(Pm4Execution& execution) {
 		const auto        opcode        = (packet_header >> 8u) & 0xffu;
 		EXIT_NOT_IMPLEMENTED(remaining_dw > total_dw);

-		if (packet_header == 0x80000000u || Pm4::PacketSizeDw(packet_header) == 1u) {
+		if (packet_header == 0x80000000u) {
 			cursor.offset_dw++;
 			execution.m_made_progress = true;
 			continue;
@@ -752,19 +773,14 @@ 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 = {};
-		}
 	}
 }

@@ -799,68 +815,68 @@ 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) {
-	args.index_type_and_size = m_index_type_and_size;
-	if (args.instance_count == 0) {
-		args.instance_count = m_num_instances;
+void CommandProcessor::DrawIndex(uint32_t index_count, const void* index_addr, uint32_t flags,
+                                 uint32_t type, uint32_t instance_count, const void* object_ids,
+                                 uint32_t render_target_slice_offset, int32_t vertex_offset_add,
+                                 uint32_t first_instance) {
+	CheckBuffer();
+
+	if (instance_count == 0) {
+		instance_count = m_num_instances;
 	}
-	if (GraphicsRunDebugDumpEnabled() && (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);
+	if (object_ids != nullptr) {
+		LOGF("\t draw indexed multi-instanced objectIds = 0x%016" PRIx64 "\n",
+		     reinterpret_cast<uint64_t>(object_ids));
 	}
-	m_renderer.GetRenderExecutor().DrawIndex(m_submit_id, CurrentBuffer(), args);
+	if (render_target_slice_offset != 0) {
+		LOGF("\t draw render target slice offset = %" PRIu32 "\n", render_target_slice_offset);
+	}
+	if (vertex_offset_add != 0 || first_instance != 0) {
+		LOGF("\t draw indexed offsets: vertex_offset_add = %" PRId32 ", first_instance = %" PRIu32
+		     "\n",
+		     vertex_offset_add, first_instance);
+	}
+	m_renderer.GetRenderExecutor().DrawIndex(
+	    m_submit_id, CurrentBuffer(), m_index_type_and_size, index_count, index_addr, flags, type,
+	    instance_count, render_target_slice_offset, vertex_offset_add, first_instance);
 }

-void CommandProcessor::DrawIndexOffset(uint32_t index_offset, uint32_t index_count) {
+void CommandProcessor::DrawIndexOffset(uint32_t index_offset, uint32_t index_count,
+                                       uint32_t flags) {
+	CheckBuffer();
+
 	uint64_t index_size = 0;
 	switch (m_index_type_and_size) {
 		case 0: index_size = 2; break;
@@ -872,36 +888,111 @@ void CommandProcessor::DrawIndexOffset(uint32_t index_offset, uint32_t index_cou
 	auto* index_addr = reinterpret_cast<const void*>(
 	    m_index_base_addr + static_cast<uint64_t>(index_offset) * index_size);

-	DrawIndex({.index_count = index_count, .index_addr = index_addr});
+	m_renderer.GetRenderExecutor().DrawIndex(m_submit_id, CurrentBuffer(), m_index_type_and_size,
+	                                         index_count, index_addr, flags, 1, m_num_instances);
 }

-static void PatchIndirectDrawOffsets(CommandProcessor& cp, IndirectDrawRegisters registers,
-                                     uint32_t vertex_offset, uint32_t instance_offset,
-                                     uint32_t first_index) {
-	const auto write = [&](uint32_t location, uint32_t value) {
-		if (location == Pm4::SH_NOP) {
-			return;
-		}
-		EXIT_NOT_IMPLEMENTED(location >= Pm4::SH_NUM ||
-		                     g_hw_sh_indirect_func[location] == nullptr);
-		g_hw_sh_indirect_func[location](cp, location, value);
+void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiator, bool indexed) {
+	struct DrawIndirectArgs {
+		uint32_t vertex_count_per_instance;
+		uint32_t instance_count;
+		uint32_t start_vertex_location;
+		uint32_t start_instance_location;
+	};
+	struct DrawIndexedIndirectArgs {
+		uint32_t index_count_per_instance;
+		uint32_t instance_count;
+		uint32_t start_index_location;
+		uint32_t base_vertex_location;
+		uint32_t start_instance_location;
 	};
-	write(registers.vertex_offset, vertex_offset);
-	write(registers.instance_offset, instance_offset);
-	write(registers.index_offset, first_index);
-}

-void CommandProcessor::DrawIndirect(uint32_t data_offset, IndirectDrawRegisters registers,
-                                    uint32_t draw_initiator, bool indexed) {
-	const auto args_size = indexed ? sizeof(DrawIndexedIndirectArgs) : sizeof(DrawIndirectArgs);
-	DrawIndirectMulti(data_offset, 1, nullptr, args_size, registers, draw_initiator, indexed);
+	EXIT_NOT_IMPLEMENTED((draw_initiator & ~0x20u) != 2u);
+	EXIT_NOT_IMPLEMENTED(m_draw_indirect_args_base_addr == 0);
+
+	const auto* args_addr =
+	    reinterpret_cast<const void*>(m_draw_indirect_args_base_addr + data_offset);
+
+	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;
+		SubmitNonIndexedDraw(args.vertex_count_per_instance, 0, 0, args.start_vertex_location,
+		                     args.start_instance_location);
+		return;
+	}
+
+	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) {
+		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);
+
+	const uint32_t index_count =
+	    (m_index_buffer_size != 0 ? std::min(args.index_count_per_instance, m_index_buffer_size)
+	                              : args.index_count_per_instance);
+	if (GraphicsRunDebugDumpEnabled() && index_count != args.index_count_per_instance) {
+		static std::atomic<uint32_t> log_count {0};
+		if (log_count.fetch_add(1, std::memory_order_relaxed) < 64) {
+			LOGF("\t DrawIndexIndirect: clamped index_count from %" PRIu32 " to %" PRIu32
+			     " using INDEX_BUFFER_SIZE\n",
+			     args.index_count_per_instance, index_count);
+		}
+	}
+
+	m_num_instances = args.instance_count;
+	DrawIndex(index_count, index_addr, 0, 1, args.instance_count, nullptr, 0,
+	          static_cast<int32_t>(args.base_vertex_location), args.start_instance_location);
 }

 void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_count_or_count,
                                         const volatile uint32_t* count_addr,
-                                         uint32_t stride_in_bytes, IndirectDrawRegisters registers,
-                                         uint32_t draw_initiator, bool indexed) {
-	EXIT_NOT_IMPLEMENTED((draw_initiator & ~0x20u) != (indexed ? 0u : 2u));
+                                         uint32_t stride_in_bytes, uint32_t draw_initiator,
+                                         bool indexed) {
+	struct DrawIndirectArgs {
+		uint32_t vertex_count_per_instance;
+		uint32_t instance_count;
+		uint32_t start_vertex_location;
+		uint32_t start_instance_location;
+	};
+	struct DrawIndexedIndirectArgs {
+		uint32_t index_count_per_instance;
+		uint32_t instance_count;
+		uint32_t start_index_location;
+		uint32_t base_vertex_location;
+		uint32_t start_instance_location;
+	};
+
+	EXIT_NOT_IMPLEMENTED((draw_initiator & ~0x20u) != 2u);
 	EXIT_NOT_IMPLEMENTED(m_draw_indirect_args_base_addr == 0);

 	uint32_t draw_count = max_count_or_count;
@@ -919,36 +1010,49 @@ 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);
-			PatchIndirectDrawOffsets(*this, registers, args->start_vertex_location,
-			                         args->start_instance_location, 0);
+			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,
-			               .first_vertex   = args->start_vertex_location,
-			               .first_instance = args->start_instance_location,
-			               .offset_source  = DrawOffsetSource::IndirectArgs});
+			SubmitNonIndexedDraw(args->vertex_count_per_instance, 0, 0, args->start_vertex_location,
+			                     args->start_instance_location);
 			continue;
 		}

 		auto* args = reinterpret_cast<const DrawIndexedIndirectArgs*>(args_addr);
-		PatchIndirectDrawOffsets(*this, registers, args->base_vertex_location,
-		                         args->start_instance_location, args->start_index_location);
+		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);
@@ -967,12 +1071,8 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 		}

 		m_num_instances = args->instance_count;
-		DrawIndex({.index_count    = index_count,
-		           .index_addr     = index_addr,
-		           .instance_count = args->instance_count,
-		           .base_vertex    = static_cast<int32_t>(args->base_vertex_location),
-		           .first_instance = args->start_instance_location,
-		           .offset_source  = DrawOffsetSource::IndirectArgs});
+		DrawIndex(index_count, index_addr, 0, 1, args->instance_count, nullptr, 0,
+		          static_cast<int32_t>(args->base_vertex_location), args->start_instance_location);
 	}
 }

@@ -981,8 +1081,12 @@ 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};
@@ -1002,28 +1106,76 @@ 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, IsAsyncComputeQueue());
+		                                              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,
-	                                                IsAsyncComputeQueue());
+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) {
-	if (args.instance_count == 0) {
-		args.instance_count = m_num_instances;
-	}
-	m_renderer.GetRenderExecutor().DrawAuto(m_submit_id, CurrentBuffer(), args);
+void CommandProcessor::DrawIndexAuto(uint32_t index_count, uint32_t flags,
+                                     uint32_t render_target_slice_offset) {
+	SubmitNonIndexedDraw(index_count, flags, render_target_slice_offset, 0, 0);
+}
+
+void CommandProcessor::SubmitNonIndexedDraw(uint32_t vertex_count, uint32_t flags,
+                                            uint32_t render_target_slice_offset,
+                                            uint32_t first_vertex, uint32_t first_instance) {
+	CheckBuffer();
+
+	m_renderer.GetRenderExecutor().DrawAuto(m_submit_id, CurrentBuffer(), vertex_count, flags,
+	                                        render_target_slice_offset, m_num_instances,
+	                                        first_vertex, first_instance);
 }

 void CommandProcessor::WaitFlipDone(uint32_t video_out_handle, uint32_t display_buffer_index) {
@@ -1041,7 +1193,7 @@ 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);
@@ -1072,7 +1224,7 @@ 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;
 			}
@@ -1089,17 +1241,17 @@ 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);
 		}
 	};

@@ -1111,7 +1263,7 @@ 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);
@@ -1124,9 +1276,8 @@ 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;
@@ -1134,9 +1285,6 @@ 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));
@@ -1144,18 +1292,18 @@ 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);
 					}
 				};

@@ -1170,7 +1318,6 @@ 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;
@@ -1178,6 +1325,7 @@ 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);
@@ -1193,7 +1341,7 @@ 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;
@@ -1225,6 +1373,45 @@ 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;
 	}

@@ -1254,6 +1441,8 @@ void CommandProcessor::WriteAtEndOfPipe64(uint32_t cache_policy, uint32_t event_
 }

 void CommandProcessor::EmitGlobalBarrier() {
+	CheckBuffer();
+
 	Common::LockGuard lock(m_renderer.GetMutex());

 	vk::MemoryBarrier2 barrier {};
@@ -1270,6 +1459,8 @@ void CommandProcessor::EmitGlobalBarrier() {
 }

 void CommandProcessor::TriggerEopEventAtEndOfPipe(uint32_t interrupt_context_id) {
+	CheckBuffer();
+
 	Sync::TriggerEopEventAtEndOfPipe(CurrentBuffer(), m_interrupt_event_id, interrupt_context_id);
 }

@@ -1353,6 +1544,8 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index,
 }

 void CommandProcessor::Flip() {
+	CheckBuffer();
+
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::Flip()\n");
 	}
@@ -1366,7 +1559,7 @@ void CommandProcessor::Flip() {
 }

 void CommandProcessor::Flip(void* dst_gpu_addr, uint32_t value) {
-	auto& command = CurrentBuffer();
+	CheckBuffer();

 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::Flip()\n"
@@ -1376,6 +1569,7 @@ 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),
@@ -1386,7 +1580,7 @@ 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"
@@ -1401,6 +1595,7 @@ 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(
@@ -1410,7 +1605,7 @@ 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");
 	}
@@ -1420,10 +1615,9 @@ 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() {
@@ -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;
@@ -202,13 +202,6 @@ struct RenderControl {
 	uint8_t copy_sample              = 0;
 };

-struct DepthRenderOverride {
-	bool force_z_valid       = false;
-	bool force_z_dirty       = false;
-	bool force_stencil_valid = false;
-	bool force_stencil_dirty = false;
-};
-
 struct GdsOaCounter {
 	uint32_t counter = 0;
 	uint32_t address = 0;
@@ -369,16 +362,16 @@ struct AaConfig {
 struct Viewport {
 	float zmin                                  = 0.0f;
 	float zmax                                  = 0.0f;
-	float xscale                                = 1.0f;
+	float xscale                                = 0.0f;
 	float xoffset                               = 0.0f;
-	float yscale                                = 1.0f;
+	float yscale                                = 0.0f;
 	float yoffset                               = 0.0f;
-	float zscale                                = 1.0f;
+	float zscale                                = 0.0f;
 	float zoffset                               = 0.0f;
 	int   viewport_scissor_left                 = 0;
 	int   viewport_scissor_top                  = 0;
-	int   viewport_scissor_right                = 16384;
-	int   viewport_scissor_bottom               = 16384;
+	int   viewport_scissor_right                = 0;
+	int   viewport_scissor_bottom               = 0;
 	bool  viewport_scissor_window_offset_enable = false;
 };

@@ -387,26 +380,26 @@ struct ScreenViewport {
 	uint32_t transform_control                    = 1087;
 	int      screen_scissor_left                  = 0;
 	int      screen_scissor_top                   = 0;
-	int      screen_scissor_right                 = 16384;
-	int      screen_scissor_bottom                = 16384;
+	int      screen_scissor_right                 = 0;
+	int      screen_scissor_bottom                = 0;
 	int      window_scissor_left                  = 0;
 	int      window_scissor_top                   = 0;
-	int      window_scissor_right                 = 16384;
-	int      window_scissor_bottom                = 16384;
+	int      window_scissor_right                 = 0;
+	int      window_scissor_bottom                = 0;
 	bool     window_scissor_window_offset_enable  = false;
 	int      generic_scissor_left                 = 0;
 	int      generic_scissor_top                  = 0;
-	int      generic_scissor_right                = 16384;
-	int      generic_scissor_bottom               = 16384;
+	int      generic_scissor_right                = 0;
+	int      generic_scissor_bottom               = 0;
 	bool     generic_scissor_window_offset_enable = false;
 	int      window_offset_x                      = 0;
 	int      window_offset_y                      = 0;
 	uint32_t hw_offset_x                          = 0;
 	uint32_t hw_offset_y                          = 0;
-	float    guard_band_horz_clip                 = 1.0f;
-	float    guard_band_vert_clip                 = 1.0f;
-	float    guard_band_horz_discard              = 1.0f;
-	float    guard_band_vert_discard              = 1.0f;
+	float    guard_band_horz_clip                 = 0.0f;
+	float    guard_band_vert_clip                 = 0.0f;
+	float    guard_band_horz_discard              = 0.0f;
+	float    guard_band_vert_discard              = 0.0f;
 	uint16_t clip_rect_rule                       = 0xffffu;
 	int      clip_rect_left[4]                    = {};
 	int      clip_rect_top[4]                     = {};
@@ -458,9 +451,6 @@ 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 +491,7 @@ struct LsStageRegisters {
 };

 struct HsStageRegisters {
-	uint64_t          data_addr      = 0;
-	uint64_t          user_data_addr = 0;
+	uint64_t          data_addr = 0;
 	HsShaderResource1 rsrc1;
 	HsShaderResource2 rsrc2;
 };
@@ -531,8 +520,7 @@ struct GsShaderResource2 {
 };

 struct GsStageRegisters {
-	uint64_t          data_addr      = 0;
-	uint64_t          user_data_addr = 0;
+	uint64_t          data_addr = 0;
 	GsShaderResource1 rsrc1;
 	GsShaderResource2 rsrc2;
 };
@@ -820,12 +808,6 @@ public:
 	void SetClipControl(const ClipControl& control) { m_clip_control = control; }
 	[[nodiscard]] const RenderControl& GetRenderControl() const { return m_render_control; }
 	void SetRenderControl(const RenderControl& control) { m_render_control = control; }
-	[[nodiscard]] const DepthRenderOverride& GetDepthRenderOverride() const {
-		return m_depth_render_override;
-	}
-	void SetDepthRenderOverride(const DepthRenderOverride& control) {
-		m_depth_render_override = control;
-	}
 	[[nodiscard]] const DepthControl& GetDepthControl() const { return m_depth_control; }
 	void SetDepthControl(const DepthControl& control) { m_depth_control = control; }
 	[[nodiscard]] const ModeControl& GetModeControl() const { return m_mode_control; }
@@ -921,16 +903,15 @@ private:

 	uint32_t m_shader_stages = 0;

-	DepthRenderTarget   m_depth_render_target;
-	RenderControl       m_render_control;
-	DepthRenderOverride m_depth_render_override;
-	DepthControl        m_depth_control;
-	StencilControl      m_stencil_control;
-	StencilMask         m_stencil_mask;
-	float               m_depth_clear_value   = 0.0f;
-	float               m_depth_bounds_min    = 0.0f;
-	float               m_depth_bounds_max    = 1.0f;
-	uint8_t             m_stencil_clear_value = 0;
+	DepthRenderTarget m_depth_render_target;
+	RenderControl     m_render_control;
+	DepthControl      m_depth_control;
+	StencilControl    m_stencil_control;
+	StencilMask       m_stencil_mask;
+	float             m_depth_clear_value   = 0.0f;
+	float             m_depth_bounds_min    = 0.0f;
+	float             m_depth_bounds_max    = 1.0f;
+	uint8_t           m_stencil_clear_value = 0;

 	ModeControl m_mode_control;
 	PolyOffset  m_poly_offset;
@@ -939,12 +920,6 @@ 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 +949,8 @@ 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 +973,13 @@ 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);
-	}
 	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 +1016,6 @@ 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;
@@ -120,15 +120,15 @@ void DumpPm4PacketStream(Common::File* file, const uint32_t* cmd_buffer, uint32_
 		uint32_t len = 0;

 		const auto packet_type = static_cast<PacketType>(cmd_id >> 30u);
-		EXIT_NOT_IMPLEMENTED(dw < 2 && packet_type != PacketType::Type2 &&
-		                     PacketSizeDw(cmd_id) != 1u);
+		// Type-2 packets are header-only padding; every other packet type requires a body.
+		EXIT_NOT_IMPLEMENTED(dw < 2 && packet_type != PacketType::Type2);

 		switch (packet_type) {
 			case PacketType::Type3: {
 				const bool    sh_gx = (cmd_id & 0x2u) == 0;
 				const uint8_t op    = ((cmd_id >> 8u) & 0xffu);
 				const auto    r     = KYTY_PM4_R(cmd_id);
-				len                 = PacketSizeDw(cmd_id) - 1u;
+				len                 = ((cmd_id >> 16u) & 0x3fffu) + 1;

 				EXIT_NOT_IMPLEMENTED(len >= dw);

@@ -19,10 +19,6 @@ class File;

 namespace Libs::Graphics::Pm4 {

-constexpr uint32_t PacketSizeDw(uint32_t header) {
-	return (header & 0x3fffff00u) == 0x3fff1000u ? 1u : KYTY_PM4_LEN(header);
-}
-
 constexpr uint32_t IT_NOP                       = 0x10;
 constexpr uint32_t IT_SET_BASE                  = 0x11;
 constexpr uint32_t IT_CLEAR_STATE               = 0x12;
@@ -701,8 +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;
@@ -840,7 +834,8 @@ constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_0                   = 0xB2;
 constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_3                   = 0xB5;
 constexpr uint32_t SPI_SHADER_PGM_LO_ES                           = 0xC8;
 constexpr uint32_t SPI_SHADER_PGM_HI_ES                           = 0xC9;
-// Private AGC registers used by shader binaries submitted through SET_SH_REG_INDIRECT.
+// Private AGC registers omitted from the public ShRegOffset enum. Shader binaries can still
+// include them in the ShRegister arrays submitted through SET_SH_REG_INDIRECT.
 constexpr uint32_t SPI_SHADER_PGM_CHKSUM_HS                       = 0x100;
 constexpr uint32_t SPI_SHADER_PGM_RSRC4_HS                        = 0x101;
 constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_LO_HS                = 0x102;
@@ -916,7 +911,6 @@ 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;

@@ -984,7 +978,7 @@ constexpr uint32_t GE_USER_VGPR1  = 0x25C;
 constexpr uint32_t GE_USER_VGPR2  = 0x25D;
 constexpr uint32_t GE_USER_VGPR3  = 0x25E;
 constexpr uint32_t GE_STEREO_CNTL = 0x25F;
-constexpr uint32_t UC_PARAMETER_OVERSUBSCRIPTION = 0x260;
+constexpr uint32_t GE_PC_ALLOC    = 0x260;

 constexpr uint32_t GE_USER_VGPR_EN                     = 0x262;
 constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR1_SHIFT = 0;
@@ -994,12 +988,6 @@ 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;
@@ -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) {
@@ -1213,44 +1221,36 @@ 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) {
-		return false;
-	}
-	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;
-	if (blocks > UINT32_MAX / 4096u / slices) {
-		return false;
-	}
-	total_size.size  = static_cast<uint32_t>(blocks * slices * 4096u);
-	total_size.align = 4096;
-	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 +1302,13 @@ 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 +1347,7 @@ 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 +1358,7 @@ 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;
 		}
@@ -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 {
@@ -123,11 +122,10 @@ uint32_t TileGetDepthPitch(uint32_t width, uint32_t bytes_per_element,
 bool     TileGetRenderTargetSize(uint32_t width, uint32_t height, uint32_t pitch,
                                 uint32_t bytes_per_element, TileSizeAlign& total_size,
                                 uint32_t num_fragments_log2 = 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 = 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);
@@ -14,7 +14,9 @@

 namespace Libs::Graphics {

+struct VulkanBuffer;
 struct VulkanImage;
+struct VulkanMemory;

 inline constexpr uint32_t VULKAN_TARGET_API_VERSION = VK_API_VERSION_1_3;

@@ -27,15 +29,10 @@ 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                               compute_wave64_supported              = false;
 	bool                               sample_rate_shading_enabled           = false;
-	bool                               shader_image_int64_atomics_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;
-	bool                                      mesh_shader_enabled                   = false;
-	vk::PhysicalDeviceMeshShaderPropertiesEXT mesh_shader_properties                = {};
 	uint32_t                           subgroup_size                         = 0;
 	uint32_t                           min_subgroup_size                     = 0;
 	uint32_t                           max_subgroup_size                     = 0;
@@ -83,10 +80,6 @@ struct GraphicContext {
 		return it->second.first;
 	}

-	[[nodiscard]] bool SupportsComputeWave64() const noexcept {
-		return subgroup_size == 64u || compute_subgroup_size_control_enabled;
-	}
-
 	[[nodiscard]] vk::DeviceSize StorageMinAlignment() const {
 		const auto alignment = physical_device_properties.limits.minStorageBufferOffsetAlignment;
 		return alignment != 0 ? alignment : 1;
@@ -98,8 +91,15 @@ struct GraphicContext {
 	[[nodiscard]] bool CanReportMemoryUsage() const noexcept { return memory_budget_ext_enabled; }
 	[[nodiscard]] uint64_t GetDeviceMemoryUsage() const;
 	[[nodiscard]] uint64_t GetTotalMemoryBudget() const;
+	void                   CreateBuffer(uint64_t size, VulkanBuffer& buffer);
 	[[nodiscard]] bool     CreateImage(const vk::ImageCreateInfo& info, VulkanImage& image);
 	void                   DeleteImage(VulkanImage& image);
+	void                   MapMemory(VulkanMemory& memory, void*& data);
+	void                   UnmapMemory(VulkanMemory& memory);
+	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;
@@ -114,6 +114,18 @@ private:
 	    m_image_format_properties;
 };

+struct VulkanMemory {
+	vk::MemoryRequirements  requirements       = {};
+	vk::MemoryPropertyFlags property           = {};
+	vk::MemoryPropertyFlags preferred_property = {};
+	vk::DeviceMemory        memory             = nullptr;
+	VmaAllocation           allocation         = nullptr;
+	VmaAllocationInfo       allocation_info    = {};
+	vk::DeviceSize          offset             = 0;
+	uint32_t                type               = 0;
+	uint64_t                unique_id          = 0;
+};
+
 struct VulkanImageState {
 	vk::PipelineStageFlags2 pl_stage    = vk::PipelineStageFlagBits2::eAllCommands;
 	vk::AccessFlags2        access_mask = vk::AccessFlagBits2::eNone;
@@ -127,6 +139,7 @@ struct VulkanImage {
 	vk::Format                    format      = vk::Format::eUndefined;
 	vk::ImageType                 image_type  = vk::ImageType::e2D;
 	vk::Extent3D                  extent      = {1, 1, 1};
+	uint32_t                      guest_pitch = 0;
 	uint32_t                      layers      = 1;
 	uint32_t                      mip_levels  = 1;
 	uint32_t                      samples     = 1;
@@ -135,10 +148,15 @@ struct VulkanImage {
 	vk::Image                     image       = nullptr;
 	VulkanImageState              state;
 	std::vector<VulkanImageState> subresource_states;
-	VmaAllocation                allocation = nullptr;
+	Graphics::VulkanMemory        memory;
 };

-
+struct VulkanBuffer {
+	vk::Buffer           buffer = nullptr;
+	VulkanMemory         memory;
+	vk::BufferUsageFlags usage       = {};
+	uint64_t             buffer_size = 0;
+};

 } // namespace Libs::Graphics

@@ -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;
@@ -21,7 +23,10 @@ void MemoryTracker::ValidateGpuDirtyPages(const RangeSet& dirty, uint64_t vaddr,
 		EXIT("MemoryTracker: invalid dirty-page validation range\n");
 	}
 	for (auto page = vaddr; page < vaddr + size; page += TRACKER_PAGE_SIZE) {
-		if (!dirty.Intersects(page, TRACKER_PAGE_SIZE)) {
+		bool found = false;
+		dirty.ForEachIntersection(page, TRACKER_PAGE_SIZE,
+		                          [&found](RangeSet::Range) { found = true; });
+		if (!found) {
 			EXIT("MemoryTracker: GPU-dirty tracker page has no dirty bytes, operation=%s "
 			     "addr=0x%016" PRIx64 "\n",
 			     operation, page);
@@ -32,10 +37,12 @@ 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);
+		bool has_dirty_bytes = false;
+		dirty.ForEachIntersection(page, TRACKER_PAGE_SIZE,
+		                          [&has_dirty_bytes](RangeSet::Range) { has_dirty_bytes = true; });
 		if (IsRegionGpuModified(page, TRACKER_PAGE_SIZE) != has_dirty_bytes) {
 			EXIT("MemoryTracker: tracker and byte ownership disagree, operation=%s "
 			     "addr=0x%016" PRIx64 "\n",
@@ -106,8 +113,7 @@ void MemoryTracker::UnmarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	});
 }

-void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
-	CheckNotInUploadCallback();
+void MemoryTracker::UntrackMemoryImpl(uint64_t vaddr, uint64_t size) {
 	std::vector<RegionManager*> managers;
 	managers.reserve((vaddr % TRACKER_REGION_SIZE + size + TRACKER_REGION_SIZE - 1) /
 	                 TRACKER_REGION_SIZE);
@@ -128,6 +134,12 @@ void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
 	Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
 	});
+	locks.clear();
+}
+
+void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
+	CheckNotInUploadCallback();
+	UntrackMemoryImpl(vaddr, size);
 }

 } // namespace Libs::Graphics
@@ -34,6 +34,7 @@ public:
 	void InvalidateRegion(uint64_t vaddr, uint64_t size, Flush&& on_flush) noexcept {
 		static_assert(std::is_invocable_v<Flush&>);
 		CheckNotInUploadCallback();
+		ValidateRange(vaddr, size);

 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 			const bool should_flush = [&] {
@@ -62,18 +63,43 @@ 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>
@@ -82,8 +108,9 @@ public:
 		static_assert(std::is_nothrow_invocable_v<RangeFunc&, uint64_t, uint64_t>);
 		static_assert(std::is_nothrow_invocable_v<UploadFunc&>);
 		CheckNotInUploadCallback();
+		Iterate<true>(vaddr, size, [](RegionManager*, uint64_t, uint64_t) {});
 		const auto* previous_upload_owner = std::exchange(s_upload_owner, this);
-		Iterate<true>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
+		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 			manager->lock.lock();
 			manager->ForEachModifiedRange<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset,
 			                                                      bytes, range_func);
@@ -144,6 +171,7 @@ private:
 	}

 	static void    ValidateRange(uint64_t vaddr, uint64_t size);
+	void           UntrackMemoryImpl(uint64_t vaddr, uint64_t size);
 	RegionManager* GetOrCreateRegion(uint64_t index);

 	std::unique_ptr<std::atomic<RegionManager*>[]> m_regions;
@@ -1,7 +1,6 @@
 #include "graphics/host_gpu/pageManager.h"

-#include "common/alignment.h"
-#include "common/virtualMemory.h"
+#include "common/platform/sysDbg.h"
 #include "graphics/host_gpu/regionDefinitions.h"
 #include "kernel/memory.h"

@@ -15,12 +14,6 @@
 #include <mutex>
 #include <vector>

-#if defined(_MSC_VER)
-#include <intrin.h>
-#elif defined(__x86_64__)
-#include <xmmintrin.h>
-#endif
-
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #ifndef NOMINMAX
 #define NOMINMAX
@@ -28,7 +21,10 @@
 #include <windows.h>
 #undef min
 #undef max
+#elif defined(__APPLE__)
+#include <unistd.h>
 #else
+#include <execinfo.h>
 #include <unistd.h>
 #endif

@@ -40,12 +36,38 @@ 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);
 	std::fputc('\n', stderr);
+#if !defined(__APPLE__)
+	void* frames[16] {};
+	int   frame_count = static_cast<int>(std::size(frames));
+	SysStackWalk(frames, &frame_count);
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	const auto image_base = reinterpret_cast<uintptr_t>(GetModuleHandleW(nullptr));
+	for (int i = 0; i < frame_count; i++) {
+		const auto address = reinterpret_cast<uintptr_t>(frames[i]);
+		std::fprintf(stderr, "  frame[%d]=0x%016" PRIxPTR " image_rva=0x%016" PRIxPTR "\n", i,
+		             address, address >= image_base ? address - image_base : 0);
+	}
+#else
+	::backtrace_symbols_fd(frames, frame_count, STDERR_FILENO);
+#endif
+#endif
 	std::fflush(stderr);
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	TerminateProcess(GetCurrentProcess(), static_cast<UINT>(EXCEPTION_NONCONTINUABLE_EXCEPTION));
@@ -64,17 +86,20 @@ 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) {
 		while (m_lock.test_and_set(std::memory_order_acquire)) {
-#if defined(__x86_64__) || defined(_M_X64)
-			_mm_pause();
-#elif defined(_M_ARM64)
-			__yield();
-#elif defined(__aarch64__)
-			asm volatile("yield");
-#endif
+			std::atomic_signal_fence(std::memory_order_seq_cst);
 		}
 	}
 	~SpinGuard() { m_lock.clear(std::memory_order_release); }
@@ -90,6 +115,15 @@ 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 {
@@ -97,14 +131,14 @@ 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>
@@ -168,6 +202,9 @@ 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() {
@@ -202,10 +239,11 @@ 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);
 		}
 	}

@@ -262,12 +300,11 @@ 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);
@@ -334,4 +371,8 @@ template void PageManager::UpdatePageWatchersForRegion<true, false>(uint64_t, Re
 template void PageManager::UpdatePageWatchersForRegion<false, true>(uint64_t, RegionBits&);
 template void PageManager::UpdatePageWatchersForRegion<false, false>(uint64_t, RegionBits&);

+void PageManager::OnGpuMap(uint64_t, uint64_t) {}
+
+void PageManager::OnGpuUnmap(uint64_t, uint64_t) {}
+
 } // namespace Libs::Graphics
@@ -24,6 +24,8 @@ public:
 	void UpdatePageWatchers(uint64_t vaddr, uint64_t size);
 	template <bool track, bool is_read = false>
 	void UpdatePageWatchersForRegion(uint64_t base_addr, RegionBits& mask);
+	void OnGpuMap(uint64_t vaddr, uint64_t size);
+	void OnGpuUnmap(uint64_t vaddr, uint64_t size);

 private:
 	struct Impl;
@@ -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});
 			}
 		}
 	}
@@ -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; }
@@ -9,12 +9,6 @@
 #include <mutex>
 #include <utility>

-#if defined(_MSC_VER)
-#include <intrin.h>
-#elif defined(__x86_64__)
-#include <xmmintrin.h>
-#endif
-
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #ifndef NOMINMAX
 #define NOMINMAX
@@ -22,7 +16,6 @@
 #include <windows.h>
 #undef min
 #undef max
-#undef MemoryBarrier
 #elif defined(__APPLE__)
 #include <pthread.h>
 #elif defined(__linux__)
@@ -36,20 +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 defined(__x86_64__) || defined(_M_X64)
-			_mm_pause();
-#elif defined(_M_ARM64)
-			__yield();
-#elif defined(__aarch64__)
-			asm volatile("yield");
-#endif
+			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);
 	}
@@ -60,9 +54,7 @@ private:
 		return GetCurrentThreadId();
 #elif defined(__APPLE__)
 		// mach thread port is a nonzero per-thread id (0 is the "no owner" sentinel).
-		static thread_local const uint32_t tid =
-		    static_cast<uint32_t>(pthread_mach_thread_np(pthread_self()));
-		return tid;
+		return static_cast<uint32_t>(pthread_mach_thread_np(pthread_self()));
 #elif defined(__linux__)
 		static thread_local const uint32_t tid = static_cast<uint32_t>(::syscall(SYS_gettid));
 		return tid;
@@ -95,7 +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>
@@ -118,46 +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>();
-		if (bits.FirstRangeFrom(start).first >= end) {
+		RegionBits mask(GetBits<source>(), start, end);
+		if constexpr (clear) {
+			GetBits<source>().UnsetRange(start, end);
+		}
+		if constexpr (source == DirtySource::Cpu && clear) {
+			UpdateCpuProtection<true>();
+			ForEachRange(mask, std::forward<Func>(func));
 			return;
 		}
-		RegionBits mask(bits, start, end);
-		if constexpr (clear) {
-			bits.UnsetRange(start, end);
-			if constexpr (source == DirtySource::Cpu) {
-				UpdateProtection<true, false>();
-			} else {
-				UpdateProtection<false, true>();
-			}
-		}
-		for (const auto [first, last]: mask) {
-			func(m_cpu_addr + first * TRACKER_PAGE_SIZE, (last - first) * TRACKER_PAGE_SIZE);
+		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>
@@ -188,6 +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;
@@ -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"
@@ -14,6 +13,7 @@
 #include "kernel/memory.h"

 #include <algorithm>
+#include <array>
 #include <cinttypes>
 #include <cstring>
 #include <memory>
@@ -34,42 +34,22 @@ void BufferCache::WriteDataBuffer(Buffer& buffer, uint64_t address, const void*
 	auto* bytes = static_cast<const uint8_t*>(source);
 	while (size != 0) {
 		const auto chunk  = std::min(size, m_staging_buffer.Size());
-		// Copy waits for ring reuse and flushes the fresh host data before submission.
 		const auto offset = m_staging_buffer.Copy(bytes, chunk, 4);
-		const auto destination_offset = buffer.Offset(address);
-		EXIT_IF(destination_offset > buffer.Size() || chunk > buffer.Size() - destination_offset);
-		m_scheduler.EndRendering();
-		const auto command = m_scheduler.Current().Handle();
-		vk::BufferMemoryBarrier2 before {};
-		before.srcStageMask  = vk::PipelineStageFlagBits2::eAllCommands;
-		before.srcAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite;
-		before.dstStageMask  = vk::PipelineStageFlagBits2::eTransfer;
-		before.dstAccessMask = vk::AccessFlagBits2::eTransferWrite;
-		before.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-		before.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
-		before.buffer        = buffer.Handle();
-		before.offset        = destination_offset;
-		before.size          = chunk;
-		vk::DependencyInfo dependency {};
-		dependency.dependencyFlags          = vk::DependencyFlagBits::eByRegion;
-		dependency.bufferMemoryBarrierCount = 1;
-		dependency.pBufferMemoryBarriers    = &before;
-		command.pipelineBarrier2(dependency);
-		const vk::BufferCopy copy {offset, destination_offset, chunk};
-		command.copyBuffer(m_staging_buffer.Handle(), buffer.Handle(), 1, &copy);
-		auto after          = before;
-		after.srcStageMask  = vk::PipelineStageFlagBits2::eTransfer;
-		after.srcAccessMask = vk::AccessFlagBits2::eTransferWrite;
-		after.dstStageMask  = vk::PipelineStageFlagBits2::eAllCommands;
-		after.dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite;
-		dependency.pBufferMemoryBarriers = &after;
-		command.pipelineBarrier2(dependency);
+		buffer.CopyFrom(m_scheduler.Current(), m_staging_buffer, offset, buffer.Offset(address),
+		                chunk, vk::AccessFlagBits::eHostWrite);
 		bytes += chunk;
 		address += chunk;
 		size -= chunk;
 	}
 }

+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);
 }
@@ -82,8 +62,7 @@ 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;
@@ -92,7 +71,6 @@ 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;
@@ -104,7 +82,7 @@ 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());
@@ -113,7 +91,7 @@ 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;
 	}
@@ -126,7 +104,8 @@ 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);
@@ -137,98 +116,94 @@ 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,
                         PageManager& page_manager, TextureCache& texture_cache)
-    : m_graphics(graphics), m_scheduler(scheduler), m_fault_manager(graphics, scheduler, *this),
-      m_gds_buffer(graphics, scheduler, MemoryUsage::Stream, 0, AllFlags, GdsBufferSize),
-      m_bda_pagetable_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 0, AllFlags,
-                             BDA_PAGETABLE_SIZE),
-      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_device_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 128 * MiB),
-      m_texture_cache(texture_cache) {
+	: m_graphics(graphics), m_scheduler(scheduler),
+	  m_fault_manager(graphics, scheduler, *this, CACHING_PAGEBITS, CACHING_NUMPAGES),
+	  m_gds_buffer(graphics, scheduler, MemoryUsage::Stream, 0, AllFlags, GdsBufferSize),
+	  m_bda_pagetable_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 0, AllFlags,
+	                         BDA_PAGETABLE_SIZE),
+	  m_memory_tracker(page_manager),
+	  m_staging_buffer(graphics, scheduler, MemoryUsage::Upload, 512 * MiB),
+	  m_stream_buffer(graphics, scheduler, MemoryUsage::Stream, 64 * MiB),
+	  m_download_buffer(graphics, scheduler, MemoryUsage::Download, 32 * MiB),
+	  m_device_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 128 * MiB),
+	  m_texture_cache(texture_cache) {
 	std::memset(m_gds_buffer.Mapped().data(), 0, static_cast<size_t>(m_gds_buffer.Size()));
 	m_gds_buffer.Flush(0, m_gds_buffer.Size());
 	SetVulkanObjectNameF(m_graphics.device, m_bda_pagetable_buffer.Handle(),
@@ -279,19 +254,59 @@ 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)) {
+	m_scheduler.Context().GetGpu().SendCommandSync(
+	    [this, vaddr, size, is_write] { ReadMemoryOnGpu(vaddr, size, is_write); });
+}
+
+void BufferCache::ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write) {
+	// CPU invalidation reaches this point only for a GPU-owned tracker page. Resolve the exact
+	// Buffer owner on the GPU thread so the cache index remains single-thread-owned.
+	if (is_write && !IsRegionRegistered(vaddr, size)) {
+		return;
+	}
+	std::vector<DownloadCopy> copies;
+	m_memory_tracker.ForEachDownloadRange<false>(
+	    vaddr, size,
+	    [&](uint64_t address, uint64_t bytes) noexcept {
+		    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
+		                                           "memory invalidation");
+	    },
+		[&](uint64_t address, uint64_t bytes) noexcept {
+		    for (const auto range: m_gpu_modified_ranges.Intersections(address, bytes)) {
+			    for (uint64_t copied = 0; copied < range.size;) {
+				    const auto copy_address = range.address + copied;
+				    const auto* owner = m_page_table.Find(copy_address >> PageTable::kPageBits);
+				    if (owner == nullptr || !*owner) {
+					    EXIT("BufferCache: invalidation readback has no buffer owner\n");
+				    }
+				    auto& buffer = m_slot_buffers[*owner];
+				    if (!buffer.IsInBounds(copy_address, 1)) {
+					    EXIT("BufferCache: invalidation readback is outside its buffer owner\n");
+				    }
+				    const auto copy_size = std::min(
+				        range.size - copied, buffer.CpuAddress() + buffer.Size() - copy_address);
+				    copies.push_back(
+				        {&buffer, buffer.Offset(copy_address), copy_address, copy_size});
+				    copied += copy_size;
+			    }
+		    }
+	    });
+	if (copies.empty()) {
+		if (!is_write) {
 			return;
 		}
-		auto& buffer = m_slot_buffers[FindBuffer(vaddr, size)];
-
-		if (DownloadBufferMemory<false>(buffer, vaddr, size)) {
-			m_memory_tracker.UnmarkRegionAsGpuModified(vaddr, size);
-		}
-		if (is_write) {
-			m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
-		}
-	});
+		// A preceding read fault can consume the last GPU-owned copy after this write invalidation
+		// has already chosen to flush. Complete the CPU ownership handoff even though this callback
+		// no longer has bytes to download.
+		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
+		return;
+	}
+	DownloadBufferMemory(copies);
+	// The enumeration above covered whole dirty pages and every exact interval on them.
+	m_memory_tracker.UnmarkRegionAsGpuModified(vaddr, size);
+	if (is_write) {
+		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
+	}
 }

 BufferId BufferCache::FindBuffer(uint64_t vaddr, uint64_t size) {
@@ -311,87 +326,38 @@ BufferId BufferCache::FindBuffer(uint64_t vaddr, uint64_t size) {
 	return CreateBuffer(vaddr, size);
 }

-BufferCache::OverlapResult BufferCache::ResolveOverlaps(uint64_t vaddr, uint64_t size) {
-	static constexpr int      StreamLeapThreshold = 16;
-	static constexpr uint64_t StreamLeapSize      = CACHING_PAGESIZE * 128;
-
-	auto       begin      = vaddr;
-	auto       end        = vaddr + size;
-	const auto find_first = [&](uint64_t address) {
-		auto first = m_buffers.lower_bound(address);
-		if (first != m_buffers.begin()) {
-			const auto  previous = std::prev(first);
-			const auto& buffer   = m_slot_buffers[previous->second];
-			if (buffer.CpuAddress() + buffer.Size() > address) {
-				first = previous;
-			}
-		}
-		return first;
-	};
-	auto first           = find_first(begin);
-	auto last            = first;
-	int  stream_score    = 0;
-	bool has_stream_leap = false;
-	for (; last != m_buffers.end() && last->first < end; ++last) {
-		const auto& buffer        = m_slot_buffers[last->second];
-		const auto  buffer_begin  = buffer.CpuAddress();
-		const auto  buffer_end    = buffer_begin + buffer.Size();
-		const bool  expands_left  = buffer_begin < begin;
-		const bool  expands_right = buffer_end > end;
-		begin                     = std::min(begin, buffer_begin);
-		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.
-			// 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);
-			}
-			if (expands_right) {
-				const auto minimum = vaddr < LOWER_ADDRESS_SIZE
-				                         ? CACHING_PAGESIZE * 2
-				                         : LibKernel::Memory::kExtendedMemoryBase;
-				if (begin > minimum) {
-					begin -= std::min(StreamLeapSize, begin - minimum);
-				}
-				first = find_first(begin);
-				begin = std::min(begin, first->first);
-			}
-		}
-	}
-	return {first, last, begin, end, has_stream_leap};
-}
-
-void BufferCache::JoinOverlap(BufferId new_id, BufferId overlap_id, bool accumulate_stream_score) {
-	auto& new_buffer = m_slot_buffers[new_id];
-	auto& overlap    = m_slot_buffers[overlap_id];
-	if (accumulate_stream_score) {
-		new_buffer.IncreaseStreamScore(overlap.StreamScore() + 1);
-	}
-	new_buffer.CopyFrom(m_scheduler.Current(), overlap, 0,
-	                    overlap.CpuAddress() - new_buffer.CpuAddress(), overlap.Size());
-	DeleteBuffer(overlap_id);
-}
-
 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);
-	size               = end - vaddr;
-	const auto overlap = ResolveOverlaps(vaddr, size);
+	auto& command = m_scheduler.Current();
+	EXIT_IF(command.IsInvalid());
+	auto       begin = vaddr & ~(CACHING_PAGESIZE - 1);
+	auto       end   = (vaddr + size + CACHING_PAGESIZE - 1) & ~(CACHING_PAGESIZE - 1);
+	auto       first = m_buffers.lower_bound(begin);
+	if (first != m_buffers.begin()) {
+		const auto previous = std::prev(first);
+		if (const auto& buffer = m_slot_buffers[previous->second];
+		    buffer.CpuAddress() + buffer.Size() > begin) {
+			first = previous;
+		}
+	}
+	auto last = first;
+	for (; last != m_buffers.end() && last->first < end; ++last) {
+		const auto& buffer = m_slot_buffers[last->second];
+		begin              = std::min(begin, buffer.CpuAddress());
+		end                = std::max(end, buffer.CpuAddress() + buffer.Size());
+	}

 	const auto id = m_slot_buffers.insert(
-	    m_graphics, m_scheduler, MemoryUsage::DeviceLocal, overlap.begin,
-	    AllFlags | vk::BufferUsageFlagBits::eShaderDeviceAddress, overlap.end - overlap.begin);
-	const auto& buffer = m_slot_buffers[id];
+	    m_graphics, m_scheduler, MemoryUsage::DeviceLocal, begin,
+	    AllFlags | vk::BufferUsageFlagBits::eShaderDeviceAddress, end - begin);
+	auto&      buffer = m_slot_buffers[id];
 	SetVulkanObjectNameF(m_graphics.device, buffer.Handle(),
-	                     "Kyty.GameBuffer[guest=0x{:016x} size=0x{:x}]", overlap.begin,
-	                     overlap.end - overlap.begin);
-	for (auto it = overlap.first; it != overlap.last;) {
-		const auto old_id = (it++)->second;
-		JoinOverlap(id, old_id, !overlap.has_stream_leap);
+	                     "Kyty.GameBuffer[guest=0x{:016x} size=0x{:x}]", begin, end - begin);
+	for (auto overlap = first; overlap != last;) {
+		const auto current = overlap++;
+		const auto old_id  = current->second;
+		const auto& old    = m_slot_buffers[old_id];
+		buffer.CopyFrom(command, old, 0, old.CpuAddress() - begin, old.Size());
+		DeleteBuffer(old_id);
 	}
 	Register(id);
 	return id;
@@ -414,6 +380,7 @@ bool BufferCache::SynchronizeBuffer(Buffer& buffer, uint64_t vaddr, uint64_t siz
 		command.EndRendering();
 		const auto native = command.Handle();
 		vk::BufferMemoryBarrier before {};
+		before.sType         = vk::StructureType::eBufferMemoryBarrier;
 		before.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite |
 		                       vk::AccessFlagBits::eTransferRead |
 		                       vk::AccessFlagBits::eTransferWrite;
@@ -485,48 +452,113 @@ 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) {
 	if (!GuestRange {vaddr, size}.Valid()) {
 		EXIT("BufferCache: invalid image source\n");
 	}
-	const auto* owner = m_page_table.Find(vaddr >> PageTable::kPageBits);
-	if (owner != nullptr && *owner) {
+	auto find_owner = [&]() -> Buffer* {
+		const auto* owner = m_page_table.Find(vaddr >> PageTable::kPageBits);
+		if (owner == nullptr || !*owner) {
+			return nullptr;
+		}
 		auto& buffer = m_slot_buffers[*owner];
-		if (buffer.IsInBounds(vaddr, size)) {
-			TouchBuffer(buffer);
-			(void)SynchronizeBuffer(buffer, vaddr, size, false, false);
-			return {&buffer, buffer.Offset(vaddr)};
+		return buffer.IsInBounds(vaddr, size) ? &buffer : nullptr;
+	};
+
+	{
+		const bool cpu_modified            = m_memory_tracker.IsRegionCpuModified(vaddr, size);
+		const bool gpu_modified            = m_memory_tracker.IsRegionGpuModified(vaddr, size);
+		const bool has_dirty_buffer_source = m_gpu_modified_ranges.Intersects(vaddr, size);
+		m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size,
+		                                           "image source");
+
+		auto* owner = find_owner();
+		if (has_dirty_buffer_source && owner == nullptr) {
+			if (!IsRegionRegistered(vaddr, size)) {
+				EXIT("BufferCache: GPU-dirty image source has no native buffer\n");
+			}
+			owner = &m_slot_buffers[FindBuffer(vaddr, size)];
+		}
+		if (owner != nullptr && !cpu_modified && (!gpu_modified || has_dirty_buffer_source)) {
+			TouchBuffer(*owner);
+			return {owner, owner->Offset(vaddr)};
+		}
+		if (has_dirty_buffer_source && owner == nullptr) {
+			EXIT("BufferCache: GPU-dirty image source could not resolve its native owner\n");
 		}
-	}
-	if (IsRegionGpuModified(vaddr, size)) {
-		return ObtainBuffer(vaddr, size, false, false);
 	}

 	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();
-	return {&m_staging_buffer, stage_offset};
+
+	const bool has_dirty_buffer_source = m_gpu_modified_ranges.Intersects(vaddr, size);
+	auto*      owner                   = find_owner();
+	if (has_dirty_buffer_source && owner == nullptr) {
+		EXIT("BufferCache: GPU-dirty image source lost its native owner\n");
+	}
+	if (owner == nullptr ||
+	    (m_memory_tracker.IsRegionGpuModified(vaddr, size) && !has_dirty_buffer_source)) {
+		return {&m_staging_buffer, stage_offset};
+	}
+
+	TouchBuffer(*owner);
+	std::vector<std::pair<uint64_t, uint64_t>> uploads;
+	m_memory_tracker.ForEachUploadRange(
+	    vaddr, size, false,
+	    [&](uint64_t address, uint64_t upload_size) noexcept {
+		    uploads.emplace_back(address, upload_size);
+	    },
+	    [&]() noexcept {
+		    for (const auto& [address, upload_size]: uploads) {
+			    owner->CopyFrom(m_scheduler.Current(), m_staging_buffer,
+			                    stage_offset + address - vaddr, owner->Offset(address), upload_size,
+			                    vk::AccessFlagBits::eHostWrite);
+		    }
+	    });
+	return {owner, owner->Offset(vaddr)};
+}
+
+void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data) {
+	if (vaddr == 0 || data.empty() || data.size() > UINT64_MAX - vaddr) {
+		EXIT("BufferCache: invalid host DMA write\n");
+	}
+	Libs::LibKernel::Memory::WriteBacking(vaddr, data.data(), data.size());
+
+	const auto end = vaddr + data.size();
+	for (const auto& [address, id]: m_buffers) {
+		auto&      buffer     = m_slot_buffers[id];
+		const auto buffer_end = address + buffer.Size();
+		const auto begin      = std::max(vaddr, address);
+		const auto range_end  = std::min(end, buffer_end);
+		if (begin >= range_end) {
+			continue;
+		}
+		WriteDataBuffer(buffer, begin, data.data() + begin - vaddr, range_end - begin);
+		TouchBuffer(buffer);
+	}
 }

 void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool is_gds) {
@@ -544,15 +576,29 @@ void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool
 		EXIT("BufferCache: invalid fill memory address\n");
 	}
 	(void)m_texture_cache.ClearMeta(vaddr);
-	if (!IsRegionGpuModified(vaddr, size)) {
-		// Access the guest mapping so write faults invalidate cached buffers and images.
-		auto* destination = reinterpret_cast<uint32_t*>(vaddr);
-		std::fill(destination, destination + size / sizeof(uint32_t), value);
-		return;
+	{
+		const auto region = m_texture_cache.QueryRegion(vaddr, size);
+		if (!HasGpuDirtyBytes(vaddr, size) && !region.gpu_image_bytes) {
+			if (region.image_bytes) {
+				m_texture_cache.InvalidateMemory(vaddr, size);
+			}
+			std::array<uint32_t, 4096> values;
+			values.fill(value);
+			const std::span<const uint8_t> bytes {reinterpret_cast<const uint8_t*>(values.data()),
+			                                      sizeof(values)};
+			for (uint64_t offset = 0; offset < size;) {
+				const auto chunk = std::min<uint64_t>(size - offset, bytes.size());
+				WriteHostMemory(vaddr + offset, bytes.first(chunk));
+				offset += chunk;
+			}
+			return;
+		}
 	}

 	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);
 }

@@ -569,11 +615,29 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 		     " size=0x%016" PRIx64 " src_gds=%d dst_gds=%d\n",
 		     src_vaddr, dst_vaddr, size, static_cast<int>(src_gds), static_cast<int>(dst_gds));
 	}
-	if (src_memory && dst_memory && !IsRegionGpuModified(dst_vaddr, size) &&
-	    !IsRegionGpuModified(src_vaddr, size) && !m_texture_cache.FindImageFromRange(src_vaddr, size)) {
-		std::memcpy(reinterpret_cast<void*>(dst_vaddr), reinterpret_cast<const void*>(src_vaddr),
-		            size);
-		return;
+	if (src_memory || dst_memory) {
+		const auto src_region =
+		    src_memory ? m_texture_cache.QueryRegion(src_vaddr, size) : TextureCache::RegionInfo {};
+		const auto dst_region =
+		    dst_memory ? m_texture_cache.QueryRegion(dst_vaddr, size) : TextureCache::RegionInfo {};
+		if (src_memory && dst_memory && !HasGpuDirtyBytes(src_vaddr, size) &&
+		    !HasGpuDirtyBytes(dst_vaddr, size) && !src_region.gpu_image_bytes &&
+		    !dst_region.gpu_image_bytes) {
+			if (dst_region.image_bytes) {
+				m_texture_cache.InvalidateMemory(dst_vaddr, size);
+			}
+			std::array<uint8_t, 64 * 1024> bytes;
+			for (uint64_t offset = 0; offset < size;) {
+				const auto chunk = std::min<uint64_t>(size - offset, bytes.size());
+				if (!Libs::LibKernel::Memory::TryReadBacking(src_vaddr + offset, bytes.data(),
+				                                             chunk)) {
+					EXIT("BufferCache: host DMA source has no direct backing\n");
+				}
+				WriteHostMemory(dst_vaddr + offset, std::span {bytes}.first(chunk));
+				offset += chunk;
+			}
+			return;
+		}
 	}

 	auto& command = m_scheduler.Current();
@@ -586,6 +650,10 @@ 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);
 }

@@ -629,7 +697,8 @@ 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);
@@ -640,7 +709,19 @@ 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());
@@ -652,10 +733,8 @@ 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());
@@ -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();
@@ -86,26 +74,16 @@ 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;
-		BufferMap::iterator last;
-		uint64_t            begin;
-		uint64_t            end;
-		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);
-	void JoinOverlap(BufferId new_id, BufferId overlap_id, bool accumulate_stream_score);
 	[[nodiscard]] BufferId CreateBuffer(uint64_t vaddr, uint64_t size);
 	void                   Register(BufferId id);
 	void Unregister(BufferId id);
@@ -117,9 +95,9 @@ 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 WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data);
+	void ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write);

 	GraphicContext&                                   m_graphics;
 	CommandScheduler&                                 m_scheduler;
@@ -128,7 +106,7 @@ private:
 	Buffer                                            m_bda_pagetable_buffer;
 	Common::SlotVector<Buffer>                        m_slot_buffers;
 	Common::LeastRecentlyUsedCache<BufferId, uint64_t> m_lru_cache;
-	BufferMap                                         m_buffers;
+	std::map<uint64_t, BufferId>                      m_buffers;
 	PageTable                                         m_page_table;
 	RangeSet                                          m_gpu_modified_ranges;
 	MemoryTracker                                     m_memory_tracker;
@@ -23,10 +23,13 @@ constexpr size_t PageFaultAreaSize = MaxPageFaults * sizeof(uint64_t);
 } // namespace

 FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler,
-                           BufferCache& buffer_cache)
+                           BufferCache& buffer_cache, uint32_t caching_pagebits,
+                           uint64_t caching_num_pages)
    : m_graphics(graphics), m_scheduler(scheduler), m_buffer_cache(buffer_cache),
+      m_caching_pagesize(uint64_t {1} << caching_pagebits),
+      m_caching_num_pages(caching_num_pages), m_fault_buffer_size(caching_num_pages / 8),
      m_fault_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 0, AllFlags,
-                     BufferCache::CACHING_NUMPAGES / 8),
+                     m_fault_buffer_size),
      m_download_buffer(graphics, scheduler, MemoryUsage::Download, 0, AllFlags,
                        MaxPendingFaults * PageFaultAreaSize) {
 	SetVulkanObjectNameF(m_graphics.device, m_fault_buffer.Handle(), "Fault Buffer");
@@ -36,6 +39,7 @@ FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler
 	    {1, vk::DescriptorType::eStorageBuffer, 1, vk::ShaderStageFlagBits::eCompute, nullptr},
 	};
 	vk::DescriptorSetLayoutCreateInfo layout_info {};
+	layout_info.sType        = vk::StructureType::eDescriptorSetLayoutCreateInfo;
 	layout_info.flags        = vk::DescriptorSetLayoutCreateFlagBits::ePushDescriptorKHR;
 	layout_info.bindingCount = std::size(bindings);
 	layout_info.pBindings    = bindings;
@@ -44,9 +48,16 @@ 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.sType    = vk::StructureType::eShaderModuleCreateInfo;
+	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.sType          = vk::StructureType::ePipelineLayoutCreateInfo;
 	pipeline_layout_info.setLayoutCount = 1;
 	pipeline_layout_info.pSetLayouts    = &m_fault_process_desc_layout;
 	RequireVulkanSuccess(
@@ -55,10 +66,12 @@ FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler
 	    "create fault-buffer pipeline layout");

 	vk::PipelineShaderStageCreateInfo stage {};
+	stage.sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
 	stage.stage  = vk::ShaderStageFlagBits::eCompute;
 	stage.module = module;
 	stage.pName  = "main";
 	vk::ComputePipelineCreateInfo pipeline_info {};
+	pipeline_info.sType  = vk::StructureType::eComputePipelineCreateInfo;
 	pipeline_info.stage  = stage;
 	pipeline_info.layout = m_fault_process_pipeline_layout;
 	const auto result = m_graphics.device.createComputePipelines(
@@ -86,13 +99,14 @@ void FaultManager::ProcessFaultBuffer() {
 	m_download_buffer.Flush(offset, PageFaultAreaSize);

 	vk::BufferMemoryBarrier2 pre_barrier {};
+	pre_barrier.sType         = vk::StructureType::eBufferMemoryBarrier2;
 	pre_barrier.srcStageMask  = vk::PipelineStageFlagBits2::eAllCommands;
 	pre_barrier.srcAccessMask = vk::AccessFlagBits2::eShaderWrite;
 	pre_barrier.dstStageMask  = vk::PipelineStageFlagBits2::eComputeShader;
 	pre_barrier.dstAccessMask = vk::AccessFlagBits2::eShaderRead;
 	pre_barrier.buffer        = m_fault_buffer.Handle();
 	pre_barrier.offset        = 0;
-	pre_barrier.size           = m_fault_buffer.Size();
+	pre_barrier.size          = m_fault_buffer_size;
 	auto post_barrier         = pre_barrier;
 	post_barrier.srcStageMask  = vk::PipelineStageFlagBits2::eComputeShader;
 	post_barrier.srcAccessMask = vk::AccessFlagBits2::eShaderWrite;
@@ -100,11 +114,12 @@ void FaultManager::ProcessFaultBuffer() {
 	post_barrier.dstAccessMask = vk::AccessFlagBits2::eShaderWrite;

 	const vk::DescriptorBufferInfo infos[] {
-	    {m_fault_buffer.Handle(), 0, m_fault_buffer.Size()},
+	    {m_fault_buffer.Handle(), 0, m_fault_buffer_size},
 	    {m_download_buffer.Handle(), offset, PageFaultAreaSize},
 	};
 	std::array<vk::WriteDescriptorSet, 2> writes {};
 	for (uint32_t index = 0; index < writes.size(); ++index) {
+		writes[index].sType           = vk::StructureType::eWriteDescriptorSet;
 		writes[index].dstBinding      = index;
 		writes[index].descriptorCount = 1;
 		writes[index].descriptorType  = vk::DescriptorType::eStorageBuffer;
@@ -114,6 +129,7 @@ void FaultManager::ProcessFaultBuffer() {
 	m_scheduler.EndRendering();
 	auto command = m_scheduler.Current().Handle();
 	vk::DependencyInfo dependency {};
+	dependency.sType                    = vk::StructureType::eDependencyInfo;
 	dependency.dependencyFlags          = vk::DependencyFlagBits::eByRegion;
 	dependency.bufferMemoryBarrierCount = 1;
 	dependency.pBufferMemoryBarriers    = &pre_barrier;
@@ -121,7 +137,7 @@ void FaultManager::ProcessFaultBuffer() {
 	command.bindPipeline(vk::PipelineBindPoint::eCompute, m_fault_process_pipeline);
 	command.pushDescriptorSetKHR(vk::PipelineBindPoint::eCompute,
 	                             m_fault_process_pipeline_layout, 0, writes);
-	const auto num_threads    = BufferCache::CACHING_NUMPAGES / 32;
+	const auto num_threads    = m_caching_num_pages / 32;
 	const auto num_workgroups = (num_threads + 63) / 64;
 	command.dispatch(static_cast<uint32_t>(num_workgroups), 1, 1);
 	dependency.pBufferMemoryBarriers = &post_barrier;
@@ -130,15 +146,14 @@ void FaultManager::ProcessFaultBuffer() {
 	const auto area = m_current_area;
 	m_scheduler.DeferOperation([this, mapped, offset, area] {
 		m_download_buffer.Invalidate(offset, PageFaultAreaSize);
-		RangeSet    fault_ranges;
+		m_fault_ranges.Clear();
 		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);
+			m_fault_ranges.Add(faults[index], m_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) {
+		m_fault_ranges.ForEach([this](uint64_t start, uint64_t end) {
 			EXIT_IF(end - start > std::numeric_limits<uint32_t>::max());
 			(void)m_buffer_cache.FindBuffer(start, end - start);
 		});
@@ -2,6 +2,7 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_CACHE_FAULTMANAGER_H_

 #include "common/abi.h"
+#include "graphics/host_gpu/rangeSet.h"
 #include "graphics/host_gpu/renderer/cache/streamBuffer.h"

 #include <array>
@@ -15,7 +16,8 @@ class FaultManager {
 	static constexpr size_t MaxPendingFaults = 8;

 public:
-	FaultManager(GraphicContext& graphics, CommandScheduler& scheduler, BufferCache& buffer_cache);
+	FaultManager(GraphicContext& graphics, CommandScheduler& scheduler, BufferCache& buffer_cache,
+	             uint32_t caching_pagebits, uint64_t caching_num_pages);
 	~FaultManager();
 	KYTY_CLASS_NO_COPY(FaultManager);

@@ -26,6 +28,10 @@ private:
 	GraphicContext&                            m_graphics;
 	CommandScheduler&                          m_scheduler;
 	BufferCache&                               m_buffer_cache;
+	RangeSet                                   m_fault_ranges;
+	uint64_t                                   m_caching_pagesize;
+	uint64_t                                   m_caching_num_pages;
+	uint64_t                                   m_fault_buffer_size;
 	Buffer                                     m_fault_buffer;
 	Buffer                                     m_download_buffer;
 	std::array<uint64_t, MaxPendingFaults>      m_fault_areas {};
@@ -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 {
+	constexpr uint64_t fault_size = 8;
+	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);
+	}
+	m_page_manager.OnGpuMap(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);
+		m_page_manager.OnGpuUnmap(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
@@ -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_
@@ -256,6 +256,17 @@ private:
 	size_type                              m_inline_size = 0;
 };

+// Removes only the requested owner. Other owners in the same page entry remain intact.
+template <typename Container, typename Value>
+[[nodiscard]] bool EraseExact(Container& owners, const Value& owner) {
+	const auto it = std::find(owners.begin(), owners.end(), owner);
+	if (it == owners.end()) {
+		return false;
+	}
+	owners.erase(it);
+	return true;
+}
+
 } // namespace Libs::Graphics

 #endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_MULTILEVELPAGETABLE_H_
@@ -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]);
 		}
 	}

@@ -93,27 +95,30 @@ 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()));
 	}

+	sampler_info.sType     = vk::StructureType::eSamplerCreateInfo;
+	sampler_info.pNext     = nullptr;
+	sampler_info.flags     = {};
 	sampler_info.magFilter = to_vk_filter(mag_filter);
 	sampler_info.minFilter = to_vk_filter(min_filter);
 	sampler_info.mipmapMode =
@@ -123,11 +128,8 @@ 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);
@@ -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 {
@@ -1,12 +1,13 @@
 #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 "graphics/host_gpu/vma.h"

 #include <cstring>
+#include <limits>
 #include <numeric>
 #include <vk_mem_alloc.h>

@@ -45,11 +46,16 @@ 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;
 }

@@ -58,13 +64,14 @@ constexpr size_t WATCHES_RESERVE_CHUNK   = 0x1000;
 Buffer::Buffer(GraphicContext& graphics, CommandScheduler& scheduler, MemoryUsage usage,
               uint64_t cpu_address, vk::BufferUsageFlags flags, uint64_t size)
    : m_graphics(&graphics), m_scheduler(&scheduler), m_usage(usage), m_cpu_address(cpu_address),
-      m_size(size) {
+      m_size(size), m_buffer(std::make_unique<VulkanBuffer>()) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(graphics.allocator == nullptr || size == 0);

 	vk::BufferCreateInfo buffer_info {};
 	buffer_info.size        = size;
 	buffer_info.usage       = flags;
+	buffer_info.sharingMode = vk::SharingMode::eExclusive;

 	const bool with_bda = bool(flags & vk::BufferUsageFlagBits::eShaderDeviceAddress);
 	const VmaAllocationCreateFlags bda_flag =
@@ -81,70 +88,93 @@ Buffer::Buffer(GraphicContext& graphics, CommandScheduler& scheduler, MemoryUsag
 	VkBuffer          native_buffer = VK_NULL_HANDLE;
 	const auto        result        = static_cast<vk::Result>(vmaCreateBuffer(
 	    graphics.allocator, static_cast<const VkBufferCreateInfo*>(buffer_info), &allocation_info,
-	    &native_buffer, &m_allocation, &allocation_result));
+	    &native_buffer, &m_buffer->memory.allocation, &allocation_result));
 	if (result != vk::Result::eSuccess) {
 		graphics.LogMemoryBudget();
 	}
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);

-	m_buffer = native_buffer;
-	if (with_bda) {
+	m_buffer->buffer                 = native_buffer;
+	m_buffer->usage                  = flags;
+	m_buffer->buffer_size            = size;
+	m_buffer->memory.allocation_info = allocation_result;
+	m_buffer->memory.memory          = allocation_result.deviceMemory;
+	m_buffer->memory.offset          = allocation_result.offset;
+	m_buffer->memory.type            = allocation_result.memoryType;
+	m_buffer->memory.unique_id       = VulkanNextMemoryUniqueId();
+	graphics.device.getBufferMemoryRequirements(m_buffer->buffer, &m_buffer->memory.requirements);
+	if (static_cast<bool>(flags & vk::BufferUsageFlagBits::eShaderDeviceAddress)) {
 		vk::BufferDeviceAddressInfo address_info {};
-		address_info.buffer = m_buffer;
+		address_info.sType  = vk::StructureType::eBufferDeviceAddressInfo;
+		address_info.buffer = m_buffer->buffer;
 		m_device_address    = graphics.device.getBufferAddress(address_info);
 		EXIT_IF(m_device_address == 0);
 	}

 	VkMemoryPropertyFlags properties = 0;
-	vmaGetAllocationMemoryProperties(graphics.allocator, m_allocation, &properties);
-	m_coherent = (properties & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
+	vmaGetAllocationMemoryProperties(graphics.allocator, m_buffer->memory.allocation, &properties);
+	m_buffer->memory.property = vk::MemoryPropertyFlags(properties);
+	m_is_coherent             = (properties & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
 	if (allocation_result.pMappedData != nullptr) {
 		m_mapped = {static_cast<uint8_t*>(allocation_result.pMappedData),
 		            static_cast<size_t>(size)};
 	}
+	VulkanTrackAllocation(m_buffer->memory);
 }

 Buffer::~Buffer() {
-	if (m_buffer != nullptr) {
-		vmaDestroyBuffer(m_graphics->allocator, m_buffer, m_allocation);
+	if (m_buffer->buffer != nullptr) {
+		VulkanUntrackAllocation(m_buffer->memory);
+		vmaDestroyBuffer(m_graphics->allocator, m_buffer->buffer, m_buffer->memory.allocation);
 	}
 }

+vk::Buffer Buffer::Handle() const noexcept {
+	return m_buffer->buffer;
+}
+
 vk::DeviceAddress Buffer::BufferDeviceAddress() const noexcept {
 	EXIT_IF(m_device_address == 0);
 	return m_device_address;
 }

 bool Buffer::IsInBounds(uint64_t address, uint64_t size) const noexcept {
-	return address >= m_cpu_address && size <= Size() && address - m_cpu_address <= Size() - size;
+	return address >= m_cpu_address && size <= m_size && address - m_cpu_address <= m_size - size;
+}
+
+void Buffer::Write(uint64_t offset, const void* source, uint64_t size) {
+	EXIT_IF(source == nullptr || m_mapped.empty() || offset > m_size || size > m_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) {
+	EXIT_IF(m_mapped.empty() || offset > m_size || size > m_size - offset);
+	if (!m_is_coherent && size != 0) {
 		const auto result =
-		    vmaFlushAllocation(m_graphics->allocator, m_allocation, offset, size);
+		    vmaFlushAllocation(m_graphics->allocator, m_buffer->memory.allocation, offset, size);
 		EXIT_NOT_IMPLEMENTED(static_cast<vk::Result>(result) != vk::Result::eSuccess);
 	}
 }

 void Buffer::Invalidate(uint64_t offset, uint64_t size) {
-	EXIT_IF(m_usage != MemoryUsage::Download || offset > Size() || size > Size() - offset);
-	if (!IsCoherent() && size != 0) {
+	EXIT_IF(m_usage != MemoryUsage::Download || offset > m_size || size > m_size - offset);
+	if (!m_is_coherent && size != 0) {
 		const auto result =
-		    vmaInvalidateAllocation(m_graphics->allocator, m_allocation, offset, size);
+		    vmaInvalidateAllocation(m_graphics->allocator, m_buffer->memory.allocation, offset, size);
 		EXIT_NOT_IMPLEMENTED(static_cast<vk::Result>(result) != vk::Result::eSuccess);
 	}
 }

 vk::BufferMemoryBarrier Buffer::Barrier(uint64_t offset, uint64_t size, vk::AccessFlags source,
                                        vk::AccessFlags destination) const {
-	if (Handle() == nullptr || size == 0 || offset > Size() || size > Size() - offset) {
+	if (Handle() == nullptr || size == 0 || offset > m_size || size > m_size - offset) {
 		EXIT("Buffer: invalid DMA barrier, handle=%p offset=0x%016" PRIx64 " size=0x%016" PRIx64
 		     " capacity=0x%016" PRIx64 "\n",
-		     static_cast<const void*>(Handle()), offset, size, Size());
+		     static_cast<const void*>(Handle()), offset, size, m_size);
 	}
 	vk::BufferMemoryBarrier barrier {};
+	barrier.sType               = vk::StructureType::eBufferMemoryBarrier;
 	barrier.srcAccessMask       = source;
 	barrier.dstAccessMask       = destination;
 	barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
@@ -159,8 +189,8 @@ void Buffer::CopyFrom(CommandBuffer& command, const Buffer& source, uint64_t sou
                      uint64_t destination_offset, uint64_t size, vk::AccessFlags source_before,
                      vk::AccessFlags destination_before, vk::AccessFlags source_after,
                      vk::AccessFlags destination_after) {
-	if (size == 0 || source_offset > source.Size() || size > source.Size() - source_offset ||
-	    destination_offset > Size() || size > Size() - destination_offset) {
+	if (size == 0 || source_offset > source.m_size || size > source.m_size - source_offset ||
+	    destination_offset > m_size || size > m_size - destination_offset) {
 		EXIT("Buffer: invalid copy range\n");
 	}
 	if (source.Handle() == Handle() && source_offset < destination_offset + size &&
@@ -217,8 +247,10 @@ void Buffer::Fill(uint64_t offset, uint64_t size, uint32_t value) {

 StreamBuffer::StreamBuffer(GraphicContext& graphics, CommandScheduler& scheduler, MemoryUsage usage,
                           uint64_t size)
-    : Buffer(graphics, scheduler, usage, 0, AllFlags, size),
-      m_current_watches(WATCHES_INITIAL_RESERVE), m_previous_watches(WATCHES_INITIAL_RESERVE) {}
+    : Buffer(graphics, scheduler, usage, 0, AllFlags, size) {
+	ReserveWatches(m_current_watches, WATCHES_INITIAL_RESERVE);
+	ReserveWatches(m_previous_watches, WATCHES_INITIAL_RESERVE);
+}

 bool StreamBuffer::NormalizeReservation(bool coherent, uint64_t atom, uint64_t& size,
                                        uint64_t& alignment) {
@@ -283,8 +315,10 @@ std::pair<uint8_t*, uint64_t> StreamBuffer::Map(uint64_t size, uint64_t alignmen
 }

 void StreamBuffer::Commit() {
-	if (Usage() != MemoryUsage::Download && m_mapped_size != 0) {
-		Flush(m_offset, m_mapped_size);
+	if (!IsCoherent() && Usage() != MemoryUsage::Download && m_mapped_size != 0) {
+		const auto result = vmaFlushAllocation(
+		    Graphics().allocator, NativeBuffer().memory.allocation, m_offset, m_mapped_size);
+		EXIT_NOT_IMPLEMENTED(static_cast<vk::Result>(result) != vk::Result::eSuccess);
 	}

 	m_offset += m_mapped_size;
@@ -294,7 +328,7 @@ void StreamBuffer::Commit() {
 		return;
 	}
 	if (m_current_watch_cursor + 1 >= m_current_watches.size()) {
-		m_current_watches.resize(m_current_watches.size() + WATCHES_RESERVE_CHUNK);
+		ReserveWatches(m_current_watches, WATCHES_RESERVE_CHUNK);
 	}
 	auto& watch       = m_current_watches[m_current_watch_cursor++];
 	watch.upper_bound = m_offset;
@@ -310,6 +344,10 @@ uint64_t StreamBuffer::Copy(const void* source, uint64_t size, uint64_t alignmen
 	return offset;
 }

+void StreamBuffer::ReserveWatches(std::vector<Watch>& watches, size_t grow_size) {
+	watches.resize(watches.size() + grow_size);
+}
+
 bool StreamBuffer::WaitPendingOperations(const std::vector<Watch>& watches,
                                         std::optional<size_t>     invalidation_mark,
                                         uint64_t requested_upper_bound, bool allow_wait,