Cache CI builds with sccache

Co-authored-by: Adrian Endlicher <godlike93@gmail.com>

.github/workflows/build.yml

@@ -6,7 +6,12 @@ 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 }}
 
@@ -78,6 +83,9 @@ jobs:
           arch: win64_msvc2022_64
           cache: true
 
+      - name: Setup sccache
+        uses: mozilla-actions/sccache-action@v0.0.11
+
       - name: Verify toolchain
         shell: cmd
         run: |
@@ -100,6 +108,8 @@ 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
@@ -156,6 +166,9 @@ jobs:
           arch: clang_64
           cache: true
 
+      - name: Setup sccache
+        uses: mozilla-actions/sccache-action@v0.0.11
+
       - name: Verify toolchain
         shell: bash
         run: |
@@ -180,6 +193,8 @@ 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
@@ -289,9 +304,15 @@ jobs:
             clang \
             glslang-tools \
             libasound2-dev \
+            libavcodec-dev \
+            libavfilter-dev \
+            libavformat-dev \
+            libavutil-dev \
             libdbus-1-dev \
             libgl1-mesa-dev \
             libpulse-dev \
+            libswresample-dev \
+            libswscale-dev \
             libudev-dev \
             libwayland-dev \
             libx11-dev \
@@ -304,6 +325,7 @@ jobs:
             libxss-dev \
             lld \
             ninja-build \
+            pkg-config \
             wayland-protocols
 
       - name: Install Qt 6.10.3
@@ -315,6 +337,9 @@ jobs:
           arch: linux_gcc_64
           cache: true
 
+      - name: Setup sccache
+        uses: mozilla-actions/sccache-action@v0.0.11
+
       - name: Verify toolchain
         shell: bash
         run: |
@@ -339,6 +364,8 @@ 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]}"
 

3rdparty/CMakeLists.txt

@@ -88,6 +88,13 @@ set(SPDLOG_FMT_EXTERNAL ON CACHE BOOL "" FORCE)
 set(SPDLOG_NO_EXCEPTIONS ON CACHE BOOL "" FORCE)
 add_subdirectory(spdlog EXCLUDE_FROM_ALL)
 
+if (NOT TARGET FFmpeg::ffmpeg AND CMAKE_SYSTEM_NAME STREQUAL "Linux")
+	find_package(PkgConfig REQUIRED)
+	pkg_check_modules(FFMPEG REQUIRED IMPORTED_TARGET GLOBAL
+		libavformat libavcodec libswscale libavutil libavfilter libswresample)
+	add_library(FFmpeg::ffmpeg ALIAS PkgConfig::FFMPEG)
+endif()
+
 if (NOT TARGET FFmpeg::ffmpeg)
 	# On macOS the target arch is driven by CMAKE_OSX_ARCHITECTURES, not the host
 	# CMAKE_SYSTEM_PROCESSOR (which reports arm64 even for an x86_64/Rosetta build).

CMakeLists.txt

@@ -317,24 +317,10 @@ endif()
 
 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 required by the static FFmpeg archive.
+# Linux system libraries.
 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
@@ -342,6 +328,7 @@ set(inc_headers
 	${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
@@ -550,6 +537,10 @@ if(BUILD_TESTING)
 	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
@@ -567,6 +558,8 @@ 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

README.md

@@ -90,8 +90,14 @@ starting a large change.
 
 Set up the clang-format hook after cloning:
 
-```powershell
-python -m pip install pre-commit
+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
 python -m pre_commit install --install-hooks
 ```
 
@@ -159,13 +165,14 @@ The finished application and its runtime dependencies will be placed in
 
 ### Building on Linux
 
-Install the toolchain and the libraries the bundled SDL2 needs. Without the audio, Wayland and
+Install the toolchain, system FFmpeg, 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 \
+  clang lld ninja-build cmake git glslang-tools pkg-config \
+  libavformat-dev libavcodec-dev libswscale-dev libavutil-dev libavfilter-dev libswresample-dev \
   libgl1-mesa-dev libx11-dev libxcursor-dev libxext-dev libxfixes-dev \
   libxi-dev libxrandr-dev libxss-dev libxkbcommon-dev \
   libasound2-dev libpulse-dev libudev-dev libdbus-1-dev libwayland-dev wayland-protocols
@@ -186,7 +193,8 @@ 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.
+`_Build/linux/install` runs without a matching system Qt. The emulator still requires
+the system FFmpeg runtime libraries matching those used to build it.
 
 As on Windows, the MSVC compiler is not used; Clang is required. `cl.exe` is rejected at configure
 time.

src/common/alignment.h

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

src/common/assert.cpp

@@ -1,6 +1,5 @@
 #include "common/assert.h"
 
-#include "common/debug.h"
 #include "common/logging/log.h"
 #include "common/subsystems.h"
 #include "kytyGitVersion.h"
@@ -12,25 +11,10 @@
 
 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) {
-	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;
+	return fmt::format("--- Build ---\n{}\n{}\n{} in {}:{}\n", KYTY_BUILD_LABEL, title, text, file,
+	                   line);
 }
 
 static int DbgReport(const char* title, std::string_view text, const char* file, int line) {

src/common/debug.cpp

@@ -1,20 +1,7 @@
 #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";

src/common/debug.h

@@ -2,42 +2,17 @@
 #define KYTY_COMMON_DEBUG_H_
 
 #include "common/common.h"
-#include "common/stringUtils.h"
 
-#include <array>
-#include <cstring>
+#include <string>
 
 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_ */

src/common/emulatorConfig.cpp

@@ -43,6 +43,10 @@ int32_t GetUserId() {
 	return g_config->user_id;
 }
 
+const std::string& GetAudioInputDevice() {
+	return g_config->audio_input_device;
+}
+
 PresentMode GetPresentMode() {
 	return g_config->present_mode;
 }
@@ -55,6 +59,10 @@ bool FullscreenEnabled() {
 	return g_config->fullscreen_enabled;
 }
 
+bool VrEnabled() {
+	return g_config->vr_enabled;
+}
+
 uint32_t GetVblankFrequency() {
 	return std::clamp(g_config->vblank_frequency, 30u, 360u);
 }
@@ -75,7 +83,7 @@ ShaderOptimizationType GetShaderOptimizationType() {
 	return g_config->shader_optimization_type;
 }
 
-ShaderLogDirection GetShaderLogDirection() {
+LogDirection GetShaderLogDirection() {
 	return g_config->shader_log_direction;
 }
 
@@ -95,7 +103,7 @@ bool GraphicsDebugDumpEnabled() {
 	return g_config->graphics_debug_dump_enabled;
 }
 
-OutputDirection GetPrintfDirection() {
+LogDirection GetPrintfDirection() {
 	return g_config->printf_direction;
 }
 
@@ -103,8 +111,8 @@ std::filesystem::path GetPrintfOutputFile() {
 	return g_config->printf_output_file;
 }
 
-ProfilerDirection GetProfilerDirection() {
-	return g_config->profiler_direction;
+bool ProfilerEnabled() {
+	return g_config->profiler_enabled;
 }
 
 bool SpirvDebugPrintfEnabled() {

src/common/emulatorConfig.h

@@ -21,11 +21,7 @@ struct Lifecycle {
 
 enum class ShaderOptimizationType { None, Size, Performance };
 
-enum class ShaderLogDirection { Silent, Console, File };
-
-enum class ProfilerDirection { None, Network };
-
-enum class OutputDirection { Silent, Console, File };
+enum class LogDirection { Silent, Console, File };
 
 enum class PresentMode { Fifo, Mailbox, Immediate };
 
@@ -47,22 +43,24 @@ struct ConfigOptions {
 	uint32_t               screen_height               = 720;
 	std::string            user_name                   = "Kyty";
 	int32_t                user_id                     = DEFAULT_USER_ID;
-	PresentMode            present_mode                = PresentMode::Fifo;
+	std::string            audio_input_device;
+	PresentMode            present_mode                = PresentMode::Mailbox;
 	int32_t                gpu_index                   = -1;
 	bool                   fullscreen_enabled          = false;
+	bool                   vr_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;
-	ShaderLogDirection     shader_log_direction        = ShaderLogDirection::Silent;
+	LogDirection           shader_log_direction        = LogDirection::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;
-	OutputDirection        printf_direction            = OutputDirection::Silent;
+	LogDirection           printf_direction            = LogDirection::Silent;
 	std::filesystem::path  printf_output_file          = "_kyty.txt";
-	ProfilerDirection      profiler_direction          = ProfilerDirection::None;
+	bool                   profiler_enabled            = false;
 	bool                   spirv_debug_printf_enabled  = false;
 	bool                   gpu_assisted_validation_enabled = false;
 	bool                   renderdoc_enabled           = false;
@@ -80,16 +78,18 @@ uint32_t GetScreenWidth();
 uint32_t GetScreenHeight();
 const std::string& GetUserName();
 int32_t  GetUserId();
+const std::string& GetAudioInputDevice();
 PresentMode GetPresentMode();
 int32_t GetGpuIndex();
 bool     FullscreenEnabled();
+bool     VrEnabled();
 uint32_t GetVblankFrequency();
 uint32_t GetConsoleLanguage();
 bool     VulkanValidationEnabled();
 
 bool                   ShaderValidationEnabled();
 ShaderOptimizationType GetShaderOptimizationType();
-ShaderLogDirection     GetShaderLogDirection();
+LogDirection           GetShaderLogDirection();
 std::filesystem::path  GetShaderLogFolder();
 
 bool                  CommandBufferDumpEnabled();
@@ -97,10 +97,10 @@ std::filesystem::path GetCommandBufferDumpFolder();
 
 bool GraphicsDebugDumpEnabled();
 
-OutputDirection       GetPrintfDirection();
+LogDirection          GetPrintfDirection();
 std::filesystem::path GetPrintfOutputFile();
 
-ProfilerDirection GetProfilerDirection();
+bool ProfilerEnabled();
 
 bool SpirvDebugPrintfEnabled();
 

src/common/hostException.cpp

@@ -5,16 +5,17 @@
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #include <windows.h> // IWYU pragma: keep
-#elif defined(__APPLE__)
-#include <csignal>
-#include <sys/ucontext.h>
 #else
 #include <algorithm>
 #include <csignal>
 #include <cstdlib>
 #include <initializer_list>
-#include <ucontext.h> // IWYU pragma: keep
 #include <unistd.h>
+#if defined(__APPLE__)
+#include <sys/ucontext.h>
+#else
+#include <ucontext.h> // IWYU pragma: keep
+#endif
 #endif
 
 // IWYU pragma: no_include <errhandlingapi.h>
@@ -34,6 +35,51 @@ 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 {
@@ -166,46 +212,6 @@ static void SignalHandler(int sig, siginfo_t* si, void* uctx) {
 
 #else
 
-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();
-}
-
 // x86-64 page-fault error bits.
 constexpr uint64_t PAGE_FAULT_ERROR_WRITE       = 0x02;
 constexpr uint64_t PAGE_FAULT_ERROR_INSTRUCTION = 0x10;
@@ -295,7 +301,7 @@ bool InstallHandler(Handler handler) {
 #elif defined(__APPLE__)
 	struct sigaction sa {};
 	sa.sa_sigaction = SignalHandler;
-	sa.sa_flags     = SA_SIGINFO;
+	sa.sa_flags     = SA_SIGINFO | SA_ONSTACK;
 	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

src/common/logging/log.cpp

@@ -129,6 +129,14 @@ 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);
@@ -152,9 +160,9 @@ void WriteFatal(fmt::text_style style, std::string_view text) {
 void Initialize() {
 	g_initialized = true;
 	switch (Config::GetPrintfDirection()) {
-		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;
+		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;
 	}
 	g_output_file =
 	    (g_direction == Direction::File ? Config::GetPrintfOutputFile() : std::filesystem::path {});

src/common/logging/log.h

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

src/common/platform/sysDbg.h

@@ -30,8 +30,6 @@ struct sys_dbg_stack_info_t {
 #endif
 };
 
-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)
+void SysStackUsage(sys_dbg_stack_info_t& s); // NOLINT(google-runtime-references)
 
 #endif /* KYTY_COMMON_PLATFORM_SYSDBG_H_ */

src/common/platform/sysLinuxDbg.cpp

@@ -8,7 +8,6 @@
 
 #include <cstdlib>
 #include <cstring>
-#include <execinfo.h>
 #include <pthread.h>
 #include <sys/param.h>
 #include <sys/types.h>
@@ -17,56 +16,6 @@
 #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();
 

src/common/platform/sysLinuxVirtual.cpp

@@ -31,8 +31,7 @@
 namespace Common {
 
 static pthread_mutex_t              g_virtual_mutex {};
-static std::map<uintptr_t, size_t>* g_allocs   = nullptr;
-static std::map<uintptr_t, int>*    g_protects = nullptr;
+static std::map<uintptr_t, size_t>* g_allocs = nullptr;
 
 void SysVirtualInit() {
 	pthread_mutexattr_t attr {};
@@ -46,8 +45,7 @@ void SysVirtualInit() {
 	pthread_mutex_init(&g_virtual_mutex, &attr);
 	pthread_mutexattr_destroy(&attr);
 
-	g_allocs   = new std::map<uintptr_t, size_t>;
-	g_protects = new std::map<uintptr_t, int>;
+	g_allocs = new std::map<uintptr_t, size_t>;
 }
 
 static int get_protection_flag(VirtualMemory::Mode mode) {
@@ -68,21 +66,6 @@ static int get_protection_flag(VirtualMemory::Mode mode) {
 	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
@@ -161,11 +144,6 @@ uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mo
 	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);
 	}
 
@@ -251,11 +229,6 @@ uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::
 
 	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;
@@ -337,11 +310,6 @@ bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode m
 	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;
@@ -516,18 +484,7 @@ bool SysVirtualFree(uint64_t address) {
 		return false;
 	}
 
-	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;
+	return munmap(reinterpret_cast<void*>(addr), size) == 0;
 }
 
 bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
@@ -582,40 +539,16 @@ bool SysVirtualFreeRange(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 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 SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::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 SysVirtualFlushInstructionCache(uint64_t /*address*/, uint64_t /*size*/) {

src/common/platform/sysVirtual.h

@@ -19,8 +19,7 @@ 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     SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
 bool     SysVirtualFlushInstructionCache(uint64_t address, uint64_t size);
 
 } // namespace Common

src/common/platform/sysWindowsDbg.cpp

@@ -15,18 +15,6 @@
 #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));

src/common/platform/sysWindowsVirtual.cpp

@@ -45,18 +45,6 @@ static DWORD GetProtectionFlag(VirtualMemory::Mode mode) {
 	return protect;
 }
 
-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;
-	}
-}
-
 void SysVirtualInit() {}
 
 uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
@@ -328,17 +316,13 @@ bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
 	return current - address == size && SysVirtualFree(address);
 }
 
-bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
-                       VirtualMemory::Mode* old_mode) {
+bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode 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;
 }
 

src/common/profiler.cpp

@@ -55,19 +55,13 @@ void SetThreadName(const char* name) {
 }
 
 void Initialize() {
-	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;
+	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);
 	}
 }
 

src/common/stringUtils.h

@@ -317,8 +317,11 @@ inline std::string SafeCsv(std::string_view text) {
 }
 
 inline std::string Utf16ToUtf8(std::u16string_view utf16) {
+	if (utf16.empty()) {
+		return {};
+	}
 	std::wstring_convert<std::codecvt_utf8_utf16<char16_t>, char16_t> convert;
-	return convert.to_bytes(utf16.begin(), utf16.end());
+	return convert.to_bytes(utf16.data(), utf16.data() + utf16.size());
 }
 
 inline ByteBuffer HexToBin(std::string_view text) {

src/common/threads.cpp

@@ -1,7 +1,6 @@
 #include "common/threads.h"
 
 #include "common/assert.h"
-#include "common/debug.h"
 
 #include <algorithm>
 #include <atomic>
@@ -9,7 +8,6 @@
 #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
@@ -198,37 +196,7 @@ struct CondVarPrivate {
 #endif
 };
 
-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);
-	}
-}
+static wait_poll_func_t g_cond_wait_poll_callback = nullptr;
 
 struct ThreadPrivate {
 	ThreadPrivate(thread_func_t f, void* a): func(f), arg(a), m_thread(&Run, this) {}
@@ -366,7 +334,6 @@ 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
@@ -407,7 +374,6 @@ void CondVar::Wait(Mutex* mutex) {
 	}
 	cpp_lock.release();
 #endif
-	UnregisterCondWaiter(m_cond_var.get());
 }
 
 void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
@@ -416,7 +382,6 @@ 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
@@ -431,7 +396,6 @@ bool CondVar::WaitFor(Mutex* mutex, uint32_t micros) {
 	      std::cv_status::no_timeout);
 	cpp_lock.release();
 #endif
-	UnregisterCondWaiter(m_cond_var.get());
 	return ok;
 }
 
@@ -446,16 +410,13 @@ void CondVar::Signal() {
 }
 
 void CondVar::SignalAll() {
-	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);
-		}
-	}
+#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
 }
 
 int Thread::GetThreadIdUnique() {

src/common/threads.h

@@ -77,7 +77,6 @@ public:
 	void Signal();
 	void SignalAll();
 
-	static void SignalThread(int thread_id);
 	static void SetWaitPollCallback(wait_poll_func_t callback);
 
 	KYTY_CLASS_NO_COPY(CondVar);

src/common/virtualMemory.cpp

@@ -50,8 +50,8 @@ 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 Protect(uint64_t address, uint64_t size, Mode mode) {
+	return SysVirtualProtect(address, size, mode);
 }
 
 bool FlushInstructionCache(uint64_t address, uint64_t size) {

src/common/virtualMemory.h

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

src/emulator.cpp

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

src/graphics/guest_gpu/command_processor/commandProcessor.h

@@ -32,11 +32,12 @@ private:
 
 	struct BufferCursor {
 		std::span<const uint32_t> commands;
-		uint32_t                  offset_dw           = 0;
-		uint32_t                  deferred_advance_dw = 0;
+		uint32_t                  offset_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;
 };
@@ -129,7 +130,7 @@ 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);
+	void             ProcessIndirectBuffer(std::span<const uint32_t> commands, bool chain);
 
 	void SetFlip(const FlipInfo& flip) { m_flip = flip; }
 
@@ -143,12 +144,11 @@ 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, size_t stop_depth);
+	void ProcessPm4(Pm4Execution& execution);
 	void SuspendPm4();
 	CommandScheduler&   GetScheduler() const { return m_renderer.GetCommandScheduler(); }
 	CommandBuffer&      CurrentBuffer() { return GetScheduler().Current(); }
 	void                CheckBuffer() const { GetScheduler().CheckActive(); }
-	GpuResourceManager& GetGpuResources() const { return m_renderer.GetGpuResources(); }
 
 	RenderContext&   m_renderer;
 	HW::Context      m_ctx;

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -315,8 +315,17 @@ static void HwCtxSetDepthBoundsRegister(CommandProcessor& cp, uint32_t cmd_offse
 	}
 }
 
-static void HwCtxIgnoreDepthMetadataRegister([[maybe_unused]] uint32_t cmd_offset,
-                                             [[maybe_unused]] uint32_t value) {}
+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 HwCtxIgnoreFovWindow([[maybe_unused]] uint32_t value) {}
 
@@ -389,7 +398,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetDepthMetadataRegisters) {
 	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
 
 	for (uint32_t i = 0; i < num_values; i++) {
-		HwCtxIgnoreDepthMetadataRegister(cmd_offset + i, buffer[i]);
+		HwCtxSetDepthMetadataRegister(cp, cmd_offset + i, buffer[i]);
 	}
 
 	return num_values;
@@ -1486,10 +1495,10 @@ KYTY_CP_OP_PARSER(CpOpBranch) {
 	     reinterpret_cast<uint64_t>(else_buffer), else_num_dw);
 
 	if (take_then) {
-		cp.ProcessIndirectBuffer({then_buffer, then_num_dw});
+		cp.ProcessIndirectBuffer({then_buffer, then_num_dw}, true);
 	} else if (mode == 2 && else_num_dw != 0) {
 		EXIT_NOT_IMPLEMENTED(else_buffer == nullptr);
-		cp.ProcessIndirectBuffer({else_buffer, else_num_dw});
+		cp.ProcessIndirectBuffer({else_buffer, else_num_dw}, true);
 	}
 
 	return payload_dw;
@@ -1950,7 +1959,7 @@ KYTY_CP_OP_PARSER(CpOpIndirectBuffer) {
 
 	GraphicsDbgDumpDcb("ci", indirect_num_dw, indirect_buffer);
 
-	cp.ProcessIndirectBuffer({indirect_buffer, indirect_num_dw});
+	cp.ProcessIndirectBuffer({indirect_buffer, indirect_num_dw}, (control & (1u << 20u)) != 0);
 
 	return 3;
 }
@@ -3003,28 +3012,28 @@ void GraphicsInitJmpTablesCxIndirect() {
 		HwCtxIgnoreCbDccControl(value);
 	};
 	g_hw_ctx_indirect_func[Pm4::DB_COUNT_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
+		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
 	};
 	for (auto cmd_offset = Pm4::DB_SRESULTS_COMPARE_STATE0;
 	     cmd_offset <= Pm4::DB_SRESULTS_COMPARE_STATE1; cmd_offset++) {
 		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-			HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
+			HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
 		};
 	}
 	g_hw_ctx_indirect_func[Pm4::DB_RENDER_OVERRIDE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
+		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::DB_RENDER_OVERRIDE2] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
+		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::DB_DFSM_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
+		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::DB_RMI_L2_CACHE_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
+		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::CB_RMI_GL2_CACHE_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnoreDepthMetadataRegister(cmd_offset, value);
+		HwCtxSetDepthMetadataRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::TA_BC_BASE_ADDR] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 		HwCtxIgnoreBorderColorTableAddr(cmd_offset, value);
@@ -3626,6 +3635,11 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_RSRC4_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
+	// Toolkit state restoration emits the compiler's GS-front allocation
+	// metadata here. Native GS resource registers already carry that allocation.
+	g_hw_sh_indirect_func[0x0ca] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+		HwShIgnoreShaderRegister(cmd_offset, value);
+	};
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
@@ -3645,12 +3659,13 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, 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);
-	};
+	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_PGM_LO_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		auto base = cp.GetShCtx().GetVs().hs_regs.data_addr;
 		base &= 0xFFFFFF00000000FFull;
@@ -3809,6 +3824,25 @@ void GraphicsInitJmpTablesUcIndirect() {
 	for (auto& func: g_hw_uc_indirect_func) {
 		func = nullptr;
 	}
+	for (uint32_t i = 0; i < 4; ++i) {
+		g_hw_uc_indirect_func[Pm4::FSR_CONTROL_POINTS_LEFT_X + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrControlPoint(0, cmd_offset - Pm4::FSR_CONTROL_POINTS_LEFT_X, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_CONTROL_POINTS_LEFT_Y + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrControlPoint(1, cmd_offset - Pm4::FSR_CONTROL_POINTS_LEFT_Y, value);
+		};
+	}
+	for (uint32_t i = 0; i < 2; ++i) {
+		g_hw_uc_indirect_func[Pm4::FSR_ALPHA_LEFT_X + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrAlpha(0, cmd_offset - Pm4::FSR_ALPHA_LEFT_X, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_ALPHA_LEFT_Y + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrAlpha(1, cmd_offset - Pm4::FSR_ALPHA_LEFT_Y, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_WINDOW_LEFT + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrWindow(cmd_offset - Pm4::FSR_WINDOW_LEFT, value);
+		};
+	}
 
 	g_hw_uc_indirect_func[Pm4::GE_CNTL] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		HW::GeControl r;

src/graphics/guest_gpu/graphicsRun.cpp

@@ -77,7 +77,7 @@ private:
 
 static bool GraphicsRunDebugDumpEnabled() {
 	return Config::GraphicsDebugDumpEnabled() &&
-	       Config::GetPrintfDirection() != Config::OutputDirection::Silent;
+	       Config::GetPrintfDirection() != Config::LogDirection::Silent;
 }
 
 GuestGpu::GuestGpu(RenderContext& renderer): m_renderer(renderer) {
@@ -427,7 +427,7 @@ void CommandProcessor::DmaData(uint8_t engine, uint8_t dst_sel, uint8_t dst_cach
 	if (!decode_gds(dst_sel, dst_gds)) {
 		EXIT("unsupported dmaData destination selector 0x%02" PRIx8 "\n", dst_sel);
 	}
-	auto& buffer_cache = GetGpuResources().GetBufferCache();
+	auto& buffer_cache = m_renderer.GetBufferCache();
 	if (src_sel == 2) {
 		buffer_cache.FillBuffer(
 		    dst_address_or_offset, num_bytes,
@@ -603,11 +603,11 @@ bool GuestGpu::Process(Submission& submission) {
 			}
 			if (progressed) {
 				if (complete) {
-					m_renderer.GetGpuResources().RunGarbageCollector();
+					m_renderer.RunGarbageCollector();
 				}
 				cp.BufferFlush();
 			} else if (complete) {
-				m_renderer.GetGpuResources().RunGarbageCollector();
+				m_renderer.RunGarbageCollector();
 			}
 			break;
 		}
@@ -629,16 +629,16 @@ bool GuestGpu::Process(Submission& submission) {
 			           Pm4ProcessResult::Complete;
 			if (submission.command_execution.MadeProgress()) {
 				if (complete) {
-					m_renderer.GetGpuResources().RunGarbageCollector();
+					m_renderer.RunGarbageCollector();
 				}
 				cp.BufferFlush();
 			} else if (complete) {
-				m_renderer.GetGpuResources().RunGarbageCollector();
+				m_renderer.RunGarbageCollector();
 			}
 			break;
 		}
 		case SubmissionType::FlipPreparation:
-			m_renderer.GetGpuResources().RunGarbageCollector();
+			m_renderer.RunGarbageCollector();
 			cp.PrepareCpuFlip(submission.flip_request_id);
 			break;
 	}
@@ -672,20 +672,16 @@ Pm4ProcessResult CommandProcessor::Process(Pm4Execution&             execution,
 		Pm4Execution*     previous_execution;
 	} execution_scope(*this, execution);
 
-	ProcessPm4(execution, 0);
+	ProcessPm4(execution);
 	return execution.m_buffer_stack.empty() ? Pm4ProcessResult::Complete
 	                                        : Pm4ProcessResult::Blocked;
 }
 
-void CommandProcessor::ProcessIndirectBuffer(std::span<const uint32_t> commands) {
+void CommandProcessor::ProcessIndirectBuffer(std::span<const uint32_t> commands, bool chain) {
 	EXIT_IF(g_current_execution == nullptr);
-	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);
+	EXIT_IF(!g_current_execution->m_next_buffer.empty());
+	g_current_execution->m_next_buffer = commands;
+	g_current_execution->m_chain       = chain;
 }
 
 void CommandProcessor::SuspendPm4() {
@@ -693,21 +689,13 @@ void CommandProcessor::SuspendPm4() {
 	g_current_execution->m_suspended = true;
 }
 
-void CommandProcessor::ProcessPm4(Pm4Execution& execution, size_t stop_depth) {
-	while (execution.m_buffer_stack.size() > stop_depth) {
+void CommandProcessor::ProcessPm4(Pm4Execution& execution) {
+	while (!execution.m_buffer_stack.empty()) {
 		if (g_gpu_state != nullptr) {
 			g_gpu_state->ProcessCommands();
 		}
-		const auto buffer_index = execution.m_buffer_stack.size() - 1;
-		auto&      cursor       = execution.m_buffer_stack[buffer_index];
+		auto& cursor = execution.m_buffer_stack.back();
 		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;
@@ -783,14 +771,19 @@ void CommandProcessor::ProcessPm4(Pm4Execution& execution, size_t stop_depth) {
 		    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;
 		}
-		EXIT_IF(execution.m_buffer_stack.size() != buffer_index + 1);
-		execution.m_buffer_stack[buffer_index].offset_dw += packet_dw;
+		cursor.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 = {};
+		}
 	}
 }
 
@@ -825,36 +818,55 @@ 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;
-		} break;
-		case 0x03: {
+			return;
+		case 0x01: {
 			EXIT_NOT_IMPLEMENTED(address == nullptr);
-
-			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);
+			// 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;
 			}
 		} break;
+		case 0x03:
+			if (wait_op != 0) {
+				BufferFlushAndWait();
+			}
+			EXIT_NOT_IMPLEMENTED(address == nullptr);
+			value = *reinterpret_cast<const volatile uint64_t*>(address);
+			break;
 		default: EXIT("unknown predication op: 0x%08" PRIx32 "\n", op);
 	}
+	switch (condition) {
+		case 0x00: m_predicate_skip = (value != 0); break;
+		case 0x01: m_predicate_skip = (value == 0); break;
+		default: EXIT("unknown predication condition: 0x%08" PRIx32 "\n", condition);
+	}
+	if (op == 0x03) {
+		static std::atomic<uint32_t> log_count {0};
+		if (log_count.fetch_add(1) < 128) {
+			LOGF("\t bool predication: addr=0x%016" PRIx64 ", value=0x%016" PRIx64
+			     ", condition=%" PRIu32 ", skip=%u, wait_op=%" PRIu32 "\n",
+			     reinterpret_cast<uint64_t>(address), value, condition,
+			     m_predicate_skip ? 1u : 0u, wait_op);
+		}
+	}
 }
 
 void CommandProcessor::DrawIndex(DrawIndexArgs args) {
@@ -1090,17 +1102,6 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 		// local_x        = std::max(cs.num_thread_x, 1u);
 		// local_y        = std::max(cs.num_thread_y, 1u);
 		// local_z        = std::max(cs.num_thread_z, 1u);
-		if (cs.wave_size == 64u) {
-			static std::atomic_bool logged_wave64_shader {false};
-			if (!logged_wave64_shader.exchange(true, std::memory_order_relaxed)) {
-				LOGF("warning: executing wave64 compute shader cs=0x%016" PRIx64 "\n",
-				     cs.data_addr);
-				std::printf("warning: executing wave64 compute shader cs=0x%016" PRIx64 "\n",
-				            cs.data_addr);
-				std::fflush(stdout);
-			}
-		}
-
 		m_renderer.GetRenderExecutor().DispatchDirect(m_submit_id, CurrentBuffer(), thread_group_x,
 		                                              thread_group_y, thread_group_z, mode);
 	}
@@ -1292,6 +1293,7 @@ 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;
@@ -1299,7 +1301,6 @@ 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);

src/graphics/guest_gpu/hardwareContext.h

@@ -202,6 +202,13 @@ 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;
@@ -491,7 +498,8 @@ struct LsStageRegisters {
 };
 
 struct HsStageRegisters {
-	uint64_t          data_addr = 0;
+	uint64_t          data_addr      = 0;
+	uint64_t          user_data_addr = 0;
 	HsShaderResource1 rsrc1;
 	HsShaderResource2 rsrc2;
 };
@@ -809,6 +817,12 @@ 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; }
@@ -904,15 +918,16 @@ private:
 
 	uint32_t m_shader_stages = 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;
+	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;
 
 	ModeControl m_mode_control;
 	PolyOffset  m_poly_offset;
@@ -921,6 +936,12 @@ 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;
@@ -950,8 +971,17 @@ 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;
@@ -974,14 +1004,14 @@ public:
 	void SetEsShaderBase(uint64_t addr) { m_vs.es_regs.data_addr = addr; }
 	void SetLsShaderBase(uint64_t addr) { m_vs.ls_regs.data_addr = addr; }
 	void SetHsShaderBase(uint64_t addr) { m_vs.hs_regs.data_addr = addr; }
+	void SetHsUserDataAddress(uint32_t word, uint32_t value) {
+		SetUserDataAddressWord(m_vs.hs_regs.user_data_addr, word, value);
+	}
 	void SetHsShaderResource1(const HsShaderResource1& rsrc1) { m_vs.hs_regs.rsrc1 = rsrc1; }
 	void SetHsShaderResource2(const HsShaderResource2& rsrc2) { m_vs.hs_regs.rsrc2 = rsrc2; }
 	void SetGsShaderBase(uint64_t addr) { m_vs.gs_regs.data_addr = addr; }
 	void SetGsUserDataAddress(uint32_t word, uint32_t value) {
-		const auto shift   = word * 32u;
-		auto&      address = m_vs.gs_regs.user_data_addr;
-		address = (address & ~(uint64_t {0xffffffffu} << shift)) |
-		          (static_cast<uint64_t>(value) << shift);
+		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; }
@@ -1023,6 +1053,12 @@ public:
 	[[nodiscard]] const ComputeShaderInfo& GetCs() const { return m_cs; }
 
 private:
+	static void SetUserDataAddressWord(uint64_t& address, uint32_t word, uint32_t value) {
+		const auto shift = word * 32u;
+		address = (address & ~(uint64_t {0xffffffffu} << shift)) |
+		          (static_cast<uint64_t>(value) << shift);
+	}
+
 	VertexShaderInfo  m_vs;
 	PixelShaderInfo   m_ps;
 	ComputeShaderInfo m_cs;

src/graphics/guest_gpu/pm4.h

@@ -987,6 +987,12 @@ constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR2_MASK  = 0x1;
 constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR3_SHIFT = 2;
 constexpr uint32_t GE_USER_VGPR_EN_EN_USER_VGPR3_MASK  = 0x1;
 
+constexpr uint32_t FSR_WINDOW_LEFT           = 0x2BC;
+constexpr uint32_t FSR_CONTROL_POINTS_LEFT_X = 0x2C4;
+constexpr uint32_t FSR_CONTROL_POINTS_LEFT_Y = 0x2CC;
+constexpr uint32_t FSR_ALPHA_LEFT_X          = 0x2E4;
+constexpr uint32_t FSR_ALPHA_LEFT_Y          = 0x2E8;
+
 constexpr uint32_t TA_CS_BC_BASE_ADDR       = 0x380;
 constexpr uint32_t TA_CS_BC_BASE_ADDR_HI    = 0x381;
 constexpr uint32_t TEXTURE_GRADIENT_FACTORS = 0x383;

src/graphics/guest_gpu/tile.cpp

@@ -1,5 +1,6 @@
 #include "graphics/guest_gpu/tile.h"
 
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
@@ -15,10 +16,6 @@
 
 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);
 }
@@ -33,7 +30,7 @@ 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 =
-	    AlignUp(std::max(level_width, 1u), CalcLinearBlockWidth(bytes_per_element));
+	    Common::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);
@@ -61,7 +58,7 @@ static uint32_t SetLinearMipChainLayout(uint32_t levels, const uint32_t* mip_pit
 		offset += mip_size[level];
 	}
 
-	return AlignUp(offset, 256u);
+	return Common::AlignUp(offset, 256u);
 }
 
 struct Log2BlockDimensions {
@@ -366,8 +363,8 @@ 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  = AlignUp(std::max(ShiftCeil(width0, level), 1u), block.block_width);
-		mip.padded_height = AlignUp(std::max(ShiftCeil(height0, level), 1u), block.block_height);
+		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.size = static_cast<uint64_t>(block.block_depth) * mip.padded_width * mip.padded_height *
 		           block.bytes_per_element;
 		result.block_slice_size += mip.size;
@@ -1121,8 +1118,8 @@ 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>(AlignUp(width, 1024u) / 1024u) *
-	                      (AlignUp(height, 512u) / 512u) * 32768u;
+	const uint64_t size = static_cast<uint64_t>(Common::AlignUp(width, 1024u) / 1024u) *
+	                      (Common::AlignUp(height, 512u) / 512u) * 32768u;
 	if (size == 0 || size > UINT32_MAX) {
 		return false;
 	}
@@ -1154,13 +1151,13 @@ 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>(AlignUp(width, depth_block_width)) *
-		                       AlignUp(height, depth_block_height) * depth_bytes * fragments
+		    valid_blocks ? static_cast<uint64_t>(Common::AlignUp(width, depth_block_width)) *
+		                       Common::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>(AlignUp(width, stencil_block_width)) *
-		              AlignUp(height, stencil_block_height) * fragments
+		        ? static_cast<uint64_t>(Common::AlignUp(width, stencil_block_width)) *
+		              Common::AlignUp(height, stencil_block_height) * fragments
 		        : 0;
 		TileSizeAlign calculated_htile {};
 		const bool    htile_valid = !htile || TileGetHtileSize(width, height, calculated_htile);
@@ -1187,8 +1184,7 @@ uint32_t TileGetRenderTargetPitch(uint32_t width, uint32_t bytes_per_element,
 	                                                         &block_width, &block_height)) {
 		return 0;
 	}
-	const uint64_t pitch = (static_cast<uint64_t>(width) + block_width - 1u) &
-	                       ~static_cast<uint64_t>(block_width - 1u);
+	const uint64_t pitch = Common::AlignUp(static_cast<uint64_t>(width), block_width);
 	return pitch <= UINT32_MAX ? static_cast<uint32_t>(pitch) : 0;
 }
 
@@ -1209,8 +1205,7 @@ 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 = (static_cast<uint64_t>(height) + block_height - 1u) &
-	                               ~static_cast<uint64_t>(block_height - 1u);
+	const uint64_t padded_height = Common::AlignUp(static_cast<uint64_t>(height), block_height);
 	const uint64_t size = static_cast<uint64_t>(pitch) * padded_height * bytes_per_element *
 	                      (1u << num_fragments_log2);
 	if (size == 0 || size > UINT32_MAX) {
@@ -1372,7 +1367,7 @@ uint32_t TileGetTexturePitch(Prospero::BufferFormat format, uint32_t width,
 			TileTextureElementLayout element {};
 			if (TileGetTextureElementLayout(format, element) && element.texel_width == 1 &&
 			    element.texel_height == 1) {
-				pitch = AlignUp(pitch, CalcLinearBlockWidth(element.bytes));
+				pitch = Common::AlignUp(pitch, CalcLinearBlockWidth(element.bytes));
 			}
 			break;
 		}
@@ -1383,7 +1378,7 @@ uint32_t TileGetTexturePitch(Prospero::BufferFormat format, uint32_t width,
 		case Prospero::TileMode::kRenderTarget: {
 			TileTextureBlockLayout layout {};
 			if (TileGetTextureBlockLayout(format, tile, false, layout)) {
-				pitch = AlignUp(pitch, layout.block.block_width * layout.texel_width);
+				pitch = Common::AlignUp(pitch, layout.block.block_width * layout.texel_width);
 			}
 			break;
 		}

src/graphics/host_gpu/graphicContext.h

@@ -14,9 +14,7 @@
 
 namespace Libs::Graphics {
 
-struct VulkanBuffer;
 struct VulkanImage;
-struct VulkanMemory;
 
 inline constexpr uint32_t VULKAN_TARGET_API_VERSION = VK_API_VERSION_1_3;
 
@@ -29,10 +27,11 @@ struct GraphicContext {
 	vk::Device                         device                                = nullptr;
 	VmaAllocator                       allocator                             = nullptr;
 	bool                               memory_budget_ext_enabled             = false;
-	bool                               rt_extensions_enabled                 = false;
 	bool                               compute_subgroup_size_control_enabled = false;
 	bool                               sample_rate_shading_enabled           = false;
 	bool                               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;
@@ -99,10 +98,6 @@ struct GraphicContext {
 	[[nodiscard]] uint64_t GetTotalMemoryBudget() const;
 	[[nodiscard]] bool     CreateImage(const vk::ImageCreateInfo& info, VulkanImage& image);
 	void                   DeleteImage(VulkanImage& image);
-	void                   AppendHardwareRayTracingDeviceExtensions(
-	    const std::vector<vk::ExtensionProperties>& available_extensions,
-	    std::vector<const char*>&                   device_extensions);
-	void LoadHardwareRayTracingFunctions() const;
 
 	uint32_t screen_width  = 0;
 	uint32_t screen_height = 0;
@@ -117,14 +112,6 @@ private:
 	    m_image_format_properties;
 };
 
-struct VulkanMemory {
-	vk::MemoryRequirements  requirements       = {};
-	vk::MemoryPropertyFlags property           = {};
-	vk::MemoryPropertyFlags preferred_property = {};
-	VmaAllocation           allocation         = nullptr;
-	uint32_t                type               = 0;
-};
-
 struct VulkanImageState {
 	vk::PipelineStageFlags2 pl_stage    = vk::PipelineStageFlagBits2::eAllCommands;
 	vk::AccessFlags2        access_mask = vk::AccessFlagBits2::eNone;
@@ -138,7 +125,6 @@ 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;
@@ -147,15 +133,10 @@ struct VulkanImage {
 	vk::Image                     image       = nullptr;
 	VulkanImageState              state;
 	std::vector<VulkanImageState> subresource_states;
-	Graphics::VulkanMemory        memory;
+	VmaAllocation                allocation = nullptr;
 };
 
-struct VulkanBuffer {
-	vk::Buffer           buffer = nullptr;
-	VulkanMemory         memory;
-	vk::BufferUsageFlags usage       = {};
-	uint64_t             buffer_size = 0;
-};
+
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/memoryTracker.cpp

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

src/graphics/host_gpu/memoryTracker.h

@@ -62,43 +62,18 @@ public:
 	void ValidateGpuDirtyOwnership(const RangeSet&, uint64_t, uint64_t, const char*) {}
 #endif
 
-	template <bool clear, typename Preflight, typename Func>
-	void ForEachDownloadRange(uint64_t vaddr, uint64_t size, Preflight&& preflight, Func&& func) {
-		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();
-		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));
+		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);
+			}
+		});
 	}
 
 	template <typename RangeFunc, typename UploadFunc>

src/graphics/host_gpu/pageManager.cpp

@@ -1,6 +1,7 @@
 #include "graphics/host_gpu/pageManager.h"
 
-#include "common/platform/sysDbg.h"
+#include "common/alignment.h"
+#include "common/virtualMemory.h"
 #include "graphics/host_gpu/regionDefinitions.h"
 #include "kernel/memory.h"
 
@@ -21,10 +22,7 @@
 #include <windows.h>
 #undef min
 #undef max
-#elif defined(__APPLE__)
-#include <unistd.h>
 #else
-#include <execinfo.h>
 #include <unistd.h>
 #endif
 
@@ -36,38 +34,12 @@ 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));
@@ -86,15 +58,6 @@ constexpr uint32_t READ_WRITE_PROTECTION = PAGE_READWRITE;
 	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) {
@@ -115,15 +78,6 @@ 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 {
@@ -131,14 +85,14 @@ struct PageManager::Impl {
 		uint8_t write_watchers  : 7 = 0;
 		uint8_t access_watchers : 1 = 0;
 
-		[[nodiscard]] uint32_t Perms() const noexcept {
+		[[nodiscard]] Common::VirtualMemory::Mode Perms() const noexcept {
 			if (access_watchers != 0) {
-				return NO_ACCESS_PROTECTION;
+				return Common::VirtualMemory::Mode::NoAccess;
 			}
 			if (write_watchers != 0) {
-				return READ_ONLY_PROTECTION;
+				return Common::VirtualMemory::Mode::Read;
 			}
-			return READ_WRITE_PROTECTION;
+			return Common::VirtualMemory::Mode::ReadWrite;
 		}
 
 		template <int delta, bool is_read>
@@ -202,9 +156,6 @@ 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() {
@@ -239,11 +190,10 @@ struct PageManager::Impl {
 		return ptr;
 	}
 
-	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);
+	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));
 		}
 	}
 
@@ -300,11 +250,12 @@ struct PageManager::Impl {
 
 	template <bool track, bool is_read>
 	void UpdatePageWatchers(uint64_t vaddr, uint64_t size) {
-		const auto begin = PageStart(vaddr);
-		const auto end   = PageEnd(vaddr, size);
+		ValidateRange(vaddr, size);
+		const auto begin = Common::AlignDown(vaddr, PAGE_SIZE);
+		const auto end   = Common::AlignUp(vaddr + size, PAGE_SIZE);
 		for (auto chunk_begin = begin; chunk_begin < end;) {
-			const auto chunk_end   = std::min(end, (chunk_begin / REGION_SIZE + 1) * REGION_SIZE);
-			const auto region_base = chunk_begin / REGION_SIZE * REGION_SIZE;
+			const auto chunk_end = std::min(end, Common::AlignUp(chunk_begin + 1, REGION_SIZE));
+			const auto region_base = Common::AlignDown(chunk_begin, REGION_SIZE);
 			auto*      region = track ? GetOrCreateRegion(chunk_begin) : FindRegion(chunk_begin);
 			if (region == nullptr) {
 				Fatal("untracking unknown page 0x%016" PRIx64, chunk_begin);

src/graphics/host_gpu/rangeSet.h

@@ -6,17 +6,11 @@
 #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);
@@ -62,12 +56,6 @@ 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);
@@ -88,7 +76,7 @@ public:
 	}
 
 	template <typename Func>
-	void ForEachIntersection(uint64_t address, uint64_t size, Func&& func) const {
+	void ForEachInRange(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()) {
@@ -98,7 +86,7 @@ public:
 			const auto begin = std::max(address, it->first);
 			const auto last  = std::min(end, it->second);
 			if (begin < last) {
-				func(Range {begin, last - begin});
+				func(begin, last);
 			}
 		}
 	}

src/graphics/host_gpu/regionManager.h

@@ -111,56 +111,43 @@ public:
 			bits.UnsetRange(start, end);
 		}
 		if constexpr (source == DirtySource::Cpu) {
-			UpdateCpuProtection<!enable>();
+			UpdateProtection<!enable, false>();
 		} else {
-			UpdateGpuProtection<enable>();
+			UpdateProtection<enable, true>();
 		}
 	}
 
 	template <DirtySource source, bool clear, typename Func>
 	void ForEachModifiedRange(uint64_t vaddr, uint64_t size, Func&& func) {
 		const auto [start, end] = GetPageRange(vaddr, size);
-		RegionBits mask(GetBits<source>(), start, end);
+		auto&      bits         = GetBits<source>();
+		RegionBits mask(bits, start, end);
 		if constexpr (clear) {
-			GetBits<source>().UnsetRange(start, end);
+			bits.UnsetRange(start, end);
+			if constexpr (source == DirtySource::Cpu) {
+				UpdateProtection<true, false>();
+			} else {
+				UpdateProtection<false, true>();
+			}
 		}
-		if constexpr (source == DirtySource::Cpu && clear) {
-			UpdateCpuProtection<true>();
-			ForEachRange(mask, std::forward<Func>(func));
-			return;
+		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>
-	void UpdateCpuProtection() {
-		auto mask  = m_cpu_dirty ^ m_writable;
-		m_writable = m_cpu_dirty;
+	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;
 		if (mask.None()) {
 			return;
 		}
-		m_page_manager.UpdatePageWatchersForRegion<track>(m_cpu_addr, mask);
-	}
-
-	template <bool track>
-	void UpdateGpuProtection() {
-		auto readable = ~m_gpu_dirty;
-		auto mask     = readable ^ m_readable;
-		m_readable    = readable;
-		if (mask.None()) {
-			return;
-		}
-		if constexpr (track) {
-			m_page_manager.UpdatePageWatchersForRegion<true, true>(m_cpu_addr, mask);
-		} else {
-			m_page_manager.UpdatePageWatchersForRegion<false, true>(m_cpu_addr, mask);
-		}
+		previous = protection;
+		m_page_manager.UpdatePageWatchersForRegion<track, is_read>(m_cpu_addr, mask);
 	}
 
 	template <DirtySource source>
@@ -191,13 +178,6 @@ private:
 		        static_cast<size_t>((offset + size + TRACKER_PAGE_SIZE - 1) / TRACKER_PAGE_SIZE)};
 	}
 
-	template <typename Func>
-	void ForEachRange(const RegionBits& bits, Func&& func) const {
-		for (const auto [start, end]: bits) {
-			func(m_cpu_addr + start * TRACKER_PAGE_SIZE, (end - start) * TRACKER_PAGE_SIZE);
-		}
-	}
-
 	PageManager& m_page_manager;
 	uint64_t     m_cpu_addr = 0;
 	RegionBits   m_cpu_dirty;

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

@@ -1,5 +1,6 @@
 #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"
@@ -42,13 +43,6 @@ void BufferCache::WriteDataBuffer(Buffer& buffer, uint64_t address, const void*
 	}
 }
 
-struct BufferCache::DownloadCopy {
-	Buffer*  buffer        = nullptr;
-	uint64_t source_offset = 0;
-	uint64_t address       = 0;
-	uint64_t size          = 0;
-};
-
 void BufferCache::Register(BufferId id) {
 	ChangeRegister<true>(id);
 }
@@ -103,8 +97,7 @@ void BufferCache::TouchBuffer(const Buffer& buffer) {
 }
 
 void BufferCache::DeleteBuffer(BufferId id) {
-	auto* buffer = m_slot_buffers.try_get(id);
-	if (buffer == nullptr || buffer->is_deleted) {
+	if (IsBufferInvalid(id)) {
 		return;
 	}
 	Unregister(id);
@@ -115,79 +108,70 @@ void BufferCache::DeleteBuffer(BufferId id) {
 	}
 }
 
-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");
+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;
 	}
-	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 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};
-}
 
-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;
-	};
-	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();
-			}
-		}
+	const auto [mapped, offset] = m_download_buffer.Map(total_size, 64);
+	if (mapped == nullptr) {
+		EXIT("BufferCache: download exceeds 32 MiB staging buffer capacity\n");
 	}
-	if (!batch.empty()) {
-		flush();
-	}
-	for (const auto& copy: copies) {
-		m_gpu_modified_ranges.Subtract(copy.address, copy.size);
+	m_download_buffer.Commit();
+	for (auto& copy: copies) {
+		copy.dstOffset += offset;
 	}
+
+	auto& command = m_scheduler.Current();
+	command.EndRendering();
+	const auto              native = command.Handle();
+	vk::BufferMemoryBarrier before {};
+	before.srcAccessMask       = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
+	before.dstAccessMask       = vk::AccessFlagBits::eTransferRead;
+	before.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+	before.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+	before.buffer              = buffer.Handle();
+	before.offset              = 0;
+	before.size                = buffer.Size();
+	native.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands,
+	                       vk::PipelineStageFlagBits::eTransfer, {}, 0, nullptr, 1, &before, 0,
+	                       nullptr);
+	native.copyBuffer(buffer.Handle(), m_download_buffer.Handle(),
+	                  static_cast<uint32_t>(copies.size()), copies.data());
+
+	auto after          = before;
+	after.srcAccessMask = vk::AccessFlagBits::eTransferWrite;
+	after.dstAccessMask = vk::AccessFlagBits::eHostRead;
+	after.buffer        = m_download_buffer.Handle();
+	after.offset        = offset;
+	after.size          = total_size;
+	native.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer,
+	                       vk::PipelineStageFlagBits::eAllCommands |
+	                           vk::PipelineStageFlagBits::eHost,
+	                       {}, 0, nullptr, 1, &after, 0, nullptr);
+	m_scheduler.DeferPriorityOperation([this, mapped, offset, total_size, buffer_address,
+	                                    copies = std::move(copies)] {
+		m_download_buffer.Invalidate(offset, total_size);
+		for (const auto& copy: copies) {
+			Libs::LibKernel::Memory::WriteBacking(buffer_address + copy.srcOffset,
+			                                      mapped + (copy.dstOffset - offset), copy.size);
+		}
+	});
+	return true;
 }
 
 BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
@@ -252,44 +236,29 @@ 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] { ReadMemoryOnGpu(vaddr, size, is_write); });
-}
+	m_scheduler.Context().GetGpu().SendCommandSync([this, vaddr, size, is_write] {
+		if (is_write && !IsRegionRegistered(vaddr, size)) {
+			return;
+		}
+		auto& buffer = m_slot_buffers[FindBuffer(vaddr, size)];
 
-void BufferCache::ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write) {
-	if (is_write && !IsRegionRegistered(vaddr, size)) {
-		return;
-	}
-	auto& buffer = m_slot_buffers[FindBuffer(vaddr, size)];
+		// Widen nearby CPU reads so they share one GPU drain.
+		constexpr uint64_t WindowSize   = 512 * 1024;
+		const auto         buffer_begin = buffer.CpuAddress();
+		const auto         buffer_end   = buffer_begin + buffer.Size();
+		const auto window_begin = std::max(Common::AlignDown(vaddr, WindowSize), buffer_begin);
+		const auto window_end = std::min(std::max(window_begin + WindowSize, vaddr + size), buffer_end);
 
-	// Widen nearby CPU reads so they share one GPU drain.
-	constexpr uint64_t WindowSize   = 512 * 1024;
-	const auto         buffer_begin = buffer.CpuAddress();
-	const auto         buffer_end   = buffer_begin + buffer.Size();
-	const auto         window_begin = std::max(vaddr & ~(WindowSize - 1), buffer_begin);
-	const auto window_end = std::min(std::max(window_begin + WindowSize, vaddr + size), buffer_end);
-
-	std::vector<DownloadCopy> copies;
-	m_memory_tracker.ForEachDownloadRange<false>(
-	    window_begin, window_end - window_begin,
-	    [&](uint64_t address, uint64_t bytes) noexcept {
-		    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
-		                                           "memory invalidation");
-	    },
-	    [&](uint64_t address, uint64_t bytes) noexcept {
-		    for (const auto range: m_gpu_modified_ranges.Intersections(address, bytes)) {
-			    copies.push_back(
-			        {&buffer, buffer.Offset(range.address), range.address, range.size});
-		    }
-	    });
-	if (!copies.empty()) {
-		DownloadBufferMemory(copies);
-		// The enumeration covered whole dirty pages and every exact interval on them.
-		m_memory_tracker.UnmarkRegionAsGpuModified(window_begin, window_end - window_begin);
-	}
-	if (is_write) {
-		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
-	}
+		if (DownloadBufferMemory(buffer, window_begin, window_end - window_begin)) {
+			const auto tick = m_scheduler.CurrentTick();
+			m_scheduler.Wait(tick);
+			m_scheduler.WaitPriorityOperations(tick);
+			m_memory_tracker.UnmarkRegionAsGpuModified(window_begin, window_end - window_begin);
+		}
+		if (is_write) {
+			m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
+		}
+	});
 }
 
 BufferId BufferCache::FindBuffer(uint64_t vaddr, uint64_t size) {
@@ -340,10 +309,12 @@ BufferCache::OverlapResult BufferCache::ResolveOverlaps(uint64_t vaddr, uint64_t
 		end                       = std::max(end, buffer_end);
 		if (!has_stream_leap && (stream_score += buffer.StreamScore()) > StreamLeapThreshold) {
 			has_stream_leap = true;
-			if (expands_right) {
+			// 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, PageTable::kAddressSpaceSize - end);
 			}
-			if (expands_left) {
+			if (expands_right) {
 				const auto minimum = CACHING_PAGESIZE * 2;
 				if (begin > minimum) {
 					begin -= std::min(StreamLeapSize, begin - minimum);
@@ -369,8 +340,8 @@ void BufferCache::JoinOverlap(BufferId new_id, BufferId overlap_id, bool accumul
 
 BufferId BufferCache::CreateBuffer(uint64_t vaddr, uint64_t size) {
 	EXIT_IF(m_scheduler.Current().IsInvalid());
-	const auto end = (vaddr + size + CACHING_PAGESIZE - 1) & ~(CACHING_PAGESIZE - 1);
-	vaddr &= ~(CACHING_PAGESIZE - 1);
+	const auto end = Common::AlignUp(vaddr + size, CACHING_PAGESIZE);
+	vaddr = Common::AlignDown(vaddr, CACHING_PAGESIZE);
 	size               = end - vaddr;
 	const auto overlap = ResolveOverlaps(vaddr, size);
 
@@ -483,17 +454,16 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBuffer(uint64_t vaddr, uint64_t
 		}
 	}
 
-	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];
+	if (IsBufferInvalid(id) || !m_slot_buffers[id].IsInBounds(vaddr, size)) {
+		id = FindBuffer(vaddr, size);
 	}
-	TouchBuffer(*buffer);
-	(void)SynchronizeBuffer(*buffer, vaddr, size, is_written, is_texel_buffer);
+	auto& buffer = m_slot_buffers[id];
+	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) {
@@ -545,9 +515,7 @@ void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool
 	}
 
 	m_texture_cache.InvalidateMemoryFromGPU(vaddr, size);
-	const auto id          = FindBuffer(vaddr, size);
-	auto [dst, dst_offset] = ObtainBuffer(vaddr, size, true, true, id);
-	EXIT_IF(dst == nullptr);
+	auto [dst, dst_offset] = ObtainBuffer(vaddr, size, true, true);
 	dst->Fill(dst_offset, size, value);
 }
 
@@ -581,10 +549,6 @@ 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);
 }
 
@@ -628,8 +592,7 @@ void BufferCache::RunGarbageCollector() {
 	const size_t   limit      = aggressive ? 64 : 32;
 
 	std::vector<BufferId> dirty_buffers;
-	std::vector<DownloadCopy> copies;
-	size_t                    retire_count = 0;
+	size_t                retire_count = 0;
 	m_lru_cache.ForEachItemBelow(tick - age, [&](BufferId id) {
 		auto& buffer = m_slot_buffers[id];
 		EXIT_IF(buffer.is_deleted);
@@ -640,19 +603,7 @@ void BufferCache::RunGarbageCollector() {
 			return false;
 		}
 		if (dirty) {
-			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});
-				        });
-				});
+			EXIT_IF(!DownloadBufferMemory(buffer, buffer.CpuAddress(), buffer.Size()));
 			dirty_buffers.push_back(id);
 		} else {
 			m_memory_tracker.UntrackMemory(buffer.CpuAddress(), buffer.Size());
@@ -664,8 +615,10 @@ void BufferCache::RunGarbageCollector() {
 		return;
 	}
 
-	EXIT_IF(copies.empty());
-	DownloadBufferMemory(copies);
+	// 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);
 	for (const auto id: dirty_buffers) {
 		auto& buffer = m_slot_buffers[id];
 		m_memory_tracker.UnmarkRegionAsGpuModified(buffer.CpuAddress(), buffer.Size());

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

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

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

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

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

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

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

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

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

@@ -1,13 +1,12 @@
 #include "graphics/host_gpu/renderer/cache/streamBuffer.h"
 
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/profiler.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
-#include "graphics/host_gpu/vma.h"
 
 #include <cstring>
-#include <limits>
 #include <numeric>
 #include <vk_mem_alloc.h>
 
@@ -46,16 +45,11 @@ constexpr size_t WATCHES_RESERVE_CHUNK   = 0x1000;
 		result = value;
 		return true;
 	}
-	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) {
+	const auto aligned = Common::AlignUp(value, alignment);
+	if (aligned < value) {
 		return false;
 	}
-	result = value + increment;
+	result = aligned;
 	return true;
 }
 
@@ -64,7 +58,7 @@ 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_buffer(std::make_unique<VulkanBuffer>()) {
+      m_size(size) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(graphics.allocator == nullptr || size == 0);
 
@@ -87,53 +81,35 @@ 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_buffer->memory.allocation, &allocation_result));
+	    &native_buffer, &m_allocation, &allocation_result));
 	if (result != vk::Result::eSuccess) {
 		graphics.LogMemoryBudget();
 	}
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
-	m_buffer->buffer                 = native_buffer;
-	m_buffer->usage                  = flags;
-	m_buffer->buffer_size            = size;
-	m_buffer->memory.type            = allocation_result.memoryType;
-	graphics.device.getBufferMemoryRequirements(m_buffer->buffer, &m_buffer->memory.requirements);
-	if (static_cast<bool>(flags & vk::BufferUsageFlagBits::eShaderDeviceAddress)) {
+	m_buffer = native_buffer;
+	if (with_bda) {
 		vk::BufferDeviceAddressInfo address_info {};
-		address_info.buffer = m_buffer->buffer;
+		address_info.buffer = m_buffer;
 		m_device_address    = graphics.device.getBufferAddress(address_info);
 		EXIT_IF(m_device_address == 0);
 	}
 
 	VkMemoryPropertyFlags properties = 0;
-	vmaGetAllocationMemoryProperties(graphics.allocator, m_buffer->memory.allocation, &properties);
-	m_buffer->memory.property = vk::MemoryPropertyFlags(properties);
+	vmaGetAllocationMemoryProperties(graphics.allocator, m_allocation, &properties);
+	m_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->buffer != nullptr) {
-		VulkanUntrackAllocation(m_buffer->memory);
-		vmaDestroyBuffer(m_graphics->allocator, m_buffer->buffer, m_buffer->memory.allocation);
+	if (m_buffer != nullptr) {
+		vmaDestroyBuffer(m_graphics->allocator, m_buffer, m_allocation);
 	}
 }
 
-vk::Buffer Buffer::Handle() const noexcept {
-	return m_buffer->buffer;
-}
-
-uint64_t Buffer::Size() const noexcept {
-	return m_buffer->buffer_size;
-}
-
-bool Buffer::IsCoherent() const noexcept {
-	return bool(m_buffer->memory.property & vk::MemoryPropertyFlagBits::eHostCoherent);
-}
-
 vk::DeviceAddress Buffer::BufferDeviceAddress() const noexcept {
 	EXIT_IF(m_device_address == 0);
 	return m_device_address;
@@ -143,17 +119,11 @@ bool Buffer::IsInBounds(uint64_t address, uint64_t size) const noexcept {
 	return address >= m_cpu_address && size <= Size() && address - m_cpu_address <= Size() - size;
 }
 
-void Buffer::Write(uint64_t offset, const void* source, uint64_t size) {
-	EXIT_IF(source == nullptr || m_mapped.empty() || offset > Size() || size > Size() - offset);
-	std::memcpy(m_mapped.data() + offset, source, static_cast<size_t>(size));
-	Flush(offset, size);
-}
-
 void Buffer::Flush(uint64_t offset, uint64_t size) {
 	EXIT_IF(m_mapped.empty() || offset > Size() || size > Size() - offset);
 	if (!IsCoherent() && size != 0) {
 		const auto result =
-		    vmaFlushAllocation(m_graphics->allocator, m_buffer->memory.allocation, offset, size);
+		    vmaFlushAllocation(m_graphics->allocator, m_allocation, offset, size);
 		EXIT_NOT_IMPLEMENTED(static_cast<vk::Result>(result) != vk::Result::eSuccess);
 	}
 }
@@ -162,7 +132,7 @@ void Buffer::Invalidate(uint64_t offset, uint64_t size) {
 	EXIT_IF(m_usage != MemoryUsage::Download || offset > Size() || size > Size() - offset);
 	if (!IsCoherent() && size != 0) {
 		const auto result =
-		    vmaInvalidateAllocation(m_graphics->allocator, m_buffer->memory.allocation, offset, size);
+		    vmaInvalidateAllocation(m_graphics->allocator, m_allocation, offset, size);
 		EXIT_NOT_IMPLEMENTED(static_cast<vk::Result>(result) != vk::Result::eSuccess);
 	}
 }

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

@@ -6,19 +6,19 @@
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <cstdint>
-#include <memory>
 #include <optional>
 #include <span>
 #include <utility>
 #include <vector>
 
+VK_DEFINE_HANDLE(VmaAllocation)
+
 namespace Libs::Graphics {
 
 class CommandBuffer;
 class CommandScheduler;
 struct StreamBufferTestAccess;
 struct GraphicContext;
-struct VulkanBuffer;
 
 enum class MemoryUsage : uint8_t {
 	DeviceLocal,
@@ -42,10 +42,10 @@ public:
 	~Buffer();
 	KYTY_CLASS_NO_COPY(Buffer);
 
-	[[nodiscard]] vk::Buffer         Handle() const noexcept;
-	[[nodiscard]] uint64_t           Size() const noexcept;
+	[[nodiscard]] vk::Buffer         Handle() const noexcept { return m_buffer; }
+	[[nodiscard]] uint64_t           Size() const noexcept { return m_size; }
 	[[nodiscard]] std::span<uint8_t> Mapped() const noexcept { return m_mapped; }
-	[[nodiscard]] bool               IsCoherent() const noexcept;
+	[[nodiscard]] bool               IsCoherent() const noexcept { return m_coherent; }
 	[[nodiscard]] MemoryUsage        Usage() const noexcept { return m_usage; }
 	[[nodiscard]] uint64_t           CpuAddress() const noexcept { return m_cpu_address; }
 	[[nodiscard]] vk::DeviceAddress BufferDeviceAddress() const noexcept;
@@ -55,7 +55,6 @@ public:
 	[[nodiscard]] bool IsInBounds(uint64_t address, uint64_t size) const noexcept;
 	void               IncreaseStreamScore(int score) noexcept { stream_score += score; }
 	[[nodiscard]] int  StreamScore() const noexcept { return stream_score; }
-	void               Write(uint64_t offset, const void* source, uint64_t size);
 	void               Flush(uint64_t offset, uint64_t size);
 	void               Invalidate(uint64_t offset, uint64_t size);
 	void CopyFrom(CommandBuffer& command, const Buffer& source, uint64_t source_offset,
@@ -88,7 +87,10 @@ private:
 	MemoryUsage                   m_usage       = MemoryUsage::DeviceLocal;
 	uint64_t                      m_cpu_address = 0;
 	vk::DeviceAddress             m_device_address = 0;
-	std::unique_ptr<VulkanBuffer> m_buffer;
+	vk::Buffer                    m_buffer     = nullptr;
+	VmaAllocation                 m_allocation = nullptr;
+	uint64_t                      m_size;
+	bool                          m_coherent = false;
 	std::span<uint8_t>            m_mapped;
 };
 

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

@@ -1,5 +1,6 @@
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
 
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/emulatorConfig.h"
 #include "common/logging/log.h"
@@ -32,22 +33,24 @@ namespace {
 
 constexpr uint64_t NumFramesBeforeRemoval = 32;
 
-[[nodiscard]] bool DecodeDccClear(const TextureCache::ImageDesc& desc, vk::Format format,
-                                  uint32_t fill, vk::ClearColorValue& clear) {
-	const auto code = static_cast<uint8_t>(fill);
-	if (fill != static_cast<uint32_t>(code) * 0x01010101u) {
-		return false;
+[[nodiscard]] bool DecodeDccClear(const TextureCache::ImageDesc& desc, uint8_t code,
+                                  vk::ClearColorValue& clear) {
+	switch (code) {
+		case 0x00:
+		case 0x20:
+		case 0x40:
+		case 0x80:
+		case 0xc0: break;
+		default: return false;
 	}
 	const auto& metadata = desc.info.metadata;
+	const auto  format   = desc.view_info.format;
 	if (code == 0x20) {
 		// Clear-to-register is a color-buffer operation; the texture pipe cannot decode it.
 		return desc.type == TextureCache::BindingType::RenderTarget &&
 		       metadata.dcc_clear_register_valid &&
 		       DecodePackedColorClear(format, metadata.dcc_clear_word, clear);
 	}
-	if (code != 0x00 && code != 0x40 && code != 0x80 && code != 0xc0) {
-		return false;
-	}
 	clear = {};
 	if (code == 0x00) {
 		return true;
@@ -187,15 +190,11 @@ TextureCache::TextureCache(GraphicContext& graphics, CommandScheduler& scheduler
 }
 
 TextureCache::~TextureCache() {
-	std::vector<ImageId> registered;
 	m_slot_images.ForEach([&](ImageId id, const Image& image) {
 		if (image.registered) {
-			registered.push_back(id);
+			UnregisterImage(id);
 		}
 	});
-	for (const auto id: registered) {
-		UnregisterImage(id);
-	}
 }
 
 bool TextureCache::SameBacking(const ImageInfo& cached, const ImageInfo& requested,
@@ -270,9 +269,9 @@ void TextureCache::RegisterImage(ImageId id) {
 	if (!ImagePageTable::TryGetPageRange(image.info.data.address, image.info.data.size, pages)) {
 		EXIT("TextureCache: image registration is outside the guest address space\n");
 	}
-	for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
+	ForEachPage(image.info.data.address, image.info.data.size, [this, id](uint64_t page) {
 		m_image_page_table[page].push_back(id);
-	}
+	});
 	image.registered = true;
 	image.lru_id     = m_lru_cache.Insert(id, m_gc_tick);
 	m_total_used_memory += image.AccountedSize();
@@ -288,12 +287,12 @@ void TextureCache::UnregisterImage(ImageId id) {
 	if (!ImagePageTable::TryGetPageRange(image.info.data.address, image.info.data.size, pages)) {
 		EXIT("TextureCache: registered image is outside the guest address space\n");
 	}
-	for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
+	ForEachPage(image.info.data.address, image.info.data.size, [this, id](uint64_t page) {
 		auto* owners = m_image_page_table.Find(page);
 		if (owners == nullptr || !owners->Erase(id)) {
 			EXIT("TextureCache: image missing from page owner index\n");
 		}
-	}
+	});
 	m_lru_cache.Free(image.lru_id);
 	const auto accounted = image.AccountedSize();
 	if (accounted > m_total_used_memory) {
@@ -316,11 +315,7 @@ void TextureCache::DeleteImage(ImageId id) {
 			}
 		});
 		for (const auto association: associations) {
-			auto& associated = m_slot_images[association];
-			if (associated.IsGpuModified()) {
-				associated.ClearGpuModified();
-			}
-			DeleteImage(association);
+			FreeImage(association);
 		}
 	}
 	if (image->IsGpuModified()) {
@@ -330,10 +325,8 @@ void TextureCache::DeleteImage(ImageId id) {
 	if (image->info.HasMetadata()) {
 		const auto metadata = m_surface_metas.find(image->info.metadata.range.address);
 		if (metadata != m_surface_metas.end() &&
-		    ((image->info.metadata.kind == ImageMetadataKind::Dcc &&
-		      metadata->second.type == MetaDataInfo::Type::Dcc) ||
-		     (image->info.metadata.kind == ImageMetadataKind::Htile &&
-		      metadata->second.type == MetaDataInfo::Type::HTile))) {
+		    image->info.metadata.kind == ImageMetadataKind::Htile &&
+		    metadata->second.type == MetaDataInfo::Type::HTile) {
 			// A later binding may have reused this address for another metadata type.
 			m_surface_metas.erase(metadata);
 		}
@@ -447,7 +440,7 @@ void TextureCache::UntrackImageHead(ImageId id) {
 	if (!image.IsTracked() || begin < image.track_addr) {
 		return;
 	}
-	const auto address = (begin + TRACKER_PAGE_SIZE) & ~(TRACKER_PAGE_SIZE - 1);
+	const auto address = Common::AlignDown(begin + TRACKER_PAGE_SIZE, TRACKER_PAGE_SIZE);
 	const auto size    = address - begin;
 	image.track_addr   = address;
 	if (image.track_addr == image.track_addr_end) {
@@ -464,7 +457,7 @@ void TextureCache::UntrackImageTail(ImageId id) {
 	if (!image.IsTracked() || image.track_addr_end < end) {
 		return;
 	}
-	const auto address   = end & ~(TRACKER_PAGE_SIZE - 1);
+	const auto address   = Common::AlignDown(end, TRACKER_PAGE_SIZE);
 	const auto size      = end - address;
 	image.track_addr_end = address;
 	if (image.track_addr == image.track_addr_end) {
@@ -498,10 +491,10 @@ TextureCache::ImageIds TextureCache::FindImagesInRegion(uint64_t address, uint64
 	}
 
 	ImageIds result;
-	for (size_t page = pages.first; page < pages.last_exclusive; ++page) {
+	ForEachPage(address, size, [&](uint64_t page) {
 		const auto* owners = m_image_page_table.Find(page);
 		if (owners == nullptr) {
-			continue;
+			return;
 		}
 		owners->ForEach([&](ImageId id) {
 			auto* image = m_slot_images.try_get(id);
@@ -516,7 +509,7 @@ TextureCache::ImageIds TextureCache::FindImagesInRegion(uint64_t address, uint64
 				result.push_back(id);
 			}
 		});
-	}
+	});
 	return result;
 }
 
@@ -817,17 +810,10 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 	const bool safe_to_delete =
 	    current_tick - std::min(current_tick, cached.tick_accessed_last) > NumFramesBeforeRemoval;
 
-	if (requested.data.address == cached.info.data.address) {
-		const uint32_t requested_block = requested.bytes_per_block * requested.samples;
-		const uint32_t cached_block    = cached.info.bytes_per_block * cached.info.samples;
-		if (requested.BlockExtent() != cached.info.BlockExtent() ||
-		    requested_block != cached_block) {
-			if (safe_to_delete) {
-				FreeImage(cached_id);
-			}
-			return {merged_id};
-		}
-
+	const uint32_t requested_block = requested.bytes_per_block * requested.samples;
+	const uint32_t cached_block    = cached.info.bytes_per_block * cached.info.samples;
+	if (requested.data.address == cached.info.data.address &&
+	    requested.BlockExtent() == cached.info.BlockExtent() && requested_block == cached_block) {
 		if (const auto depth_id = ResolveDepthOverlap(requested, binding, cached_id)) {
 			return {depth_id};
 		}
@@ -880,38 +866,27 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		     static_cast<uint32_t>(cached.info.tile_mode));
 	}
 
-	if (requested.data.address > cached.info.data.address) {
-		const int32_t mip = requested.MipOf(cached.info);
-		if (mip >= 0) {
-			const int32_t layer = requested.SliceOf(cached.info, mip);
-			if (layer >= 0) {
-				return {cached_id, mip, layer};
-			}
-		}
-		if (safe_to_delete) {
-			FreeImage(cached_id);
-		}
-		return {};
+	const int32_t requested_mip = requested.MipOf(cached.info);
+	if (requested_mip >= 0) {
+		const int32_t layer = requested.SliceOf(cached.info, requested_mip);
+		return {cached_id, requested_mip, layer};
 	}
 
 	const int32_t mip = cached.info.MipOf(requested);
 	if (mip >= 0) {
 		const int32_t layer = cached.info.SliceOf(requested, mip);
-		if (layer >= 0) {
-			if (cached.binding.is_target) {
-				cached.binding.needs_rebind = true;
-				if (merged_id) {
-					m_slot_images[merged_id].binding.is_target = true;
-				}
-				FreeImage(cached_id);
-				return {merged_id};
-			}
-			if (merged_id) {
-				CopyImageMip(merged_id, cached_id, static_cast<uint32_t>(mip),
-				             static_cast<uint32_t>(layer));
-				FreeImage(cached_id);
-			}
+		if (!merged_id) {
+			return {ExpandImage(requested, cached_id)};
 		}
+		cached.binding.needs_rebind |= cached.binding.is_bound || cached.binding.is_target;
+		m_slot_images[merged_id].binding.is_target |= cached.binding.is_target;
+		CopyImageMip(merged_id, cached_id, static_cast<uint32_t>(mip),
+		             static_cast<uint32_t>(layer));
+		FreeImage(cached_id);
+		return {merged_id};
+	}
+	if (requested.data.address >= cached.info.data.address && safe_to_delete) {
+		FreeImage(cached_id);
 	}
 	return {merged_id};
 }
@@ -926,12 +901,19 @@ ImageId TextureCache::ExpandImage(const ImageInfo& info, ImageId source_id) {
 		source.binding.needs_rebind = true;
 	}
 	InitializeImage(expanded_id);
-	CopyImage(expanded_id, source_id);
+	const int32_t mip = source.info.MipOf(info);
+	const int32_t layer = source.info.SliceOf(info, mip);
+	if (layer >= 0) {
+		CopyImageMip(expanded_id, source_id, static_cast<uint32_t>(mip),
+		             static_cast<uint32_t>(layer));
+	} else {
+		CopyImage(expanded_id, source_id);
+	}
 	FreeImage(source_id);
 	return expanded_id;
 }
 
-struct TextureCache::TextureTransferPlan {
+struct TextureCache::TextureTransfer {
 	TextureUploadLayout              layout;
 	std::vector<vk::BufferImageCopy> regions;
 	std::vector<GpuTileInfo>         tiles;
@@ -947,13 +929,13 @@ struct TextureCache::TextureTransferPlan {
 	}
 };
 
-struct TextureCache::DownloadPlan {
-	TextureTransferPlan texture;
+struct TextureCache::ImageDownload {
+	TextureTransfer texture;
 	bool                depth_target = false;
 	bool                valid        = false;
 };
 
-TextureCache::TextureTransferPlan
+TextureCache::TextureTransfer
 TextureCache::BuildTextureTransfer(const Image& image, BindingType binding,
                                     TransferDirection direction) const {
 	const auto& info             = image.info;
@@ -966,8 +948,8 @@ TextureCache::BuildTextureTransfer(const Image& image, BindingType binding,
 	bool        allow_depth_tile = upload;
 	const char* owner            = "TextureCache readback";
 
-	TextureTransferPlan plan;
-	plan.swap_bgra16 = info.bgra16 && (!upload || render_target || video_out);
+	TextureTransfer transfer;
+	transfer.swap_bgra16 = info.bgra16 && (!upload || render_target || video_out);
 	if (render_target) {
 		format = ImageOps::RenderTargetTransferFormat(info.bytes_per_block);
 	}
@@ -997,44 +979,44 @@ TextureCache::BuildTextureTransfer(const Image& image, BindingType binding,
 		}
 	}
 
-	plan.layout  = TextureCalcUploadLayout(format, info.extent.width, info.extent.height,
+	transfer.layout  = TextureCalcUploadLayout(format, info.extent.width, info.extent.height,
 	                                       info.resources.levels, layers, info.tile_mode,
 	                                       info.data.size, allow_depth_tile, volume, owner);
-	plan.regions = TextureBuildImageCopies(plan.layout);
+	transfer.regions = TextureBuildImageCopies(transfer.layout);
 	if (info.IsDepth()) {
-		for (auto& region: plan.regions) {
+		for (auto& region: transfer.regions) {
 			region.imageSubresource.aspectMask = vk::ImageAspectFlagBits::eDepth;
 		}
 	}
-	if (plan.layout.surface.description.tile_mode != Prospero::TileMode::kLinear) {
-		if (!TextureBuildGpuTileInfos(info.data.size, plan.regions, plan.layout,
-		                              info.resources.levels, plan.tiles)) {
-			return plan;
+	if (transfer.layout.surface.description.tile_mode != Prospero::TileMode::kLinear) {
+		if (!TextureBuildGpuTileInfos(info.data.size, transfer.regions, transfer.layout,
+		                              info.resources.levels, transfer.tiles)) {
+			return transfer;
 		}
 	}
-	plan.valid = true;
-	return plan;
+	transfer.valid = true;
+	return transfer;
 }
 
-TextureCache::DownloadPlan TextureCache::BuildDownload(const Image& image) const {
+TextureCache::ImageDownload TextureCache::BuildDownload(const Image& image) const {
 	const auto&  info    = image.info;
 	const auto   binding = UploadBinding(image);
-	DownloadPlan plan {.depth_target = binding == BindingType::DepthTarget};
+	ImageDownload transfer {.depth_target = binding == BindingType::DepthTarget};
 	if (info.samples != 1 || image.backing.samples != 1) {
-		return plan;
+		return transfer;
 	}
-	if (plan.depth_target) {
-		plan.valid = IsSupportedDepthPlaneReadback(info) && info.resources.layers != 0 &&
+	if (transfer.depth_target) {
+		transfer.valid = IsSupportedDepthPlaneReadback(info) && info.resources.layers != 0 &&
 		             info.data.size % info.resources.layers == 0 &&
 		             Prospero::NumBytesPerElement(info.guest_format) == info.bytes_per_block;
-		return plan;
+		return transfer;
 	}
 	if (info.metadata.compression != VideoOutCompression::Uncompressed) {
-		return plan;
+		return transfer;
 	}
-	plan.texture = BuildTextureTransfer(image, binding, TransferDirection::Download);
-	plan.valid   = plan.texture.valid;
-	return plan;
+	transfer.texture = BuildTextureTransfer(image, binding, TransferDirection::Download);
+	transfer.valid   = transfer.texture.valid;
+	return transfer;
 }
 
 void TextureCache::UploadImage(Image& image, Buffer& source, uint64_t source_offset) {
@@ -1048,26 +1030,26 @@ void TextureCache::UploadImage(Image& image, Buffer& source, uint64_t source_off
 	};
 
 	if (binding != BindingType::DepthTarget) {
-		auto plan = BuildTextureTransfer(image, binding, TransferDirection::Upload);
-		if (!plan.valid) {
+		auto transfer = BuildTextureTransfer(image, binding, TransferDirection::Upload);
+		if (!transfer.valid) {
 			EXIT("TextureCache: invalid texture upload: binding=%u addr=0x%016" PRIx64
 			     " size=0x%016" PRIx64 " format=%u tile=%u family=%u extent=%ux%ux%u "
 			     "pitch=%u levels=%u layers=%u samples=%u\n",
 			     static_cast<uint32_t>(binding), info.data.address, info.data.size,
 			     static_cast<uint32_t>(info.guest_format), static_cast<uint32_t>(info.tile_mode),
-			     static_cast<uint32_t>(plan.layout.surface.texture.block.family), info.extent.width,
+			     static_cast<uint32_t>(transfer.layout.surface.texture.block.family), info.extent.width,
 			     info.extent.height, info.extent.depth, info.pitch, info.resources.levels,
 			     info.resources.layers, info.samples);
 		}
 		TileManager::Result linear {source.Handle(), source_offset, info.data.size};
-		if (!plan.tiles.empty()) {
+		if (!transfer.tiles.empty()) {
 			linear = m_tiler.Detile(source.Handle(), source_offset, info.data.size,
-			                        plan.LinearSize(), plan.tiles);
+			                        transfer.LinearSize(), transfer.tiles);
 		}
-		if (plan.swap_bgra16) {
+		if (transfer.swap_bgra16) {
 			linear = m_tiler.SwapBgra16(linear);
 		}
-		upload(plan.regions, linear);
+		upload(transfer.regions, linear);
 		return;
 	}
 
@@ -1142,53 +1124,69 @@ void TextureCache::InitializeImage(ImageId id) {
 	}
 }
 
-void TextureCache::PrepareDccClear(ImageId id, const ImageDesc& desc) {
+void TextureCache::MaterializeDccClear(ImageId id, const ImageDesc& desc,
+                                       uint32_t metadata_base_layer) {
 	if (desc.info.metadata.kind != ImageMetadataKind::Dcc) {
 		return;
 	}
-	auto& image            = m_slot_images[id];
-	image.info.metadata    = desc.info.metadata;
-	auto [entry, inserted] = m_surface_metas.try_emplace(
-	    desc.info.metadata.range.address, MetaDataInfo {.type = MetaDataInfo::Type::Dcc});
-	auto& metadata = entry->second;
-	if (!inserted && metadata.type == MetaDataInfo::Type::PendingDcc) {
-		metadata.type = MetaDataInfo::Type::Dcc;
-	} else if (metadata.type != MetaDataInfo::Type::Dcc) {
-		EXIT("TextureCache: image reuses non-DCC metadata\n");
+	const auto range = desc.info.metadata.range;
+	{
+		std::scoped_lock lock {m_lock};
+		auto& image         = m_slot_images[id];
+		image.info.metadata = desc.info.metadata;
+		// A native DCC allocation must not retain a reused HTile/CMask/FMask interpretation.
+		m_surface_metas.erase(range.address);
+		if (range.size == 0 || desc.info.resources.levels != 1 || image.info.resources.levels != 1) {
+			return;
+		}
 	}
-	if (metadata.clear_mask == 0 || image.info.resources.levels != 1 ||
-	    desc.info.metadata.range.size == 0 || metadata.fill_size < desc.info.metadata.range.size) {
-		return;
-	}
-	vk::ClearValue clear {};
-	if (!DecodeDccClear(desc, image.backing.format, metadata.fill_value, clear.color)) {
-		return;
+	const auto layers = desc.info.TransferLayers();
+	// These one-mip surfaces use complete 4 KiB DCC metadata blocks.
+	constexpr uint64_t MetadataBlockSize = 0x1000;
+	if (!range.Valid() || range.address % MetadataBlockSize != 0 || layers == 0 ||
+	    range.size % layers != 0 || (range.size / layers) % MetadataBlockSize != 0) {
+		EXIT("TextureCache: DCC slices must contain aligned 4 KiB blocks\n");
 	}
 	const auto& view           = desc.view_info;
-	const bool  volume_texture = image.info.IsVolume() && view.type == vk::ImageViewType::e3D;
-	const auto  first          = volume_texture ? 0u : view.base_layer;
-	const auto  count = volume_texture ? std::max(image.info.extent.depth >> view.base_level, 1u)
-	                                   : view.layer_count;
-	if (first >= 32 || count > 32 - first) {
-		return;
+	const bool  volume_texture = desc.info.IsVolume() && view.type == vk::ImageViewType::e3D;
+	const auto  first          = volume_texture ? 0u : metadata_base_layer;
+	const auto  image_first    = volume_texture ? 0u : view.base_layer;
+	const auto  count          = volume_texture ? desc.info.extent.depth : view.layer_count;
+	if (first >= layers || count > layers - first) {
+		EXIT("TextureCache: DCC view exceeds its native metadata slices\n");
 	}
-	// The metadata fill covers the complete allocation. Consume each layer only after its
-	// native image contents exist; already materialized layers may have been rendered since.
-	for (uint32_t layer = first; layer < first + count;) {
-		if ((metadata.clear_mask & (1u << layer)) == 0) {
-			layer++;
+	// Finish native metadata writes before reading backing bytes. This can submit the scheduler,
+	// so discovery runs before final draw uploads and never holds the texture lock across it.
+	if (m_buffer_cache.IsRegionGpuModified(range.address, range.size)) {
+		m_buffer_cache.ReadMemory(range.address, range.size, false);
+	}
+	const auto slice_size = range.size / layers;
+	for (uint32_t slice = 0; slice < count; slice++) {
+		const auto address = range.address + slice_size * (first + slice);
+		uint8_t code = 0;
+		if (!LibKernel::Memory::TryReadBacking(address, &code, sizeof(code))) {
+			EXIT("TextureCache: failed to read DCC metadata backing\n");
+		}
+		vk::ClearValue clear {};
+		if (!DecodeDccClear(desc, code, clear.color)) {
 			continue;
 		}
-		const auto start = layer;
-		uint32_t   mask  = 0;
-		do {
-			mask |= 1u << layer++;
-		} while (layer < first + count && (metadata.clear_mask & (1u << layer)) != 0);
-		ClearImage(m_scheduler.Current(), id,
-		           {vk::ImageAspectFlagBits::eColor, view.base_level, view.level_count, start,
-		            layer - start},
-		           clear);
-		metadata.clear_mask &= ~mask;
+		std::vector<uint8_t> bytes(slice_size);
+		if (!LibKernel::Memory::TryReadBacking(address, bytes.data(), bytes.size())) {
+			EXIT("TextureCache: failed to read DCC metadata slice\n");
+		}
+		if (!std::all_of(bytes.begin(), bytes.end(), [code](uint8_t byte) { return byte == code; })) {
+			continue;
+		}
+		{
+			std::scoped_lock lock {m_lock};
+			ClearImage(m_scheduler.Current(), id, view.format,
+			           {vk::ImageAspectFlagBits::eColor, view.base_level, view.level_count,
+			            image_first + slice, 1}, clear);
+		}
+		// Native expanded keys own consumption. Existing buffer tracking publishes this CPU
+		// write to future GPU readers; FillBuffer can fault and must run outside the texture lock.
+		m_buffer_cache.FillBuffer(address, slice_size, UINT32_MAX, false);
 	}
 }
 
@@ -1253,6 +1251,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		std::scoped_lock lock {m_lock};
 		return GetNullImage(desc);
 	}
+	const auto metadata_base_layer = desc.view_info.base_layer;
 
 	ImageId result {};
 	{
@@ -1320,6 +1319,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		image.tick_accessed_last = m_scheduler.CurrentTick();
 		TouchImage(image);
 	}
+	MaterializeDccClear(result, desc, metadata_base_layer);
 	return result;
 }
 
@@ -1334,8 +1334,8 @@ ImageId TextureCache::FindImageFromRange(uint64_t address, uint64_t size, bool e
 	if (!GuestRange {address, size}.Valid()) {
 		return {};
 	}
-	std::scoped_lock     lock {m_lock};
-	std::vector<ImageId> matches;
+	std::scoped_lock lock {m_lock};
+	ImageIds         matches;
 	for (const auto id: FindImagesInRegion(address, size, false)) {
 		auto owner = m_slot_images.try_get(id);
 		if (owner == nullptr || owner->info.data.address != address) {
@@ -1383,16 +1383,12 @@ vk::ImageView TextureCache::FindTexture(ImageId id, const ImageDesc& desc) {
 		image.MarkGpuModified();
 	}
 	if (!image.info.data.Empty()) {
-		PrepareDccClear(id, desc);
 		RefreshImage(id);
 	}
 	switch (desc.type) {
 		case BindingType::Texture: break;
 		case BindingType::Storage:
 			if (!image.info.data.Empty()) {
-				if (!image.registered || image.depth_id) {
-					EXIT("TextureCache: cannot acquire an unavailable storage image\n");
-				}
 				CommitGpuWrite(image);
 			}
 			TrackImageDownload(id, image);
@@ -1416,7 +1412,6 @@ vk::ImageView TextureCache::FindRenderTarget(ImageId id, const ImageDesc& desc)
 	TouchImage(image);
 	image.MarkGpuModified();
 	image.usage.render_target = true;
-	PrepareDccClear(id, desc);
 	RefreshImage(id);
 	CommitGpuWrite(image);
 	TrackImageDownload(id, image);
@@ -1440,16 +1435,9 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 	RefreshImage(id);
 	if (desc.info.HasMetadata()) {
 		image.info.metadata = desc.info.metadata;
-		auto [metadata, inserted] =
-		    m_surface_metas.try_emplace(desc.info.metadata.range.address,
-		                                MetaDataInfo {.type       = MetaDataInfo::Type::HTile,
-		                                              .clear_mask = image.info.htile_clear_mask});
-		if (!inserted && metadata->second.type != MetaDataInfo::Type::HTile) {
-			// PS5 allocations can reuse DCC storage as HTile while the old color image is cached.
-			// The depth binding defines the new type; incompatible fill state cannot carry over.
-			metadata->second = {.type       = MetaDataInfo::Type::HTile,
-			                    .clear_mask = image.info.htile_clear_mask};
-		}
+		m_surface_metas.emplace(desc.info.metadata.range.address,
+		                        MetaDataInfo {.type       = MetaDataInfo::Type::HTile,
+		                                      .clear_mask = image.info.htile_clear_mask});
 	}
 	CommitGpuWrite(image);
 	if (desc.info.HasStencil()) {
@@ -1537,12 +1525,12 @@ bool TextureCache::ClearImageFromBuffer(CommandBuffer& command, uint64_t address
 		}
 		clear.depthStencil.stencil = stencil_clear;
 	}
-	ClearImage(command, selected,
+	ClearImage(command, selected, image.backing.format,
 	           {aspect, 0, image.info.resources.levels, 0, image.info.TransferLayers()}, clear);
 	return true;
 }
 
-void TextureCache::ClearImage(CommandBuffer& command, ImageId id,
+void TextureCache::ClearImage(CommandBuffer& command, ImageId id, vk::Format format,
                               const vk::ImageSubresourceRange& range, const vk::ClearValue& clear) {
 	auto& image = m_slot_images[id];
 	const auto aspects = image.info.IsDepth() ? ImageViewOps::DepthAspectMask(image.backing.format)
@@ -1567,11 +1555,12 @@ void TextureCache::ClearImage(CommandBuffer& command, ImageId id,
 		}
 	}
 	command.EndRendering();
-	if (image.info.IsVolume() && !full_image) {
+	// Transfer clears use the backing format; aliased clears must encode through their view.
+	if (format != image.backing.format || (image.info.IsVolume() && !full_image)) {
 		EXIT_NOT_IMPLEMENTED(range.aspectMask != vk::ImageAspectFlagBits::eColor ||
 		                     range.levelCount != 1);
 		ImageViewInfo view {};
-		view.format = image.backing.format;
+		view.format = format;
 		view.type   = range.layerCount == 1 ? vk::ImageViewType::e2D : vk::ImageViewType::e2DArray;
 		view.base_level  = range.baseMipLevel;
 		view.base_layer  = range.baseArrayLayer;
@@ -1679,12 +1668,12 @@ void TextureCache::DownloadDepth(Image& image, Buffer& destination, uint64_t des
 	             destination_offset, info.data.size, tiles);
 }
 
-void TextureCache::DownloadImageData(Image& image, Buffer& destination, uint64_t destination_offset,
-                                     uint64_t destination_size, DownloadPlan plan) {
-	if (!plan.valid) {
-		EXIT("TextureCache: invalid image download plan\n");
+void TextureCache::DownloadImage(Image& image, Buffer& destination, uint64_t destination_offset,
+                                     uint64_t destination_size, ImageDownload transfer) {
+	if (!transfer.valid) {
+		EXIT("TextureCache: invalid image download transfer\n");
 	}
-	if (plan.depth_target) {
+	if (transfer.depth_target) {
 		if (destination_size != image.info.data.size) {
 			EXIT("TextureCache: partial depth image download is unsupported\n");
 		}
@@ -1692,7 +1681,7 @@ void TextureCache::DownloadImageData(Image& image, Buffer& destination, uint64_t
 		return;
 	}
 
-	auto&      texture   = plan.texture;
+	auto&      texture   = transfer.texture;
 	const auto transform = texture.swap_bgra16 ? TileManager::ColorTransform::SwapBgra16
 	                                           : TileManager::ColorTransform::None;
 	if (texture.tiles.empty()) {
@@ -1753,15 +1742,15 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uin
 	if (copy_size == 0) {
 		return false;
 	}
-	auto plan = m_texture_cache.BuildDownload(image);
-	if (!plan.valid) {
+	auto transfer = m_texture_cache.BuildDownload(image);
+	if (!transfer.valid) {
 		return false;
 	}
-	if (plan.depth_target && copy_size != image.info.data.size) {
+	if (transfer.depth_target && copy_size != image.info.data.size) {
 		return false;
 	}
-	if (!plan.depth_target && levels < image.info.resources.levels) {
-		auto& texture = plan.texture;
+	if (!transfer.depth_target && levels < image.info.resources.levels) {
+		auto& texture = transfer.texture;
 		std::erase_if(texture.regions, [levels](const vk::BufferImageCopy& region) {
 			return region.imageSubresource.mipLevel >= levels;
 		});
@@ -1776,17 +1765,17 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uin
 			}
 		}
 	}
-	m_texture_cache.DownloadImageData(image, buffer, buf_offset, copy_size, std::move(plan));
+	m_texture_cache.DownloadImage(image, buffer, buf_offset, copy_size, std::move(transfer));
 	return true;
 }
 
-bool TextureCache::TryDownloadImage(ImageId id) {
+bool TextureCache::DownloadImageMemory(ImageId id) {
 	auto& image = m_slot_images[id];
 	if (image.depth_id) {
 		return false;
 	}
-	auto plan = BuildDownload(image);
-	if (!plan.valid || !SafeToDownload(image)) {
+	auto transfer = BuildDownload(image);
+	if (!transfer.valid || !SafeToDownload(image)) {
 		return false;
 	}
 	const auto range    = image.info.data;
@@ -1802,7 +1791,7 @@ bool TextureCache::TryDownloadImage(ImageId id) {
 	}
 	download.Flush(offset, range.size);
 
-	DownloadImageData(image, download, offset, range.size, std::move(plan));
+	DownloadImage(image, download, offset, range.size, std::move(transfer));
 	vk::BufferMemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eMemoryWrite | vk::AccessFlagBits::eTransferWrite |
 	                        vk::AccessFlagBits::eShaderWrite;
@@ -1855,8 +1844,8 @@ bool TextureCache::IsRegionGpuModified(uint64_t address, uint64_t size) {
 }
 
 void TextureCache::InvalidateCpuAliases(uint64_t address, uint64_t size) {
-	const auto page_begin = address & ~(TRACKER_PAGE_SIZE - 1);
-	const auto page_end   = (address + size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
+	const auto page_begin = Common::AlignDown(address, TRACKER_PAGE_SIZE);
+	const auto page_end   = Common::AlignUp(address + size, TRACKER_PAGE_SIZE);
 	for (const auto id: FindImagesInRegion(address, size, true)) {
 		auto owner = m_slot_images.try_get(id);
 		if (owner == nullptr) {
@@ -1882,72 +1871,32 @@ void TextureCache::InvalidateCpuAliases(uint64_t address, uint64_t size) {
 bool TextureCache::IsMeta(uint64_t address) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	return found != m_surface_metas.end() && found->second.type != MetaDataInfo::Type::PendingDcc;
+	return found != m_surface_metas.end();
 }
 
-bool TextureCache::IsMetaCleared(uint64_t address, uint32_t slice, uint32_t* fill_value) {
+bool TextureCache::IsMetaCleared(uint64_t address, uint32_t slice) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    slice >= 32) {
+	if (found == m_surface_metas.end() || slice >= 32) {
 		return false;
 	}
-	if (fill_value != nullptr) {
-		*fill_value = found->second.fill_value;
-	}
 	return (found->second.clear_mask & (1u << slice)) != 0;
 }
 
 bool TextureCache::ClearMeta(uint64_t address) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    found->second.type == MetaDataInfo::Type::Dcc) {
-		// Preserve the broad metadata-clear operation for CMask/FMask/HTile. DCC requires a
-		// validated fill value, so an arbitrary compute write must not clear it.
+	if (found == m_surface_metas.end()) {
 		return false;
 	}
 	found->second.clear_mask = UINT32_MAX;
 	return true;
 }
 
-void TextureCache::TrackDccFill(uint64_t address, uint64_t size, uint32_t fill_value) {
-	if (!GuestRange {address, size}.Valid()) {
-		EXIT("TextureCache: invalid DCC fill range\n");
-	}
-	// DCC fills use a repeated byte code. Require all four bytes of the detected dword to agree,
-	// and mark only recognized deferred-clear encodings as logically clear.
-	const auto dcc_clear_mask = [fill_value] {
-		const auto code = static_cast<uint8_t>(fill_value);
-		if (fill_value != static_cast<uint32_t>(code) * 0x01010101u) {
-			return 0u;
-		}
-		switch (code) {
-			case 0x00:
-			case 0x20:
-			case 0x40:
-			case 0x80:
-			case 0xc0: return UINT32_MAX;
-			default: return 0u;
-		}
-	}();
-	std::scoped_lock lock {m_lock};
-	// The guest dispatch still writes metadata. An unknown address remains PendingDcc until an
-	// image descriptor confirms its role; never reinterpret CMask/FMask/HTile as DCC.
-	const auto found = m_surface_metas.try_emplace(address).first;
-	if (found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    found->second.type == MetaDataInfo::Type::Dcc) {
-		found->second.clear_mask = dcc_clear_mask;
-		found->second.fill_value = fill_value;
-		found->second.fill_size  = size;
-	}
-}
-
 bool TextureCache::TouchMeta(uint64_t address, uint32_t slice, bool is_clear) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    slice >= 32) {
+	if (found == m_surface_metas.end() || slice >= 32) {
 		return false;
 	}
 	if (is_clear) {
@@ -1963,10 +1912,13 @@ void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 		EXIT("TextureCache: invalid unmap range\n");
 	}
 	std::scoped_lock lock {m_lock};
-	const auto       end = address + size;
-	for (auto metadata = m_surface_metas.lower_bound(address);
-	     metadata != m_surface_metas.end() && metadata->first < end;) {
-		metadata = m_surface_metas.erase(metadata);
+	for (auto metadata = m_surface_metas.begin(); metadata != m_surface_metas.end();) {
+		const auto base = metadata->first;
+		if (base >= address && base < address + size) {
+			metadata = m_surface_metas.erase(metadata);
+		} else {
+			++metadata;
+		}
 	}
 	auto images = FindImagesInRegion(address, size, false);
 	for (const auto id: images) {
@@ -1974,10 +1926,7 @@ void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 		if (owner == nullptr) {
 			continue;
 		}
-		if (owner->IsGpuModified()) {
-			owner->ClearGpuModified();
-		}
-		DeleteImage(id);
+		FreeImage(id);
 	}
 }
 
@@ -2020,12 +1969,11 @@ void TextureCache::RunGarbageCollector() {
 				if (safe && !pressured) {
 					continue;
 				}
-				if (safe && !TryDownloadImage(id)) {
+				if (safe && !DownloadImageMemory(id)) {
 					continue;
 				}
-				owner->ClearGpuModified();
 			}
-			DeleteImage(id);
+			FreeImage(id);
 			if (m_total_used_memory < m_critical_gc_memory && aggressive) {
 				deletions >>= 2;
 				aggressive = false;
@@ -2047,7 +1995,7 @@ void TextureCache::ProcessDownloadImages() {
 	for (const auto id: m_download_images) {
 		const auto owner = m_slot_images.try_get(id);
 		if (owner != nullptr && owner->registered && owner->IsGpuModified()) {
-			(void)TryDownloadImage(id);
+			(void)DownloadImageMemory(id);
 		}
 	}
 	m_download_images.clear();

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

@@ -13,6 +13,7 @@
 #include "graphics/host_gpu/renderer/image/tiler.h"
 
 #include <map>
+#include <type_traits>
 #include <unordered_map>
 #include <unordered_set>
 #include <vector>
@@ -63,11 +64,8 @@ public:
 	[[nodiscard]] bool IsRegionGpuModified(uint64_t address, uint64_t size);
 
 	[[nodiscard]] bool IsMeta(uint64_t address);
-	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice,
-	                                 uint32_t* fill_value = nullptr);
+	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice);
 	[[nodiscard]] bool ClearMeta(uint64_t address);
-	// Record deferred DCC state while the original guest dispatch writes the metadata.
-	void               TrackDccFill(uint64_t address, uint64_t size, uint32_t fill_value);
 	[[nodiscard]] bool TouchMeta(uint64_t address, uint32_t slice, bool is_clear);
 
 	void UnmapMemory(uint64_t address, uint64_t size);
@@ -76,21 +74,14 @@ public:
 
 private:
 	enum class TransferDirection { Upload, Download };
-	struct TextureTransferPlan;
-	struct DownloadPlan;
+	struct TextureTransfer;
+	struct ImageDownload;
 
 	struct MetaDataInfo {
-		// A guest metadata-fill dispatch may initialize DCC before its render target is bound.
-		// PendingDcc retains that exact fill until an image binding classifies the address,
-		// without exposing an unconfirmed buffer address to the normal metadata heuristics.
-		// Keep all surface metadata in one entry so CMask/FMask can be
-		// registered beside HTile and DCC without introducing parallel tracking paths.
-		enum class Type : uint8_t { PendingDcc, CMask, FMask, HTile, Dcc };
+		enum class Type : uint8_t { CMask, FMask, HTile };
 
-		Type     type       = Type::PendingDcc;
-		uint32_t clear_mask = 0;
-		uint32_t fill_value = 0xffffffffu;
-		uint64_t fill_size  = 0;
+		Type     type;
+		uint32_t clear_mask = UINT32_MAX;
 	};
 
 	struct OverlapResult {
@@ -102,6 +93,23 @@ private:
 	using ImageIds       = InlinePageOwnerList<ImageId, 16>;
 	using ImagePageTable = MultiLevelPageTable<ImageIds, 20, 40, 10>;
 
+	// Callers have validated the nonempty 40-bit range with TryGetPageRange.
+	template <typename Func>
+	static void ForEachPage(uint64_t address, size_t size, Func&& func) {
+		using FuncReturn = typename std::invoke_result<Func, uint64_t>::type;
+		static constexpr bool RETURNS_BOOL = std::is_same_v<FuncReturn, bool>;
+		const uint64_t page_end = (address + size - 1) >> ImagePageTable::kPageBits;
+		for (uint64_t page = address >> ImagePageTable::kPageBits; page <= page_end; ++page) {
+			if constexpr (RETURNS_BOOL) {
+				if (func(page)) {
+					break;
+				}
+			} else {
+				func(page);
+			}
+		}
+	}
+
 	[[nodiscard]] ImageId     InsertImage(const ImageInfo& info);
 	[[nodiscard]] ImageId     GetNullImage(const ImageDesc& desc);
 	void                      RegisterImage(ImageId id);
@@ -131,19 +139,20 @@ private:
 	                                                ImageId cached);
 	[[nodiscard]] ImageId       ExpandImage(const ImageInfo& info, ImageId source);
 	void                        RefreshImage(ImageId id);
-	void                        PrepareDccClear(ImageId id, const ImageDesc& desc);
+	void                        MaterializeDccClear(ImageId id, const ImageDesc& desc,
+	                                                uint32_t metadata_base_layer);
 	void                        InitializeImage(ImageId id);
-	[[nodiscard]] TextureTransferPlan
+	[[nodiscard]] TextureTransfer
 	BuildTextureTransfer(const Image& image, BindingType binding, TransferDirection direction) const;
-	[[nodiscard]] DownloadPlan BuildDownload(const Image& image) const;
+	[[nodiscard]] ImageDownload BuildDownload(const Image& image) const;
 	void UploadImage(Image& image, Buffer& source, uint64_t source_offset);
-	void DownloadImageData(Image& image, Buffer& destination, uint64_t destination_offset,
-	                       uint64_t destination_size, DownloadPlan plan);
+	void DownloadImage(Image& image, Buffer& destination, uint64_t destination_offset,
+	                       uint64_t destination_size, ImageDownload transfer);
 	void DownloadDepth(Image& image, Buffer& destination, uint64_t destination_offset);
 	void CommitGpuWrite(Image& image);
 	// Caller holds m_lock. Volume layer ranges select depth slices.
-	void ClearImage(CommandBuffer& command, ImageId id, const vk::ImageSubresourceRange& range,
-	                const vk::ClearValue& clear);
+	void ClearImage(CommandBuffer& command, ImageId id, vk::Format format,
+	                const vk::ImageSubresourceRange& range, const vk::ClearValue& clear);
 	void PrepareImageCopy(Image& image);
 	void RefreshCopySource(ImageId id);
 	[[nodiscard]] bool CopyD16(Image& destination, Image& source);
@@ -153,7 +162,7 @@ private:
 	void ValidateImageDesc(const ImageDesc& desc) const;
 
 	void               InvalidateCpuAliases(uint64_t address, uint64_t size);
-	[[nodiscard]] bool TryDownloadImage(ImageId id);
+	[[nodiscard]] bool DownloadImageMemory(ImageId id);
 
 	GraphicContext&                                   m_graphics;
 	CommandScheduler&                                 m_scheduler;

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -41,16 +41,13 @@ static bool DccAlphaOnMsb(const HW::ColorInfo& info) {
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer& buffer,
-                                              RenderColorInfo& r,
+void RenderExecutor::ResolveRenderColorTarget(CommandBuffer& buffer, RenderColorInfo& r,
                                               uint32_t         render_target_slice_offset,
-                                              uint32_t render_target_slot, bool ignore_target_mask,
+                                              uint32_t rt_slot, bool ignore_target_mask,
                                               bool exact_format) {
 	KYTY_PROFILER_FUNCTION();
 	const auto& hw = buffer.GetRegisters();
 
-	const auto  rt_slot = (render_target_slot == UINT32_MAX ? render_target_first_bound_slot(buffer)
-	                                                        : render_target_slot);
 	const auto& rt      = hw.GetRenderTarget(rt_slot);
 	auto        mask    = render_target_mask_slot(hw.GetRenderTargetMask(), rt_slot);
 	if (ignore_target_mask && rt.base.addr != 0 && mask == 0) {
@@ -206,17 +203,14 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer&
 		     " layer=%u/%u\n",
 		     rt.attrib3.dimension, rt.attrib3.depth, view.base_layer, view.image_layers);
 	}
-	if (tile) {
-		if (volume || texture_tile) {
-			pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
-		} else {
-			pitch = TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
-		}
-		if (pitch == 0) {
-			EXIT("unsupported render-target pitch: width=%u bytes=%u\n", width, bytes_per_element);
-		}
+	// PPSA28068: the linear EULA target and its sampled view must share the padded pitch.
+	if (!tile || volume || texture_tile) {
+		pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
 	} else {
-		pitch = width;
+		pitch = TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
+	}
+	if (pitch == 0) {
+		EXIT("unsupported render-target pitch: width=%u bytes=%u\n", width, bytes_per_element);
 	}
 
 	TileSizeOffset    mip_sizes[16] {};

src/graphics/host_gpu/renderer/commandScheduler.cpp

@@ -208,13 +208,7 @@ void CommandScheduler::Wait(uint64_t tick) {
 }
 
 void CommandScheduler::PopPendingOperations() {
-	PopPendingOperations(true);
-}
-
-void CommandScheduler::PopPendingOperations(bool refresh_gpu_tick) {
-	if (refresh_gpu_tick) {
-		m_master.Refresh();
-	}
+	m_master.Refresh();
 	for (;;) {
 		PendingOperation operation;
 		{

src/graphics/host_gpu/renderer/commandScheduler.h

@@ -81,7 +81,6 @@ private:
 	};
 
 	void BeginNext();
-	void PopPendingOperations(bool refresh_gpu_tick);
 	void PriorityOperationsThread(std::stop_token stop);
 	void RunOperation(Common::UniqueFunction<void>&& operation);
 

src/graphics/host_gpu/renderer/context.cpp

@@ -9,7 +9,6 @@
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
-#include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <algorithm>
@@ -62,11 +61,11 @@ void CommandBuffer::SetDebugInfo(uint32_t op, uint64_t submit_id, uint32_t arg0,
 }
 
 void CommandBuffer::BeginRendering(const RenderState& state) const {
-	EXIT_IF(state.width == 0 || state.height == 0 || state.num_layers == 0 ||
-	        state.num_color_attachments > RENDER_COLOR_ATTACHMENTS_MAX);
 	if (m_rendering && m_render_state == state) {
 		return;
 	}
+	EXIT_IF(state.width == 0 || state.height == 0 || state.num_layers == 0 ||
+	        state.num_color_attachments > RENDER_COLOR_ATTACHMENTS_MAX);
 	EndRendering();
 
 	std::array<vk::RenderingAttachmentInfo, RENDER_COLOR_ATTACHMENTS_MAX> colors {};

src/graphics/host_gpu/renderer/debug.cpp

@@ -4,7 +4,6 @@
 #include "common/common.h"
 #include "common/emulatorConfig.h"
 #include "common/logging/log.h"
-#include "common/stringUtils.h"
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/host_gpu/renderer/render.h"
@@ -57,7 +56,7 @@ uint32_t render_target_first_bound_slot(const CommandBuffer& buffer) {
 
 bool graphics_debug_dump_enabled() {
 	return Config::GraphicsDebugDumpEnabled() &&
-	       Config::GetPrintfDirection() != Config::OutputDirection::Silent;
+	       Config::GetPrintfDirection() != Config::LogDirection::Silent;
 }
 
 void uc_print(const char* func, const HW::UserConfig& uc) {
@@ -89,91 +88,75 @@ void uc_check(const HW::UserConfig& uc) {
 	EXIT_NOT_IMPLEMENTED(user_en.vgpr3 != false);
 }
 
-void sh_print(const char* func, const HW::Shader& /*uc*/) {
-	LOGF("%s\n", func);
-}
+std::string rt_print(const char* func, const HW::RenderTarget& rt) {
+	std::string dst;
+	dst.reserve(4096);
 
-std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt) {
-	std::vector<std::string> dst;
-	dst.reserve(40);
+	dst += fmt::format("{}\n", func);
 
-	dst.push_back(fmt::format("{}\n", func));
+	dst += fmt::format("\t base.addr                       = 0x{:016x}\n", rt.base.addr);
+	dst += fmt::format("\t view.base_array_slice_index     = 0x{:08x}\n",
+	                  rt.view.base_array_slice_index);
+	dst += fmt::format("\t view.last_array_slice_index     = 0x{:08x}\n",
+	                  rt.view.last_array_slice_index);
+	dst += fmt::format("\t view.current_mip_level          = 0x{:08x}\n", rt.view.current_mip_level);
+	dst += fmt::format("\t info.fmask_compression_enable   = {}\n",
+	                  rt.info.fmask_compression_enable ? "true" : "false");
 
-	dst.push_back(fmt::format("\t base.addr                       = 0x{:016x}\n", rt.base.addr));
-	dst.push_back(fmt::format("\t view.base_array_slice_index     = 0x{:08x}\n",
-	                          rt.view.base_array_slice_index));
-	dst.push_back(fmt::format("\t view.last_array_slice_index     = 0x{:08x}\n",
-	                          rt.view.last_array_slice_index));
-	dst.push_back(
-	    fmt::format("\t view.current_mip_level          = 0x{:08x}\n", rt.view.current_mip_level));
-	dst.push_back(fmt::format("\t info.fmask_compression_enable   = {}\n",
-	                          rt.info.fmask_compression_enable ? "true" : "false"));
+	dst += fmt::format("\t info.fmask_data_compression_disable = {}\n",
+	                  rt.info.fmask_data_compression_disable ? "true" : "false");
+	dst += fmt::format("\t info.fmask_one_frag_mode        = {}\n",
+	                  rt.info.fmask_one_frag_mode ? "true" : "false");
 
-	// dst.push_back(fmt::format("\t info.fmask_compression_mode     = 0x{:08x}\n", //
-	// rt.info.fmask_compression_mode));
-	dst.push_back(fmt::format("\t info.fmask_data_compression_disable = {}\n",
-	                          rt.info.fmask_data_compression_disable ? "true" : "false"));
-	dst.push_back(fmt::format("\t info.fmask_one_frag_mode        = {}\n",
-	                          rt.info.fmask_one_frag_mode ? "true" : "false"));
-
-	dst.push_back(fmt::format("\t info.cmask_fast_clear_enable    = {}\n",
-	                          rt.info.cmask_fast_clear_enable ? "true" : "false"));
-	dst.push_back(fmt::format("\t info.dcc_compression_enable     = {}\n",
-	                          rt.info.dcc_compression_enable ? "true" : "false"));
-	dst.push_back(fmt::format("\t info.format                     = 0x{:08x}\n",
-	                          static_cast<uint32_t>(rt.info.format)));
-	dst.push_back(fmt::format("\t info.channel_type               = 0x{:08x}\n",
-	                          static_cast<uint32_t>(rt.info.channel_type)));
-	dst.push_back(fmt::format("\t info.channel_order              = 0x{:08x}\n",
-	                          static_cast<uint32_t>(rt.info.channel_order)));
-	dst.push_back(fmt::format("\t info.blend_bypa                 = {}\n",
-	                          rt.info.blend_bypass ? "true" : "false"));
-	dst.push_back(fmt::format("\t info.blend_clamp                = {}\n",
-	                          rt.info.blend_clamp ? "true" : "false"));
-	dst.push_back(fmt::format("\t info.round_mode                 = {}\n",
-	                          rt.info.round_mode ? "true" : "false"));
-	dst.push_back(fmt::format("\t attrib.force_dest_alpha_to_one  = {}\n",
-	                          rt.attrib.force_dest_alpha_to_one ? "true" : "false"));
-	dst.push_back(
-	    fmt::format("\t attrib.num_samples              = 0x{:08x}\n", rt.attrib.num_samples));
-	dst.push_back(
-	    fmt::format("\t attrib.num_fragments            = 0x{:08x}\n", rt.attrib.num_fragments));
-	dst.push_back(fmt::format("\t attrib2.width                   = 0x{:08x}\n", rt.attrib2.width));
-	dst.push_back(
-	    fmt::format("\t attrib2.height                  = 0x{:08x}\n", rt.attrib2.height));
-	dst.push_back(
-	    fmt::format("\t attrib2.num_mip_levels          = 0x{:08x}\n", rt.attrib2.num_mip_levels));
-	dst.push_back(fmt::format("\t attrib3.depth                   = 0x{:08x}\n", rt.attrib3.depth));
-	dst.push_back(fmt::format("\t attrib3.tile_mode               = 0x{:08x}\n",
-	                          static_cast<uint32_t>(rt.attrib3.tile_mode)));
-	dst.push_back(
-	    fmt::format("\t attrib3.dimension               = 0x{:08x}\n", rt.attrib3.dimension));
-	dst.push_back(fmt::format("\t attrib3.metadata_pipe_aligned   = {}\n",
-	                          rt.attrib3.metadata_pipe_aligned ? "true" : "false"));
-	dst.push_back(fmt::format("\t attrib3.write_vrs_rate_hint_to_cmask = {}\n",
-	                          rt.attrib3.write_vrs_rate_hint_to_cmask ? "true" : "false"));
-	dst.push_back(fmt::format("\t dcc.max_uncompressed_block_size = 0x{:08x}\n",
-	                          rt.dcc.max_uncompressed_block_size));
-	dst.push_back(fmt::format("\t dcc.max_compressed_block_size   = 0x{:08x}\n",
-	                          rt.dcc.max_compressed_block_size));
-	dst.push_back(
-	    fmt::format("\t dcc.color_transform             = 0x{:08x}\n", rt.dcc.color_transform));
-	dst.push_back(fmt::format("\t dcc.overwrite_combiner_disable  = {}\n",
-	                          rt.dcc.overwrite_combiner_disable ? "true" : "false"));
-	dst.push_back(fmt::format("\t dcc.independent_block_size      = 0x{:02x}\n",
-	                          static_cast<uint8_t>(rt.dcc.independent_block_size)));
-	dst.push_back(fmt::format("\t data_write_on_dcc_clear_to_reg  = {}\n",
-	                          rt.dcc.data_write_on_dcc_clear_to_reg ? "true" : "false"));
-	dst.push_back(fmt::format("\t dcc.dcc_clear_key_enable        = {}\n",
-	                          rt.dcc.dcc_clear_key_enable ? "true" : "false"));
-	dst.push_back(fmt::format("\t cmask.addr                      = 0x{:016x}\n", rt.cmask.addr));
-	dst.push_back(fmt::format("\t fmask.addr                      = 0x{:016x}\n", rt.fmask.addr));
-	dst.push_back(
-	    fmt::format("\t clear_word0.word0               = 0x{:08x}\n", rt.clear_word0.word0));
-	dst.push_back(
-	    fmt::format("\t clear_word1.word1               = 0x{:08x}\n", rt.clear_word1.word1));
-	dst.push_back(
-	    fmt::format("\t dcc_addr.addr                   = 0x{:016x}\n", rt.dcc_addr.addr));
+	dst += fmt::format("\t info.cmask_fast_clear_enable    = {}\n",
+	                  rt.info.cmask_fast_clear_enable ? "true" : "false");
+	dst += fmt::format("\t info.dcc_compression_enable     = {}\n",
+	                  rt.info.dcc_compression_enable ? "true" : "false");
+	dst += fmt::format("\t info.format                     = 0x{:08x}\n",
+	                  static_cast<uint32_t>(rt.info.format));
+	dst += fmt::format("\t info.channel_type               = 0x{:08x}\n",
+	                  static_cast<uint32_t>(rt.info.channel_type));
+	dst += fmt::format("\t info.channel_order              = 0x{:08x}\n",
+	                  static_cast<uint32_t>(rt.info.channel_order));
+	dst += fmt::format("\t info.blend_bypa                 = {}\n",
+	                  rt.info.blend_bypass ? "true" : "false");
+	dst += fmt::format("\t info.blend_clamp                = {}\n",
+	                  rt.info.blend_clamp ? "true" : "false");
+	dst += fmt::format("\t info.round_mode                 = {}\n",
+	                  rt.info.round_mode ? "true" : "false");
+	dst += fmt::format("\t attrib.force_dest_alpha_to_one  = {}\n",
+	                  rt.attrib.force_dest_alpha_to_one ? "true" : "false");
+	dst += fmt::format("\t attrib.num_samples              = 0x{:08x}\n", rt.attrib.num_samples);
+	dst += fmt::format("\t attrib.num_fragments            = 0x{:08x}\n", rt.attrib.num_fragments);
+	dst += fmt::format("\t attrib2.width                   = 0x{:08x}\n", rt.attrib2.width);
+	dst += fmt::format("\t attrib2.height                  = 0x{:08x}\n", rt.attrib2.height);
+	dst += fmt::format("\t attrib2.num_mip_levels          = 0x{:08x}\n", rt.attrib2.num_mip_levels);
+	dst += fmt::format("\t attrib3.depth                   = 0x{:08x}\n", rt.attrib3.depth);
+	dst += fmt::format("\t attrib3.tile_mode               = 0x{:08x}\n",
+	                  static_cast<uint32_t>(rt.attrib3.tile_mode));
+	dst += fmt::format("\t attrib3.dimension               = 0x{:08x}\n", rt.attrib3.dimension);
+	dst += fmt::format("\t attrib3.metadata_pipe_aligned   = {}\n",
+	                  rt.attrib3.metadata_pipe_aligned ? "true" : "false");
+	dst += fmt::format("\t attrib3.write_vrs_rate_hint_to_cmask = {}\n",
+	                  rt.attrib3.write_vrs_rate_hint_to_cmask ? "true" : "false");
+	dst += fmt::format("\t dcc.max_uncompressed_block_size = 0x{:08x}\n",
+	                  rt.dcc.max_uncompressed_block_size);
+	dst += fmt::format("\t dcc.max_compressed_block_size   = 0x{:08x}\n",
+	                  rt.dcc.max_compressed_block_size);
+	dst += fmt::format("\t dcc.color_transform             = 0x{:08x}\n", rt.dcc.color_transform);
+	dst += fmt::format("\t dcc.overwrite_combiner_disable  = {}\n",
+	                  rt.dcc.overwrite_combiner_disable ? "true" : "false");
+	dst += fmt::format("\t dcc.independent_block_size      = 0x{:02x}\n",
+	                  static_cast<uint8_t>(rt.dcc.independent_block_size));
+	dst += fmt::format("\t data_write_on_dcc_clear_to_reg  = {}\n",
+	                  rt.dcc.data_write_on_dcc_clear_to_reg ? "true" : "false");
+	dst += fmt::format("\t dcc.dcc_clear_key_enable        = {}\n",
+	                  rt.dcc.dcc_clear_key_enable ? "true" : "false");
+	dst += fmt::format("\t cmask.addr                      = 0x{:016x}\n", rt.cmask.addr);
+	dst += fmt::format("\t fmask.addr                      = 0x{:016x}\n", rt.fmask.addr);
+	dst += fmt::format("\t clear_word0.word0               = 0x{:08x}\n", rt.clear_word0.word0);
+	dst += fmt::format("\t clear_word1.word1               = 0x{:08x}\n", rt.clear_word1.word1);
+	dst += fmt::format("\t dcc_addr.addr                   = 0x{:016x}\n", rt.dcc_addr.addr);
 
 	return dst;
 }
@@ -478,7 +461,6 @@ static void McCheck(const HW::ModeControl& c) {
 			logged = true;
 		}
 	}
-	EXIT_NOT_IMPLEMENTED(c.provoking_vtx_last != false);
 	EXIT_NOT_IMPLEMENTED(c.persp_corr_dis != false);
 }
 
@@ -505,16 +487,7 @@ static void BcPrint(const char* func, const HW::BlendControl& c, const HW::Blend
 	     color.red, color.green, color.blue, color.alpha, cc.mode, cc.op);
 }
 
-static void BcCheck(const HW::BlendControl& /*c*/, const HW::BlendColor& color,
-                    const HW::ColorControl& cc) {
-	// EXIT_NOT_IMPLEMENTED(c.color_srcblend != 0);
-	// EXIT_NOT_IMPLEMENTED(c.color_comb_fcn != 0);
-	// EXIT_NOT_IMPLEMENTED(c.color_destblend != 0);
-	// EXIT_NOT_IMPLEMENTED(c.alpha_srcblend != 0);
-	// EXIT_NOT_IMPLEMENTED(c.alpha_comb_fcn != 0);
-	// EXIT_NOT_IMPLEMENTED(c.alpha_destblend != 0);
-	// EXIT_NOT_IMPLEMENTED(c.separate_alpha_blend != false);
-	// EXIT_NOT_IMPLEMENTED(c.enable != false);
+static void BcCheck(const HW::BlendColor& color, const HW::ColorControl& cc) {
 	if (color.red != 0.0f || color.green != 0.0f || color.blue != 0.0f || color.alpha != 0.0f) {
 		static bool logged = false;
 		if (!logged) {
@@ -577,29 +550,6 @@ static void DPrint(const char* func, const HW::DepthControl& c, const HW::Stenci
 	     sm.stencil_opval_bf);
 }
 
-static void DCheck(const HW::DepthControl& c, const HW::StencilControl& s,
-                   const HW::StencilMask& /*sm*/) {
-	// EXIT_NOT_IMPLEMENTED(c.stencil_enable != false);
-	// EXIT_NOT_IMPLEMENTED(c.z_enable != false);
-	// EXIT_NOT_IMPLEMENTED(c.z_write_enable != false);
-	// EXIT_NOT_IMPLEMENTED(c.zfunc != 0);
-	// Back-face stencil state is handled separately when enabled.
-	// EXIT_NOT_IMPLEMENTED(c.stencilfunc != 0);
-	// EXIT_NOT_IMPLEMENTED(c.stencilfunc_bf != 0);
-	// EXIT_NOT_IMPLEMENTED(s.stencil_fail != 0);
-	// EXIT_NOT_IMPLEMENTED(s.stencil_zpass != 0);
-	// EXIT_NOT_IMPLEMENTED(s.stencil_zfail != 0);
-	// Back-face stencil ops may legitimately differ from the front-face ops.
-	// EXIT_NOT_IMPLEMENTED(sm.stencil_testval != 0);
-	// EXIT_NOT_IMPLEMENTED(sm.stencil_mask != 0);
-	// EXIT_NOT_IMPLEMENTED(sm.stencil_writemask != 0);
-	// EXIT_NOT_IMPLEMENTED(sm.stencil_opval != 0);
-	// EXIT_NOT_IMPLEMENTED(sm.stencil_testval_bf != 0);
-	// EXIT_NOT_IMPLEMENTED(sm.stencil_mask_bf != 0);
-	// EXIT_NOT_IMPLEMENTED(sm.stencil_writemask_bf != 0);
-	// EXIT_NOT_IMPLEMENTED(sm.stencil_opval_bf != 0);
-}
-
 static void EqaaPrint(const char* func, const HW::EqaaControl& c) {
 	LOGF("%s\n", func);
 
@@ -905,14 +855,10 @@ void hw_check(const CommandBuffer& buffer) {
 	const auto& hw      = buffer.GetRegisters();
 	const auto  rt_slot = render_target_first_bound_slot(buffer);
 	const auto& rt      = hw.GetRenderTarget(rt_slot);
-	const auto& bc      = hw.GetBlendControl(rt_slot);
 	const auto& bclr    = hw.GetBlendColor();
 	const auto& vp      = hw.GetScreenViewport();
 	const auto& c       = hw.GetClipControl();
 	const auto& rc      = hw.GetRenderControl();
-	const auto& d       = hw.GetDepthControl();
-	const auto& s       = hw.GetStencilControl();
-	const auto& sm      = hw.GetStencilMask();
 	const auto& mc      = hw.GetModeControl();
 	const auto& eqaa    = hw.GetEqaaControl();
 	const auto& cc      = hw.GetColorControl();
@@ -938,11 +884,10 @@ void hw_check(const CommandBuffer& buffer) {
 	log_phase("rc");
 	RcCheck(rc);
 	log_phase("depth");
-	DCheck(d, s, sm);
 	log_phase("mode");
 	McCheck(mc);
 	log_phase("blend");
-	BcCheck(bc, bclr, cc);
+	BcCheck(bclr, cc);
 	log_phase("eqaa");
 	EqaaCheck(eqaa, ac);
 	log_phase("aa");
@@ -1001,7 +946,7 @@ void hw_print(const CommandBuffer& buffer) {
 		     hw.GetRenderTargetMask(), hw.GetDepthClearValue(), hw.GetStencilClearValue(),
 		     hw.GetLineWidth(), hw.GetPrimitiveResetIndex());
 
-		LOGF("%s", Common::Concat(rt_print("RenderTraget:", rt), "").c_str());
+		LOGF("%s", rt_print("RenderTraget:", rt).c_str());
 
 		ZPrint("DepthRenderTraget:", z);
 		VpPrint("ScreenViewport:", vp, smc);

src/graphics/host_gpu/renderer/debug.h

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

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -10,6 +10,7 @@
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
+#include "graphics/host_gpu/renderer/commandScheduler.h"
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
@@ -18,6 +19,7 @@
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <algorithm>
+#include <array>
 #include <atomic>
 #include <cmath>
 #include <cstdarg>
@@ -44,8 +46,12 @@ static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t
 	switch (static_cast<Prospero::StencilOp>(value)) {
 		case Prospero::StencilOp::kKeep: return vk::StencilOp::eKeep;
 		case Prospero::StencilOp::kZero: return vk::StencilOp::eZero;
-		case Prospero::StencilOp::kReplaceTest:
-		case Prospero::StencilOp::kReplaceOp: return vk::StencilOp::eReplace;
+		case Prospero::StencilOp::kReplaceTest: return vk::StencilOp::eReplace;
+		case Prospero::StencilOp::kReplaceOp:
+			if ((op_value & write_mask) == 0) {
+				return vk::StencilOp::eZero;
+			}
+			return vk::StencilOp::eReplace;
 		case Prospero::StencilOp::kAddClamp: return vk::StencilOp::eIncrementAndClamp;
 		case Prospero::StencilOp::kSubClamp: return vk::StencilOp::eDecrementAndClamp;
 		case Prospero::StencilOp::kInvert: return vk::StencilOp::eInvert;
@@ -65,9 +71,38 @@ static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t
 	}
 }
 
-static bool UsesStencilOpValue(uint8_t fail, uint8_t pass, uint8_t depth_fail) {
-	constexpr auto replace_op = static_cast<uint8_t>(Prospero::StencilOp::kReplaceOp);
-	return fail == replace_op || pass == replace_op || depth_fail == replace_op;
+static vk::StencilOpState ConvertStencilState(
+    uint8_t compare, const std::array<uint8_t, 3>& operations, uint8_t op_value,
+    const vk::StencilOpState& state) {
+	const auto test_value = state.reference;
+	auto reference       = test_value;
+	auto required_bits   = state.compareMask;
+	if (compare == static_cast<uint8_t>(vk::CompareOp::eAlways) ||
+	    compare == static_cast<uint8_t>(vk::CompareOp::eNever)) {
+		required_bits = 0;
+	}
+	std::array<vk::StencilOp, 3> converted {};
+	for (size_t i = 0; i < operations.size(); i++) {
+		converted[i] = ConvertStencilOp(operations[i], state.writeMask, op_value);
+		if (converted[i] != vk::StencilOp::eReplace) {
+			continue;
+		}
+		auto replacement = test_value;
+		if (static_cast<Prospero::StencilOp>(operations[i]) == Prospero::StencilOp::kReplaceOp) {
+			replacement = op_value;
+		}
+		if (((reference ^ replacement) & required_bits & state.writeMask) != 0) {
+			DepthFatal("unsupported stencil replacement: compare=%u, compare mask=0x%02" PRIx32
+			           ", write mask=0x%02" PRIx32 ", operation value=0x%02" PRIx8
+			           ", test value=0x%02" PRIx32,
+			           compare, state.compareMask, state.writeMask, op_value, test_value);
+		}
+		// Vulkan shares one reference between comparison and every replacement on this face.
+		reference = (reference & ~state.writeMask) | (replacement & state.writeMask);
+		required_bits |= state.writeMask;
+	}
+	return {converted[0], converted[1], converted[2], static_cast<vk::CompareOp>(compare),
+	        state.compareMask, state.writeMask, reference};
 }
 
 [[nodiscard]] static vk::Format ResolveHostDepthAttachmentFormat(const CommandBuffer&     buffer,
@@ -96,53 +131,17 @@ static bool UsesStencilOpValue(uint8_t fail, uint8_t pass, uint8_t depth_fail) {
 	return vk::Format::eUndefined;
 }
 
-// NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer& buffer,
-                                              RenderDepthInfo& r) {
-	KYTY_PROFILER_FUNCTION();
-	(void)submit_id;
-	const auto& hw          = buffer.GetRegisters();
-	const auto& z           = hw.GetDepthRenderTarget();
-	const auto& rc          = hw.GetRenderControl();
-	const auto& dc          = hw.GetDepthControl();
-	const auto& sc          = hw.GetStencilControl();
-	const auto& sm          = hw.GetStencilMask();
-	const bool  has_stencil = z.stencil_info.format != Prospero::StencilFormat::kInvalid;
-	const bool depth_active = dc.z_enable || dc.depth_bounds_enable ||
-	                          rc.depth_clear_enable || rc.copy_depth_to_color;
-	const bool stencil_active =
-	    has_stencil && (dc.stencil_enable || rc.stencil_clear_enable || rc.copy_stencil_to_color);
-	if (!depth_active && !stencil_active) {
-		return;
-	}
+static TextureCache::ImageDesc MakeDepthTargetDesc(const CommandBuffer& buffer,
+                                                  const HW::DepthRenderTarget& z,
+                                                  bool write_buffer = false) {
+	const bool has_stencil = z.stencil_info.format != Prospero::StencilFormat::kInvalid;
+	const auto depth_address = write_buffer ? z.z_write_base_addr : z.z_read_base_addr;
+	const auto stencil_address =
+	    write_buffer ? z.stencil_write_base_addr : z.stencil_read_base_addr;
 	if (!z.z_info.HasValidTextureCompatibility() ||
 	    !z.stencil_info.HasValidTextureCompatibility()) {
 		DepthFatal("invalid PS5 depth texture-compatibility encoding");
 	}
-	const bool attachment_unbound =
-	    z.z_info.format == Prospero::DepthFormat::kInvalid &&
-	    z.stencil_info.format == Prospero::StencilFormat::kInvalid && z.z_info.num_samples == 0 &&
-	    z.z_info.texture_compatibility == Prospero::TextureCompatiblePlaneCompression::kDisable &&
-	    !z.z_info.expclear_enabled && !z.z_info.partially_resident && z.z_info.max_mip_level == 0 &&
-	    z.stencil_info.texture_compatibility == Prospero::TextureCompatibleStencil::kDisable &&
-	    !z.stencil_info.expclear_enabled && !z.stencil_info.partially_resident &&
-	    z.depth_view.slice_start == 0 && z.depth_view.slice_max == 0 &&
-	    z.depth_view.current_mip_level == 0 && !z.depth_view.depth_write_disable &&
-	    !z.depth_view.stencil_write_disable && z.z_read_base_addr == 0 && z.z_write_base_addr == 0 &&
-	    z.stencil_read_base_addr == 0 && z.stencil_write_base_addr == 0 &&
-	    z.htile_data_base_addr == 0 &&
-	    // DB_DEPTH_SIZE_XY is independent state and may remain programmed after the attachment
-	    // formats and addresses are unbound. A zero encoding is the valid 1x1 value, so its
-	    // presence alone must not manufacture a depth attachment.
-	    !z.z_info.htile_acceleration && z.shading_rate_encoding == 0 && z.size.x_max == 0 &&
-	    z.size.y_max == 0;
-	if (attachment_unbound) {
-		static std::atomic_bool logged = false;
-		if (!logged.exchange(true, std::memory_order_relaxed)) {
-			LOGF("DepthTarget: ignoring enabled depth/stencil state without a bound attachment\n");
-		}
-		return;
-	}
 	const bool has_htile = z.z_info.htile_acceleration;
 	const bool unsupported_shading_rate_encoding =
 	    z.shading_rate_encoding > 1 || (z.shading_rate_encoding != 0 && !has_htile);
@@ -160,23 +159,16 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 			DepthFatal("invalid depth view: base=%u last=%u", z.depth_view.slice_start,
 			           z.depth_view.slice_max);
 	}
-	if (rc.copy_depth_to_color || rc.copy_stencil_to_color || rc.copy_centroid ||
-	    rc.copy_sample != 0 || z.z_info.expclear_enabled ||
-	    z.stencil_info.expclear_enabled || z.z_info.partially_resident ||
+	if (z.z_info.expclear_enabled || z.stencil_info.expclear_enabled ||
+	    z.z_info.partially_resident ||
 	    z.stencil_info.partially_resident || z.z_info.max_mip_level != 0 ||
 	    z.depth_view.current_mip_level != 0 || unsupported_shading_rate_encoding ||
-	    z.z_read_base_addr == 0 ||
-	    (!z.depth_view.depth_write_disable && z.z_write_base_addr != z.z_read_base_addr) ||
-	    (z.z_read_base_addr & 0xffffu) != 0 ||
-	    dc.zfunc > static_cast<uint8_t>(vk::CompareOp::eAlways)) {
+	    depth_address == 0 || (depth_address & 0xffffu) != 0) {
 		DepthFatal("unsupported depth register state");
 	}
 	if (has_stencil) {
 		if (z.stencil_info.format != Prospero::StencilFormat::k8UInt || !htile_stencil_compat ||
-		    z.stencil_read_base_addr == 0 ||
-		    (!z.depth_view.stencil_write_disable &&
-		     z.stencil_write_base_addr != z.stencil_read_base_addr) ||
-		    (z.stencil_read_base_addr & 0xffffu) != 0) {
+		    stencil_address == 0 || (stencil_address & 0xffffu) != 0) {
 			DepthFatal("unsupported stencil attachment state");
 		}
 	} else if (z.stencil_read_base_addr != 0 || z.stencil_write_base_addr != 0) {
@@ -229,67 +221,16 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 	const auto depth_backing_size   = depth_size.size * view.image_layers;
 	const auto stencil_backing_size = stencil_size.size * view.image_layers;
 	const auto htile_backing_size   = htile_size.size * view.image_layers;
-	if (!GuestRange {z.z_read_base_addr, depth_backing_size}.Valid() ||
-	    (has_stencil && !GuestRange {z.stencil_read_base_addr, stencil_backing_size}.Valid()) ||
+	if (!GuestRange {depth_address, depth_backing_size}.Valid() ||
+	    (has_stencil && !GuestRange {stencil_address, stencil_backing_size}.Valid()) ||
 	    (has_htile && !GuestRange {z.htile_data_base_addr, htile_backing_size}.Valid())) {
 		DepthFatal("layered depth backing range is invalid");
 	}
-	r.depth_clear_enable      = rc.depth_clear_enable;
-	r.depth_meta_clear_enable = false;
-	r.depth_load_clear_enable = r.depth_clear_enable;
-	r.depth_clear_value       = hw.GetDepthClearValue();
-	r.depth_test_enable       = dc.z_enable;
-	r.depth_write_enable      = r.depth_test_enable && dc.z_write_enable &&
-	                            !z.depth_view.depth_write_disable && !r.depth_clear_enable;
-	r.depth_compare_op        = static_cast<vk::CompareOp>(dc.zfunc);
-
-	r.depth_bounds_test_enable = dc.depth_bounds_enable;
-	r.depth_min_bounds         = hw.GetDepthBoundsMin();
-	r.depth_max_bounds         = hw.GetDepthBoundsMax();
-
-	r.stencil_clear_enable =
-	    has_stencil && rc.stencil_clear_enable && !z.depth_view.stencil_write_disable;
-	r.stencil_clear_value = hw.GetStencilClearValue();
-	r.stencil_test_enable = has_stencil && dc.stencil_enable;
-	if (r.stencil_test_enable) {
-		const bool stencil_ops_disabled =
-		    rc.stencil_clear_enable || z.depth_view.stencil_write_disable;
-		const uint8_t front_write_mask = stencil_ops_disabled ? 0 : sm.stencil_writemask;
-		const uint8_t back_write_mask  = stencil_ops_disabled ? 0 : sm.stencil_writemask_bf;
-		if (dc.stencilfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
-		    (dc.backface_enable &&
-		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways)) ||
-		    (front_write_mask != 0 &&
-		     UsesStencilOpValue(sc.stencil_fail, sc.stencil_zpass, sc.stencil_zfail) &&
-		     sm.stencil_opval != sm.stencil_testval) ||
-		    (dc.backface_enable && back_write_mask != 0 &&
-		     UsesStencilOpValue(sc.stencil_fail_bf, sc.stencil_zpass_bf, sc.stencil_zfail_bf) &&
-		     sm.stencil_opval_bf != sm.stencil_testval_bf)) {
-			DepthFatal("unsupported stencil compare or replacement state");
-		}
-		r.stencil_static_front = {
-		    ConvertStencilOp(sc.stencil_fail, front_write_mask, sm.stencil_opval),
-		    ConvertStencilOp(sc.stencil_zpass, front_write_mask, sm.stencil_opval),
-		    ConvertStencilOp(sc.stencil_zfail, front_write_mask, sm.stencil_opval),
-		    static_cast<vk::CompareOp>(dc.stencilfunc)};
-		r.stencil_dynamic_front = {sm.stencil_mask, front_write_mask, sm.stencil_testval};
-		if (dc.backface_enable) {
-			r.stencil_static_back = {
-			    ConvertStencilOp(sc.stencil_fail_bf, back_write_mask, sm.stencil_opval_bf),
-			    ConvertStencilOp(sc.stencil_zpass_bf, back_write_mask, sm.stencil_opval_bf),
-			    ConvertStencilOp(sc.stencil_zfail_bf, back_write_mask, sm.stencil_opval_bf),
-			    static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
-			r.stencil_dynamic_back = {sm.stencil_mask_bf, back_write_mask, sm.stencil_testval_bf};
-		} else {
-			r.stencil_static_back  = r.stencil_static_front;
-			r.stencil_dynamic_back = r.stencil_dynamic_front;
-		}
-	}
 	TextureCache::ImageDesc desc {};
 	desc.type                 = TextureCache::BindingType::DepthTarget;
-	desc.info.data            = {z.z_read_base_addr, depth_backing_size};
+	desc.info.data            = {depth_address, depth_backing_size};
 	desc.info.stencil =
-	    has_stencil ? GuestRange {z.stencil_read_base_addr, stencil_backing_size} : GuestRange {};
+	    has_stencil ? GuestRange {stencil_address, stencil_backing_size} : GuestRange {};
 	desc.info.pixel_format    = format;
 	desc.info.guest_format    = guest_format;
 	desc.info.type            = Prospero::ImageType::kColor2D;
@@ -314,12 +255,164 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 	desc.view_info.base_layer  = view.base_layer;
 	desc.view_info.layer_count = view.layer_count;
 	desc.view_info.usage       = vk::ImageUsageFlagBits::eDepthStencilAttachment;
-	r.desc                     = std::move(desc);
-	auto& cache                = m_context.GetTextureCache();
-	r.image_id                 = cache.FindImage(r.desc);
+	return desc;
+}
+
+// NOLINTNEXTLINE(readability-function-cognitive-complexity)
+void RenderExecutor::ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepthInfo& r) {
+	KYTY_PROFILER_FUNCTION();
+	const auto& hw          = buffer.GetRegisters();
+	const auto& z           = hw.GetDepthRenderTarget();
+	const auto& rc          = hw.GetRenderControl();
+	const auto& dc          = hw.GetDepthControl();
+	const auto& sc          = hw.GetStencilControl();
+	const auto& sm          = hw.GetStencilMask();
+	const bool  has_stencil = z.stencil_info.format != Prospero::StencilFormat::kInvalid;
+	const bool depth_active = dc.z_enable || dc.depth_bounds_enable ||
+	                          rc.depth_clear_enable || rc.copy_depth_to_color;
+	const bool stencil_active =
+	    has_stencil && (dc.stencil_enable || rc.stencil_clear_enable || rc.copy_stencil_to_color);
+	if (!depth_active && !stencil_active) {
+		return;
+	}
+	const bool attachment_unbound =
+	    z.z_info.format == Prospero::DepthFormat::kInvalid &&
+	    z.stencil_info.format == Prospero::StencilFormat::kInvalid && z.z_info.num_samples == 0 &&
+	    z.z_info.texture_compatibility == Prospero::TextureCompatiblePlaneCompression::kDisable &&
+	    !z.z_info.expclear_enabled && !z.z_info.partially_resident && z.z_info.max_mip_level == 0 &&
+	    z.stencil_info.texture_compatibility == Prospero::TextureCompatibleStencil::kDisable &&
+	    !z.stencil_info.expclear_enabled && !z.stencil_info.partially_resident &&
+	    z.depth_view.slice_start == 0 && z.depth_view.slice_max == 0 &&
+	    z.depth_view.current_mip_level == 0 && !z.depth_view.depth_write_disable &&
+	    !z.depth_view.stencil_write_disable && z.z_read_base_addr == 0 && z.z_write_base_addr == 0 &&
+	    z.stencil_read_base_addr == 0 && z.stencil_write_base_addr == 0 &&
+	    z.htile_data_base_addr == 0 &&
+	    // DB_DEPTH_SIZE_XY is independent state and may remain programmed after the attachment
+	    // formats and addresses are unbound. A zero encoding is the valid 1x1 value, so its
+	    // presence alone must not manufacture a depth attachment.
+	    !z.z_info.htile_acceleration && z.shading_rate_encoding == 0 && z.size.x_max == 0 &&
+	    z.size.y_max == 0;
+	if (attachment_unbound) {
+		static std::atomic_bool logged = false;
+		if (!logged.exchange(true, std::memory_order_relaxed)) {
+			LOGF("DepthTarget: ignoring enabled depth/stencil state without a bound attachment\n");
+		}
+		return;
+	}
+	if (rc.copy_depth_to_color || rc.copy_stencil_to_color || rc.copy_centroid ||
+	    rc.copy_sample != 0 || dc.zfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
+	    (!z.depth_view.depth_write_disable && z.z_write_base_addr != z.z_read_base_addr) ||
+	    (has_stencil && !z.depth_view.stencil_write_disable &&
+	     z.stencil_write_base_addr != z.stencil_read_base_addr)) {
+		DepthFatal("unsupported depth register state");
+	}
+	r.desc = MakeDepthTargetDesc(buffer, z);
+	r.depth_clear_enable      = rc.depth_clear_enable;
+	r.depth_load_clear_enable = r.depth_clear_enable;
+	r.depth_clear_value       = hw.GetDepthClearValue();
+	r.depth_test_enable       = dc.z_enable;
+	r.depth_write_enable      = r.depth_test_enable && dc.z_write_enable &&
+	                            !z.depth_view.depth_write_disable && !r.depth_clear_enable;
+	r.depth_compare_op        = static_cast<vk::CompareOp>(dc.zfunc);
+
+	r.depth_bounds_test_enable = dc.depth_bounds_enable;
+	r.depth_min_bounds         = hw.GetDepthBoundsMin();
+	r.depth_max_bounds         = hw.GetDepthBoundsMax();
+
+	r.stencil_clear_enable =
+	    has_stencil && rc.stencil_clear_enable && !z.depth_view.stencil_write_disable;
+	r.stencil_clear_value = hw.GetStencilClearValue();
+	r.stencil_test_enable = has_stencil && dc.stencil_enable;
+	if (r.stencil_test_enable) {
+		const bool stencil_ops_disabled =
+		    rc.stencil_clear_enable || z.depth_view.stencil_write_disable;
+		const uint8_t front_write_mask = stencil_ops_disabled ? 0 : sm.stencil_writemask;
+		const uint8_t back_write_mask  = stencil_ops_disabled ? 0 : sm.stencil_writemask_bf;
+		if (dc.stencilfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
+		    (dc.backface_enable &&
+		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways))) {
+			DepthFatal("unsupported stencil compare state");
+		}
+		r.stencil_front = ConvertStencilState(
+		    dc.stencilfunc, {sc.stencil_fail, sc.stencil_zpass, sc.stencil_zfail},
+		    sm.stencil_opval, {.compareMask = sm.stencil_mask, .writeMask = front_write_mask,
+		                       .reference = sm.stencil_testval});
+		if (dc.backface_enable) {
+			r.stencil_back = ConvertStencilState(
+			    dc.stencilfunc_bf, {sc.stencil_fail_bf, sc.stencil_zpass_bf, sc.stencil_zfail_bf},
+			    sm.stencil_opval_bf, {.compareMask = sm.stencil_mask_bf, .writeMask = back_write_mask,
+			                          .reference = sm.stencil_testval_bf});
+		} else {
+			r.stencil_back = r.stencil_front;
+		}
+	}
+	auto& cache = m_context.GetTextureCache();
+	r.image_id = cache.FindImage(r.desc);
 	BindRenderTarget(r.image_id);
 }
 
+bool RenderExecutor::DepthStencilCopy(CommandBuffer& buffer) {
+	const auto& hw       = buffer.GetRegisters();
+	const auto& z        = hw.GetDepthRenderTarget();
+	const auto& override = hw.GetDepthRenderOverride();
+	if (hw.GetColorControl().mode != 0) {
+		return false;
+	}
+	const bool depth_copy = override.force_z_dirty && override.force_z_valid &&
+	                        z.z_info.format != Prospero::DepthFormat::kInvalid &&
+	                        z.z_read_base_addr != 0 && z.z_write_base_addr != 0 &&
+	                        z.z_read_base_addr != z.z_write_base_addr;
+	const bool stencil_copy = override.force_stencil_dirty && override.force_stencil_valid &&
+	                          z.stencil_info.format != Prospero::StencilFormat::kInvalid &&
+	                          z.stencil_read_base_addr != 0 && z.stencil_write_base_addr != 0 &&
+	                          z.stencil_read_base_addr != z.stencil_write_base_addr;
+	if (!depth_copy && !stencil_copy) {
+		return false;
+	}
+
+	auto  read_desc  = MakeDepthTargetDesc(buffer, z);
+	auto  write_desc = MakeDepthTargetDesc(buffer, z, true);
+	auto& cache      = m_context.GetTextureCache();
+	const auto read_id = cache.FindImage(read_desc);
+	BindRenderTarget(read_id);
+	const auto write_id = cache.FindImage(write_desc);
+	BindRenderTarget(write_id);
+	cache.UpdateImage(read_id);
+	cache.UpdateImage(write_id);
+	cache.MarkGpuWritten(write_id);
+	auto& source      = cache.GetImage(read_id);
+	auto& destination = cache.GetImage(write_id);
+	EXIT_IF(read_id == write_id || source.backing.format != destination.backing.format);
+
+	auto& scheduler = m_context.GetCommandScheduler();
+	scheduler.EndRendering();
+	const auto command = scheduler.Current().Handle();
+	const ImageSubresourceRange range {read_desc.view_info.base_level, 1,
+	                                  read_desc.view_info.base_layer,
+	                                  read_desc.view_info.layer_count};
+	source.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead,
+	               range, command);
+	destination.Transit(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite,
+	                    range, command);
+	std::array<vk::ImageCopy, 2> regions {};
+	uint32_t                   count = 0;
+	for (const auto aspect: {vk::ImageAspectFlagBits::eDepth, vk::ImageAspectFlagBits::eStencil}) {
+		if ((aspect == vk::ImageAspectFlagBits::eDepth && !depth_copy) ||
+		    (aspect == vk::ImageAspectFlagBits::eStencil && !stencil_copy)) {
+			continue;
+		}
+		auto& region = regions[count++];
+		region.srcSubresource = {aspect, range.base_level, range.base_layer,
+		                         range.layer_count};
+		region.dstSubresource = region.srcSubresource;
+		region.extent = write_desc.info.extent;
+	}
+	command.copyImage(source.backing.image, vk::ImageLayout::eTransferSrcOptimal,
+	                  destination.backing.image, vk::ImageLayout::eTransferDstOptimal,
+	                  count, regions.data());
+	return true;
+}
+
 vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
 	const auto format = desc.view_info.format;
 	if (format == vk::Format::eUndefined) {
@@ -336,14 +429,13 @@ vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
 		return writes;
 	}
 
-	const auto face_writes = [&](const PipelineStencilStaticState&  state,
-	                             const PipelineStencilDynamicState& dynamic) {
-		if (dynamic.writeMask == 0) {
+	const auto face_writes = [&](const vk::StencilOpState& state) {
+		if (state.writeMask == 0) {
 			return false;
 		}
 		bool can_pass = state.compareOp != vk::CompareOp::eNever;
 		bool can_fail = state.compareOp != vk::CompareOp::eAlways;
-		if (dynamic.compareMask == 0) {
+		if (state.compareMask == 0) {
 			switch (state.compareOp) {
 				case vk::CompareOp::eEqual:
 				case vk::CompareOp::eLessOrEqual:
@@ -369,8 +461,7 @@ vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
 		       (can_pass && depth_fail && state.depthFailOp != vk::StencilOp::eKeep);
 	};
 	if (stencil_clear_enable ||
-	    (stencil_test_enable && (face_writes(stencil_static_front, stencil_dynamic_front) ||
-	                             face_writes(stencil_static_back, stencil_dynamic_back)))) {
+	    (stencil_test_enable && (face_writes(stencil_front) || face_writes(stencil_back)))) {
 		writes |= vk::ImageAspectFlagBits::eStencil;
 	}
 	return writes;

src/graphics/host_gpu/renderer/depthRenderTarget.h

@@ -23,7 +23,6 @@ struct RenderDepthInfo {
 	TextureCache::ImageDesc     desc;
 	bool                        depth_clear_enable       = false;
 	bool                        depth_load_clear_enable  = false;
-	bool                        depth_meta_clear_enable  = false;
 	float                       depth_clear_value        = 0.0f;
 	bool                        depth_test_enable        = false;
 	// Effective draw writes; discovery applies test, target-write and clear controls.
@@ -35,10 +34,8 @@ struct RenderDepthInfo {
 	bool                        stencil_clear_enable     = false;
 	uint8_t                     stencil_clear_value      = 0;
 	bool                        stencil_test_enable      = false;
-	PipelineStencilStaticState  stencil_static_front;
-	PipelineStencilStaticState  stencil_static_back;
-	PipelineStencilDynamicState stencil_dynamic_front;
-	PipelineStencilDynamicState stencil_dynamic_back;
+	vk::StencilOpState          stencil_front;
+	vk::StencilOpState          stencil_back;
 	ImageId                     image_id;
 
 	[[nodiscard]] vk::ImageAspectFlags AttachmentWriteAspects() const;

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

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

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

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

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

@@ -26,12 +26,13 @@ namespace {
 	}
 }
 
-[[nodiscard]] vk::ImageCreateFlags ImageCreateFlags(const ImageInfo& info) {
+[[nodiscard]] vk::ImageCreateFlags ImageCreateFlags(const GraphicContext& graphics,
+                                                   const ImageInfo& info) {
 	vk::ImageCreateFlags flags {};
 	if (DepthAspectTransferFormat(info.pixel_format) == vk::Format::eUndefined) {
 		flags |= vk::ImageCreateFlagBits::eMutableFormat;
 		flags |= vk::ImageCreateFlagBits::eExtendedUsage;
-		if (Prospero::BlockCompressedBytesPerBlock(info.guest_format) != 0) {
+		if (info.IsBlock() && graphics.supports_block_texel_view) {
 			flags |= vk::ImageCreateFlagBits::eBlockTexelViewCompatible;
 		}
 	}
@@ -47,6 +48,10 @@ namespace {
 }
 
 [[nodiscard]] vk::ImageUsageFlags ImageUsageFlags(GraphicContext& graphics, const ImageInfo& info) {
+	if (info.IsBlock()) {
+		return vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eTransferDst |
+		       vk::ImageUsageFlagBits::eSampled;
+	}
 	const auto properties = graphics.GetFormatProperties(info.pixel_format);
 	auto       usage = vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eTransferDst;
 	if (HasFormatFeature(properties, vk::FormatFeatureFlagBits::eSampledImage)) {
@@ -656,27 +661,17 @@ Image::Image(GraphicContext& graphics, CommandScheduler& scheduler, const ImageI
 		return;
 	}
 
-	backing.format      = info.pixel_format;
-	backing.image_type  = HostImageType(info.type);
-	backing.extent      = info.extent;
-	backing.guest_pitch = info.pitch;
-	backing.layers      = info.IsVolume() ? 1u : info.resources.layers;
-	backing.mip_levels  = info.resources.levels;
-	backing.samples     = info.samples;
-	backing.flags       = ImageCreateFlags(info);
-	backing.usage       = ImageUsageFlags(graphics, info);
-
 	vk::ImageCreateInfo create {};
-	create.flags         = backing.flags;
-	create.imageType     = backing.image_type;
-	create.extent        = backing.extent;
-	create.mipLevels     = backing.mip_levels;
-	create.arrayLayers   = backing.layers;
-	create.format        = backing.format;
+	create.flags         = ImageCreateFlags(graphics, info);
+	create.imageType     = HostImageType(info.type);
+	create.extent        = info.extent;
+	create.mipLevels     = info.resources.levels;
+	create.arrayLayers   = info.IsVolume() ? 1u : info.resources.layers;
+	create.format        = info.pixel_format;
 	create.tiling        = vk::ImageTiling::eOptimal;
-	create.initialLayout = backing.state.layout;
-	create.usage         = backing.usage;
-	create.samples       = vulkan_sample_count(backing.samples);
+	create.initialLayout = vk::ImageLayout::eUndefined;
+	create.usage         = ImageUsageFlags(graphics, info);
+	create.samples       = vulkan_sample_count(info.samples);
 
 	vk::ImageFormatProperties properties {};
 	if (graphics.GetImageFormatProperties(create.format, create.imageType, create.tiling,
@@ -687,10 +682,9 @@ Image::Image(GraphicContext& graphics, CommandScheduler& scheduler, const ImageI
 		     "flags=0x%x samples=%u\n",
 		     static_cast<int>(create.format), static_cast<int>(create.imageType),
 		     static_cast<vk::ImageUsageFlags::MaskType>(create.usage),
-		     static_cast<vk::ImageCreateFlags::MaskType>(create.flags), backing.samples);
+		     static_cast<vk::ImageCreateFlags::MaskType>(create.flags), info.samples);
 	}
 
-	backing.memory.property = vk::MemoryPropertyFlagBits::eDeviceLocal;
 	if (!graphics.CreateImage(create, backing)) {
 		EXIT("failed to create image: extent=%ux%ux%u format=%d layers=%u levels=%u\n",
 		     create.extent.width, create.extent.height, create.extent.depth,
@@ -699,11 +693,12 @@ Image::Image(GraphicContext& graphics, CommandScheduler& scheduler, const ImageI
 }
 
 uint64_t Image::HashGuestEdges() const {
-	constexpr uint64_t                         page_mask = TRACKER_PAGE_SIZE - 1;
 	std::array<uint8_t, TRACKER_PAGE_SIZE * 2> bytes {};
 	const auto                                 range = info.data;
-	const uint64_t head_end     = std::min(range.End(), (range.address + page_mask) & ~page_mask);
-	const uint64_t tail_begin   = std::max(range.address, range.End() & ~page_mask);
+	const uint64_t head_end =
+	    std::min(range.End(), Common::AlignUp(range.address, TRACKER_PAGE_SIZE));
+	const uint64_t tail_begin =
+	    std::max(range.address, Common::AlignDown(range.End(), TRACKER_PAGE_SIZE));
 	const uint64_t head_size    = head_end - range.address;
 	const uint64_t tail_address = tail_begin < head_end ? head_end : tail_begin;
 	const uint64_t tail_size    = range.End() - tail_address;

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

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

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

@@ -82,9 +82,6 @@ struct ImageInfo {
 	[[nodiscard]] constexpr bool IsVolume() const noexcept {
 		return type == Prospero::ImageType::kColor3D;
 	}
-	[[nodiscard]] constexpr bool IsLayered() const noexcept {
-		return !IsVolume() && resources.layers > 1;
-	}
 	[[nodiscard]] constexpr uint32_t TransferLayers() const noexcept {
 		return IsVolume() ? extent.depth : resources.layers;
 	}
@@ -97,9 +94,24 @@ struct ImageInfo {
 		       bytes_per_block == other.bytes_per_block;
 	}
 	[[nodiscard]] int32_t MipOf(const ImageInfo& container) const noexcept {
-		if (!IsCompatible(container) || tile_mode != container.tile_mode || resources.levels != 1 ||
-		    container.resources.layers == 0 ||
-		    container.resources.levels > container.mip_layout.size()) {
+		if (container.resources.levels > container.mip_layout.size()) {
+			return -1;
+		}
+		for (uint32_t level = 0; level < container.resources.levels; level++) {
+			if (SliceOf(container, static_cast<int32_t>(level)) >= 0) {
+				return static_cast<int32_t>(level);
+			}
+		}
+		return -1;
+	}
+	[[nodiscard]] int32_t SliceOf(const ImageInfo& container, int32_t mip) const noexcept {
+		if (!IsCompatible(container) || tile_mode != container.tile_mode || type != container.type ||
+		    IsVolume() || resources.levels != 1 || resources.layers == 0 ||
+		    container.resources.layers == 0 || mip < 0 ||
+		    static_cast<uint32_t>(mip) >= container.resources.levels ||
+		    container.resources.levels > container.mip_layout.size() ||
+		    !data.Valid() || !container.data.Valid() || data.address < container.data.address ||
+		    data.End() > container.data.End()) {
 			return -1;
 		}
 		if (HasStencil() != container.HasStencil() ||
@@ -108,75 +120,36 @@ struct ImageInfo {
 			return -1;
 		}
 
-		int32_t mip = -1;
-		for (uint32_t level = 0; level < container.resources.levels; level++) {
-			const auto& layout = container.mip_layout[level];
-			if (layout.size == 0 || layout.size % container.resources.layers != 0 ||
-			    container.data.address > UINT64_MAX - layout.offset) {
-				continue;
-			}
-			const auto mip_base   = container.data.address + layout.offset;
-			const auto slice_size = layout.size / container.resources.layers;
-			if (slice_size == 0 || mip_base > UINT64_MAX - layout.size) {
-				continue;
-			}
-			const auto mip_end = mip_base + layout.size;
-			if (data.address >= mip_base && data.address < mip_end &&
-			    (data.address - mip_base) % slice_size == 0) {
-				mip = static_cast<int32_t>(level);
-				break;
-			}
-		}
-		if (mip < 0) {
+		const auto  level  = static_cast<uint32_t>(mip);
+		const auto& layout = container.mip_layout[level];
+		if (extent.width != std::max(container.extent.width >> level, 1u) ||
+		    extent.height != std::max(container.extent.height >> level, 1u) ||
+		    extent.depth != container.extent.depth || mip_layout[0].pitch != layout.pitch ||
+		    mip_layout[0].height != layout.height || layout.size == 0 ||
+		    layout.size % container.resources.layers != 0 ||
+		    container.data.size % container.resources.layers != 0 ||
+		    data.size % resources.layers != 0) {
 			return -1;
 		}
 
-		const auto level = static_cast<uint32_t>(mip);
-		if (extent.width != std::max(container.extent.width >> level, 1u) ||
-		    extent.height != std::max(container.extent.height >> level, 1u)) {
+		// PS5 block slices contain the entire reversed mip chain.
+		const auto slice_stride = container.data.size / container.resources.layers;
+		const auto child_stride = data.size / resources.layers;
+		if (child_stride != layout.size / container.resources.layers ||
+		    layout.offset >= slice_stride || child_stride > slice_stride - layout.offset ||
+		    (resources.layers > 1 && child_stride != slice_stride)) {
 			return -1;
 		}
-		const auto mip_depth = std::max(container.extent.depth >> level, 1u);
-		if (container.type == Prospero::ImageType::kColor3D &&
-		    type == Prospero::ImageType::kColor2D) {
-			if (resources.layers != mip_depth) {
-				return -1;
-			}
-		} else if (type != container.type) {
+		const auto address_delta = data.address - container.data.address;
+		if (address_delta < layout.offset || (address_delta - layout.offset) % slice_stride != 0) {
 			return -1;
 		}
-		return mip;
-	}
-	[[nodiscard]] int32_t SliceOf(const ImageInfo& container, int32_t mip) const noexcept {
-		if (!IsCompatible(container) || type != container.type || mip < 0 ||
-		    static_cast<uint32_t>(mip) >= container.resources.levels ||
-		    container.resources.levels > container.mip_layout.size() ||
-		    container.resources.layers == 0 || data.size == 0) {
+		const auto layer = (address_delta - layout.offset) / slice_stride;
+		if (layer > INT32_MAX || layer >= container.resources.layers ||
+		    resources.layers > container.resources.layers - layer) {
 			return -1;
 		}
-		const auto level = static_cast<uint32_t>(mip);
-		if (extent.width != std::max(container.extent.width >> level, 1u) ||
-		    extent.height != std::max(container.extent.height >> level, 1u)) {
-			return -1;
-		}
-		const auto& layout = container.mip_layout[level];
-		if (layout.size == 0 || layout.size % container.resources.layers != 0 ||
-		    container.data.address > UINT64_MAX - layout.offset) {
-			return -1;
-		}
-		const auto slice_size = layout.size / container.resources.layers;
-		if (slice_size == 0 || data.size % slice_size != 0) {
-			return -1;
-		}
-		const auto mip_base = container.data.address + layout.offset;
-		if (data.address < mip_base) {
-			return -1;
-		}
-		const auto address_delta = data.address - mip_base;
-		if (address_delta % data.size != 0 || address_delta / data.size > INT32_MAX) {
-			return -1;
-		}
-		return static_cast<int32_t>(address_delta / data.size);
+		return static_cast<int32_t>(layer);
 	}
 };
 
@@ -342,14 +315,6 @@ ClassifyVideoOutCompression(bool compressed, uint64_t metadata_address, uint32_t
 	}
 }
 
-[[nodiscard]] inline constexpr bool
-CanUseVideoOutNativeWithoutUpload(VideoOutCompression compression, bool render_target,
-                                  bool gpu_modified, bool guest_modified) noexcept {
-	return compression != VideoOutCompression::Uncompressed &&
-	       compression != VideoOutCompression::Unsupported && !guest_modified &&
-	       (render_target || gpu_modified);
-}
-
 struct VideoOutPixelFormatInfo {
 	vk::Format             format            = vk::Format::eUndefined;
 	Prospero::BufferFormat guest_format      = Prospero::BufferFormat::kInvalid;
@@ -400,32 +365,6 @@ inline constexpr std::array<VideoOutFormatPolicy, 6> VIDEO_OUT_FORMAT_POLICIES {
 	return false;
 }
 
-[[nodiscard]] inline constexpr bool
-IsSupportedDisplayRenderTargetTileMode(Prospero::TileMode tile_mode) noexcept {
-	return tile_mode == Prospero::TileMode::kRenderTarget;
-}
-
-[[nodiscard]] inline constexpr bool IsSupportedStandard64RenderTarget(const ImageInfo& info) {
-	if (info.tile_mode != Prospero::TileMode::kStandard64KB || info.data.address == 0 ||
-	    (info.data.address & 0xffffu) != 0 || info.extent.width == 0 || info.extent.height == 0 ||
-	    info.bytes_per_block != 4 || info.resources.levels != 1 || info.resources.layers != 1 ||
-	    info.samples != 1) {
-		return false;
-	}
-	const auto expected_pitch =
-	    (static_cast<uint64_t>(info.extent.width) + 127u) & ~uint64_t {127u};
-	const auto padded_height =
-	    (static_cast<uint64_t>(info.extent.height) + 127u) & ~uint64_t {127u};
-	return expected_pitch <= UINT32_MAX && info.pitch == expected_pitch &&
-	       expected_pitch <= UINT64_MAX / padded_height / info.bytes_per_block &&
-	       info.data.size == expected_pitch * padded_height * info.bytes_per_block;
-}
-
-[[nodiscard]] inline constexpr bool IsTiledRenderTarget(const ImageInfo& info) noexcept {
-	return info.tile_mode == Prospero::TileMode::kRenderTarget ||
-	       IsSupportedStandard64RenderTarget(info);
-}
-
 [[nodiscard]] inline bool DecodePackedColorClear(vk::Format format, uint32_t packed,
                                                  vk::ClearColorValue& clear) {
 	vk::ClearColorValue next {};

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

@@ -322,6 +322,12 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 		normalized.aspect = vk::ImageAspectFlagBits::eStencil;
 	}
 	normalized.usage = is_storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlags {};
+	for (const auto& cached: views) {
+		if (cached.info == normalized) {
+			return cached.view;
+		}
+	}
+
 	const bool format_compatible = normalized.format != vk::Format::eUndefined &&
 	                               ImageViewOps::FormatsCompatible(image.format, normalized.format);
 	const bool slice_view =
@@ -351,12 +357,6 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 		     image.layers);
 	}
 
-	for (const auto& cached: views) {
-		if (cached.info == normalized) {
-			return cached.view;
-		}
-	}
-
 	vk::ImageViewUsageCreateInfo usage {};
 	usage.usage = image.usage;
 	if (!is_storage) {

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

@@ -1,15 +1,16 @@
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
 
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/gpu_format.h"
 #include "graphics/host_gpu/renderer/image/tiler.h"
 #include "graphics/host_gpu/vulkanCommon.h"
+#include "graphics/shader/shader.h"
 
 #include <algorithm>
-#include <bit>
+#include <array>
 #include <cinttypes>
-#include <cstring>
 
 namespace Libs::Graphics {
 namespace {
@@ -27,13 +28,13 @@ constexpr Prospero::ColorComponentMapping kRenderTargetColorMappings[4][4] = {
      Prospero::ColorMappingArgb},
 };
 
-struct RenderTargetHostFormatInfo {
+struct HostFormatInfo {
 	vk::Format                      format = vk::Format::eUndefined;
 	Prospero::ColorComponentMapping host_to_storage;
 };
 
-RenderTargetHostFormatInfo ResolveRenderTargetHostFormat(Prospero::BufferFormat guest_format,
-                                                         Prospero::ChannelOrder order) {
+HostFormatInfo ResolveHostFormat(Prospero::BufferFormat guest_format,
+                                 Prospero::ChannelOrder order) {
 	if (order == Prospero::ChannelOrder::kAlt) {
 		switch (guest_format) {
 			case Prospero::BufferFormat::k8_8_8_8UNorm:
@@ -47,17 +48,77 @@ RenderTargetHostFormatInfo ResolveRenderTargetHostFormat(Prospero::BufferFormat
 			default: break;
 		}
 	}
+	const auto format = VulkanFormat(guest_format);
 	switch (guest_format) {
-		case Prospero::BufferFormat::k5_5_5_1UNorm:
-			return {vk::Format::eA1R5G5B5UnormPack16, Prospero::ColorMappingBgra};
+		case Prospero::BufferFormat::k5_5_5_1UNorm: return {format, Prospero::ColorMappingBgra};
 		case Prospero::BufferFormat::k1_5_5_5UNorm:
-			return {vk::Format::eR5G5B5A1UnormPack16, Prospero::ColorMappingAbgr};
-		case Prospero::BufferFormat::k4_4_4_4UNorm:
-			return {vk::Format::eR4G4B4A4UnormPack16, Prospero::ColorMappingAbgr};
-		default: return {VulkanFormat(guest_format), {}};
+		case Prospero::BufferFormat::k4_4_4_4UNorm: return {format, Prospero::ColorMappingAbgr};
+		default: return {format, {}};
 	}
 }
 
+uint32_t GetTextureLevelDepth(uint32_t depth, uint32_t level, bool volume_texture) {
+	return volume_texture ? std::max(depth >> level, 1u) : depth;
+}
+
+size_t GetTextureRegionCount(uint32_t depth, uint32_t levels, bool volume_texture) {
+	size_t count = 0;
+	for (uint32_t level = 0; level < levels; level++) {
+		count += GetTextureLevelDepth(depth, level, volume_texture);
+	}
+	return count;
+}
+
+uint64_t CalcTextureSliceStride(const TextureUploadMipLayout* mips, uint32_t levels,
+                                uint64_t total_size, uint32_t depth) {
+	uint64_t stride = 0;
+	for (uint32_t mip_level = 0; mip_level < levels; ++mip_level) {
+		stride = std::max(stride, mips[mip_level].offset + mips[mip_level].size);
+	}
+
+	if (depth > 1 && total_size != 0 && total_size % depth == 0) {
+		stride = std::max(stride, total_size / depth);
+	}
+
+	return stride;
+}
+
+uint64_t CalcLinearUploadLevelSize(const TileTextureElementLayout& element, uint32_t pitch,
+                                   uint32_t height) {
+	const uint32_t element_columns =
+	    std::max((pitch + element.texel_width - 1u) / element.texel_width, 1u);
+	const uint32_t element_rows =
+	    std::max((height + element.texel_height - 1u) / element.texel_height, 1u);
+	return static_cast<uint64_t>(element_columns) * element_rows * element.bytes;
+}
+
+uint64_t SetLinearUploadLevels(TextureUploadMipLayout*         mips,
+                               const TileTextureElementLayout& element, uint32_t height,
+                               uint32_t levels, uint32_t base_pitch) {
+	uint64_t offset     = 0;
+	auto     mip_pitch  = base_pitch;
+	auto     mip_height = height;
+
+	for (uint32_t mip_level = 0; mip_level < levels; ++mip_level) {
+		const auto size = CalcLinearUploadLevelSize(element, mip_pitch, mip_height);
+		mips[mip_level] = {offset, size};
+
+		offset += size;
+		if (mip_pitch > 1) {
+			mip_pitch /= 2;
+		}
+		if (mip_height > 1) {
+			mip_height /= 2;
+		}
+	}
+
+	return offset;
+}
+
+bool FitsBufferRange(uint64_t offset, uint64_t size, uint64_t capacity) {
+	return offset <= capacity && size <= capacity - offset;
+}
+
 } // namespace
 
 RenderTargetFormatInfo TextureGetRenderTargetFormat(Prospero::ChannelLayout layout,
@@ -65,135 +126,61 @@ RenderTargetFormatInfo TextureGetRenderTargetFormat(Prospero::ChannelLayout layo
                                                     Prospero::ChannelOrder  order) {
 	const auto encoding = Prospero::ResolveRenderTargetFormat(layout, type);
 	if (encoding.IsValid() && encoding.SupportsOrder(order)) {
-		const auto host  = ResolveRenderTargetHostFormat(encoding.buffer_format, order);
-		const auto bytes = Prospero::RenderTargetBytesPerElement(encoding.buffer_format);
-		if (host.format != vk::Format::eUndefined && bytes != 0) {
+		const auto host_format = ResolveHostFormat(encoding.buffer_format, order);
+		const auto bytes       = Prospero::RenderTargetBytesPerElement(encoding.buffer_format);
+		if (host_format.format != vk::Format::eUndefined && bytes != 0) {
 			const auto order_mapping =
 			    kRenderTargetColorMappings[static_cast<size_t>(order)][encoding.components - 1u];
-			return {host.format, bytes, host.host_to_storage.Then(order_mapping)};
+			return {host_format.format, bytes, host_format.host_to_storage.Then(order_mapping)};
 		}
 	}
 	EXIT("unsupported render-target format combination: layout=%u type=%u order=%u\n",
 	     static_cast<uint32_t>(layout), static_cast<uint32_t>(type), static_cast<uint32_t>(order));
 }
 
-namespace {
-
-static uint32_t GetTextureLevelDepth(uint32_t depth, uint32_t level, bool volume_texture) {
-	return volume_texture ? std::max(depth >> level, 1u) : depth;
-}
-
-static size_t GetTextureRegionCount(uint32_t depth, uint64_t levels, bool volume_texture) {
-	size_t count = 0;
-	for (uint32_t level = 0; level < levels; level++) {
-		count += GetTextureLevelDepth(depth, level, volume_texture);
+vk::ComponentMapping TextureGetComponentMapping(uint32_t                        swizzle,
+                                                Prospero::ColorComponentMapping host_to_storage) {
+	constexpr std::array host_components {vk::ComponentSwizzle::eR, vk::ComponentSwizzle::eG,
+	                                      vk::ComponentSwizzle::eB, vk::ComponentSwizzle::eA};
+	std::array<vk::ComponentSwizzle, 4> storage_components {};
+	for (uint32_t host_component = 0; host_component < host_components.size(); ++host_component) {
+		storage_components[host_to_storage.Map(host_component)] = host_components[host_component];
 	}
-	return count;
-}
-
-vk::ComponentSwizzle TextureGetComponentSwizzle(uint8_t s) {
-	switch (static_cast<Prospero::CompSwizzle>(s)) {
-		case Prospero::CompSwizzle::kZero: return vk::ComponentSwizzle::eZero;
-		case Prospero::CompSwizzle::kOne: return vk::ComponentSwizzle::eOne;
-		case Prospero::CompSwizzle::kRed: return vk::ComponentSwizzle::eR;
-		case Prospero::CompSwizzle::kGreen: return vk::ComponentSwizzle::eG;
-		case Prospero::CompSwizzle::kBlue: return vk::ComponentSwizzle::eB;
-		case Prospero::CompSwizzle::kAlpha: return vk::ComponentSwizzle::eA;
-		default: EXIT("unknown swizzle: %d\n", static_cast<int>(s));
-	}
-	return vk::ComponentSwizzle::eIdentity;
-}
-
-static uint32_t TextureGetDstSel(uint32_t swizzle, uint32_t channel) {
-	return (swizzle >> (channel * 3u)) & 0x7u;
-}
-
-} // namespace
-
-vk::ComponentMapping TextureGetComponentMapping(uint32_t swizzle) {
-	vk::ComponentMapping components {};
-	components.r = TextureGetComponentSwizzle(static_cast<uint8_t>(TextureGetDstSel(swizzle, 0)));
-	components.g = TextureGetComponentSwizzle(static_cast<uint8_t>(TextureGetDstSel(swizzle, 1)));
-	components.b = TextureGetComponentSwizzle(static_cast<uint8_t>(TextureGetDstSel(swizzle, 2)));
-	components.a = TextureGetComponentSwizzle(static_cast<uint8_t>(TextureGetDstSel(swizzle, 3)));
-	return components;
+	const auto resolve = [&](uint32_t component) {
+		const auto selector = static_cast<Prospero::CompSwizzle>(GetDstSel(swizzle, component));
+		switch (selector) {
+			case Prospero::CompSwizzle::kZero: return vk::ComponentSwizzle::eZero;
+			case Prospero::CompSwizzle::kOne: return vk::ComponentSwizzle::eOne;
+			case Prospero::CompSwizzle::kRed: return storage_components[0];
+			case Prospero::CompSwizzle::kGreen: return storage_components[1];
+			case Prospero::CompSwizzle::kBlue: return storage_components[2];
+			case Prospero::CompSwizzle::kAlpha: return storage_components[3];
+			default: EXIT("unknown swizzle: %u\n", static_cast<uint32_t>(selector));
+		}
+	};
+	return {resolve(0), resolve(1), resolve(2), resolve(3)};
 }
 
 SurfaceFormatInfo TextureGetSurfaceFormatInfo(Prospero::BufferFormat format) {
-	const auto backing_format    = Prospero::RemapTextureFormat(format);
-	const auto vk_format         = VulkanFormat(backing_format);
+	const auto backing_format = Prospero::RemapTextureFormat(format);
+	const auto host_format = ResolveHostFormat(backing_format, Prospero::ChannelOrder::kStandard);
 	const auto conversion_format =
 	    backing_format != format ? format : Prospero::BufferFormat::kInvalid;
-	if (vk_format != vk::Format::eUndefined) {
-		return SurfaceFormatInfo(vk_format, conversion_format);
+	if (host_format.format != vk::Format::eUndefined) {
+		return {host_format.format, conversion_format, host_format.host_to_storage};
 	}
 	EXIT("unknown format: fmt = %u\n", static_cast<uint32_t>(format));
-	return SurfaceFormatInfo(vk::Format::eUndefined, Prospero::BufferFormat::kInvalid);
 }
 
-namespace {
-
-uint64_t CalcTextureSliceStride(const TextureUploadMipLayout* mips, uint64_t levels,
-                                uint64_t total_size, uint32_t depth) {
-	uint64_t stride = 0;
-	for (uint32_t i = 0; i < levels; i++) {
-		stride = std::max(stride, mips[i].offset + mips[i].size);
-	}
-
-	if (depth > 1 && total_size != 0 && total_size % depth == 0) {
-		const auto guest_stride = total_size / depth;
-		if (guest_stride >= stride) {
-			stride = guest_stride;
-		}
-	}
-
-	return stride;
-}
-
-uint64_t CalcLinearUploadLevelSize(const TileTextureElementLayout& element, uint32_t pitch,
-                                   uint32_t height) {
-	const uint32_t elements_w =
-	    std::max((pitch + element.texel_width - 1u) / element.texel_width, 1u);
-	const uint32_t elements_h =
-	    std::max((height + element.texel_height - 1u) / element.texel_height, 1u);
-	return static_cast<uint64_t>(elements_w) * elements_h * element.bytes;
-}
-
-uint64_t SetLinearUploadLevels(TextureUploadMipLayout*         mips,
-                               const TileTextureElementLayout& element, uint64_t height,
-                               uint64_t levels, uint32_t base_pitch) {
-	uint64_t offset = 0;
-	auto     pitch  = base_pitch;
-	auto     h      = static_cast<uint32_t>(height);
-
-	for (uint32_t i = 0; i < levels; i++) {
-		const auto size = CalcLinearUploadLevelSize(element, pitch, h);
-		mips[i].size    = size;
-		mips[i].offset  = offset;
-
-		offset += size;
-		if (pitch > 1) {
-			pitch /= 2;
-		}
-		if (h > 1) {
-			h /= 2;
-		}
-	}
-
-	return offset;
-}
-
-} // namespace
-
 TextureUploadLayout TextureCalcUploadLayout(Prospero::BufferFormat format, uint32_t width,
                                             uint32_t height, uint32_t levels, uint32_t depth,
-                                            Prospero::TileMode tile, uint64_t upload_size,
+                                            Prospero::TileMode tile_mode, uint64_t upload_size,
                                             bool allow_depth_tile, bool volume_texture,
                                             const char* owner) {
 	TextureUploadLayout layout {};
 	layout.surface.description = {
 	    format,
-	    tile,
+	    tile_mode,
 	    volume_texture ? TileSurfaceDimension::Dim3D : TileSurfaceDimension::Dim2D,
 	    width,
 	    height,
@@ -202,7 +189,6 @@ TextureUploadLayout TextureCalcUploadLayout(Prospero::BufferFormat format, uint3
 	    volume_texture ? 1u : depth,
 	};
 	const auto&              description = layout.surface.description;
-	const auto               tile_mode   = description.tile_mode;
 	TileTextureElementLayout element {};
 
 	if (format == Prospero::BufferFormat::kInvalid) {
@@ -212,34 +198,19 @@ TextureUploadLayout TextureCalcUploadLayout(Prospero::BufferFormat format, uint3
 		     levels);
 	}
 
-	switch (tile_mode) {
-		case Prospero::TileMode::kLinear:
-			if (!TileGetTextureElementLayout(format, element)) {
-				EXIT("%s: unsupported linear texture format: fmt=%u\n", owner,
-				     static_cast<uint32_t>(format));
-			}
-			break;
-		case Prospero::TileMode::kDepth:
-			if (!allow_depth_tile || !TileGetTiledTextureLayout(description, layout.surface)) {
-				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
-				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
-				     owner, static_cast<uint32_t>(format),
-				     static_cast<uint32_t>(description.tile_mode), upload_size, width, height,
-				     levels);
-			}
-			break;
-		default:
-			if (!TileGetTiledTextureLayout(description, layout.surface)) {
-				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
-				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
-				     owner, static_cast<uint32_t>(format),
-				     static_cast<uint32_t>(description.tile_mode), upload_size, width, height,
-				     levels);
-			}
-			break;
-	}
-
-	if (tile_mode != Prospero::TileMode::kLinear) {
+	if (tile_mode == Prospero::TileMode::kLinear) {
+		if (!TileGetTextureElementLayout(format, element)) {
+			EXIT("%s: unsupported linear texture format: fmt=%u\n", owner,
+			     static_cast<uint32_t>(format));
+		}
+	} else {
+		if ((tile_mode == Prospero::TileMode::kDepth && !allow_depth_tile) ||
+		    !TileGetTiledTextureLayout(description, layout.surface)) {
+			EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
+			     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
+			     owner, static_cast<uint32_t>(format), static_cast<uint32_t>(tile_mode),
+			     upload_size, width, height, levels);
+		}
 		element = {layout.surface.texture.block.bytes_per_element,
 		           layout.surface.texture.texel_width, layout.surface.texture.texel_height};
 	}
@@ -272,40 +243,33 @@ TextureUploadLayout TextureCalcUploadLayout(Prospero::BufferFormat format, uint3
 std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout) {
 	const auto& description    = layout.surface.description;
 	const bool  volume_texture = description.dimension == TileSurfaceDimension::Dim3D;
+	const bool  linear_texture = description.tile_mode == Prospero::TileMode::kLinear;
 	const auto  depth          = volume_texture ? description.depth : description.layers;
 	uint32_t    mip_width      = description.width;
 	uint32_t    mip_height     = description.height;
-	uint32_t    mip_pitch = volume_texture && description.tile_mode != Prospero::TileMode::kLinear
-	                            ? description.width
-	                            : layout.pitch;
+	uint32_t    mip_pitch = volume_texture && !linear_texture ? description.width : layout.pitch;
 
 	std::vector<vk::BufferImageCopy> regions;
 	regions.reserve(GetTextureRegionCount(depth, description.levels, volume_texture));
-	for (uint32_t i = 0; i < description.levels; i++) {
-		EXIT_NOT_IMPLEMENTED(layout.mips[i].size == 0);
-
-		const auto mip_depth = GetTextureLevelDepth(depth, i, volume_texture);
-
-		for (uint32_t z = 0; z < mip_depth; z++) {
-			const auto          slice_offset = z * layout.slice_stride;
-			vk::BufferImageCopy region {};
-			region.bufferOffset     = layout.mips[i].offset + slice_offset;
-			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i, volume_texture ? 0u : z,
-			                           1};
-			region.imageOffset.z    = volume_texture ? static_cast<int>(z) : 0;
-			region.imageExtent      = {mip_width, mip_height, 1};
-			const bool linear       = description.tile_mode == Prospero::TileMode::kLinear;
-			if (linear) {
-				region.bufferRowLength   = layout.mips[i].row_length;
-				region.bufferImageHeight = layout.mips[i].image_height;
-			} else {
-				const auto texel_width = layout.surface.texture.texel_width;
-				const auto align       = [](uint32_t value, uint32_t block) {
-					return ((value + block - 1u) / block) * block;
-				};
-				const auto pitch       = align(mip_pitch, texel_width);
-				region.bufferRowLength = pitch > align(mip_width, texel_width) ? pitch : 0;
-			}
+	for (uint32_t mip_level = 0; mip_level < description.levels; ++mip_level) {
+		EXIT_NOT_IMPLEMENTED(layout.mips[mip_level].size == 0);
+		const auto          mip_depth = GetTextureLevelDepth(depth, mip_level, volume_texture);
+		vk::BufferImageCopy region {};
+		region.imageSubresource = {vk::ImageAspectFlagBits::eColor, mip_level, 0, 1};
+		region.imageExtent      = {mip_width, mip_height, 1};
+		if (linear_texture) {
+			region.bufferRowLength   = layout.mips[mip_level].row_length;
+			region.bufferImageHeight = layout.mips[mip_level].image_height;
+		} else {
+			const auto texel_width   = layout.surface.texture.texel_width;
+			const auto aligned_pitch = Common::AlignUp(mip_pitch, texel_width);
+			region.bufferRowLength =
+			    aligned_pitch > Common::AlignUp(mip_width, texel_width) ? aligned_pitch : 0;
+		}
+		for (uint32_t slice_index = 0; slice_index < mip_depth; ++slice_index) {
+			region.bufferOffset = layout.mips[mip_level].offset + slice_index * layout.slice_stride;
+			region.imageSubresource.baseArrayLayer = volume_texture ? 0u : slice_index;
+			region.imageOffset.z = volume_texture ? static_cast<int>(slice_index) : 0;
 			regions.push_back(region);
 		}
 
@@ -323,17 +287,9 @@ std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayo
 	return regions;
 }
 
-static bool SetGpuTileSize(uint64_t offset, uint64_t length, uint64_t capacity, uint64_t& size) {
-	if (offset > capacity || length > capacity - offset) {
-		return false;
-	}
-	size = length;
-	return true;
-}
-
 bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
                               const TextureUploadLayout& layout, uint32_t levels,
-                              std::vector<GpuTileInfo>& out_infos) {
+                              std::vector<GpuTileInfo>& out_tile_infos) {
 	const auto& description    = layout.surface.description;
 	const bool  volume_texture = description.dimension == TileSurfaceDimension::Dim3D;
 	const auto  depth          = volume_texture ? description.depth : description.layers;
@@ -351,111 +307,69 @@ bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferI
 		return false;
 	}
 
-	std::vector<GpuTileInfo> infos;
-	infos.reserve(regions.size());
-	if (volume_texture) {
-		size_t region_base = 0;
-		for (uint32_t level = 0; level < levels; ++level) {
-			const uint32_t mip_depth     = GetTextureLevelDepth(depth, level, true);
-			const bool     tail          = level >= surface.first_tail_level;
-			const uint64_t linear_stride = layout.slice_stride;
-			for (uint32_t z = 0; z < mip_depth; z += block.block_depth) {
-				const uint32_t copy_depth = std::min(block.block_depth, mip_depth - z);
-				const auto&    region     = regions[region_base + z];
-				const auto     pitch =
-				    region.bufferRowLength != 0 ? region.bufferRowLength : region.imageExtent.width;
-				const auto  logical_height = region.bufferImageHeight != 0
-				                                 ? region.bufferImageHeight
-				                                 : region.imageExtent.height;
-				GpuTileInfo info {};
-				info.family            = block.family;
-				info.bytes_per_element = block.bytes_per_element;
-				info.linear_offset     = region.bufferOffset;
-				info.tiled_offset =
-				    static_cast<uint64_t>(z / block.block_depth) * surface.block_slice_size +
-				    surface.mips[level].offset;
-				const uint64_t linear_span =
-				    static_cast<uint64_t>(copy_depth - 1u) * linear_stride +
-				    layout.mips[level].size;
-				if (!SetGpuTileSize(info.linear_offset, linear_span, UINT64_MAX,
-				                    info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, surface.mips[level].size, tiled_size,
-				                    info.tiled_size)) {
-					return false;
-				}
-				info.linear_slice_stride = linear_stride;
-				info.width  = std::max((region.imageExtent.width + texture.texel_width - 1u) /
-				                           texture.texel_width,
-				                       1u);
-				info.height = std::max(
-				    (logical_height + texture.texel_height - 1u) / texture.texel_height, 1u);
-				info.depth = copy_depth;
-				info.surface_z =
+	std::vector<GpuTileInfo> tile_infos;
+	tile_infos.reserve(regions.size());
+	const auto slices_per_copy = volume_texture ? block.block_depth : 1u;
+	size_t     region_base     = 0;
+	for (uint32_t mip_level = 0; mip_level < levels; ++mip_level) {
+		const auto& mip       = surface.mips[mip_level];
+		const auto  mip_depth = GetTextureLevelDepth(depth, mip_level, volume_texture);
+		for (uint32_t slice_index = 0; slice_index < mip_depth; slice_index += slices_per_copy) {
+			const auto& region = regions[region_base + slice_index];
+			GpuTileInfo tile_info {};
+			tile_info.family            = block.family;
+			tile_info.bytes_per_element = block.bytes_per_element;
+			tile_info.linear_offset     = region.bufferOffset;
+			tile_info.linear_size       = layout.mips[mip_level].size;
+			tile_info.tiled_size        = mip.size;
+			if (volume_texture) {
+				tile_info.depth               = std::min(slices_per_copy, mip_depth - slice_index);
+				tile_info.linear_slice_stride = layout.slice_stride;
+				tile_info.linear_size += (tile_info.depth - 1u) * layout.slice_stride;
+				tile_info.tiled_offset =
+				    mip.offset +
+				    static_cast<uint64_t>(slice_index / slices_per_copy) * surface.block_slice_size;
+				tile_info.surface_z =
 				    block.block_depth == 1 ? static_cast<uint32_t>(region.imageOffset.z) : 0;
-				info.pitch = std::max((pitch + texture.texel_width - 1u) / texture.texel_width, 1u);
-				info.tail_x       = tail ? surface.mips[level].tail_x : 0;
-				info.tail_y       = tail ? surface.mips[level].tail_y : 0;
-				info.tail         = tail;
-				info.tiled_width  = surface.mips[level].padded_width;
-				info.tiled_height = surface.mips[level].padded_height;
-				infos.push_back(info);
-			}
-			region_base += mip_depth;
-		}
-	} else {
-		size_t region_index = 0;
-		for (uint32_t level = 0; level < levels; level++) {
-			const auto& level_size  = layout.mips[level];
-			const auto& mip         = surface.mips[level];
-			const bool  tail        = level >= surface.first_tail_level;
-			const auto  level_depth = GetTextureLevelDepth(depth, level, false);
-			for (uint32_t z = 0; z < level_depth; z++) {
-				const auto& region = regions[region_index++];
-				const auto  pitch =
-				    region.bufferRowLength != 0 ? region.bufferRowLength : region.imageExtent.width;
-				const auto  logical_height = region.bufferImageHeight != 0
-				                                 ? region.bufferImageHeight
-				                                 : region.imageExtent.height;
-				GpuTileInfo info {};
-				info.family              = block.family;
-				info.bytes_per_element   = block.bytes_per_element;
-				info.linear_offset       = region.bufferOffset;
+			} else {
 				const auto source_stride = layout.source_slice_stride != 0
 				                               ? layout.source_slice_stride
 				                               : surface.block_slice_size;
-				if (z > (UINT64_MAX - mip.offset) / source_stride) {
+				if (slice_index > (UINT64_MAX - mip.offset) / source_stride) {
 					return false;
 				}
-				info.tiled_offset = mip.offset + static_cast<uint64_t>(z) * source_stride;
-				if (!SetGpuTileSize(info.linear_offset, level_size.size, UINT64_MAX,
-				                    info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, mip.size, tiled_size, info.tiled_size)) {
-					return false;
-				}
-				info.width  = std::max((region.imageExtent.width + texture.texel_width - 1u) /
-				                           texture.texel_width,
-				                       1u);
-				info.height = std::max(
-				    (logical_height + texture.texel_height - 1u) / texture.texel_height, 1u);
-				info.surface_z = block.family == TileBlockFamily::RenderTarget64KB ||
-				                         block.family == TileBlockFamily::Depth64KB
-				                     ? region.imageSubresource.baseArrayLayer
-				                     : 0;
-				info.pitch = std::max((pitch + texture.texel_width - 1u) / texture.texel_width, 1u);
-				info.tail  = tail;
-				info.tail_x       = tail ? mip.tail_x : 0;
-				info.tail_y       = tail ? mip.tail_y : 0;
-				info.tiled_width  = mip.padded_width;
-				info.tiled_height = mip.padded_height;
-				infos.push_back(info);
+				tile_info.tiled_offset =
+				    mip.offset + static_cast<uint64_t>(slice_index) * source_stride;
+				tile_info.surface_z = block.family == TileBlockFamily::RenderTarget64KB ||
+				                              block.family == TileBlockFamily::Depth64KB
+				                          ? region.imageSubresource.baseArrayLayer
+				                          : 0;
 			}
+			if (!FitsBufferRange(tile_info.linear_offset, tile_info.linear_size, UINT64_MAX) ||
+			    !FitsBufferRange(tile_info.tiled_offset, tile_info.tiled_size, tiled_size)) {
+				return false;
+			}
+			const auto row_length =
+			    region.bufferRowLength != 0 ? region.bufferRowLength : region.imageExtent.width;
+			const auto image_height = region.bufferImageHeight != 0 ? region.bufferImageHeight
+			                                                        : region.imageExtent.height;
+			tile_info.width         = std::max(
+			    (region.imageExtent.width + texture.texel_width - 1u) / texture.texel_width, 1u);
+			tile_info.height =
+			    std::max((image_height + texture.texel_height - 1u) / texture.texel_height, 1u);
+			tile_info.pitch =
+			    std::max((row_length + texture.texel_width - 1u) / texture.texel_width, 1u);
+			tile_info.tail         = mip_level >= surface.first_tail_level;
+			tile_info.tail_x       = tile_info.tail ? mip.tail_x : 0;
+			tile_info.tail_y       = tile_info.tail ? mip.tail_y : 0;
+			tile_info.tiled_width  = mip.padded_width;
+			tile_info.tiled_height = mip.padded_height;
+			tile_infos.push_back(tile_info);
 		}
+		region_base += mip_depth;
 	}
 
-	if (infos.empty()) {
-		return false;
-	}
-	out_infos = std::move(infos);
+	out_tile_infos = std::move(tile_infos);
 	return true;
 }
 

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

@@ -1,9 +1,7 @@
 #ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_HOST_GPU_RENDERER_IMAGE_TEXTURECOMMON_H_
 #define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_HOST_GPU_RENDERER_IMAGE_TEXTURECOMMON_H_
 
-#include "common/abi.h"
-#include "common/common.h"
-#include "graphics/guest_gpu/gpu_format.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 
@@ -20,12 +18,9 @@ struct RenderTargetFormatInfo {
 };
 
 struct SurfaceFormatInfo {
-	explicit constexpr SurfaceFormatInfo(vk::Format format,
-	                                     Prospero::BufferFormat conversion_format)
-	    : vk_format(format), conversion_format(conversion_format) {}
-
-	vk::Format             vk_format;
-	Prospero::BufferFormat conversion_format;
+	vk::Format                      vk_format;
+	Prospero::BufferFormat          conversion_format;
+	Prospero::ColorComponentMapping host_to_storage;
 };
 
 struct TextureUploadMipLayout {
@@ -43,20 +38,21 @@ struct TextureUploadLayout {
 	TextureUploadMipLayout mips[16] = {};
 };
 
-vk::ComponentMapping   TextureGetComponentMapping(uint32_t swizzle);
+vk::ComponentMapping   TextureGetComponentMapping(uint32_t                        swizzle,
+                                                  Prospero::ColorComponentMapping host_to_storage);
 SurfaceFormatInfo      TextureGetSurfaceFormatInfo(Prospero::BufferFormat format);
 RenderTargetFormatInfo TextureGetRenderTargetFormat(Prospero::ChannelLayout layout,
                                                     Prospero::ChannelType   type,
                                                     Prospero::ChannelOrder  order);
 TextureUploadLayout    TextureCalcUploadLayout(Prospero::BufferFormat format, uint32_t width,
                                                uint32_t height, uint32_t levels, uint32_t depth,
-                                               Prospero::TileMode tile, uint64_t upload_size,
+                                               Prospero::TileMode tile_mode, uint64_t upload_size,
                                                bool allow_depth_tile, bool volume_texture,
                                                const char* owner);
 std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout);
 bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
                               const TextureUploadLayout& layout, uint32_t levels,
-                              std::vector<GpuTileInfo>& infos);
+                              std::vector<GpuTileInfo>& out_tile_infos);
 
 } // namespace Libs::Graphics
 

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

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

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

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

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

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

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

@@ -1,5 +1,6 @@
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
 
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "common/common.h"
 #include "common/file.h"
@@ -14,13 +15,13 @@
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/hostMemory.h"
+#include "graphics/host_gpu/renderer/colorRenderTarget.h"
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
-#include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 #include "graphics/shader/recompiler/ir/passes/BindingLayout.h"
@@ -93,22 +94,15 @@ static const char* ShaderStageResourceName(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Vertex: return "Vertex";
 		case ShaderType::Mesh: return "Mesh";
+		case ShaderType::Local: return "Local";
+		case ShaderType::TessellationControl: return "Hull";
+		case ShaderType::TessellationEvaluation: return "Domain";
 		case ShaderType::Pixel: return "Pixel";
 		case ShaderType::Compute: return "Compute";
 		default: return "Unknown";
 	}
 }
 
-static vk::ShaderStageFlags NativeShaderStage(ShaderType stage) {
-	switch (stage) {
-		case ShaderType::Vertex: return vk::ShaderStageFlagBits::eVertex;
-		case ShaderType::Mesh: return vk::ShaderStageFlagBits::eMeshEXT;
-		case ShaderType::Pixel: return vk::ShaderStageFlagBits::eFragment;
-		case ShaderType::Compute: return vk::ShaderStageFlagBits::eCompute;
-		default: EXIT("unknown native shader stage\n");
-	}
-}
-
 static Prospero::ImageType TextureType(const ShaderTextureResource& descriptor) {
 	const auto type = descriptor.Type();
 	return type == Prospero::ImageType::kCube ? Prospero::ImageType::kColor2DArray : type;
@@ -146,7 +140,7 @@ NativeStorageBuffer(RenderContext& context, const PreparedBindings::BufferSource
 	}
 	auto [buffer, offset] = context.GetBufferCache().ObtainBuffer(address, size, resource.written,
 	                                                              resource.formatted, id);
-	const auto aligned_offset = offset - offset % alignment;
+	const auto aligned_offset = Common::AlignDown(offset, alignment);
 	const auto adjustment     = offset - aligned_offset;
 	const auto max_range      = graphics.GetPhysicalDeviceProperties().limits.maxStorageBufferRange;
 	if (adjustment % sizeof(uint32_t) != 0 || adjustment >= 256 || size > max_range - adjustment) {
@@ -154,7 +148,7 @@ NativeStorageBuffer(RenderContext& context, const PreparedBindings::BufferSource
 	}
 	buffer_offset = static_cast<uint32_t>(adjustment);
 	const vk::DescriptorBufferInfo result {buffer->Handle(), aligned_offset, size + adjustment};
-	if (resource.formatted && resource.written) {
+	if (resource.written) {
 		context.GetTextureCache().InvalidateMemoryFromGPU(address, size);
 	}
 	const char* access = "Read";
@@ -170,34 +164,6 @@ NativeStorageBuffer(RenderContext& context, const PreparedBindings::BufferSource
 	return result;
 }
 
-static bool IsSupportedSampledColorResource(const ShaderRecompiler::IR::ImageResource& resource) {
-	bool supported_dimension = false;
-	switch (resource.dimension) {
-		case ShaderRecompiler::Decoder::ImageDimension::Dim1D:
-		case ShaderRecompiler::Decoder::ImageDimension::Dim1DArray:
-		case ShaderRecompiler::Decoder::ImageDimension::Dim2D:
-		case ShaderRecompiler::Decoder::ImageDimension::Dim2DArray:
-		case ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaa:
-		case ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaaArray:
-			supported_dimension = true;
-			break;
-		default: break;
-	}
-	return resource.resource_class == ShaderRecompiler::IR::ImageResourceClass::Sampled &&
-	       resource.numeric_class != Prospero::TextureNumericClass::Unsupported &&
-	       supported_dimension && resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None &&
-	       resource.read && !resource.written && !resource.atomic && !resource.depth_compare;
-}
-
-bool IsSupportedSampledVideoOutView(const ShaderRecompiler::IR::ImageResource& resource,
-                                    const ShaderTextureResource& descriptor, const Image& image) {
-	return image.usage.video_out && image.info.resources.layers == 1 &&
-	       IsSupportedSampledColorResource(resource) &&
-	       resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D &&
-	       descriptor.Type() == Prospero::ImageType::kColor2D && descriptor.Depth() == 0 &&
-	       descriptor.BaseArray5() == 0;
-}
-
 bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor, bool r128) {
 	constexpr uint32_t field1_reserved_mask = 0x200fff00u;
 	constexpr uint32_t field2_reserved_mask = 0xf0003000u;
@@ -405,30 +371,22 @@ void ValidateStorageTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	     descriptor.fields[5], descriptor.fields[6], descriptor.fields[7]);
 }
 
-struct NullImageSpec {
-	vk::Format             format;
-	Prospero::BufferFormat guest_format;
-};
-
-static NullImageSpec NullTextureSpec(const ShaderRecompiler::IR::ImageResource& resource) {
-	switch (resource.numeric_class) {
-		case Prospero::TextureNumericClass::Float:
-			return {vk::Format::eR32Sfloat, Prospero::BufferFormat::k32Float};
-		case Prospero::TextureNumericClass::Uint:
-			return {vk::Format::eR32Uint, Prospero::BufferFormat::k32UInt};
-		case Prospero::TextureNumericClass::Sint:
-			return {vk::Format::eR32Sint, Prospero::BufferFormat::k32SInt};
-		case Prospero::TextureNumericClass::Unsupported: break;
-	}
-	EXIT("null image has unsupported numeric class\n");
-}
-
 static TextureCache::ImageDesc NullTextureDesc(const ShaderRecompiler::IR::ImageResource& resource,
                                                TextureCache::BindingType                  binding) {
-	const auto              spec = NullTextureSpec(resource);
 	TextureCache::ImageDesc desc {};
-	desc.info.pixel_format    = spec.format;
-	desc.info.guest_format    = spec.guest_format;
+	switch (resource.numeric_class) {
+		case Prospero::TextureNumericClass::Float:
+			desc.info.guest_format = Prospero::BufferFormat::k32Float;
+			break;
+		case Prospero::TextureNumericClass::Uint:
+			desc.info.guest_format = Prospero::BufferFormat::k32UInt;
+			break;
+		case Prospero::TextureNumericClass::Sint:
+			desc.info.guest_format = Prospero::BufferFormat::k32SInt;
+			break;
+		default: EXIT("null image has unsupported numeric class\n");
+	}
+	desc.info.pixel_format    = VulkanFormat(desc.info.guest_format);
 	desc.info.type            = Prospero::ImageType::kColor2D;
 	desc.info.extent          = {1, 1, 1};
 	desc.info.resources       = {1, 1};
@@ -492,17 +450,18 @@ static void PopulateTextureMipLayout(ImageInfo& info) {
 
 static ImageViewInfo TextureViewInfo(const ShaderRecompiler::IR::ImageResource& resource,
                                      const ShaderTextureResource& descriptor, vk::Format format,
-                                     bool shader_conversion, bool storage, uint32_t view_levels,
-                                     uint32_t image_layers) {
+                                     const SurfaceFormatInfo& surface_format, bool storage,
+                                     uint32_t view_levels, uint32_t image_layers) {
 	ImageViewInfo view {};
 	view.format      = format;
 	view.aspect      = vk::ImageAspectFlagBits::eColor;
 	view.base_level  = descriptor.BaseLevel();
 	view.level_count = view_levels;
-	view.usage   = storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlagBits::eSampled;
-	view.mapping = storage || shader_conversion
-	                   ? vk::ComponentMapping {}
-	                   : TextureGetComponentMapping(descriptor.DstSelXYZW());
+	view.usage = storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlagBits::eSampled;
+	view.mapping =
+	    storage || surface_format.conversion_format != Prospero::BufferFormat::kInvalid
+	        ? vk::ComponentMapping {}
+	        : TextureGetComponentMapping(descriptor.DstSelXYZW(), surface_format.host_to_storage);
 	switch (resource.dimension) {
 		case ShaderRecompiler::Decoder::ImageDimension::Dim1D:
 			view.type       = vk::ImageViewType::e1D;
@@ -681,7 +640,7 @@ TextureBinding RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageR
 	} else {
 		PopulateTextureMipLayout(desc.info);
 	}
-	desc.view_info = TextureViewInfo(resource, descriptor, view_format, shader_conversion, storage,
+	desc.view_info = TextureViewInfo(resource, descriptor, view_format, surface_format, storage,
 	                                 view_levels, desc.info.resources.layers);
 	desc.type = storage ? TextureCache::BindingType::Storage : TextureCache::BindingType::Texture;
 
@@ -729,13 +688,6 @@ static vk::DescriptorBufferInfo NativeUpload(RenderContext&            context,
 	return {buffer.Handle(), offset, data.size_bytes()};
 }
 
-void RenderExecutor::TrackImageBinding(ImageId id) {
-	EXIT_IF(m_context.GetTextureCache().m_slot_images.try_get(id) == nullptr);
-	if (std::ranges::find(m_bound_images, id) == m_bound_images.end()) {
-		m_bound_images.push_back(id);
-	}
-}
-
 void RenderExecutor::BindImage(ImageId id, bool storage) {
 	auto& image = m_context.GetTextureCache().GetImage(id);
 	if (image.info.data.Empty()) {
@@ -746,13 +698,13 @@ void RenderExecutor::BindImage(ImageId id, bool storage) {
 	}
 	image.binding.is_bound = true;
 	image.binding.shader_write |= storage;
-	TrackImageBinding(id);
+	m_bound_images.push_back(id);
 }
 
 void RenderExecutor::BindRenderTarget(ImageId id) {
 	auto& image             = m_context.GetTextureCache().GetImage(id);
 	image.binding.is_target = true;
-	TrackImageBinding(id);
+	m_bound_images.push_back(id);
 }
 
 void RenderExecutor::ResetBindings() {
@@ -770,8 +722,7 @@ PreparedBindings RenderExecutor::PrepareBindings(const ShaderStageRuntime& runti
 	const auto& program  = *runtime.program;
 	const auto& snapshot = runtime.resources;
 	PreparedBindings prepared;
-	prepared.program  = runtime.program;
-	prepared.snapshot = &runtime.resources;
+	prepared.runtime = &runtime;
 	prepared.images.reserve(program.info.images.size());
 	for (uint32_t i = 0; i < program.info.images.size(); i++) {
 		auto binding = ResolveTexture(program.info.images[i], snapshot.images[i]);
@@ -796,9 +747,9 @@ PreparedBindings RenderExecutor::PrepareBindings(const ShaderStageRuntime& runti
 
 void RenderExecutor::FindBuffers(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(prepared.program == nullptr || prepared.snapshot == nullptr);
-	const auto& program  = *prepared.program;
-	const auto& snapshot = *prepared.snapshot;
+	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
+	const auto& program  = *prepared.runtime->program;
+	const auto& snapshot = prepared.runtime->resources;
 	auto&       cache    = m_context.GetBufferCache();
 
 	prepared.buffer_sources.clear();
@@ -806,10 +757,7 @@ void RenderExecutor::FindBuffers(PreparedBindings& prepared) {
 	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {
 		auto descriptor = DecodeNativeDescriptor<ShaderBufferResource>(snapshot.buffers[i]);
 		const auto address = descriptor.Base48();
-		const auto stride  = descriptor.Stride();
-		const auto records = descriptor.NumRecords();
-		// The descriptor has a 14-bit stride and 32-bit record count, so the product fits u64.
-		const auto requested_size = stride != 0 ? static_cast<uint64_t>(stride) * records : records;
+		const auto requested_size = descriptor.GetSize();
 		if (address == 0 || requested_size == 0) {
 			prepared.buffer_sources.push_back({});
 			continue;
@@ -821,9 +769,9 @@ void RenderExecutor::FindBuffers(PreparedBindings& prepared) {
 
 void RenderExecutor::RebindBuffers(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(prepared.program == nullptr || prepared.snapshot == nullptr);
-	const auto& program   = *prepared.program;
-	const auto& snapshot  = *prepared.snapshot;
+	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
+	const auto& program   = *prepared.runtime->program;
+	const auto& snapshot  = prepared.runtime->resources;
 	const auto& layout    = program.bindings;
 	EXIT_IF(prepared.buffer_sources.size() != program.info.buffers.size());
 
@@ -856,9 +804,9 @@ void RenderExecutor::RebindBuffers(PreparedBindings& prepared) {
 
 void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(prepared.program == nullptr || prepared.snapshot == nullptr);
-	const auto& program  = *prepared.program;
-	const auto& snapshot = *prepared.snapshot;
+	EXIT_IF(prepared.runtime == nullptr || !*prepared.runtime);
+	const auto& program  = *prepared.runtime->program;
+	const auto& snapshot = prepared.runtime->resources;
 	auto&       images   = prepared.images;
 	EXIT_IF(images.size() != program.info.images.size());
 	auto& texture_cache = m_context.GetTextureCache();
@@ -903,32 +851,39 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 	}
 }
 
-RenderExecutor::GraphicsBindings
-RenderExecutor::PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
-                                        const ShaderStageRuntime& pixel, bool pixel_active) {
-	GraphicsBindings bindings {
-	    .vertex = PrepareBindings(vertex),
-	};
-	if (pixel_active) {
-		bindings.pixel.emplace(PrepareBindings(pixel));
+void RenderExecutor::PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
+                                             std::span<RenderColorInfo> colors) {
+	bool uses_dma = false;
+	for (auto* stage: stages) {
+		FindBuffers(*stage);
+		uses_dma |= stage->runtime->program->info.uses_dma;
 	}
-	FindBuffers(bindings.vertex);
-	if (bindings.pixel) {
-		FindBuffers(*bindings.pixel);
+	if (uses_dma) {
+		m_context.PrepareBda();
 	}
-	if (bindings.vertex.program->info.uses_dma ||
-	    (bindings.pixel && bindings.pixel->program->info.uses_dma)) {
-		m_context.GetGpuResources().PrepareBda();
+	for (auto* stage: stages) {
+		RebindImages(*stage);
 	}
-	RebindBuffers(bindings.vertex);
-	if (bindings.pixel) {
-		RebindBuffers(*bindings.pixel);
+	auto& cache = m_context.GetTextureCache();
+	for (auto& target: colors) {
+		EXIT_IF(!target.image_id);
+		const auto old_image = cache.m_slot_images.try_get(target.image_id);
+		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
+		    old_image->binding.needs_rebind) {
+			if (old_image != nullptr) {
+				old_image->binding = {};
+			}
+			target.desc.view_info.base_level = target.guest_mip_level;
+			target.desc.view_info.base_layer = target.guest_array_layer;
+			target.image_id = cache.FindImage(target.desc);
+			BindRenderTarget(target.image_id);
+		}
 	}
-	RebindImages(bindings.vertex);
-	if (bindings.pixel) {
-		RebindImages(*bindings.pixel);
+	// Discovery can read back PS5 metadata and submit the scheduler. Reserve draw buffers only
+	// after image identities are final; attachment layout transitions follow buffer alias copies.
+	for (auto* stage: stages) {
+		RebindBuffers(*stage);
 	}
-	return bindings;
 }
 
 void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
@@ -941,20 +896,21 @@ void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
 	size_t write_count      = 0;
 	ShaderRecompiler::IR::PushData push_data;
 	bool                           has_push_data = false;
-	constexpr auto                 GraphicsStages = vk::ShaderStageFlagBits::eVertex |
-	                                                vk::ShaderStageFlagBits::eMeshEXT |
-	                                                vk::ShaderStageFlagBits::eFragment;
+	constexpr auto                 GraphicsStages =
+	    vk::ShaderStageFlagBits::eVertex | vk::ShaderStageFlagBits::eMeshEXT |
+	    vk::ShaderStageFlagBits::eTessellationControl |
+	    vk::ShaderStageFlagBits::eTessellationEvaluation | vk::ShaderStageFlagBits::eFragment;
 	vk::ShaderStageFlags push_stages = pipeline_bind_point == vk::PipelineBindPoint::eGraphics
 	                                       ? vk::ShaderStageFlagBits::eFragment
 	                                       : vk::ShaderStageFlags {};
 	for (const auto* prepared: prepared_bindings) {
-		EXIT_IF(prepared == nullptr || prepared->program == nullptr ||
-		        prepared->snapshot == nullptr);
-		write_count += prepared->program->bindings.descriptors.size();
-		for (const auto& binding: prepared->program->bindings.descriptors) {
+		EXIT_IF(prepared == nullptr || prepared->runtime == nullptr || !*prepared->runtime);
+		const auto& program = *prepared->runtime->program;
+		write_count += program.bindings.descriptors.size();
+		for (const auto& binding: program.bindings.descriptors) {
 			descriptor_count += NativeDescriptorCount(binding);
 		}
-		const auto shader_stage = NativeShaderStage(prepared->program->stage);
+		const auto shader_stage = NativeShaderStage(program.stage);
 		push_stages |= shader_stage;
 		EXIT_IF((pipeline_bind_point == vk::PipelineBindPoint::eGraphics &&
 		         (shader_stage & GraphicsStages) == vk::ShaderStageFlags {}) ||
@@ -969,7 +925,7 @@ void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
 	m_descriptor_writes.reserve(write_count);
 
 	for (auto* prepared: prepared_bindings) {
-		const auto& program       = *prepared->program;
+		const auto& program       = *prepared->runtime->program;
 		auto&       descriptors   = *prepared;
 		const auto  shader_stage  = NativeShaderStage(program.stage);
 		const auto  shader_stages = ShaderPipelineStages(shader_stage);

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

@@ -16,9 +16,7 @@
 
 namespace Libs::Graphics {
 
-namespace ShaderRecompiler::IR {
-struct ResourceSnapshot;
-}
+struct ShaderStageRuntime;
 
 struct TextureBinding {
 	ImageId                    image_id;
@@ -35,8 +33,8 @@ struct PreparedBindings {
 		BufferId id;
 	};
 
-	const ShaderRecompiler::IR::CompiledShaderInfo* program  = nullptr;
-	const ShaderRecompiler::IR::ResourceSnapshot* snapshot = nullptr;
+	// The draw owns the immutable compiled-program/runtime-snapshot association through commit.
+	const ShaderStageRuntime* runtime = nullptr;
 	// Keep the resolved guest range through cache preparation; only the host buffer ID may
 	// become stale and need resolving again when bindings are rebound.
 	std::vector<BufferSource>             buffer_sources;
@@ -68,9 +66,6 @@ template <typename T>
 
 [[nodiscard]] bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor,
                                                    bool r128 = false);
-[[nodiscard]] bool
-IsSupportedSampledVideoOutView(const ShaderRecompiler::IR::ImageResource& resource,
-                               const ShaderTextureResource& descriptor, const Image& image);
 void ValidateStorageTexture(const ShaderRecompiler::IR::ImageResource& resource,
                             const ShaderTextureResource& descriptor, uint64_t size);
 

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

@@ -89,12 +89,7 @@ template <typename... Args>
 void PipelineCacheLog(fmt::format_string<Args...> format, Args&&... args) {
 	auto message = fmt::format(format, std::forward<Args>(args)...);
 	message += '\n';
-	if (Log::GetDirection() != Log::Direction::Console) {
-		std::fwrite(message.data(), 1, message.size(), stdout);
-		std::fflush(stdout);
-	}
-	Log::Write(message);
-	Log::Flush();
+	Log::WriteToConsoleAndLog(message);
 }
 
 bool ReadShaderGuestMemory(void*, uint64_t address, uint32_t* value) {
@@ -236,6 +231,9 @@ struct PipelineCache::ProgramCache {
 		switch (options.stage) {
 			case ShaderType::Vertex: stage_name = "vs"; break;
 			case ShaderType::Mesh: stage_name = "ms"; break;
+			case ShaderType::Local: stage_name = "ls"; break;
+			case ShaderType::TessellationControl: stage_name = "hs"; break;
+			case ShaderType::TessellationEvaluation: stage_name = "ds"; break;
 			case ShaderType::Pixel: stage_name = "ps"; break;
 			case ShaderType::Compute: stage_name = "cs"; break;
 			default: EXIT("invalid pipeline shader stage\n");
@@ -250,18 +248,9 @@ struct PipelineCache::ProgramCache {
 		}
 		DumpShaderSpirv(stage_name, options.shader_hash, result.spirv);
 
-		vk::ShaderModuleCreateInfo create_info {};
-		create_info.codeSize    = result.spirv.size() * sizeof(uint32_t);
-		create_info.pCode       = result.spirv.data();
-		vk::ShaderModule module = nullptr;
-		RequireVulkanSuccess(device.createShaderModule(&create_info, nullptr, &module),
-		                     "create recompiled shader module");
+		const auto module = CompileSPV(result.spirv, device);
 		EXIT_IF(module == nullptr);
 		if (options.dump_ir) {
-			if (!options.early_dump) {
-				LOGF("%s decoded RDNA2:\n%s", options.dump_label, result.decoded_dump.c_str());
-				LOGF("%s IR:\n%s", options.dump_label, result.ir_dump.c_str());
-			}
 			LOGF("%s SPIR-V words=%" PRIu64 " wave_size=%u\n", options.dump_label,
 			     static_cast<uint64_t>(result.spirv.size()), options.wave_size);
 		}
@@ -277,7 +266,7 @@ struct PipelineCache::ProgramCache {
 	                  uint32_t& push_data_cursor) {
 		ShaderType stage;
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
-			stage = input_info.mesh.threads_num[0] != 0 ? ShaderType::Mesh : ShaderType::Vertex;
+			stage = input_info.logical_stage;
 		} else if constexpr (std::is_same_v<InputInfo, ShaderPixelInputInfo>) {
 			stage = ShaderType::Pixel;
 		} else {
@@ -329,6 +318,9 @@ struct PipelineCache::ProgramCache {
 		switch (stage) {
 			case ShaderType::Vertex: label = "ShaderRecompiler VS"; break;
 			case ShaderType::Mesh: label = "ShaderRecompiler MS"; break;
+			case ShaderType::Local: label = "ShaderRecompiler LS"; break;
+			case ShaderType::TessellationControl: label = "ShaderRecompiler HS"; break;
+			case ShaderType::TessellationEvaluation: label = "ShaderRecompiler DS"; break;
 			case ShaderType::Pixel: label = "ShaderRecompiler PS"; break;
 			case ShaderType::Compute: label = "ShaderRecompiler CS"; break;
 			default: EXIT("invalid pipeline shader stage\n");
@@ -338,19 +330,18 @@ struct PipelineCache::ProgramCache {
 		options.shader_hash = params.hash;
 		options.user_data   = params.user_data;
 		options.back_code      = params.back_code;
-		options.dump_ir     = Config::GetShaderLogDirection() != Config::ShaderLogDirection::Silent;
+		options.dump_ir     = Config::GetShaderLogDirection() != Config::LogDirection::Silent;
 		options.early_dump  = options.dump_ir;
 		options.dump_label  = label;
 		options.input_info  = stage_input;
-		options.scratch_dwords = input_info.scratch_size_dwords;
+
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
 			options.user_data_base = 8;
-			if (stage == ShaderType::Mesh) {
+			if (stage == ShaderType::Mesh || stage == ShaderType::TessellationControl) {
 				options.user_data_base = 0;
-				options.wave_size      = input_info.mesh.wave_size;
-				options.scratch_dwords = input_info.mesh.scratch_size_dwords;
+				options.wave_size = stage == ShaderType::Mesh ? input_info.mesh.wave_size : 64u;
 			}
-		} else if constexpr (std::is_same_v<InputInfo, ShaderComputeInputInfo>) {
+		} else {
 			options.wave_size = input_info.wave_size;
 		}
 		auto translated = ShaderRecompiler::TranslateProgram(params.code, options);
@@ -366,16 +357,19 @@ struct PipelineCache::ProgramCache {
 		input_info.stage = {.program = &permutation.program, .resources = std::move(resources)};
 		permutation.program.bindings.AdvancePushData(push_data_cursor);
 
-		std::array<size_t, static_cast<size_t>(ShaderType::Mesh) + 1> counts {};
+		std::array<size_t, static_cast<size_t>(ShaderType::TessellationEvaluation) + 1> counts {};
 		for (const auto& [key, source]: programs) {
 			counts[static_cast<size_t>(key.stage)] += source.permutations.size();
 		}
 		// Guest geometry shaders are compiled through the host mesh stage.
-		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu\n",
+		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu | LS %zu | HS %zu | TES %zu\n",
 		            counts[static_cast<size_t>(ShaderType::Vertex)],
 		            counts[static_cast<size_t>(ShaderType::Pixel)],
 		            counts[static_cast<size_t>(ShaderType::Compute)],
-		            counts[static_cast<size_t>(ShaderType::Mesh)]);
+		            counts[static_cast<size_t>(ShaderType::Mesh)],
+		            counts[static_cast<size_t>(ShaderType::Local)],
+		            counts[static_cast<size_t>(ShaderType::TessellationControl)],
+		            counts[static_cast<size_t>(ShaderType::TessellationEvaluation)]);
 		return permutation.handle;
 	}
 
@@ -569,12 +563,18 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
     const HW::VertexShaderInfo& vertex_regs, const HW::PixelShaderInfo& pixel_regs,
     const HW::ShaderRegisters& sh, const HW::Context& context, const HW::UserConfig& user_config,
     std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping, bool pixel_active,
-    ShaderVertexInputInfo& vertex_info, ShaderPixelInputInfo& pixel_info) {
-	const auto vertex_params = PrepareProgram(vertex_regs, context, user_config, vertex_info);
-	const bool mesh_active   = vertex_info.mesh.threads_num[0] != 0;
+    std::array<ShaderVertexInputInfo, 3>& vertex_info, ShaderPixelInputInfo& pixel_info) {
+	const bool tess_active = user_config.GetPrimType() == Prospero::PrimitiveType::kPatch;
+	std::array<ShaderParams, 3> vertex_params;
+	if (tess_active) {
+		vertex_params = PrepareTessellationPrograms(vertex_regs, context, vertex_info);
+	} else {
+		vertex_params[0] = PrepareProgram(vertex_regs, context, user_config, vertex_info[0]);
+	}
+	const bool mesh_active = vertex_info[0].logical_stage == ShaderType::Mesh;
 	if (mesh_active) {
 		EXIT_NOT_IMPLEMENTED(!m_graphics.mesh_shader_enabled);
-		auto& mesh              = vertex_info.mesh;
+		auto& mesh              = vertex_info[0].mesh;
 		mesh.host_subgroup_size = m_graphics.subgroup_size;
 		const auto& limits      = m_graphics.mesh_shader_properties;
 		const auto  logical_threads =
@@ -597,7 +597,7 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	if (context.GetClipControl().clip_disable) {
 		const auto& viewport = context.GetScreenViewport().viewports[0];
 		const auto& limits   = m_graphics.GetPhysicalDeviceProperties().limits;
-		auto&       clip     = vertex_info.clip_space;
+		auto&       clip     = vertex_info[tess_active ? 2u : 0u].clip_space;
 		clip.scale[0]        = viewport.xscale;
 		clip.scale[1]        = viewport.yscale;
 		clip.offset[0]       = viewport.xoffset;
@@ -615,7 +615,9 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	if (pixel_active) {
 		result.pixel = m_program_cache->Get(pixel_params, pixel_info, push_data_cursor);
 	}
-	result.vertex = m_program_cache->Get(vertex_params, vertex_info, push_data_cursor);
+	for (uint32_t i = 0; i < (tess_active ? 3u : 1u); i++) {
+		result.vertex[i] = m_program_cache->Get(vertex_params[i], vertex_info[i], push_data_cursor);
+	}
 	return result;
 }
 
@@ -633,12 +635,14 @@ bool PipelineStaticParameters::operator==(const PipelineStaticParameters& other)
 	return std::memcmp(this, &other, sizeof(*this)) == 0;
 }
 
-PipelineCache::Pipeline& PipelineCache::CreateGraphicsPipeline(
+PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
     std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
-    const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
+    std::span<const ShaderVertexInputInfo> vertex_info, CommandBuffer& command,
     const ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology,
-    bool primitive_restart_enable, const ShaderProgram& vertex_program,
-    const ShaderProgram& pixel_program) {
+    bool primitive_restart_enable, const GraphicsPrograms& programs) {
+	const auto& vs_input_info  = vertex_info.front();
+	const auto& vertex_program = programs.vertex[0];
+	const auto& pixel_program  = programs.pixel;
 	KYTY_PROFILER_BLOCK("PipelineCache::CreatePipeline(Gfx)", profiler::colors::DeepOrangeA200);
 
 	EXIT_IF(colors.size() > RENDER_COLOR_ATTACHMENTS_MAX);
@@ -655,8 +659,13 @@ PipelineCache::Pipeline& PipelineCache::CreateGraphicsPipeline(
 	const auto vs_id = vertex_program.id;
 	const auto ps_id = ps_active ? pixel_program.id : 0;
 
-	PipelineStaticParameters static_params {};
-	PipelineRenderingState   rendering {};
+	GraphicsPipelineKey key {};
+	for (uint32_t i = 0; i < programs.vertex.size(); i++) {
+		key.vertex_shader_ids[i] = programs.vertex[i].id;
+	}
+	key.ps_shader_id            = ps_id;
+	auto& static_params         = key.static_params;
+	auto& rendering             = key.rendering;
 	rendering.color_count       = color_count;
 	uint32_t attachment_samples = 0;
 	for (uint32_t i = 0; i < color_count; i++) {
@@ -670,6 +679,16 @@ PipelineCache::Pipeline& PipelineCache::CreateGraphicsPipeline(
 			EXIT("mixed color attachment sample counts are unsupported: %u and %u\n",
 			     attachment_samples, colors[i].desc.info.samples);
 		}
+		const auto& rt                        = ctx.GetRenderTarget(colors[i].target_slot);
+		const auto& bc                        = ctx.GetBlendControl(colors[i].target_slot);
+		static_params.color_srcblend[i]       = bc.color_srcblend;
+		static_params.color_comb_fcn[i]       = bc.color_comb_fcn;
+		static_params.color_destblend[i]      = bc.color_destblend;
+		static_params.alpha_srcblend[i]       = bc.alpha_srcblend;
+		static_params.alpha_comb_fcn[i]       = bc.alpha_comb_fcn;
+		static_params.alpha_destblend[i]      = bc.alpha_destblend;
+		static_params.separate_alpha_blend[i] = bc.separate_alpha_blend;
+		static_params.blend_enable[i]         = bc.enable && !rt.info.blend_bypass;
 	}
 	const bool with_depth =
 	    depth.desc.view_info.format != vk::Format::eUndefined && static_cast<bool>(depth.image_id);
@@ -718,34 +737,15 @@ PipelineCache::Pipeline& PipelineCache::CreateGraphicsPipeline(
 	static_params.depth_bounds_test_enable = depth.depth_bounds_test_enable;
 	static_params.depth_min_bounds         = depth.depth_min_bounds;
 	static_params.depth_max_bounds         = depth.depth_max_bounds;
-	static_params.stencil_test_enable      = depth.stencil_test_enable;
-	static_params.stencil_front            = depth.stencil_static_front;
-	static_params.stencil_back             = depth.stencil_static_back;
-	const bool rect_list     = topology == vk::PrimitiveTopology::ePatchList;
+	const bool rect_list =
+	    command.GetUserConfig().GetPrimType() == Prospero::PrimitiveType::kRectList;
 	static_params.cull_back  = !rect_list && mc.cull_back;
 	static_params.cull_front = !rect_list && mc.cull_front;
 	static_params.face       = mc.face;
+	static_params.provoking_vtx_last = mc.provoking_vtx_last;
 	static_params.polygon_mode =
 	    ResolvePolygonMode(mc, static_params.cull_front, static_params.cull_back);
 
-	for (uint32_t i = 0; i < color_count; i++) {
-		const auto& rt                        = ctx.GetRenderTarget(colors[i].target_slot);
-		const auto& bc                        = ctx.GetBlendControl(colors[i].target_slot);
-		static_params.color_srcblend[i]       = bc.color_srcblend;
-		static_params.color_comb_fcn[i]       = bc.color_comb_fcn;
-		static_params.color_destblend[i]      = bc.color_destblend;
-		static_params.alpha_srcblend[i]       = bc.alpha_srcblend;
-		static_params.alpha_comb_fcn[i]       = bc.alpha_comb_fcn;
-		static_params.alpha_destblend[i]      = bc.alpha_destblend;
-		static_params.separate_alpha_blend[i] = bc.separate_alpha_blend;
-		static_params.blend_enable[i]         = bc.enable;
-		static_params.blend_bypass[i]         = rt.info.blend_bypass;
-	}
-	GraphicsPipelineKey key {};
-	key.rendering     = rendering;
-	key.vs_shader_id  = vs_id;
-	key.ps_shader_id  = ps_id;
-	key.static_params = static_params;
 	if (vs_input_info.stage.program->stage != ShaderType::Mesh) {
 		EXIT_IF(vs_input_info.buffers_num < 0 ||
 		        vs_input_info.buffers_num > ShaderVertexInputInfo::RES_MAX ||
@@ -789,9 +789,8 @@ PipelineCache::Pipeline& PipelineCache::CreateGraphicsPipeline(
 
 	auto cached = std::make_unique<Pipeline>();
 	LogPipelineTrace("CreatePipelineInternal begin", vs_id, ps_id);
-	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vs_input_info,
-	                       vertex_program, ps_input_info, pixel_program, static_params,
-	                       m_driver_cache);
+	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vertex_info,
+	                       ps_input_info, programs, static_params, m_driver_cache);
 	LogPipelineTrace("CreatePipelineInternal done", vs_id, ps_id);
 
 	EXIT_NOT_IMPLEMENTED(cached->pipeline == nullptr);
@@ -804,8 +803,8 @@ PipelineCache::Pipeline& PipelineCache::CreateGraphicsPipeline(
 }
 
 PipelineCache::Pipeline&
-PipelineCache::CreateComputePipeline(const ShaderComputeInputInfo& input_info,
-                                     const ShaderProgram&          compute_program) {
+PipelineCache::GetComputePipeline(const ShaderComputeInputInfo& input_info,
+                                  const ShaderProgram&          compute_program) {
 	KYTY_PROFILER_BLOCK("PipelineCache::CreatePipeline(Compute)", profiler::colors::RedA100);
 
 	EXIT_IF(!compute_program);

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

@@ -42,13 +42,11 @@ struct PipelineStaticParameters {
 	bool                       depth_bounds_test_enable = false;
 	float                      depth_min_bounds         = 0.0f;
 	float                      depth_max_bounds         = 0.0f;
-	bool                       stencil_test_enable      = false;
-	PipelineStencilStaticState stencil_front;
-	PipelineStencilStaticState stencil_back;
 	uint32_t                   color_mask[RENDER_COLOR_ATTACHMENTS_MAX]           = {};
 	bool                       cull_front                                         = false;
 	bool                       cull_back                                          = false;
 	bool                       face                                               = false;
+	bool                       provoking_vtx_last                                 = false;
 	vk::PolygonMode            polygon_mode                                       = vk::PolygonMode::eFill;
 	uint8_t                    color_srcblend[RENDER_COLOR_ATTACHMENTS_MAX]       = {};
 	uint8_t                    color_comb_fcn[RENDER_COLOR_ATTACHMENTS_MAX]       = {};
@@ -58,7 +56,6 @@ struct PipelineStaticParameters {
 	uint8_t                    alpha_destblend[RENDER_COLOR_ATTACHMENTS_MAX]      = {};
 	bool                       separate_alpha_blend[RENDER_COLOR_ATTACHMENTS_MAX] = {};
 	bool                       blend_enable[RENDER_COLOR_ATTACHMENTS_MAX]         = {};
-	bool                       blend_bypass[RENDER_COLOR_ATTACHMENTS_MAX]         = {};
 
 	bool operator==(const PipelineStaticParameters& other) const noexcept;
 };
@@ -68,7 +65,7 @@ struct PipelineStaticParameters {
 static_assert(std::is_trivially_copyable_v<PipelineStaticParameters>);
 static_assert(std::is_standard_layout_v<PipelineStaticParameters>);
 static_assert(alignof(PipelineStaticParameters) == 1);
-static_assert(sizeof(PipelineStaticParameters) == 165);
+static_assert(sizeof(PipelineStaticParameters) == 125);
 
 struct PipelineRenderingState {
 	std::array<vk::Format, RENDER_COLOR_ATTACHMENTS_MAX> color_formats {};
@@ -121,8 +118,10 @@ public:
 	};
 
 	struct GraphicsPrograms {
-		ShaderProgram vertex;
+		std::array<ShaderProgram, 3> vertex;
 		ShaderProgram pixel;
+
+		[[nodiscard]] uint32_t VertexStageCount() const { return vertex[1] ? 3u : 1u; }
 	};
 
 	GraphicsPrograms
@@ -130,33 +129,33 @@ public:
 	                    const HW::PixelShaderInfo& pixel_regs, const HW::ShaderRegisters& sh,
 	                    const HW::Context& context, const HW::UserConfig& user_config,
 	                    std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
-	                    bool pixel_active, ShaderVertexInputInfo& vertex_info,
+	                    bool pixel_active, std::array<ShaderVertexInputInfo, 3>& vertex_info,
 	                    ShaderPixelInputInfo& pixel_info);
 	ShaderProgram GetComputeProgram(const HW::ComputeShaderInfo& regs,
 	                                const HW::ShaderRegisters&   sh,
 	                                ShaderComputeInputInfo&      input_info);
 
-	Pipeline&
-	CreateGraphicsPipeline(std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
-	                       const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
-	                       const ShaderPixelInputInfo* ps_input_info,
-	                       vk::PrimitiveTopology topology, bool primitive_restart_enable,
-	                       const ShaderProgram& vertex_program, const ShaderProgram& pixel_program);
-	Pipeline& CreateComputePipeline(const ShaderComputeInputInfo& input_info,
-	                                const ShaderProgram&          compute_program);
+	Pipeline& GetGraphicsPipeline(std::span<const RenderColorInfo>       colors,
+	                              const RenderDepthInfo&                 depth,
+	                              std::span<const ShaderVertexInputInfo> vertex_info,
+	                              CommandBuffer& command, const ShaderPixelInputInfo* ps_input_info,
+	                              vk::PrimitiveTopology topology, bool primitive_restart_enable,
+	                              const GraphicsPrograms& programs);
+	Pipeline& GetComputePipeline(const ShaderComputeInputInfo& input_info,
+	                             const ShaderProgram&          compute_program);
 
 private:
 	struct ProgramCache;
 
 	struct GraphicsPipelineKey {
 		PipelineRenderingState   rendering;
-		uint64_t                 vs_shader_id = 0;
+		std::array<uint64_t, 3>  vertex_shader_ids {};
 		uint64_t                 ps_shader_id = 0;
 		PipelineVertexInputState vertex_input;
 		PipelineStaticParameters static_params;
 
 		bool operator==(const GraphicsPipelineKey& other) const {
-			return rendering == other.rendering && vs_shader_id == other.vs_shader_id &&
+			return rendering == other.rendering && vertex_shader_ids == other.vertex_shader_ids &&
 			       ps_shader_id == other.ps_shader_id && vertex_input == other.vertex_input &&
 			       static_params == other.static_params;
 		}
@@ -189,7 +188,9 @@ private:
 		std::size_t operator()(const GraphicsPipelineKey& key) const {
 			std::size_t hash = 0;
 			PipelineKeyHash::MixRendering(hash, key.rendering);
-			PipelineKeyHash::Mix(hash, key.vs_shader_id);
+			for (const auto id: key.vertex_shader_ids) {
+				PipelineKeyHash::Mix(hash, id);
+			}
 			PipelineKeyHash::Mix(hash, key.ps_shader_id);
 			PipelineKeyHash::Mix(hash, key.vertex_input.binding_count);
 			for (uint32_t i = 0; i < key.vertex_input.binding_count; i++) {
@@ -219,12 +220,14 @@ private:
 };
 
 void LogPipelineTrace(const char* phase, uint64_t vertex_program_id, uint64_t pixel_program_id);
-void CreatePipelineInternal(
-    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
-    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
-    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
-    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
-    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache);
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+                            const PipelineRenderingState&          rendering,
+                            const PipelineVertexInputState&        vertex_input,
+                            std::span<const ShaderVertexInputInfo> vertex_info,
+                            const ShaderPixelInputInfo*            ps_input_info,
+                            const PipelineCache::GraphicsPrograms& programs,
+                            const PipelineStaticParameters&        static_params,
+                            vk::PipelineCache                      driver_cache);
 void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
                             const ShaderComputeInputInfo& input_info,
                             vk::ShaderModule compute_module, vk::PipelineCache driver_cache);

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

@@ -8,6 +8,20 @@
 
 namespace Libs::Graphics {
 
+vk::ShaderStageFlagBits NativeShaderStage(ShaderType stage) {
+	switch (stage) {
+		case ShaderType::Local:
+		case ShaderType::Vertex: return vk::ShaderStageFlagBits::eVertex;
+		case ShaderType::Mesh: return vk::ShaderStageFlagBits::eMeshEXT;
+		case ShaderType::TessellationControl: return vk::ShaderStageFlagBits::eTessellationControl;
+		case ShaderType::TessellationEvaluation:
+			return vk::ShaderStageFlagBits::eTessellationEvaluation;
+		case ShaderType::Pixel: return vk::ShaderStageFlagBits::eFragment;
+		case ShaderType::Compute: return vk::ShaderStageFlagBits::eCompute;
+		default: EXIT("unknown native shader stage\n");
+	}
+}
+
 vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	vk::PipelineStageFlags result = {};
 	if (stages & vk::ShaderStageFlagBits::eVertex) {
@@ -16,6 +30,12 @@ vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	if (stages & vk::ShaderStageFlagBits::eMeshEXT) {
 		result |= vk::PipelineStageFlagBits::eMeshShaderEXT;
 	}
+	if (stages & vk::ShaderStageFlagBits::eTessellationControl) {
+		result |= vk::PipelineStageFlagBits::eTessellationControlShader;
+	}
+	if (stages & vk::ShaderStageFlagBits::eTessellationEvaluation) {
+		result |= vk::PipelineStageFlagBits::eTessellationEvaluationShader;
+	}
 	if (stages & vk::ShaderStageFlagBits::eFragment) {
 		result |= vk::PipelineStageFlagBits::eFragmentShader;
 	}

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

@@ -8,6 +8,7 @@ namespace Libs::Graphics {
 
 struct ShaderStageRuntime;
 
+vk::ShaderStageFlagBits NativeShaderStage(ShaderType stage);
 vk::PipelineStageFlags  ShaderPipelineStages(vk::ShaderStageFlags stages);
 VulkanMemoryBarrier     MakeShaderAccessDependency();
 VulkanMemoryBarrier     MakeShaderWriteHazardDependency();

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

@@ -8,10 +8,12 @@
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
+#include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/renderTarget.h"
 #include "graphics/host_gpu/vulkanCommon.h"
+#include "graphics/shader/recompiler/BufferFormat.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 #include "graphics/shader/rectListShader.h"
 #include "graphics/shader/shader.h"
@@ -115,227 +117,13 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 		}
 		format = vk::Format::eR16G16Sfloat;
 		size   = 2;
-		if (NarrowInputFormat(format, size, used_components)) {
-			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
-			     vk::to_string(format).c_str(), used_components);
-		}
 		return;
 	}
 
-	switch (fmt) {
-		case Prospero::BufferFormat::k32_32_32_32Float:
-			format = vk::Format::eR32G32B32A32Sfloat;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k32_32_32_32SInt:
-			format = vk::Format::eR32G32B32A32Sint;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k32_32_32_32UInt:
-			format = vk::Format::eR32G32B32A32Uint;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k32_32_32Float:
-			format = vk::Format::eR32G32B32Sfloat;
-			size   = 3;
-			break;
-		case Prospero::BufferFormat::k32_32_32SInt:
-			format = vk::Format::eR32G32B32Sint;
-			size   = 3;
-			break;
-		case Prospero::BufferFormat::k32_32_32UInt:
-			format = vk::Format::eR32G32B32Uint;
-			size   = 3;
-			break;
-		case Prospero::BufferFormat::k16_16_16_16Float:
-			format = vk::Format::eR16G16B16A16Sfloat;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k16_16_16_16SInt:
-			format = vk::Format::eR16G16B16A16Sint;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k16_16_16_16UInt:
-			format = vk::Format::eR16G16B16A16Uint;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k16_16_16_16SScaled:
-			format = vk::Format::eR16G16B16A16Sscaled;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k16_16_16_16UScaled:
-			format = vk::Format::eR16G16B16A16Uscaled;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k16_16_16_16SNorm:
-			format = vk::Format::eR16G16B16A16Snorm;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k16_16_16_16UNorm:
-			format = vk::Format::eR16G16B16A16Unorm;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k32_32Float:
-			format = vk::Format::eR32G32Sfloat;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k32_32SInt:
-			format = vk::Format::eR32G32Sint;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k32_32UInt:
-			format = vk::Format::eR32G32Uint;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k8_8_8_8UInt:
-			format = vk::Format::eR8G8B8A8Uint;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k8_8_8_8SScaled:
-			format = vk::Format::eR8G8B8A8Sscaled;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k8_8_8_8UScaled:
-			format = vk::Format::eR8G8B8A8Uscaled;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k8_8_8_8SNorm:
-			format = vk::Format::eR8G8B8A8Snorm;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k8_8_8_8UNorm:
-			format = vk::Format::eR8G8B8A8Unorm;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k10_10_10_2UNorm:
-			format = vk::Format::eA2B10G10R10UnormPack32;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k10_10_10_2SNorm:
-			format = vk::Format::eA2B10G10R10SnormPack32;
-			size   = 4;
-			break;
-		case Prospero::BufferFormat::k16_16Float:
-			format = vk::Format::eR16G16Sfloat;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k16_16SInt:
-			format = vk::Format::eR16G16Sint;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k16_16UInt:
-			format = vk::Format::eR16G16Uint;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k16_16SScaled:
-			format = vk::Format::eR16G16Sscaled;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k16_16UScaled:
-			format = vk::Format::eR16G16Uscaled;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k16_16SNorm:
-			format = vk::Format::eR16G16Snorm;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k16_16UNorm:
-			format = vk::Format::eR16G16Unorm;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k32Float:
-			format = vk::Format::eR32Sfloat;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k32SInt:
-			format = vk::Format::eR32Sint;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k32UInt:
-			format = vk::Format::eR32Uint;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k8_8SInt:
-			format = vk::Format::eR8G8Sint;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k8_8UInt:
-			format = vk::Format::eR8G8Uint;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k8_8SScaled:
-			format = vk::Format::eR8G8Sscaled;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k8_8UScaled:
-			format = vk::Format::eR8G8Uscaled;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k8_8SNorm:
-			format = vk::Format::eR8G8Snorm;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k8_8UNorm:
-			format = vk::Format::eR8G8Unorm;
-			size   = 2;
-			break;
-		case Prospero::BufferFormat::k16Float:
-			format = vk::Format::eR16Sfloat;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k16SInt:
-			format = vk::Format::eR16Sint;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k16UInt:
-			format = vk::Format::eR16Uint;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k16SScaled:
-			format = vk::Format::eR16Sscaled;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k16UScaled:
-			format = vk::Format::eR16Uscaled;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k16SNorm:
-			format = vk::Format::eR16Snorm;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k16UNorm:
-			format = vk::Format::eR16Unorm;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k8SInt:
-			format = vk::Format::eR8Sint;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k8UInt:
-			format = vk::Format::eR8Uint;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k8SScaled:
-			format = vk::Format::eR8Sscaled;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k8UScaled:
-			format = vk::Format::eR8Uscaled;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k8SNorm:
-			format = vk::Format::eR8Snorm;
-			size   = 1;
-			break;
-		case Prospero::BufferFormat::k8UNorm:
-			format = vk::Format::eR8Unorm;
-			size   = 1;
-			break;
-		default:
-			EXIT("unknown format: fmt = %u\n", raw_format);
-			format = vk::Format::eUndefined;
-			size   = 4;
-			break;
+	format = VulkanFormat(fmt);
+	size   = ShaderRecompiler::Format::GetFormatInfo(fmt).component_count;
+	if (format == vk::Format::eUndefined || size == 0) {
+		EXIT("unknown vertex format: fmt = %u\n", raw_format);
 	}
 
 	if (NarrowInputFormat(format, size, used_components)) {
@@ -418,12 +206,18 @@ static void CreateDescriptorLayout(GraphicContext& graphics, PipelineCache::Pipe
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void CreatePipelineInternal(
-    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
-    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
-    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
-    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
-    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache) {
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+                            const PipelineRenderingState&          rendering,
+                            const PipelineVertexInputState&        vertex_input,
+                            std::span<const ShaderVertexInputInfo> vertex_info,
+                            const ShaderPixelInputInfo*            ps_input_info,
+                            const PipelineCache::GraphicsPrograms& programs,
+                            const PipelineStaticParameters&        static_params,
+                            vk::PipelineCache                      driver_cache) {
+	const auto& vs_input_info  = vertex_info.front();
+	const auto& vertex_program = programs.vertex[0];
+	const auto& pixel_program  = programs.pixel;
+	const bool  tessellation   = vertex_info.size() == 3;
 	const bool ps_active = ps_input_info != nullptr;
 	EXIT_IF(!vertex_program || (ps_active && !pixel_program));
 	const bool with_depth = rendering.depth_format != vk::Format::eUndefined ||
@@ -431,72 +225,53 @@ void CreatePipelineInternal(
 	EXIT_IF(!vs_input_info.stage);
 	const bool mesh = vs_input_info.stage.program->stage == ShaderType::Mesh;
 	EXIT_NOT_IMPLEMENTED(mesh && !graphics.mesh_shader_enabled);
-	const auto vertex_stage =
-	    mesh ? vk::ShaderStageFlagBits::eMeshEXT : vk::ShaderStageFlagBits::eVertex;
-
-	const bool rect_list = !mesh && static_params.topology == vk::PrimitiveTopology::ePatchList;
+	const bool rect_list =
+	    !mesh && !tessellation && static_params.topology == vk::PrimitiveTopology::ePatchList;
 
 	vk::ShaderModule tess_control_shader_module = nullptr;
 	vk::ShaderModule tess_eval_shader_module    = nullptr;
 
-	vk::ShaderModuleCreateInfo create_info {};
-	vk::Result result {};
 	if (rect_list) {
 		const auto shaders =
 		    BuildRectListShaders(vs_input_info, ps_active ? ps_input_info : nullptr);
-		create_info.codeSize = shaders.control.size() * 4;
-		create_info.pCode    = shaders.control.data();
-		result =
-		    graphics.device.createShaderModule(&create_info, nullptr, &tess_control_shader_module);
+		tess_control_shader_module = CompileSPV(shaders.control, graphics.device);
 		if (graphics_debug_dump_enabled()) {
-			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done result=%s module=%p\n",
-			     vk::to_string(result).c_str(), static_cast<void*>(tess_control_shader_module));
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done module=%p\n",
+			     static_cast<void*>(tess_control_shader_module));
 		}
-		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
-		create_info.codeSize = shaders.evaluation.size() * 4;
-		create_info.pCode    = shaders.evaluation.data();
-		result =
-		    graphics.device.createShaderModule(&create_info, nullptr, &tess_eval_shader_module);
+		tess_eval_shader_module = CompileSPV(shaders.evaluation, graphics.device);
 		if (graphics_debug_dump_enabled()) {
-			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done result=%s module=%p\n",
-			     vk::to_string(result).c_str(), static_cast<void*>(tess_eval_shader_module));
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done module=%p\n",
+			     static_cast<void*>(tess_eval_shader_module));
 		}
-		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 	}
 
 	EXIT_NOT_IMPLEMENTED(
 	    rect_list && (tess_control_shader_module == nullptr || tess_eval_shader_module == nullptr));
 
-	vk::PipelineShaderStageCreateInfo vert_shader_stage_info {};
-	vert_shader_stage_info.stage  = vertex_stage;
-	vert_shader_stage_info.module = vertex_program.module;
-	vert_shader_stage_info.pName  = "main";
-
-	vk::PipelineShaderStageCreateInfo frag_shader_stage_info {};
-	frag_shader_stage_info.stage  = vk::ShaderStageFlagBits::eFragment;
-	frag_shader_stage_info.module = pixel_program.module;
-	frag_shader_stage_info.pName  = "main";
-
-	vk::PipelineShaderStageCreateInfo tess_control_shader_stage_info {};
-	tess_control_shader_stage_info.stage  = vk::ShaderStageFlagBits::eTessellationControl;
-	tess_control_shader_stage_info.module = tess_control_shader_module;
-	tess_control_shader_stage_info.pName  = "main";
-
-	vk::PipelineShaderStageCreateInfo tess_eval_shader_stage_info {};
-	tess_eval_shader_stage_info.stage  = vk::ShaderStageFlagBits::eTessellationEvaluation;
-	tess_eval_shader_stage_info.module = tess_eval_shader_module;
-	tess_eval_shader_stage_info.pName  = "main";
-
-	vk::PipelineShaderStageCreateInfo shader_stages[4]   = {};
+	vk::PipelineShaderStageCreateInfo shader_stages[4] {};
 	uint32_t                          shader_stage_count = 0;
-	shader_stages[shader_stage_count++]                  = vert_shader_stage_info;
+	for (uint32_t i = 0; i < vertex_info.size(); i++) {
+		shader_stages[shader_stage_count++] = {.stage =
+		                                           NativeShaderStage(vertex_info[i].logical_stage),
+		                                       .module = programs.vertex[i].module,
+		                                       .pName  = "main"};
+	}
 	if (rect_list) {
-		shader_stages[shader_stage_count++] = tess_control_shader_stage_info;
-		shader_stages[shader_stage_count++] = tess_eval_shader_stage_info;
+		shader_stages[shader_stage_count++] = {.stage =
+		                                           vk::ShaderStageFlagBits::eTessellationControl,
+		                                       .module = tess_control_shader_module,
+		                                       .pName  = "main"};
+		shader_stages[shader_stage_count++] = {.stage =
+		                                           vk::ShaderStageFlagBits::eTessellationEvaluation,
+		                                       .module = tess_eval_shader_module,
+		                                       .pName  = "main"};
 	}
 	if (ps_active) {
-		shader_stages[shader_stage_count++] = frag_shader_stage_info;
+		shader_stages[shader_stage_count++] = {.stage  = vk::ShaderStageFlagBits::eFragment,
+		                                       .module = pixel_program.module,
+		                                       .pName  = "main"};
 	}
 
 	vk::VertexInputAttributeDescription input_attr[ShaderVertexInputInfo::RES_MAX] {};
@@ -595,6 +370,15 @@ void CreatePipelineInternal(
 #if !defined(__APPLE__)
 	rasterizer.pNext = &clip_ext;
 #endif
+	vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT provoking_vertex {};
+	EXIT_NOT_IMPLEMENTED(static_params.provoking_vtx_last &&
+	                     !graphics.provoking_vertex_last_enabled);
+	if (graphics.provoking_vertex_last_enabled) {
+		provoking_vertex.provokingVertexMode = static_params.provoking_vtx_last
+		    ? vk::ProvokingVertexModeEXT::eLastVertex : vk::ProvokingVertexModeEXT::eFirstVertex;
+		provoking_vertex.pNext = rasterizer.pNext;
+		rasterizer.pNext = &provoking_vertex;
+	}
 	rasterizer.cullMode  = cull_mode;
 	rasterizer.frontFace = front_face;
 	rasterizer.polygonMode = static_params.polygon_mode;
@@ -610,8 +394,7 @@ void CreatePipelineInternal(
 		EXIT_NOT_IMPLEMENTED((static_params.color_mask[i] & ~0x0fu) != 0);
 		color_blend_attachment[i].colorWriteMask =
 		    vk::ColorComponentFlags {static_params.color_mask[i]};
-		color_blend_attachment[i].blendEnable =
-		    (static_params.blend_enable[i] && !static_params.blend_bypass[i]) ? VK_TRUE : VK_FALSE;
+		color_blend_attachment[i].blendEnable = static_params.blend_enable[i] ? VK_TRUE : VK_FALSE;
 		color_blend_attachment[i].srcColorBlendFactor =
 		    GetBlendFactor(static_params.color_srcblend[i]);
 		color_blend_attachment[i].dstColorBlendFactor =
@@ -649,15 +432,19 @@ void CreatePipelineInternal(
 	color_blending.pAttachments    = color_blend_attachment;
 
 	std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
-	AddLayoutBindings(descriptor_bindings, *vs_input_info.stage.program, vertex_stage);
+	vk::ShaderStageFlags graphics_stages = vk::ShaderStageFlagBits::eFragment;
+	for (const auto& stage: vertex_info) {
+		const auto native_stage = NativeShaderStage(stage.logical_stage);
+		AddLayoutBindings(descriptor_bindings, *stage.stage.program, native_stage);
+		graphics_stages |= native_stage;
+	}
 	if (ps_active) {
 		EXIT_IF(!ps_input_info->stage);
 		AddLayoutBindings(descriptor_bindings, *ps_input_info->stage.program,
 		                  vk::ShaderStageFlagBits::eFragment);
 	}
 	CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
-	const auto                  GraphicsStages = vertex_stage | vk::ShaderStageFlagBits::eFragment;
-	const vk::PushConstantRange push_constants {GraphicsStages, 0,
+	const vk::PushConstantRange push_constants {graphics_stages, 0,
 	                                            ShaderRecompiler::IR::NativePushConstantSize};
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
@@ -673,8 +460,8 @@ void CreatePipelineInternal(
 		     " set_layouts=1 push_constants=%" PRIu32 "\n",
 		     vertex_program.id, ps_active ? pixel_program.id : 0, 1u);
 	}
-	result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
-	                                              &pipeline.pipeline_layout);
+	auto result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
+	                                                   &pipeline.pipeline_layout);
 	if (graphics_debug_dump_enabled()) {
 		LOGF("PipelineTrace: vkCreatePipelineLayout done result=%s layout=%p\n",
 		     vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
@@ -690,15 +477,6 @@ void CreatePipelineInternal(
 #else
 	    (static_params.depth_bounds_test_enable ? VK_TRUE : VK_FALSE);
 #endif
-	depth_stencil_info.stencilTestEnable = (static_params.stencil_test_enable ? VK_TRUE : VK_FALSE);
-	depth_stencil_info.front.failOp      = static_params.stencil_front.failOp;
-	depth_stencil_info.front.passOp      = static_params.stencil_front.passOp;
-	depth_stencil_info.front.depthFailOp = static_params.stencil_front.depthFailOp;
-	depth_stencil_info.front.compareOp   = static_params.stencil_front.compareOp;
-	depth_stencil_info.back.failOp       = static_params.stencil_back.failOp;
-	depth_stencil_info.back.passOp       = static_params.stencil_back.passOp;
-	depth_stencil_info.back.depthFailOp  = static_params.stencil_back.depthFailOp;
-	depth_stencil_info.back.compareOp    = static_params.stencil_back.compareOp;
 	depth_stencil_info.minDepthBounds    = static_params.depth_min_bounds;
 	depth_stencil_info.maxDepthBounds    = static_params.depth_max_bounds;
 
@@ -711,6 +489,8 @@ void CreatePipelineInternal(
 	    vk::DynamicState::eDepthCompareOp,
 	    vk::DynamicState::eDepthBiasEnable,
 	    vk::DynamicState::eDepthBias,
+	    vk::DynamicState::eStencilTestEnable,
+	    vk::DynamicState::eStencilOp,
 	    vk::DynamicState::eStencilCompareMask,
 	    vk::DynamicState::eStencilReference,
 	    vk::DynamicState::eStencilWriteMask,
@@ -741,8 +521,9 @@ void CreatePipelineInternal(
 	pipeline_info.pVertexInputState        = mesh ? nullptr : &vertex_input_info;
 	pipeline_info.pInputAssemblyState      = mesh ? nullptr : &input_assembly;
 	vk::PipelineTessellationStateCreateInfo tessellation_state {};
-	tessellation_state.patchControlPoints = 3;
-	pipeline_info.pTessellationState      = (rect_list ? &tessellation_state : nullptr);
+	tessellation_state.patchControlPoints =
+	    tessellation ? vs_input_info.tess.input_control_points : 3u;
+	pipeline_info.pTessellationState = (rect_list || tessellation) ? &tessellation_state : nullptr;
 	pipeline_info.pViewportState          = &viewport_state;
 	pipeline_info.pRasterizationState     = &rasterizer;
 	pipeline_info.pMultisampleState       = &multisampling;

src/graphics/host_gpu/renderer/render.h

@@ -170,38 +170,34 @@ private:
 	void DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const DrawAutoArgs& args);
 
 	struct GraphicsBindings {
-		PreparedBindings                vertex;
+		std::array<PreparedBindings, 3> vertex;
 		std::optional<PreparedBindings> pixel;
 	};
 
 	[[nodiscard]] TextureBinding ResolveTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	                                            const ShaderRecompiler::IR::DescriptorValue& value);
-	[[nodiscard]] GraphicsBindings PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
-	                                                       const ShaderStageRuntime& pixel,
-	                                                       bool                      pixel_active);
-	void ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer& buffer,
-	                              RenderColorInfo& target, uint32_t render_target_slice_offset = 0,
-	                              uint32_t render_target_slot = UINT32_MAX,
+	void PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
+	                             std::span<RenderColorInfo> colors);
+	void ResolveRenderColorTarget(CommandBuffer& buffer, RenderColorInfo& target,
+	                              uint32_t render_target_slice_offset, uint32_t render_target_slot,
 	                              bool ignore_target_mask = false, bool exact_format = false);
-	void ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer& buffer,
-	                              RenderDepthInfo& target);
-	[[nodiscard]] bool PrepareDrawRenderState(uint64_t submit_id, CommandBuffer& buffer,
+	void ResolveRenderDepthTarget(CommandBuffer& buffer, RenderDepthInfo& target);
+	[[nodiscard]] bool DepthStencilCopy(CommandBuffer& buffer);
+	[[nodiscard]] bool PrepareDrawRenderState(CommandBuffer& buffer,
 	                                          const DrawCallInfo& draw,
 	                                          uint32_t            render_target_slice_offset,
-	                                          bool log_setup_phases, DrawRenderState& state);
+	                                          DrawRenderState& state);
 	void ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buffer, const DrawCallInfo& draw,
 	                         DrawRenderState& state, vk::PrimitiveTopology topology,
 	                         const DrawEmitInfo& emit, const DrawIndexBufferSource& index_source,
-	                         bool primitive_restart_enable, bool log_pipeline_phase,
-	                         bool set_bind_debug, bool set_auto_debug);
+	                         bool primitive_restart_enable);
 	[[nodiscard]] RenderState AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
 	                                               uint32_t color_count, RenderDepthInfo& depth,
 	                                               const std::optional<PreparedBindings>& pixel = std::nullopt);
-	[[nodiscard]] bool        ResolveColorTargets(uint64_t submit_id, CommandBuffer& buffer,
+	[[nodiscard]] bool        ResolveColorTargets(CommandBuffer& buffer,
 	                                              uint32_t render_target_slice_offset);
 	void                      BindImage(ImageId id, bool storage);
 	void                      BindRenderTarget(ImageId id);
-	void                      TrackImageBinding(ImageId id);
 	void                      ResetBindings();
 	[[nodiscard]] bool        TryConsumeComputeMetaClear(const ShaderComputeInputInfo& input,
 	                                                     const CommandBuffer&          buffer);

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -36,18 +36,12 @@
 #include <vector>
 
 namespace Libs::Graphics {
-static uint64_t BufferDescriptorSize(const ShaderBufferResource& descriptor) {
-	const uint64_t records = descriptor.NumRecords();
-	const uint64_t stride  = descriptor.Stride();
-	return stride == 0 ? records : records * stride;
-}
-
 static bool FillSourcesDisjoint(std::span<const ShaderRecompiler::IR::DescriptorValue> sources,
                                  GuestRange destination, uint32_t output_buffer = UINT32_MAX) {
 	for (uint32_t i = 0; i < sources.size(); ++i) {
 		if (i == output_buffer) continue;
 		const auto source = DecodeNativeDescriptor<ShaderBufferResource>(sources[i]);
-		const auto bytes  = BufferDescriptorSize(source);
+		const auto bytes  = source.GetSize();
 		if (source.Base48() < destination.End() && destination.address < source.Base48() + bytes)
 			return false;
 	}
@@ -67,7 +61,7 @@ bool RenderExecutor::TryConsumeComputeMetaClear(const ShaderComputeInputInfo& in
 		const auto  descriptor = DecodeNativeDescriptor<ShaderBufferResource>(resources.buffers[i]);
 		// A metadata resource that is also read is not proven to be a full overwrite. Execute it
 		// conservatively instead of replacing the dispatch with a coarse full-surface clear.
-		if (cache.IsMeta(descriptor.Base48()) && (!resource.written || resource.read)) {
+		if ((!resource.written || resource.read) && cache.IsMeta(descriptor.Base48())) {
 			return false;
 		}
 	}
@@ -115,7 +109,7 @@ bool ResolveComputeBufferFill(const ShaderComputeInputInfo& input, uint32_t grou
 	const uint64_t invocations = input.dispatch_thread_dimensions
 	                                 ? group_x
 	                                 : static_cast<uint64_t>(group_x) * input.threads_num[0];
-	const auto     size        = BufferDescriptorSize(descriptor);
+	const auto     size        = descriptor.GetSize();
 	if (invocations != descriptor.NumRecords() || size == 0 || size > UINT32_MAX ||
 	    (input.dispatch_thread_dimensions &&
 	     (group_x % input.threads_num[0] != 0 || input.dispatch_threads_num[0] != group_x ||
@@ -178,7 +172,7 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
 		                                       1, view.base_layer, view.layer_count};
 		vk::ClearValue clear {};
 		clear.depthStencil = vk::ClearDepthStencilValue {0.0f, fill.value};
-		cache.ClearImage(command, binding.image_id, range, clear);
+		cache.ClearImage(command, binding.image_id, image.backing.format, range, clear);
 		return true;
 	}
 	ShaderBufferResource descriptor;
@@ -189,14 +183,6 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
 		return false;
 	}
 	if (!cache.ClearImageFromBuffer(command, descriptor.Base48(), size, packed_clear)) {
-		// Track deferred DCC state while the original dispatch writes the metadata allocation.
-		cache.TrackDccFill(descriptor.Base48(), size, packed_clear);
-		static std::atomic<uint32_t> logged_metadata_clears {0};
-		if (logged_metadata_clears.fetch_add(1, std::memory_order_relaxed) < 32) {
-			LOGF("GraphicsRenderDispatchDirect: metadata fill shader=0x%016" PRIx64
-			     " addr=0x%016" PRIx64 " size=0x%016" PRIx64 " value=0x%08" PRIx32 "\n",
-			     input.stage.program->shader_hash, descriptor.Base48(), size, packed_clear);
-		}
 		return false;
 	}
 	static std::atomic<uint32_t> logged_clears {0};
@@ -264,9 +250,6 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		input_info.dispatch_threads_num[2]    = thread_group_z;
 	}
 
-	const uint32_t frame_num = static_cast<uint32_t>(m_context.GetGpu().GetFrameNum());
-	const bool     large_workgroup =
-	    (input_info.threads_num[0] * input_info.threads_num[1] * input_info.threads_num[2] >= 512);
 	const auto& program   = *input_info.stage.program;
 	const auto& resources = input_info.stage.resources;
 	if (TryConsumeComputeMetaClear(input_info, buffer)) {
@@ -278,14 +261,17 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		ResetBindings();
 		return;
 	}
-	const auto sampled_images = std::count_if(
-	    program.info.images.begin(), program.info.images.end(), [](const auto& image) {
-		    return image.resource_class == ShaderRecompiler::IR::ImageResourceClass::Sampled;
-	    });
+	const bool large_workgroup =
+	    (input_info.threads_num[0] * input_info.threads_num[1] * input_info.threads_num[2] >= 512);
 	const bool                   has_sampler = !program.info.samplers.empty();
 	static std::atomic<uint32_t> dispatch_log_count {0};
 	if ((large_workgroup || has_sampler) &&
 	    dispatch_log_count.fetch_add(1, std::memory_order_relaxed) < 512) {
+		const auto sampled_images = std::count_if(
+		    program.info.images.begin(), program.info.images.end(), [](const auto& image) {
+			    return image.resource_class == ShaderRecompiler::IR::ImageResourceClass::Sampled;
+		    });
+		const uint32_t frame_num = static_cast<uint32_t>(m_context.GetGpu().GetFrameNum());
 		LOGF("GraphicsRenderDispatchDirect: frame=%u shader=0x%016" PRIx64
 		     " groups=%ux%ux%u mode=0x%08" PRIx32 " local=%ux%ux%u "
 		     "buffers=%zu textures=%zu sampled=%zu storage=%zu samplers=%zu push=%u\n",
@@ -374,14 +360,14 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 
 	buffer.EndRendering();
 	auto& pipeline =
-	    m_context.GetPipelineCache().CreateComputePipeline(input_info, compute_program);
+	    m_context.GetPipelineCache().GetComputePipeline(input_info, compute_program);
 	auto bindings = PrepareBindings(input_info.stage);
 	FindBuffers(bindings);
 	if (program.info.uses_dma) {
-		m_context.GetGpuResources().PrepareBda();
+		m_context.PrepareBda();
 	}
-	RebindBuffers(bindings);
 	RebindImages(bindings);
+	RebindBuffers(bindings);
 
 	auto              vk_buffer        = buffer.Handle();
 	PreparedBindings* descriptor_stage = &bindings;

src/graphics/host_gpu/renderer/renderContext.cpp

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

src/graphics/host_gpu/renderer/renderContext.h

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

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -46,42 +46,34 @@
 
 namespace Libs::Graphics {
 
-int32_t ResolveVertexOffset(uint32_t index_offset, const ShaderVertexInputInfo& vs_input_info) {
-	if (index_offset != 0 || !vs_input_info.fetch_embedded) {
-		return static_cast<int32_t>(index_offset);
+std::pair<int32_t, uint32_t> ResolveDrawOffsets(uint32_t index_offset,
+	                                           const ShaderVertexInputInfo& vs_input_info) {
+	auto     vertex_offset   = static_cast<int32_t>(index_offset);
+	uint32_t instance_offset = 0;
+	if (!vs_input_info.fetch_embedded) {
+		return {vertex_offset, instance_offset};
 	}
 
 	EXIT_IF(!vs_input_info.stage);
 	const auto& program   = *vs_input_info.stage.program;
 	const auto& resources = vs_input_info.stage.resources;
-	if (program.info.vertex_offset_sgpr >= static_cast<int32_t>(program.user_data_base)) {
+	if (index_offset == 0 &&
+	    program.info.vertex_offset_sgpr >= static_cast<int32_t>(program.user_data_base)) {
 		const auto index =
 		    static_cast<uint32_t>(program.info.vertex_offset_sgpr) - program.user_data_base;
 		if (index < resources.user_data.size()) {
-			return static_cast<int32_t>(resources.user_data[index]);
+			vertex_offset = static_cast<int32_t>(resources.user_data[index]);
 		}
 	}
-
-	return 0;
-}
-
-uint32_t ResolveInstanceOffset(const ShaderVertexInputInfo& vs_input_info) {
-	if (!vs_input_info.fetch_embedded) {
-		return 0;
-	}
-
-	EXIT_IF(!vs_input_info.stage);
-	const auto& program   = *vs_input_info.stage.program;
-	const auto& resources = vs_input_info.stage.resources;
 	if (program.info.instance_offset_sgpr >= static_cast<int32_t>(program.user_data_base)) {
 		const auto index =
 		    static_cast<uint32_t>(program.info.instance_offset_sgpr) - program.user_data_base;
 		if (index < resources.user_data.size()) {
-			return resources.user_data[index];
+			instance_offset = resources.user_data[index];
 		}
 	}
 
-	return 0;
+	return {vertex_offset, instance_offset};
 }
 
 static std::atomic<uint32_t> g_draw_state_log_count   = 0;
@@ -411,19 +403,16 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 		vk_buffer.setDepthBias(constant_factor, poly_offset.clamp, slope_factor);
 	}
 
+	vk_buffer.setStencilTestEnable(depth.stencil_test_enable ? VK_TRUE : VK_FALSE);
 	if (depth.stencil_test_enable) {
-		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eFront,
-		                                depth.stencil_dynamic_front.compareMask);
-		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eBack,
-		                                depth.stencil_dynamic_back.compareMask);
-		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eFront,
-		                              depth.stencil_dynamic_front.writeMask);
-		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eBack,
-		                              depth.stencil_dynamic_back.writeMask);
-		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eFront,
-		                              depth.stencil_dynamic_front.reference);
-		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eBack,
-		                              depth.stencil_dynamic_back.reference);
+		const auto set_stencil = [&](vk::StencilFaceFlagBits face, const vk::StencilOpState& state) {
+			vk_buffer.setStencilOp(face, state.failOp, state.passOp, state.depthFailOp, state.compareOp);
+			vk_buffer.setStencilCompareMask(face, state.compareMask);
+			vk_buffer.setStencilWriteMask(face, state.writeMask);
+			vk_buffer.setStencilReference(face, state.reference);
+		};
+		set_stencil(vk::StencilFaceFlagBits::eFront, depth.stencil_front);
+		set_stencil(vk::StencilFaceFlagBits::eBack, depth.stencil_back);
 	}
 
 #if defined(__APPLE__)
@@ -461,17 +450,19 @@ struct DrawRenderState {
 	RenderColorInfo       color_info[RENDER_COLOR_ATTACHMENTS_MAX] = {};
 	uint32_t              color_count                              = 0;
 	bool                  ps_active                                = true;
-	ShaderVertexInputInfo vs_input_info;
+	std::array<ShaderVertexInputInfo, 3> vertex_info;
 	ShaderPixelInputInfo  ps_input_info;
 	PipelineCache::GraphicsPrograms programs;
 };
 
 struct DrawCallInfo {
-	const char*          name           = nullptr;
 	CommandBufferDebugOp debug_op       = CommandBufferDebugOp::DrawIndex;
 	uint32_t             index_count    = 0;
 	uint32_t             instance_count = 0;
 	uint32_t             first_instance = 0;
+
+	[[nodiscard]] bool IsIndexed() const { return debug_op == CommandBufferDebugOp::DrawIndex; }
+	[[nodiscard]] const char* Name() const { return IsIndexed() ? "DrawIndex" : "DrawIndexAuto"; }
 };
 
 RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
@@ -488,14 +479,10 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 	for (uint32_t i = 0; i < color_count; i++) {
 		auto& target = colors[i];
 		EXIT_IF(!target.image_id);
-		const auto old_image = cache.m_slot_images.try_get(target.image_id);
-		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
-		    old_image->binding.needs_rebind) {
-			if (old_image != nullptr) {
-				old_image->binding = {};
-			}
-			target.image_id = cache.FindImage(target.desc);
-			BindRenderTarget(target.image_id);
+		const auto owner = cache.m_slot_images.try_get(target.image_id);
+		if (owner == nullptr || (!owner->registered && !owner->info.data.Empty()) ||
+		    owner->binding.needs_rebind) {
+			EXIT("color target changed after render-state discovery\n");
 		}
 		const auto image_view = cache.FindRenderTarget(target.image_id, target.desc);
 		auto&      image      = cache.GetImage(target.image_id);
@@ -544,11 +531,11 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		    !cache.ClearMeta(metadata.range.address)) {
 			EXIT("failed to acquire HTile metadata for a depth clear\n");
 		}
-		depth.depth_meta_clear_enable =
+		const bool meta_clear =
 		    metadata.kind == ImageMetadataKind::Htile &&
 		    cache.IsMetaCleared(metadata.range.address, depth.desc.view_info.base_layer);
-		depth.depth_load_clear_enable = depth.depth_clear_enable || depth.depth_meta_clear_enable;
-		if (depth.depth_meta_clear_enable &&
+		depth.depth_load_clear_enable = depth.depth_clear_enable || meta_clear;
+		if (meta_clear &&
 		    !cache.TouchMeta(metadata.range.address, depth.desc.view_info.base_layer, false)) {
 			EXIT("failed to consume HTile clear state\n");
 		}
@@ -662,7 +649,6 @@ static bool ConsumeMetadataColorOperation(const CommandBuffer& buffer) {
 }
 
 struct DrawEmitInfo {
-	bool     indexed       = false;
 	int32_t  vertex_offset = 0;
 	uint32_t first_vertex  = 0;
 	uint32_t first_instance = 0;
@@ -716,6 +702,7 @@ struct PreparedVertexBuffers {
 
 	std::array<vk::Buffer, MaxBuffers>     buffers {};
 	std::array<vk::DeviceSize, MaxBuffers> offsets {};
+	std::array<vk::DeviceSize, MaxBuffers> sizes {};
 	uint32_t                               count = 0;
 };
 
@@ -800,6 +787,7 @@ static PreparedVertexBuffers AcquireVertexBuffers(CommandBuffer&               b
 
 		prepared.buffers[i] = range->binding.first->Handle();
 		prepared.offsets[i] = range->binding.second + vertex.addr - range->base_address;
+		prepared.sizes[i]   = std::min(size, range->acquired_end - vertex.addr);
 		SetVulkanObjectNameF(
 		    buffer.GetContext().GetGraphics().device, prepared.buffers[i],
 		    "Kyty.VertexBuffer[slot={} guest=0x{:016x} size=0x{:x} stride={} records={}]", i,
@@ -838,6 +826,7 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 		case Prospero::PrimitiveType::kTriStrip:
 			topology = vk::PrimitiveTopology::eTriangleStrip;
 			break;
+		case Prospero::PrimitiveType::kPatch:
 		case Prospero::PrimitiveType::kRectList:
 			topology = vk::PrimitiveTopology::ePatchList;
 			break;
@@ -850,14 +839,21 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			topology = vk::PrimitiveTopology::eTriangleFan;
 			break;
-		default: EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
+		default: {
+			static std::atomic_bool logged = false;
+			if (!logged.exchange(true, std::memory_order_relaxed)) {
+				std::printf("Skipping draw with unknown primitive type: %u\n",
+				            static_cast<uint32_t>(ucfg.GetPrimType()));
+			}
+			return false;
+		}
 	}
 
 	return true;
 }
 
-static bool ResolvePrimitiveRestart(const CommandBuffer& buffer, vk::PrimitiveTopology topology,
-                                    uint32_t index_type_and_size) {
+static bool ResolvePrimitiveRestart(const CommandBuffer& buffer,
+                                    const DrawIndexBufferSource& source) {
 	const auto control = buffer.GetUserConfig().GetPrimitiveResetControl();
 	EXIT_NOT_IMPLEMENTED((control & ~0x3u) != 0);
 	if ((control & 0x1u) == 0) {
@@ -869,58 +865,61 @@ static bool ResolvePrimitiveRestart(const CommandBuffer& buffer, vk::PrimitiveTo
 		case Prospero::PrimitiveType::kTriStrip: break;
 		default: return false;
 	}
-	if (topology != vk::PrimitiveTopology::eLineStrip &&
-	    topology != vk::PrimitiveTopology::eTriangleStrip &&
-	    topology != vk::PrimitiveTopology::eTriangleFan) {
-		return false;
-	}
 
-	uint32_t index_mask = 0;
-	switch (static_cast<Prospero::IndexType>(index_type_and_size)) {
-		case Prospero::IndexType::kIndex8: index_mask = 0xffu; break;
-		case Prospero::IndexType::kIndex16: index_mask = 0xffffu; break;
-		case Prospero::IndexType::kIndex32: index_mask = 0xffffffffu; break;
-		default: EXIT("unknown index_type_and_size: %u\n", index_type_and_size);
-	}
-
-	const auto reset_index = buffer.GetRegisters().GetPrimitiveResetIndex();
+	const auto element_size = source.guest_element_size;
+	const auto index_mask   = UINT32_MAX >> ((4 - element_size) * 8);
+	const auto reset_index  = buffer.GetRegisters().GetPrimitiveResetIndex();
 	if ((control & 0x2u) != 0 && (reset_index & ~index_mask) != 0) {
 		return false;
 	}
-	EXIT_NOT_IMPLEMENTED((reset_index & index_mask) != index_mask);
-	return true;
+	const auto restart_index = reset_index & index_mask;
+	if (restart_index == index_mask) {
+		// Use native restart; the 8-bit path widens its marker to 0xffff.
+		return true;
+	}
+
+	// A game can set a custom reset value without using it in the index buffer.
+	// Keep restart off in that case; fail if we actually find the value.
+	// Scan before preparing draw resources: readback can restart the command buffer.
+	EXIT_NOT_IMPLEMENTED(source.address == 0);
+	const auto* indices = reinterpret_cast<const uint8_t*>(source.address);
+	for (uint64_t offset = 0; offset < source.size; offset += element_size) {
+		uint32_t index = 0;
+		std::memcpy(&index, indices + offset, element_size);
+		EXIT_NOT_IMPLEMENTED(index == restart_index);
+	}
+	return false;
 }
 
-bool RenderExecutor::PrepareDrawRenderState(uint64_t submit_id, CommandBuffer& buffer,
-                                            const DrawCallInfo& draw,
+bool RenderExecutor::PrepareDrawRenderState(CommandBuffer& buffer, const DrawCallInfo& draw,
                                             uint32_t            render_target_slice_offset,
-                                            bool log_setup_phases, DrawRenderState& state) {
+	                                        DrawRenderState& state) {
 	auto& ctx = buffer.GetRegisters();
 
-	if (ResolveColorTargets(submit_id, buffer, render_target_slice_offset)) {
+	if (ResolveColorTargets(buffer, render_target_slice_offset)) {
 		return false;
 	}
-	if (log_setup_phases) {
-		LogDrawPhase(draw.name, "ResolveRenderColorTarget");
+	if (draw.IsIndexed()) {
+		LogDrawPhase(draw.Name(), "ResolveRenderColorTarget");
 	}
 	for (uint32_t slot = 0; slot < RENDER_COLOR_ATTACHMENTS_MAX; slot++) {
 		if (slot == 0 || (render_target_mask_slot(ctx.GetRenderTargetMask(), slot) != 0 &&
 		                  ctx.GetRenderTarget(slot).base.addr != 0)) {
-			ResolveRenderColorTarget(submit_id, buffer, state.color_info[state.color_count],
+			ResolveRenderColorTarget(buffer, state.color_info[state.color_count],
 			                         render_target_slice_offset, slot);
 			if (state.color_info[state.color_count].image_id) {
 				state.color_count++;
 			}
 		}
 	}
-	if (log_setup_phases) {
-		LogDrawPhase(draw.name, "ResolveRenderDepthTarget");
+	if (draw.IsIndexed()) {
+		LogDrawPhase(draw.Name(), "ResolveRenderDepthTarget");
 	}
-	ResolveRenderDepthTarget(submit_id, buffer, state.depth_info);
+	ResolveRenderDepthTarget(buffer, state.depth_info);
 
 	state.ps_active       = DrawHasActivePixelShader(buffer);
 	if (state.color_count == 0 && !state.depth_info.image_id && !state.ps_active) {
-		LogFramebufferSkip(draw.name, state.color_info[0], state.depth_info, buffer,
+		LogFramebufferSkip(draw.Name(), state.color_info[0], state.depth_info, buffer,
 		                   draw.index_count, 0);
 		return false;
 	}
@@ -928,7 +927,7 @@ bool RenderExecutor::PrepareDrawRenderState(uint64_t submit_id, CommandBuffer& b
 	return true;
 }
 
-static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw, bool log_phases,
+static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw,
                            DrawRenderState& state) {
 	auto& ctx    = buffer.GetRegisters();
 	auto& sh_ctx = buffer.GetShaders();
@@ -945,12 +944,12 @@ static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw, bool
 		target_export_mapping[state.color_info[i].target_slot] = state.color_info[i].export_mapping;
 	}
 	auto& pipeline_cache = buffer.GetContext().GetPipelineCache();
-	if (log_phases) {
-		LogDrawPhase(draw.name, "GetGraphicsPrograms");
+	if (draw.IsIndexed()) {
+		LogDrawPhase(draw.Name(), "GetGraphicsPrograms");
 	}
 	state.programs = pipeline_cache.GetGraphicsPrograms(
 	    vertex_shader_info, pixel_shader_info, shader_regs, ctx, buffer.GetUserConfig(),
-	    target_export_mapping, state.ps_active, state.vs_input_info, state.ps_input_info);
+	    target_export_mapping, state.ps_active, state.vertex_info, state.ps_input_info);
 }
 
 static PreparedIndexBuffer PrepareIndexBuffer(CommandBuffer&               buffer,
@@ -964,17 +963,14 @@ static PreparedIndexBuffer PrepareIndexBuffer(CommandBuffer&               buffe
 		auto& stream = buffer.GetContext().GetBufferCache().GetUtilityBuffer(MemoryUsage::Stream);
 		prepared.offset = stream.Copy(source.host_data, source.size, 16);
 		prepared.buffer = stream.Handle();
+		SetVulkanObjectNameF(buffer.GetContext().GetGraphics().device, prepared.buffer,
+		                     "Kyty.IndexBuffer[guest=transient size=0x{:x} type={}]", source.size,
+		                     static_cast<uint32_t>(source.type));
 	} else {
 		auto [buffer_ptr, offset] =
 		    buffer.GetContext().GetBufferCache().ObtainBuffer(source.address, source.size, false);
 		prepared.buffer = buffer_ptr->Handle();
 		prepared.offset = offset;
-	}
-	if (source.host_data != nullptr) {
-		SetVulkanObjectNameF(buffer.GetContext().GetGraphics().device, prepared.buffer,
-		                     "Kyty.IndexBuffer[guest=transient size=0x{:x} type={}]", source.size,
-		                     static_cast<uint32_t>(source.type));
-	} else {
 		SetVulkanObjectNameF(buffer.GetContext().GetGraphics().device, prepared.buffer,
 		                     "Kyty.IndexBuffer[guest=0x{:016x} size=0x{:x} type={}]",
 		                     source.address, source.size, static_cast<uint32_t>(source.type));
@@ -988,8 +984,9 @@ static void CommitVertexBuffers(vk::CommandBuffer            vk_buffer,
 		EXIT_IF(prepared.buffers[i] == nullptr);
 	}
 	if (prepared.count != 0) {
-		vk_buffer.bindVertexBuffers(0, prepared.count, prepared.buffers.data(),
-		                            prepared.offsets.data());
+		// Guest descriptor bounds must survive allocation merging in the cache.
+		vk_buffer.bindVertexBuffers2(0, prepared.count, prepared.buffers.data(),
+		                             prepared.offsets.data(), prepared.sizes.data(), nullptr);
 	}
 }
 
@@ -1001,22 +998,22 @@ static void CommitIndexBuffer(vk::CommandBuffer vk_buffer, const PreparedIndexBu
 }
 
 static void LogDrawStateIfNeeded(const CommandBuffer& buffer, const DrawCallInfo& draw,
-                                 const DrawRenderState& state, bool always_log,
-                                 bool force_legacy_rect_log, uint32_t index_type_and_size,
+	                             const DrawRenderState& state, uint32_t index_type_and_size,
                                  const void* index_addr) {
 	if (!graphics_debug_dump_enabled()) {
 		return;
 	}
 
-	if (!always_log && !force_legacy_rect_log) {
+	if (!draw.IsIndexed() &&
+	    buffer.GetUserConfig().GetPrimType() != Prospero::PrimitiveType::kRectListLegacy) {
 		return;
 	}
 
 	if (state.ps_active) {
-		LogDrawTargetState(draw.name, state.color_info[0], state.depth_info, buffer,
+		LogDrawTargetState(draw.Name(), state.color_info[0], state.depth_info, buffer,
 		                   state.ps_input_info, draw.index_count, 0);
 	}
-	LogDrawInputState(buffer, state.color_info[0], state.vs_input_info, index_type_and_size,
+	LogDrawInputState(buffer, state.color_info[0], state.vertex_info[0], index_type_and_size,
 	                  draw.index_count, index_addr);
 }
 
@@ -1031,7 +1028,8 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 		case Prospero::PrimitiveType::kTriFan:
 		case Prospero::PrimitiveType::kTriStrip:
 		case Prospero::PrimitiveType::kRectList:
-			if (emit.indexed) {
+		case Prospero::PrimitiveType::kPatch:
+			if (draw.IsIndexed()) {
 				vk_buffer.drawIndexed(draw.index_count, draw.instance_count, 0, emit.vertex_offset,
 				                      emit.first_instance);
 			} else {
@@ -1040,7 +1038,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 			}
 			break;
 		case Prospero::PrimitiveType::kRectListLegacy:
-			if (emit.indexed) {
+			if (draw.IsIndexed()) {
 				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 			}
 			// Sarah
@@ -1050,7 +1048,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			EXIT_NOT_IMPLEMENTED((draw.index_count & 0x3u) != 0);
 			for (uint32_t i = 0; i < draw.index_count; i += 4) {
-				if (emit.indexed) {
+				if (draw.IsIndexed()) {
 					vk_buffer.drawIndexed(4, draw.instance_count, i, emit.vertex_offset,
 					                      emit.first_instance);
 				} else {
@@ -1067,16 +1065,17 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
                                          const DrawCallInfo& draw, DrawRenderState& state,
                                          vk::PrimitiveTopology topology, const DrawEmitInfo& emit,
                                          const DrawIndexBufferSource& index_source,
-                                         bool primitive_restart_enable, bool log_pipeline_phase,
-                                         bool set_bind_debug, bool set_auto_debug) {
+	                                     bool primitive_restart_enable) {
 	auto& ucfg = buffer.GetUserConfig();
-	const bool mesh_active = state.vs_input_info.stage.program->stage == ShaderType::Mesh;
+	const auto vertex_stages =
+	    std::span {state.vertex_info.data(), state.programs.VertexStageCount()};
+	const bool mesh_active = state.vertex_info[0].stage.program->stage == ShaderType::Mesh;
 	uint32_t   mesh_groups = 0;
 	if (mesh_active) {
-		const auto& mesh = state.vs_input_info.mesh;
+		const auto& mesh = state.vertex_info[0].mesh;
 		if (primitive_restart_enable || mesh.primitives_per_group == 0) {
 			EXIT("unsupported mesh draw: primitive=%u indexed=%u restart=%u\n",
-			     static_cast<uint32_t>(ucfg.GetPrimType()), emit.indexed, primitive_restart_enable);
+			     static_cast<uint32_t>(ucfg.GetPrimType()), draw.IsIndexed(), primitive_restart_enable);
 		}
 		const auto primitives = mesh.InputPrimitiveCount(draw.index_count);
 		if (primitives == 0 || draw.instance_count == 0) {
@@ -1093,64 +1092,62 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		}
 	}
 
-	if (mesh_active && emit.indexed) {
+	if (mesh_active && draw.IsIndexed()) {
 		// Register the original guest indices for shader reads; PrepareGraphicsBindings
 		// synchronizes registered BDA ranges before any draw commands are committed.
 		(void)m_context.GetBufferCache().FindBuffer(
 		    index_source.address, static_cast<uint64_t>(draw.index_count) *
 		                              index_source.guest_element_size);
 	}
-	LogDrawPhase(draw.name, "PrepareBindings");
-	auto bindings = PrepareGraphicsBindings(state.vs_input_info.stage, state.ps_input_info.stage,
-	                                        state.ps_active);
+	LogDrawPhase(draw.Name(), "PrepareBindings");
+	GraphicsBindings                 bindings;
+	std::array<PreparedBindings*, 4> descriptor_stages {};
+	uint32_t                         stage_count = 0;
+	for (uint32_t i = 0; i < vertex_stages.size(); i++) {
+		bindings.vertex[i]               = PrepareBindings(state.vertex_info[i].stage);
+		descriptor_stages[stage_count++] = &bindings.vertex[i];
+	}
+	if (state.ps_active) {
+		bindings.pixel.emplace(PrepareBindings(state.ps_input_info.stage));
+		descriptor_stages[stage_count++] = &*bindings.pixel;
+	}
+	const auto stages = std::span {descriptor_stages.data(), stage_count};
+	PrepareGraphicsBindings(stages, std::span {state.color_info, state.color_count});
 	PreparedVertexBuffers vertex_bindings;
 	PreparedIndexBuffer   index_binding;
 	if (!mesh_active) {
-		LogDrawPhase(draw.name, "PrepareVertexBuffers");
-		vertex_bindings = AcquireVertexBuffers(buffer, state.vs_input_info);
+		LogDrawPhase(draw.Name(), "PrepareVertexBuffers");
+		vertex_bindings = AcquireVertexBuffers(buffer, state.vertex_info[0]);
 		index_binding   = PrepareIndexBuffer(buffer, index_source);
 	}
 	const auto rendering =
 	    AcquireRenderTargets(buffer, state.color_info, state.color_count, state.depth_info,
 	                         bindings.pixel);
 
-	if (log_pipeline_phase) {
-		LogDrawPhase(draw.name, "CreatePipeline");
+	if (draw.IsIndexed()) {
+		LogDrawPhase(draw.Name(), "CreatePipeline");
 	}
-	auto& pipeline = m_context.GetPipelineCache().CreateGraphicsPipeline(
-	    std::span {state.color_info, state.color_count}, state.depth_info, state.vs_input_info, buffer,
+	auto& pipeline = m_context.GetPipelineCache().GetGraphicsPipeline(
+	    std::span {state.color_info, state.color_count}, state.depth_info, vertex_stages, buffer,
 	    state.ps_active ? &state.ps_input_info : nullptr, topology, primitive_restart_enable,
-	    state.programs.vertex, state.programs.pixel);
+	    state.programs);
 
 	// Resource preparation above may synchronously finish and restart the scheduler. From this
 	// point onward, every operation targets the current command buffer and cannot touch guest
 	// memory.
 	auto vk_buffer = buffer.Handle();
-	if (set_bind_debug) {
-		SetDrawDebugPhase(buffer, submit_id, draw, 0x100u);
-	}
-	if (set_auto_debug) {
-		SetDrawDebugPhase(buffer, submit_id, draw, 0x200u);
-	}
+	SetDrawDebugPhase(buffer, submit_id, draw, draw.IsIndexed() ? 0x100u : 0x200u);
 	if (!mesh_active) {
 		CommitVertexBuffers(vk_buffer, vertex_bindings);
 	}
-	if (bindings.pixel.has_value()) {
-		if (set_auto_debug) {
-			SetDrawDebugPhase(buffer, submit_id, draw, 0x300u);
-		}
+	if (bindings.pixel && !draw.IsIndexed()) {
+		SetDrawDebugPhase(buffer, submit_id, draw, 0x300u);
 	}
-	std::array<PreparedBindings*, 2> descriptor_stages {&bindings.vertex, nullptr};
-	const size_t                     descriptor_stage_count = bindings.pixel.has_value() ? 2u : 1u;
-	if (bindings.pixel) {
-		descriptor_stages[1] = &*bindings.pixel;
-	}
-	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline,
-	               std::span {descriptor_stages.data(), descriptor_stage_count});
+	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline, stages);
 	if (mesh_active) {
 		const uint32_t draw_data[] {
 		    draw.index_count,
-		    emit.indexed ? static_cast<uint32_t>(emit.vertex_offset) : emit.first_vertex,
+		    draw.IsIndexed() ? static_cast<uint32_t>(emit.vertex_offset) : emit.first_vertex,
 		    emit.first_instance, index_source.guest_element_size,
 		    static_cast<uint32_t>(index_source.address),
 		    static_cast<uint32_t>(index_source.address >> 32u)};
@@ -1163,7 +1160,7 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		CommitIndexBuffer(vk_buffer, index_binding);
 	}
 
-	SetGraphicsDynamicParams(buffer, vk_buffer, state.vs_input_info, state.color_info,
+	SetGraphicsDynamicParams(buffer, vk_buffer, vertex_stages.back(), state.color_info,
 	                         state.color_count, state.depth_info);
 	if (m_context.GetGraphics().attachment_feedback_loop_enabled) {
 		vk_buffer.setAttachmentFeedbackLoopEnableEXT(
@@ -1173,28 +1170,29 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		        : vk::ImageAspectFlags {});
 	}
 
-	LogDrawPhase(draw.name, "BeginRendering");
-	if (set_auto_debug) {
+	LogDrawPhase(draw.Name(), "BeginRendering");
+	if (!draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x400u);
 	}
 	m_context.GetCommandScheduler().BeginRendering(rendering);
 	vk_buffer.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline.pipeline);
-	if (set_auto_debug) {
+	if (!draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x500u);
 	}
 	if (mesh_active) {
 		vk_buffer.drawMeshTasksEXT(mesh_groups, draw.instance_count, 1);
 	} else {
-		EmitDrawPrimitives(ucfg, vk_buffer, state.vs_input_info, draw, emit);
+		EmitDrawPrimitives(ucfg, vk_buffer, state.vertex_info[0], draw, emit);
 	}
 
-	if (set_auto_debug) {
+	if (!draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x600u);
 	}
 	vk::PipelineStageFlags shader_write_stages = {};
-	if (HasShaderBufferWrites(state.vs_input_info.stage)) {
-		shader_write_stages |= mesh_active ? vk::PipelineStageFlagBits::eMeshShaderEXT
-		                                   : vk::PipelineStageFlagBits::eVertexShader;
+	for (const auto& stage: vertex_stages) {
+		if (HasShaderBufferWrites(stage.stage)) {
+			shader_write_stages |= ShaderPipelineStages(NativeShaderStage(stage.logical_stage));
+		}
 	}
 	if (state.ps_active && HasShaderBufferWrites(state.ps_input_info.stage)) {
 		shader_write_stages |= vk::PipelineStageFlagBits::eFragmentShader;
@@ -1203,8 +1201,8 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		m_context.GetCommandScheduler().EndRendering();
 		ShaderWriteBarrier(vk_buffer, shader_write_stages);
 	}
-	LogDrawPhase(draw.name, "DrawComplete");
-	if (set_auto_debug) {
+	LogDrawPhase(draw.Name(), "DrawComplete");
+	if (!draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x700u);
 	}
 }
@@ -1228,7 +1226,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 		return;
 	}
 
-	if (ConsumeMetadataColorOperation(buffer)) {
+	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer)) {
 		ResetBindings();
 		return;
 	}
@@ -1238,7 +1236,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 	}
 
 	if (graphics_debug_dump_enabled()) {
-		sh_print("GraphicsRenderDrawIndex():Shader:", sh_ctx);
+		LOGF("GraphicsRenderDrawIndex():Shader:\n");
 		uc_print("GraphicsRenderDrawIndex():UserConfig:", ucfg);
 		hw_print(buffer);
 
@@ -1263,77 +1261,61 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 		return;
 	}
 
-	vk::IndexType index_type           = vk::IndexType::eUint16;
-	uint64_t      index_size           = 0;
-	bool          expand_index8_to_u16 = false;
-	const bool    primitive_restart =
-	    ResolvePrimitiveRestart(buffer, topology, args.index_type_and_size);
-
+	DrawIndexBufferSource index_source {};
+	index_source.address = reinterpret_cast<uint64_t>(args.index_addr);
 	switch (static_cast<Prospero::IndexType>(args.index_type_and_size)) {
 		case Prospero::IndexType::kIndex16:
-			index_type = vk::IndexType::eUint16;
-			index_size = 2 * static_cast<uint64_t>(args.index_count);
+			index_source.type               = vk::IndexType::eUint16;
+			index_source.guest_element_size = 2;
 			break;
 		case Prospero::IndexType::kIndex32:
-			index_type = vk::IndexType::eUint32;
-			index_size = 4 * static_cast<uint64_t>(args.index_count);
+			index_source.type               = vk::IndexType::eUint32;
+			index_source.guest_element_size = 4;
 			break;
-		// Some games use it - need vulkan extension
 		case Prospero::IndexType::kIndex8:
-			index_type           = vk::IndexType::eUint16;
-			index_size           = static_cast<uint64_t>(args.index_count);
-			expand_index8_to_u16 = true;
+			index_source.guest_element_size = 1;
 			break;
 		default: EXIT("unknown index_type_and_size: %u\n", args.index_type_and_size);
 	}
+	index_source.size = static_cast<uint64_t>(args.index_count) * index_source.guest_element_size;
+	const bool primitive_restart = ResolvePrimitiveRestart(buffer, index_source);
 
-	const DrawCallInfo    draw {"DrawIndex", CommandBufferDebugOp::DrawIndex, args.index_count,
-	                            args.instance_count, args.first_instance};
 	std::vector<uint16_t> expanded_indices;
-	if (expand_index8_to_u16) {
+	if (index_source.guest_element_size == 1) {
 		EXIT_NOT_IMPLEMENTED(args.index_addr == nullptr);
 		const auto* src = static_cast<const uint8_t*>(args.index_addr);
 		expanded_indices.resize(args.index_count);
 		for (uint32_t i = 0; i < args.index_count; i++) {
 			expanded_indices[i] = primitive_restart && src[i] == 0xffu ? 0xffffu : src[i];
 		}
+		index_source.host_data = expanded_indices.data();
+		index_source.size      = expanded_indices.size() * sizeof(uint16_t);
 	}
 
-	DrawIndexBufferSource index_source {};
-	index_source.address = reinterpret_cast<uint64_t>(args.index_addr);
-	index_source.host_data =
-	    expanded_indices.empty() ? nullptr : static_cast<const void*>(expanded_indices.data());
-	index_source.size =
-	    expanded_indices.empty() ? index_size : expanded_indices.size() * sizeof(uint16_t);
-	index_source.type = index_type;
-	index_source.guest_element_size = static_cast<uint32_t>(index_size / args.index_count);
-
+	const DrawCallInfo draw {CommandBufferDebugOp::DrawIndex, args.index_count,
+	                        args.instance_count, args.first_instance};
 	DrawRenderState state {};
-	if (!PrepareDrawRenderState(submit_id, buffer, draw, args.render_target_slice_offset, true,
-	                            state)) {
+	if (!PrepareDrawRenderState(buffer, draw, args.render_target_slice_offset, state)) {
 		ResetBindings();
 		return;
 	}
 
-	RefreshShaders(buffer, draw, true, state);
+	RefreshShaders(buffer, draw, state);
 
-	LogDrawStateIfNeeded(buffer, draw, state, true, false, args.index_type_and_size,
+	LogDrawStateIfNeeded(buffer, draw, state, args.index_type_and_size,
 	                     args.index_addr);
 
 	const bool indirect = args.offset_source == DrawOffsetSource::IndirectArgs;
-	const auto vertex_offset =
-	    indirect
-	        ? args.base_vertex
-	        : ResolveVertexOffset(ucfg.GetIndexOffset(), state.vs_input_info) + args.base_vertex;
+	const auto [vertex_offset, instance_offset] =
+	    indirect ? std::pair<int32_t, uint32_t> {0, args.first_instance}
+	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vertex_info[0]);
 
 	DrawEmitInfo emit {};
-	emit.indexed       = true;
-	emit.vertex_offset = vertex_offset;
-	emit.first_instance =
-	    indirect ? args.first_instance : ResolveInstanceOffset(state.vs_input_info);
+	emit.vertex_offset  = vertex_offset + args.base_vertex;
+	emit.first_instance = instance_offset;
 
 	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source,
-	                    primitive_restart, true, true, false);
+	                    primitive_restart);
 	ResetBindings();
 }
 
@@ -1356,7 +1338,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 		return;
 	}
 
-	if (ConsumeMetadataColorOperation(buffer)) {
+	if (ConsumeMetadataColorOperation(buffer) || DepthStencilCopy(buffer)) {
 		ResetBindings();
 		return;
 	}
@@ -1366,7 +1348,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 	}
 
 	if (graphics_debug_dump_enabled()) {
-		sh_print("GraphicsRenderDrawIndexAuto():Shader:", sh_ctx);
+		LOGF("GraphicsRenderDrawIndexAuto():Shader:\n");
 		uc_print("GraphicsRenderDrawIndexAuto():UserConfig:", ucfg);
 		hw_print(buffer);
 
@@ -1382,12 +1364,11 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 
 	hw_check(buffer);
 
-	const DrawCallInfo draw {"DrawIndexAuto", CommandBufferDebugOp::DrawIndexAuto,
+	const DrawCallInfo draw {CommandBufferDebugOp::DrawIndexAuto,
 	                         args.vertex_count, args.instance_count, args.first_instance};
 
 	DrawRenderState state {};
-	if (!PrepareDrawRenderState(submit_id, buffer, draw, args.render_target_slice_offset, false,
-	                            state)) {
+	if (!PrepareDrawRenderState(buffer, draw, args.render_target_slice_offset, state)) {
 		ResetBindings();
 		return;
 	}
@@ -1397,11 +1378,11 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 		ResetBindings();
 		return;
 	}
-	RefreshShaders(buffer, draw, false, state);
+	RefreshShaders(buffer, draw, state);
 
-	const bool rect_list = topology == vk::PrimitiveTopology::ePatchList;
-	if (rect_list && state.vs_input_info.buffers_num == 0 &&
-	    state.vs_input_info.stage.program->param_export_mask == 0 &&
+	const bool rect_list = ucfg.GetPrimType() == Prospero::PrimitiveType::kRectList;
+	if (rect_list && state.vertex_info[0].buffers_num == 0 &&
+	    state.vertex_info[0].stage.program->param_export_mask == 0 &&
 	    state.ps_input_info.input_num != 0) {
 		if (graphics_debug_dump_enabled()) {
 			LOGF("DrawIndexAuto: skipping rect-list draw with no VS param exports and PS inputs: "
@@ -1413,28 +1394,22 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 		return;
 	}
 
-	LogDrawStateIfNeeded(buffer, draw, state, false,
-	                     ucfg.GetPrimType() == Prospero::PrimitiveType::kRectListLegacy, 0,
-	                     nullptr);
+	LogDrawStateIfNeeded(buffer, draw, state, 0, nullptr);
 
 	const bool indirect = args.offset_source == DrawOffsetSource::IndirectArgs;
-	const auto vertex_offset =
-	    indirect ? static_cast<int32_t>(args.first_vertex)
-	             : ResolveVertexOffset(ucfg.GetIndexOffset(), state.vs_input_info) +
-	                   static_cast<int32_t>(args.first_vertex);
+	const auto [vertex_offset, instance_offset] =
+	    indirect ? std::pair<int32_t, uint32_t> {0, args.first_instance}
+	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vertex_info[0]);
 	DrawEmitInfo emit {};
-	emit.first_vertex = static_cast<uint32_t>(vertex_offset);
-	emit.first_instance =
-	    indirect ? args.first_instance : ResolveInstanceOffset(state.vs_input_info);
+	emit.first_vertex = static_cast<uint32_t>(vertex_offset + static_cast<int32_t>(args.first_vertex));
+	emit.first_instance = instance_offset;
 
 	DrawIndexBufferSource index_source {};
-	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source, false, false,
-	                    false, true);
+	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source, false);
 	ResetBindings();
 }
 
-bool RenderExecutor::ResolveColorTargets(uint64_t submit_id, CommandBuffer& buffer,
-                                         uint32_t render_target_slice_offset) {
+bool RenderExecutor::ResolveColorTargets(CommandBuffer& buffer, uint32_t render_target_slice_offset) {
 	const auto& hw = buffer.GetRegisters();
 	if (hw.GetColorControl().mode != 3) {
 		return false;
@@ -1448,8 +1423,8 @@ bool RenderExecutor::ResolveColorTargets(uint64_t submit_id, CommandBuffer& buff
 
 	RenderColorInfo src {};
 	RenderColorInfo dst {};
-	ResolveRenderColorTarget(submit_id, buffer, src, render_target_slice_offset, 0, true, true);
-	ResolveRenderColorTarget(submit_id, buffer, dst, render_target_slice_offset, 1, true, true);
+	ResolveRenderColorTarget(buffer, src, render_target_slice_offset, 0, true, true);
+	ResolveRenderColorTarget(buffer, dst, render_target_slice_offset, 1, true, true);
 	if (!src.image_id || !dst.image_id) {
 		return false;
 	}

src/graphics/host_gpu/renderer/renderDraw.h

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

src/graphics/host_gpu/renderer/renderTarget.h

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

src/graphics/host_gpu/renderer/sync.cpp

@@ -13,10 +13,8 @@
 #include "kernel/pthread.h"
 #include "libs/errno.h"
 
-#include <array>
 #include <cstring>
 #include <limits>
-#include <optional>
 
 namespace Libs::Graphics::Sync {
 
@@ -55,70 +53,9 @@ uint64_t ReadReferenceClock() {
 	return value;
 }
 
-enum class EndOfPipeCompletion { None, Interrupt, Flip, FlipAndInterrupt };
-
-struct EndOfPipeSignal {
-	CommandBuffer*          buffer          = nullptr;
-	uint64_t                submit_id       = 0;
-	CommandBufferDebugOp    debug_operation = CommandBufferDebugOp::Unknown;
-	std::array<uint32_t, 4> debug_args      = {};
-	uint64_t                debug_data      = 0;
-	std::optional<uint64_t> destination;
-	EndOfPipeCompletion     completion         = EndOfPipeCompletion::None;
-	uint64_t                completion_data    = 0;
-	int                     interrupt_event_id = 0;
-};
-
 enum class EndOfPipeWriteSize : uint32_t { Dword = 4, Qword = 8 };
 enum class EndOfPipeWriteAction { Write, WriteBack, Interrupt, InterruptWriteBack };
 
-static void ValidateEndOfPipeSignal(const EndOfPipeSignal& signal) {
-	if (signal.destination.has_value()) {
-		EXIT_IF(*signal.destination == 0);
-	}
-	EXIT_IF(signal.buffer == nullptr);
-	(void)signal.buffer->Handle();
-}
-
-static void RecordEndOfPipeSignal(const EndOfPipeSignal& signal) {
-	ValidateEndOfPipeSignal(signal);
-	signal.buffer->SetDebugInfo(static_cast<uint32_t>(signal.debug_operation), signal.submit_id,
-	                            signal.debug_args[0], signal.debug_args[1], signal.debug_args[2],
-	                            signal.debug_args[3], signal.debug_data);
-
-	auto& renderer  = signal.buffer->GetContext();
-	auto& scheduler = renderer.GetCommandScheduler();
-	if (signal.completion != EndOfPipeCompletion::None) {
-		EXIT_IF(!scheduler.Active() || signal.buffer != &scheduler.Current());
-	}
-	switch (signal.completion) {
-		case EndOfPipeCompletion::None: return;
-		case EndOfPipeCompletion::Interrupt: {
-			const auto context_id = static_cast<uint32_t>(signal.completion_data);
-			const auto event_id   = signal.interrupt_event_id;
-			scheduler.DeferPriorityOperation([&renderer, event_id, context_id] {
-				renderer.TriggerInterrupt(event_id, context_id);
-			});
-			return;
-		}
-		case EndOfPipeCompletion::Flip: {
-			const auto request_id = signal.completion_data;
-			scheduler.DeferPriorityOperation(
-			    [&renderer, request_id] { renderer.GetVideoOut().CompleteFlip(request_id); });
-			return;
-		}
-		case EndOfPipeCompletion::FlipAndInterrupt: {
-			const auto request_id = signal.completion_data;
-			const auto event_id   = signal.interrupt_event_id;
-			scheduler.DeferPriorityOperation([&renderer, event_id, request_id] {
-				renderer.GetVideoOut().CompleteFlip(request_id);
-				renderer.TriggerInterrupt(event_id, 0);
-			});
-			return;
-		}
-	}
-}
-
 static CommandBufferDebugOp DebugOperation(EndOfPipeWriteAction action) {
 	switch (action) {
 		case EndOfPipeWriteAction::Write: return CommandBufferDebugOp::EopWrite;
@@ -139,24 +76,20 @@ static void RecordEndOfPipeWrite(uint64_t submit_id, CommandBuffer& buffer, uint
                                  uint64_t value, EndOfPipeWriteSize size,
                                  EndOfPipeWriteAction action, int interrupt_event_id = 0,
                                  uint32_t context_id = 0) {
+	EXIT_IF(destination == 0);
+	(void)buffer.Handle();
+
 	const auto width      = static_cast<uint32_t>(size);
 	const auto value_low  = static_cast<uint32_t>(value);
 	const auto value_high = static_cast<uint32_t>(value >> 32u);
-	const bool interrupt  = TriggersInterrupt(action);
-
-	EndOfPipeSignal signal {
-	    .buffer          = &buffer,
-	    .submit_id       = submit_id,
-	    .debug_operation = DebugOperation(action),
-	    .debug_args      = interrupt ? std::array {width, context_id, value_low, value_high}
-	                                 : std::array {width, value_low, value_high, 0u},
-	    .debug_data      = destination,
-	    .destination     = destination,
-	    .completion      = interrupt ? EndOfPipeCompletion::Interrupt : EndOfPipeCompletion::None,
-	    .completion_data = interrupt ? context_id : value,
-	    .interrupt_event_id = interrupt_event_id,
-	};
-	RecordEndOfPipeSignal(signal);
+	const auto operation  = static_cast<uint32_t>(DebugOperation(action));
+	if (TriggersInterrupt(action)) {
+		buffer.SetDebugInfo(operation, submit_id, width, context_id, value_low, value_high,
+		                    destination);
+		TriggerEopEventAtEndOfPipe(buffer, interrupt_event_id, context_id);
+	} else {
+		buffer.SetDebugInfo(operation, submit_id, width, value_low, value_high, 0, destination);
+	}
 }
 
 void WriteAtEndOfPipe32(uint64_t submit_id, CommandBuffer& buffer, uint32_t* dst_gpu_addr,
@@ -167,15 +100,10 @@ void WriteAtEndOfPipe32(uint64_t submit_id, CommandBuffer& buffer, uint32_t* dst
 
 void WriteAtEndOfPipeGds32(uint64_t submit_id, CommandBuffer& buffer, uint32_t* dst_gpu_addr,
                            uint32_t dw_offset, uint32_t dw_num) {
-	const auto destination = reinterpret_cast<uint64_t>(dst_gpu_addr);
-	RecordEndOfPipeSignal({
-	    .buffer          = &buffer,
-	    .submit_id       = submit_id,
-	    .debug_operation = CommandBufferDebugOp::EopWrite,
-	    .debug_args      = {dw_offset, dw_num, 0, 0},
-	    .debug_data      = destination,
-	    .destination     = destination,
-	});
+	EXIT_IF(dst_gpu_addr == nullptr);
+	(void)buffer.Handle();
+	buffer.SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWrite), submit_id,
+	                    dw_offset, dw_num, 0, 0, reinterpret_cast<uint64_t>(dst_gpu_addr));
 }
 
 void WriteAtEndOfPipe64(uint64_t submit_id, CommandBuffer& buffer, uint64_t* dst_gpu_addr,
@@ -273,54 +201,53 @@ void WriteAtEndOfPipeWithInterruptWriteBackFlip32(uint64_t submit_id, CommandBuf
                                                   int handle, int index, int flip_mode,
                                                   int64_t flip_arg, uint64_t request_id,
                                                   int event_id) {
-	const auto destination = reinterpret_cast<uint64_t>(dst_gpu_addr);
-	RecordEndOfPipeSignal({
-	    .buffer             = &buffer,
-	    .submit_id          = submit_id,
-	    .debug_operation    = CommandBufferDebugOp::EopWriteBackFlip,
-	    .debug_args         = {static_cast<uint32_t>(handle), static_cast<uint32_t>(index),
-	                           static_cast<uint32_t>(flip_mode), value},
-	    .debug_data         = static_cast<uint64_t>(flip_arg),
-	    .destination        = destination,
-	    .completion         = EndOfPipeCompletion::FlipAndInterrupt,
-	    .completion_data    = request_id,
-	    .interrupt_event_id = event_id,
+	EXIT_IF(dst_gpu_addr == nullptr);
+	(void)buffer.Handle();
+	buffer.SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopWriteBackFlip), submit_id,
+	                    static_cast<uint32_t>(handle), static_cast<uint32_t>(index),
+	                    static_cast<uint32_t>(flip_mode), value, static_cast<uint64_t>(flip_arg));
+
+	auto& renderer  = buffer.GetContext();
+	auto& scheduler = renderer.GetCommandScheduler();
+	EXIT_IF(!scheduler.Active() || &buffer != &scheduler.Current());
+	scheduler.DeferPriorityOperation([&renderer, event_id, request_id] {
+		renderer.GetVideoOut().CompleteFlip(request_id);
+		renderer.TriggerInterrupt(event_id, 0);
 	});
 }
 
 void WriteAtEndOfPipeWithFlip32(uint64_t submit_id, CommandBuffer& buffer, uint32_t* dst_gpu_addr,
                                 uint32_t value, int handle, int index, int flip_mode,
                                 int64_t flip_arg, uint64_t request_id) {
-	const auto destination = reinterpret_cast<uint64_t>(dst_gpu_addr);
-	RecordEndOfPipeSignal({
-	    .buffer          = &buffer,
-	    .submit_id       = submit_id,
-	    .debug_operation = CommandBufferDebugOp::EopFlip,
-	    .debug_args      = {static_cast<uint32_t>(handle), static_cast<uint32_t>(index),
-	                        static_cast<uint32_t>(flip_mode), value},
-	    .debug_data      = static_cast<uint64_t>(flip_arg),
-	    .destination     = destination,
-	    .completion      = EndOfPipeCompletion::Flip,
-	    .completion_data = request_id,
-	});
+	EXIT_IF(dst_gpu_addr == nullptr);
+	(void)buffer.Handle();
+	buffer.SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopFlip), submit_id,
+	                    static_cast<uint32_t>(handle), static_cast<uint32_t>(index),
+	                    static_cast<uint32_t>(flip_mode), value, static_cast<uint64_t>(flip_arg));
+
+	auto& renderer  = buffer.GetContext();
+	auto& scheduler = renderer.GetCommandScheduler();
+	EXIT_IF(!scheduler.Active() || &buffer != &scheduler.Current());
+	scheduler.DeferPriorityOperation(
+	    [&renderer, request_id] { renderer.GetVideoOut().CompleteFlip(request_id); });
 }
 
 void WriteAtEndOfPipeOnlyFlip(uint64_t submit_id, CommandBuffer& buffer, int handle, int index,
                               int flip_mode, int64_t flip_arg, uint64_t request_id) {
-	RecordEndOfPipeSignal({
-	    .buffer          = &buffer,
-	    .submit_id       = submit_id,
-	    .debug_operation = CommandBufferDebugOp::EopOnlyFlip,
-	    .debug_args      = {static_cast<uint32_t>(handle), static_cast<uint32_t>(index),
-	                        static_cast<uint32_t>(flip_mode), 0},
-	    .debug_data      = static_cast<uint64_t>(flip_arg),
-	    .completion      = EndOfPipeCompletion::Flip,
-	    .completion_data = request_id,
-	});
+	(void)buffer.Handle();
+	buffer.SetDebugInfo(static_cast<uint32_t>(CommandBufferDebugOp::EopOnlyFlip), submit_id,
+	                    static_cast<uint32_t>(handle), static_cast<uint32_t>(index),
+	                    static_cast<uint32_t>(flip_mode), 0, static_cast<uint64_t>(flip_arg));
+
+	auto& renderer  = buffer.GetContext();
+	auto& scheduler = renderer.GetCommandScheduler();
+	EXIT_IF(!scheduler.Active() || &buffer != &scheduler.Current());
+	scheduler.DeferPriorityOperation(
+	    [&renderer, request_id] { renderer.GetVideoOut().CompleteFlip(request_id); });
 }
 
 void TriggerEopEventAtEndOfPipe(CommandBuffer& buffer, int event_id, uint32_t context_id) {
-	ValidateEndOfPipeSignal({.buffer = &buffer});
+	(void)buffer.Handle();
 	auto& renderer  = buffer.GetContext();
 	auto& scheduler = renderer.GetCommandScheduler();
 	EXIT_IF(!scheduler.Active() || &buffer != &scheduler.Current());

src/graphics/host_gpu/shaders/gpu_tiler_standard64.inc

@@ -1,29 +1,17 @@
+#include "gpu_tiler_standard.inc"
 uint standard64_offset(uvec2 p) {
 	uint x = p.x, y = p.y;
+	uint offset = standard_offset(p);
 	switch (ELEMENT_BYTES) {
-	case 1: return (x & 0x0fu) ^ ((x << 5) & 0x200u) ^ ((x << 6) & 0x800u) ^
-	               ((x << 7) & 0x2000u) ^ ((x << 8) & 0x8000u) ^
-	               ((y << 4) & 0x1f0u) ^ ((y << 5) & 0x400u) ^
-	               ((y << 6) & 0x1000u) ^ ((y << 7) & 0x4000u);
-	case 2: return ((x << 1) & 0x00eu) ^ ((x << 4) & 0x080u) ^
-	               ((x << 5) & 0x200u) ^ ((x << 6) & 0x800u) ^
-	               ((x << 7) & 0x2000u) ^ ((x << 8) & 0x8000u) ^
-	               ((y << 4) & 0x070u) ^ ((y << 5) & 0x100u) ^
-	               ((y << 6) & 0x400u) ^ ((y << 7) & 0x1000u) ^ ((y << 8) & 0x4000u);
-	case 4: return ((x << 2) & 0x00cu) ^ ((x << 5) & 0x080u) ^
-	               ((x << 6) & 0x200u) ^ ((x << 7) & 0x800u) ^
-	               ((x << 8) & 0x2000u) ^ ((x << 9) & 0x8000u) ^
-	               ((y << 4) & 0x070u) ^ ((y << 5) & 0x100u) ^
-	               ((y << 6) & 0x400u) ^ ((y << 7) & 0x1000u) ^ ((y << 8) & 0x4000u);
-	case 8: return ((x << 3) & 0x008u) ^ ((x << 5) & 0x0c0u) ^
-	               ((x << 6) & 0x200u) ^ ((x << 7) & 0x800u) ^
-	               ((x << 8) & 0x2000u) ^ ((x << 9) & 0x8000u) ^
-	               ((y << 4) & 0x030u) ^ ((y << 6) & 0x100u) ^
-	               ((y << 7) & 0x400u) ^ ((y << 8) & 0x1000u) ^ ((y << 9) & 0x4000u);
-	default: return ((x << 6) & 0x0c0u) ^ ((x << 7) & 0x200u) ^
-	                ((x << 8) & 0x800u) ^ ((x << 9) & 0x2000u) ^
-	                ((x << 10) & 0x8000u) ^ ((y << 4) & 0x030u) ^
-	                ((y << 6) & 0x100u) ^ ((y << 7) & 0x400u) ^
-	                ((y << 8) & 0x1000u) ^ ((y << 9) & 0x4000u);
+	case 1: return offset ^ ((x << 7) & 0x2000u) ^ ((x << 8) & 0x8000u) ^
+	                       ((y << 6) & 0x1000u) ^ ((y << 7) & 0x4000u);
+	case 2: return offset ^ ((x << 7) & 0x2000u) ^ ((x << 8) & 0x8000u) ^
+	                       ((y << 7) & 0x1000u) ^ ((y << 8) & 0x4000u);
+	case 4: return offset ^ ((x << 8) & 0x2000u) ^ ((x << 9) & 0x8000u) ^
+	                       ((y << 7) & 0x1000u) ^ ((y << 8) & 0x4000u);
+	case 8: return offset ^ ((x << 8) & 0x2000u) ^ ((x << 9) & 0x8000u) ^
+	                       ((y << 8) & 0x1000u) ^ ((y << 9) & 0x4000u);
+	default: return offset ^ ((x << 9) & 0x2000u) ^ ((x << 10) & 0x8000u) ^
+	                        ((y << 8) & 0x1000u) ^ ((y << 9) & 0x4000u);
 	}
 }

src/graphics/host_gpu/vma.cpp

@@ -18,37 +18,12 @@
 #include "common/logging/log.h"
 #include "common/profiler.h"
 #include "graphics/host_gpu/graphicContext.h"
-#include "graphics/host_gpu/vma.h"
 
 #include <algorithm>
-#include <atomic>
 #include <cinttypes>
 
 namespace Libs::Graphics {
 
-namespace {
-
-struct MemoryStats {
-	std::atomic_uint64_t allocated[VK_MAX_MEMORY_TYPES] {};
-	std::atomic_uint64_t count[VK_MAX_MEMORY_TYPES] {};
-};
-
-MemoryStats g_memory_stats;
-
-} // namespace
-
-void VulkanTrackAllocation(const VulkanMemory& memory) {
-	g_memory_stats.allocated[memory.type] += memory.requirements.size;
-	g_memory_stats.count[memory.type]++;
-}
-
-void VulkanUntrackAllocation(const VulkanMemory& memory) {
-	EXIT_IF(g_memory_stats.allocated[memory.type] < memory.requirements.size);
-	EXIT_IF(g_memory_stats.count[memory.type] == 0);
-	g_memory_stats.allocated[memory.type] -= memory.requirements.size;
-	g_memory_stats.count[memory.type]--;
-}
-
 bool GraphicContext::CreateAllocator() {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(instance == nullptr || physical_device == nullptr || device == nullptr ||
@@ -155,40 +130,42 @@ uint64_t GraphicContext::GetTotalMemoryBudget() const {
 
 bool GraphicContext::CreateImage(const vk::ImageCreateInfo& image_info, VulkanImage& image) {
 	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(allocator == nullptr || image.image != nullptr || image.memory.allocation != nullptr);
+	EXIT_IF(allocator == nullptr || image.image != nullptr || image.allocation != nullptr);
 
-	auto&                   memory = image.memory;
 	VmaAllocationCreateInfo alloc_info {};
-	alloc_info.requiredFlags = static_cast<vk::MemoryPropertyFlags::MaskType>(memory.property);
-	alloc_info.preferredFlags =
-	    static_cast<vk::MemoryPropertyFlags::MaskType>(memory.preferred_property);
+	alloc_info.requiredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
 
-	VmaAllocationInfo allocation_result {};
 	vk::Image::CType native_image = VK_NULL_HANDLE;
 	const auto        result       = static_cast<vk::Result>(
 	    vmaCreateImage(allocator, static_cast<const vk::ImageCreateInfo::NativeType*>(image_info),
-	                   &alloc_info, &native_image, &memory.allocation, &allocation_result));
+	                   &alloc_info, &native_image, &image.allocation, nullptr));
 	image.image = native_image;
 	if (result != vk::Result::eSuccess) {
 		LogMemoryBudget();
 		return false;
 	}
 
-	device.getImageMemoryRequirements(image.image, &memory.requirements);
-	memory.type = allocation_result.memoryType;
-	VulkanTrackAllocation(memory);
+	image.format     = image_info.format;
+	image.image_type = image_info.imageType;
+	image.extent     = image_info.extent;
+	image.layers     = image_info.arrayLayers;
+	image.mip_levels = image_info.mipLevels;
+	image.samples    = static_cast<uint32_t>(image_info.samples);
+	image.usage      = image_info.usage;
+	image.flags      = image_info.flags;
+	image.state      = {.layout = image_info.initialLayout};
+	image.subresource_states.clear();
+
 	return true;
 }
 
 void GraphicContext::DeleteImage(VulkanImage& image) {
 	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(allocator == nullptr || image.image == nullptr || image.memory.allocation == nullptr);
+	EXIT_IF(allocator == nullptr || image.image == nullptr || image.allocation == nullptr);
 
-	auto& memory = image.memory;
-	VulkanUntrackAllocation(memory);
-	vmaDestroyImage(allocator, image.image, memory.allocation);
-	image.image            = nullptr;
-	memory.allocation      = nullptr;
+	vmaDestroyImage(allocator, image.image, image.allocation);
+	image.image      = nullptr;
+	image.allocation = nullptr;
 }
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/vma.h

@@ -1,14 +0,0 @@
-#ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_VMA_H_
-#define EMULATOR_SRC_GRAPHICS_HOST_GPU_VMA_H_
-
-#include "common/common.h"
-#include "graphics/host_gpu/graphicContext.h"
-
-namespace Libs::Graphics {
-
-void     VulkanTrackAllocation(const VulkanMemory& memory);
-void     VulkanUntrackAllocation(const VulkanMemory& memory);
-
-} // namespace Libs::Graphics
-
-#endif /* EMULATOR_SRC_GRAPHICS_HOST_GPU_VMA_H_ */

src/graphics/host_gpu/vulkanCommon.cpp

@@ -17,14 +17,22 @@ struct FormatMapping {
 
 constexpr FormatMapping kFormatMappings[] = {
     {Prospero::BufferFormat::k8UNorm, vk::Format::eR8Unorm},
+    {Prospero::BufferFormat::k8SNorm, vk::Format::eR8Snorm},
+    {Prospero::BufferFormat::k8UScaled, vk::Format::eR8Uscaled},
+    {Prospero::BufferFormat::k8SScaled, vk::Format::eR8Sscaled},
     {Prospero::BufferFormat::k8UInt, vk::Format::eR8Uint},
+    {Prospero::BufferFormat::k8SInt, vk::Format::eR8Sint},
     {Prospero::BufferFormat::k16UNorm, vk::Format::eR16Unorm},
     {Prospero::BufferFormat::k16SNorm, vk::Format::eR16Snorm},
+    {Prospero::BufferFormat::k16UScaled, vk::Format::eR16Uscaled},
+    {Prospero::BufferFormat::k16SScaled, vk::Format::eR16Sscaled},
     {Prospero::BufferFormat::k16UInt, vk::Format::eR16Uint},
     {Prospero::BufferFormat::k16SInt, vk::Format::eR16Sint},
     {Prospero::BufferFormat::k16Float, vk::Format::eR16Sfloat},
     {Prospero::BufferFormat::k8_8UNorm, vk::Format::eR8G8Unorm},
     {Prospero::BufferFormat::k8_8SNorm, vk::Format::eR8G8Snorm},
+    {Prospero::BufferFormat::k8_8UScaled, vk::Format::eR8G8Uscaled},
+    {Prospero::BufferFormat::k8_8SScaled, vk::Format::eR8G8Sscaled},
     {Prospero::BufferFormat::k8_8UInt, vk::Format::eR8G8Uint},
     {Prospero::BufferFormat::k8_8SInt, vk::Format::eR8G8Sint},
     {Prospero::BufferFormat::k32UInt, vk::Format::eR32Uint},
@@ -32,14 +40,19 @@ constexpr FormatMapping kFormatMappings[] = {
     {Prospero::BufferFormat::k32Float, vk::Format::eR32Sfloat},
     {Prospero::BufferFormat::k16_16UNorm, vk::Format::eR16G16Unorm},
     {Prospero::BufferFormat::k16_16SNorm, vk::Format::eR16G16Snorm},
+    {Prospero::BufferFormat::k16_16UScaled, vk::Format::eR16G16Uscaled},
+    {Prospero::BufferFormat::k16_16SScaled, vk::Format::eR16G16Sscaled},
     {Prospero::BufferFormat::k16_16UInt, vk::Format::eR16G16Uint},
     {Prospero::BufferFormat::k16_16SInt, vk::Format::eR16G16Sint},
     {Prospero::BufferFormat::k16_16Float, vk::Format::eR16G16Sfloat},
     {Prospero::BufferFormat::k11_11_10Float, vk::Format::eB10G11R11UfloatPack32},
     {Prospero::BufferFormat::k10_10_10_2UNorm, vk::Format::eA2B10G10R10UnormPack32},
+    {Prospero::BufferFormat::k10_10_10_2SNorm, vk::Format::eA2B10G10R10SnormPack32},
     {Prospero::BufferFormat::k10_10_10_2UInt, vk::Format::eA2B10G10R10UintPack32},
     {Prospero::BufferFormat::k8_8_8_8UNorm, vk::Format::eR8G8B8A8Unorm},
     {Prospero::BufferFormat::k8_8_8_8SNorm, vk::Format::eR8G8B8A8Snorm},
+    {Prospero::BufferFormat::k8_8_8_8UScaled, vk::Format::eR8G8B8A8Uscaled},
+    {Prospero::BufferFormat::k8_8_8_8SScaled, vk::Format::eR8G8B8A8Sscaled},
     {Prospero::BufferFormat::k8_8_8_8UInt, vk::Format::eR8G8B8A8Uint},
     {Prospero::BufferFormat::k8_8_8_8SInt, vk::Format::eR8G8B8A8Sint},
     {Prospero::BufferFormat::k32_32UInt, vk::Format::eR32G32Uint},
@@ -47,6 +60,8 @@ constexpr FormatMapping kFormatMappings[] = {
     {Prospero::BufferFormat::k32_32Float, vk::Format::eR32G32Sfloat},
     {Prospero::BufferFormat::k16_16_16_16UNorm, vk::Format::eR16G16B16A16Unorm},
     {Prospero::BufferFormat::k16_16_16_16SNorm, vk::Format::eR16G16B16A16Snorm},
+    {Prospero::BufferFormat::k16_16_16_16UScaled, vk::Format::eR16G16B16A16Uscaled},
+    {Prospero::BufferFormat::k16_16_16_16SScaled, vk::Format::eR16G16B16A16Sscaled},
     {Prospero::BufferFormat::k16_16_16_16UInt, vk::Format::eR16G16B16A16Uint},
     {Prospero::BufferFormat::k16_16_16_16SInt, vk::Format::eR16G16B16A16Sint},
     {Prospero::BufferFormat::k16_16_16_16Float, vk::Format::eR16G16B16A16Sfloat},
@@ -63,7 +78,8 @@ constexpr FormatMapping kFormatMappings[] = {
     {Prospero::BufferFormat::k8_8_8_8Srgb, vk::Format::eR8G8B8A8Srgb},
     {Prospero::BufferFormat::k9_9_9_5Float, vk::Format::eE5B9G9R9UfloatPack32},
     {Prospero::BufferFormat::k5_6_5UNorm, vk::Format::eB5G6R5UnormPack16},
-    {Prospero::BufferFormat::k5_5_5_1UNorm, vk::Format::eR5G5B5A1UnormPack16},
+    {Prospero::BufferFormat::k5_5_5_1UNorm, vk::Format::eA1R5G5B5UnormPack16},
+    {Prospero::BufferFormat::k1_5_5_5UNorm, vk::Format::eR5G5B5A1UnormPack16},
     {Prospero::BufferFormat::k4_4_4_4UNorm, vk::Format::eR4G4B4A4UnormPack16},
     {Prospero::BufferFormat::kBc1UNorm, vk::Format::eBc1RgbaUnormBlock},
     {Prospero::BufferFormat::kBc1Srgb, vk::Format::eBc1RgbaSrgbBlock},
@@ -107,4 +123,14 @@ void RequireVulkanSuccess(vk::Result result, const char* operation) {
 	}
 }
 
+vk::ShaderModule CompileSPV(std::span<const uint32_t> code, vk::Device device) {
+	vk::ShaderModuleCreateInfo create_info {};
+	create_info.codeSize    = code.size_bytes();
+	create_info.pCode       = code.data();
+	vk::ShaderModule module = nullptr;
+	RequireVulkanSuccess(device.createShaderModule(&create_info, nullptr, &module),
+	                     "create SPIR-V shader module");
+	return module;
+}
+
 } // namespace Libs::Graphics

src/graphics/host_gpu/vulkanCommon.h

@@ -15,6 +15,7 @@
 
 #include <cstdint>
 #include <fmt/format.h>
+#include <span>
 #include <string>
 #include <utility>
 #include <vector>
@@ -27,6 +28,7 @@ using VulkanMemoryBarrier = vk::MemoryBarrier;
 
 vk::Format  VulkanFormat(Prospero::BufferFormat guest_format);
 void        RequireVulkanSuccess(vk::Result result, const char* operation);
+vk::ShaderModule CompileSPV(std::span<const uint32_t> code, vk::Device device);
 
 template <typename Handle, typename... Args>
 void SetVulkanObjectNameF(vk::Device device, Handle handle, fmt::format_string<Args...> format,
@@ -38,7 +40,6 @@ void SetVulkanObjectNameF(vk::Device device, Handle handle, fmt::format_string<A
 
 	const auto                      name = fmt::format(format, std::forward<Args>(args)...);
 	vk::DebugUtilsObjectNameInfoEXT info {};
-	info.sType        = vk::StructureType::eDebugUtilsObjectNameInfoEXT;
 	info.objectType   = Handle::objectType;
 	info.objectHandle = static_cast<uint64_t>(
 	    reinterpret_cast<uintptr_t>(static_cast<typename Handle::CType>(handle)));

src/graphics/presentation/videoOut.cpp

@@ -237,7 +237,9 @@ private:
 
 struct VideoOutDriver::Impl {
 public:
-	static constexpr int VIDEO_OUT_NUM_MAX = 4;
+	static constexpr std::array VIDEO_OUT_BUSES {
+	    VIDEO_OUT_BUS_TYPE_MAIN, VIDEO_OUT_BUS_TYPE_OVERLAY, VIDEO_OUT_BUS_TYPE_SUB};
+	static constexpr int VIDEO_OUT_NUM_MAX = VIDEO_OUT_BUSES.size() + 1;
 
 	Impl(uint32_t width, uint32_t height, Graphics::Presenter& presenter)
 	    : m_renderer(presenter.Renderer()), m_presenter(presenter), m_flip_queue(presenter) {
@@ -248,7 +250,7 @@ public:
 	~Impl();
 	KYTY_CLASS_NO_COPY(Impl);
 
-	int             Open(int bus_type);
+	int             Open(int bus_type, int index);
 	bool            Close(int handle);
 	VideoOutConfig* Get(int handle);
 	VideoOutConfig* Get(int handle, uint64_t& generation);
@@ -610,12 +612,17 @@ void VideoOutDriver::Impl::Init(uint32_t width, uint32_t height) {
 	}
 }
 
-int VideoOutDriver::Impl::Open(int bus_type) {
+int VideoOutDriver::Impl::Open(int bus_type, int index) {
+	const auto bus = std::find(VIDEO_OUT_BUSES.begin(), VIDEO_OUT_BUSES.end(), bus_type);
+	if (bus == VIDEO_OUT_BUSES.end()) {
+		return VIDEO_OUT_ERROR_INVALID_VALUE;
+	}
+	EXIT_NOT_IMPLEMENTED(index != 0);
 	Common::LockGuard lock(m_mutex);
 
-	const int handle = bus_type + 1;
+	const int handle = static_cast<int>(bus - VIDEO_OUT_BUSES.begin()) + 1;
 	if (m_video_out_ctx[handle].opened) {
-		return -1;
+		return VIDEO_OUT_ERROR_RESOURCE_BUSY;
 	}
 	auto&             config = m_video_out_ctx[handle];
 	Common::LockGuard config_lock(config.mutex);
@@ -1152,7 +1159,7 @@ bool FlipQueue::Flip(uint32_t micros) {
 	Graphics::RenderDocOnGuestFlip(m_presenter.Renderer());
 
 	if (Config::GraphicsDebugDumpEnabled() &&
-	    Config::GetPrintfDirection() != Config::OutputDirection::Silent) {
+	    Config::GetPrintfDirection() != Config::LogDirection::Silent) {
 		LOGF("Flip done: %d\n", r.index);
 	}
 
@@ -1169,21 +1176,9 @@ KYTY_SYSV_ABI int VideoOutOpen(int user_id, int bus_type, int index, const void*
 	PRINT_NAME();
 
 	EXIT_NOT_IMPLEMENTED(user_id != 255 && user_id != 0);
-	if (bus_type != VIDEO_OUT_BUS_TYPE_MAIN && bus_type != VIDEO_OUT_BUS_TYPE_OVERLAY &&
-	    bus_type != VIDEO_OUT_BUS_TYPE_SUB) {
-		return VIDEO_OUT_ERROR_INVALID_VALUE;
-	}
-	EXIT_NOT_IMPLEMENTED(index != 0);
-
 	LOGF("\t param = 0x%016" PRIx64 "\n", reinterpret_cast<uint64_t>(param));
 
-	int handle = DriverState().Open(bus_type);
-
-	if (handle < 0) {
-		return VIDEO_OUT_ERROR_RESOURCE_BUSY;
-	}
-
-	return handle;
+	return DriverState().Open(bus_type, index);
 }
 
 KYTY_SYSV_ABI int VideoOutClose(int handle) {

src/graphics/presentation/window/hostInput.cpp

@@ -220,6 +220,8 @@ uint32_t DefaultKeyboardButton(int key_code) {
 		case SDLK_l: return Controller::PAD_BUTTON_CIRCLE;
 		case SDLK_q: return Controller::PAD_BUTTON_L1;
 		case SDLK_e: return Controller::PAD_BUTTON_R1;
+		case SDLK_z: return Controller::PAD_BUTTON_L2;
+		case SDLK_c: return Controller::PAD_BUTTON_R2;
 		case SDLK_LSHIFT: return Controller::PAD_BUTTON_L3;
 		case SDLK_LCTRL: return Controller::PAD_BUTTON_R3;
 		case SDLK_RETURN:

src/graphics/presentation/window/swapchain.cpp

@@ -7,7 +7,6 @@
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
-#include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/presentation/presenter.h"
 #include "graphics/presentation/systemOverlay.h"
@@ -192,24 +191,15 @@ void Presenter::Frame::Configure(GraphicContext& graphics, vk::Extent2D extent,
 	}
 	if (dst.image != nullptr) {
 		graphics.DeleteImage(dst);
-		dst.memory = {};
 	}
 
-	dst.extent     = {extent.width, extent.height, 1};
-	dst.format     = format;
-	dst.layers     = 1;
-	dst.mip_levels = 1;
-	dst.state      = {};
-	dst.subresource_states.clear();
-	dst.memory.property = vk::MemoryPropertyFlagBits::eDeviceLocal;
-
 	vk::ImageCreateInfo create {};
 	create.sType         = vk::StructureType::eImageCreateInfo;
 	create.imageType     = vk::ImageType::e2D;
-	create.extent        = {dst.extent.width, dst.extent.height, 1};
+	create.extent        = {extent.width, extent.height, 1};
 	create.mipLevels     = 1;
 	create.arrayLayers   = 1;
-	create.format        = dst.format;
+	create.format        = format;
 	create.tiling        = vk::ImageTiling::eOptimal;
 	create.initialLayout = vk::ImageLayout::eUndefined;
 	create.usage = vk::ImageUsageFlagBits::eTransferSrc | vk::ImageUsageFlagBits::eTransferDst;
@@ -217,7 +207,7 @@ void Presenter::Frame::Configure(GraphicContext& graphics, vk::Extent2D extent,
 	create.samples     = vk::SampleCountFlagBits::e1;
 	if (!graphics.CreateImage(create, dst)) {
 		EXIT("failed to allocate prepared presentation image, extent=%ux%u format=%d\n",
-		     dst.extent.width, dst.extent.height, static_cast<int>(dst.format));
+		     extent.width, extent.height, static_cast<int>(format));
 	}
 }
 

src/graphics/presentation/window/vulkanWindow.cpp

@@ -27,7 +27,6 @@
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
-#include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/presentation/presenter.h"
 #include "graphics/presentation/systemOverlay.h"
@@ -65,7 +64,6 @@ struct VulkanExtensions {
 
 vk::PhysicalDeviceVulkan12Features WindowContext::RequiredVulkan12Features() noexcept {
 	vk::PhysicalDeviceVulkan12Features features {};
-	features.sType                     = vk::StructureType::ePhysicalDeviceVulkan12Features;
 	features.samplerMirrorClampToEdge  = VK_TRUE;
 	features.timelineSemaphore         = VK_TRUE;
 	features.shaderOutputLayer         = VK_TRUE;
@@ -77,7 +75,6 @@ vk::PhysicalDeviceVulkan12Features WindowContext::RequiredVulkan12Features() noe
 
 vk::PhysicalDeviceVulkan13Features WindowContext::RequiredVulkan13Features() noexcept {
 	vk::PhysicalDeviceVulkan13Features features {};
-	features.sType            = vk::StructureType::ePhysicalDeviceVulkan13Features;
 	features.dynamicRendering = VK_TRUE;
 	features.synchronization2 = VK_TRUE;
 	return features;
@@ -170,7 +167,12 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 	EXIT_NOT_IMPLEMENTED(devices.empty());
 
 	if (Config::GetGpuIndex() >= 0) {
-		devices = {devices[Config::GetGpuIndex()]};
+		if (static_cast<size_t>(Config::GetGpuIndex()) < devices.size()) {
+			devices = {devices[Config::GetGpuIndex()]};
+		} else {
+			LOGF("Vulkan GPU index %d is unavailable; selecting automatically\n",
+			     Config::GetGpuIndex());
+		}
 	}
 
 	vk::PhysicalDevice  best_device       = nullptr;
@@ -197,35 +199,24 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 		vk::PhysicalDeviceFeatures2 device_features2 {};
 
 		vk::PhysicalDeviceVulkan13Features features13 {};
-		features13.sType = vk::StructureType::ePhysicalDeviceVulkan13Features;
-		features13.pNext = nullptr;
 
 		vk::PhysicalDeviceColorWriteEnableFeaturesEXT color_write_ext {};
-		color_write_ext.sType = vk::StructureType::ePhysicalDeviceColorWriteEnableFeaturesEXT;
-		color_write_ext.pNext = nullptr;
 
 		vk::PhysicalDeviceDepthClipEnableFeaturesEXT depth_clip_enable {};
-		depth_clip_enable.sType = vk::StructureType::ePhysicalDeviceDepthClipEnableFeaturesEXT;
 		depth_clip_enable.pNext = &color_write_ext;
 
 		vk::PhysicalDeviceDepthClipControlFeaturesEXT depth_clip_control {};
-		depth_clip_control.sType = vk::StructureType::ePhysicalDeviceDepthClipControlFeaturesEXT;
 		depth_clip_control.pNext = &depth_clip_enable;
 
 		vk::PhysicalDeviceVulkan12Features features12 {};
-		features12.sType = vk::StructureType::ePhysicalDeviceVulkan12Features;
 #if defined(__APPLE__)
 		features12.pNext = &depth_clip_control;
 #else
 		vk::PhysicalDeviceFragmentShaderBarycentricFeaturesKHR fragment_barycentric {};
-		fragment_barycentric.sType =
-		    vk::StructureType::ePhysicalDeviceFragmentShaderBarycentricFeaturesKHR;
 		fragment_barycentric.pNext = &depth_clip_control;
 		features12.pNext           = &fragment_barycentric;
 #endif
 		features13.pNext = &features12;
-
-		device_features2.sType = vk::StructureType::ePhysicalDeviceFeatures2;
 		device_features2.pNext = &features13;
 
 		device.getFeatures2(&device_features2);
@@ -502,83 +493,31 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 	}
 }
 
-static void VulkanInitSubgroupSizeControl(vk::PhysicalDevice physical_device,
-                                          GraphicContext&    graphics) {
-	EXIT_IF(physical_device == nullptr);
-
-	vk::PhysicalDeviceSubgroupSizeControlProperties subgroup_size_control {};
-	subgroup_size_control.sType = vk::StructureType::ePhysicalDeviceSubgroupSizeControlProperties;
-	subgroup_size_control.pNext = nullptr;
-
-	vk::PhysicalDeviceVulkan11Properties properties11 {};
-	properties11.sType = vk::StructureType::ePhysicalDeviceVulkan11Properties;
-	properties11.pNext = &subgroup_size_control;
-
-	vk::PhysicalDeviceProperties2 properties2 {};
-	properties2.sType = vk::StructureType::ePhysicalDeviceProperties2;
-	properties2.pNext = &properties11;
-
-	physical_device.getProperties2(&properties2);
-
-	vk::PhysicalDeviceVulkan13Features features13 {};
-	features13.sType = vk::StructureType::ePhysicalDeviceVulkan13Features;
-	features13.pNext = nullptr;
-
-	vk::PhysicalDeviceFeatures2 features2 {};
-	features2.sType = vk::StructureType::ePhysicalDeviceFeatures2;
-	features2.pNext = &features13;
-
-	physical_device.getFeatures2(&features2);
-
-	graphics.subgroup_size                 = properties11.subgroupSize;
-	graphics.min_subgroup_size             = subgroup_size_control.minSubgroupSize;
-	graphics.max_subgroup_size             = subgroup_size_control.maxSubgroupSize;
-	graphics.required_subgroup_size_stages = subgroup_size_control.requiredSubgroupSizeStages;
-	graphics.compute_subgroup_size_control_enabled =
-	    features13.subgroupSizeControl == VK_TRUE &&
-	    (graphics.required_subgroup_size_stages & vk::ShaderStageFlagBits::eCompute) &&
-	    subgroup_size_control.minSubgroupSize <= 64 &&
-	    subgroup_size_control.maxSubgroupSize >= 64;
-
-	LOGF("Vulkan subgroup: default=%u min=%u max=%u stages=0x%08x size_control=%s wave64=%s\n",
-	     graphics.subgroup_size, graphics.min_subgroup_size, graphics.max_subgroup_size,
-	     static_cast<vk::ShaderStageFlags::MaskType>(graphics.required_subgroup_size_stages),
-	     graphics.compute_subgroup_size_control_enabled ? "true" : "false",
-	     graphics.SupportsComputeWave64() ? "true" : "false");
-}
-
-static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const VulkanExtensions& r,
-                                     uint32_t                        queue_family,
-                                     const std::vector<const char*>& device_extensions,
-                                     GraphicContext&                 graphics) {
+static vk::Device VulkanCreateDevice(GraphicContext& graphics,
+	                                 const std::vector<const char*>& device_extensions) {
+	const auto physical_device = graphics.physical_device;
+	const auto queue_family    = graphics.queue_family;
 	EXIT_IF(physical_device == nullptr);
 	EXIT_IF(queue_family == static_cast<uint32_t>(-1));
 
 	const float               queue_priority = 1.0f;
 	vk::DeviceQueueCreateInfo queue_create_info {};
-	queue_create_info.sType            = vk::StructureType::eDeviceQueueCreateInfo;
 	queue_create_info.queueFamilyIndex = queue_family;
 	queue_create_info.queueCount       = 1;
 	queue_create_info.pQueuePriorities = &queue_priority;
 
 	vk::PhysicalDeviceColorWriteEnableFeaturesEXT color_write_ext {};
-	color_write_ext.sType = vk::StructureType::ePhysicalDeviceColorWriteEnableFeaturesEXT;
-	color_write_ext.pNext = nullptr;
 	color_write_ext.colorWriteEnable = VK_TRUE;
 
 	vk::PhysicalDeviceDepthClipEnableFeaturesEXT depth_clip_enable {};
-	depth_clip_enable.sType = vk::StructureType::ePhysicalDeviceDepthClipEnableFeaturesEXT;
 	depth_clip_enable.pNext = &color_write_ext;
 	depth_clip_enable.depthClipEnable = VK_TRUE;
 
 	vk::PhysicalDeviceDepthClipControlFeaturesEXT depth_clip_control {};
-	depth_clip_control.sType = vk::StructureType::ePhysicalDeviceDepthClipControlFeaturesEXT;
 	// MoltenVK lacks VK_EXT_depth_clip_enable and VK_EXT_color_write_enable, so drop those
 	// feature structs from the chain on macOS (the renderer falls back to default depth
 	// clipping and static color-write masks).
-#if defined(__APPLE__)
-	depth_clip_control.pNext = nullptr;
-#else
+#if !defined(__APPLE__)
 	depth_clip_control.pNext = &depth_clip_enable;
 #endif
 	depth_clip_control.depthClipControl = VK_TRUE;
@@ -587,40 +526,19 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	features12.pNext = &depth_clip_control;
 
 	vk::PhysicalDeviceVulkan13Features supported_features13 {};
-	supported_features13.sType = vk::StructureType::ePhysicalDeviceVulkan13Features;
-
-	vk::PhysicalDeviceVulkan12Features supported_features12 {};
-	supported_features12.sType = vk::StructureType::ePhysicalDeviceVulkan12Features;
-	supported_features12.pNext = nullptr;
-	supported_features13.pNext = &supported_features12;
-
-#if !defined(__APPLE__)
-	vk::PhysicalDeviceFragmentShaderBarycentricFeaturesKHR supported_fragment_barycentric {};
-	supported_fragment_barycentric.sType =
-	    vk::StructureType::ePhysicalDeviceFragmentShaderBarycentricFeaturesKHR;
-	supported_fragment_barycentric.pNext = nullptr;
-	supported_features12.pNext           = &supported_fragment_barycentric;
-#endif
 
 	const auto robustness2_ext_enabled =
 	    HasExtension(device_extensions, VK_EXT_ROBUSTNESS_2_EXTENSION_NAME);
 
 	vk::PhysicalDeviceRobustness2FeaturesEXT supported_robustness2 {};
-	supported_robustness2.sType = vk::StructureType::ePhysicalDeviceRobustness2FeaturesEXT;
-	supported_robustness2.pNext = nullptr;
 	if (robustness2_ext_enabled) {
-#if defined(__APPLE__)
-		supported_features12.pNext = &supported_robustness2;
-#else
-		supported_fragment_barycentric.pNext = &supported_robustness2;
-#endif
+		supported_features13.pNext = &supported_robustness2;
 	}
 
 	const bool mesh_extension = HasExtension(device_extensions, VK_EXT_MESH_SHADER_EXTENSION_NAME);
 	vk::PhysicalDeviceMeshShaderFeaturesEXT supported_mesh {};
 	supported_mesh.pNext = &supported_features13;
 	vk::PhysicalDeviceFeatures2 supported_features2 {};
-	supported_features2.sType = vk::StructureType::ePhysicalDeviceFeatures2;
 	supported_features2.pNext = mesh_extension ? static_cast<void*>(&supported_mesh)
 	                                           : static_cast<void*>(&supported_features13);
 	const bool feedback_extensions =
@@ -633,53 +551,61 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 		feedback_layout.pNext  = &feedback_dynamic;
 		supported_features2.pNext = &feedback_layout;
 	}
+	const bool provoking_extension =
+	    HasExtension(device_extensions, VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME);
+	vk::PhysicalDeviceProvokingVertexFeaturesEXT provoking_vertex {};
+	if (provoking_extension) {
+		provoking_vertex.pNext = supported_features2.pNext;
+		supported_features2.pNext = &provoking_vertex;
+	}
 	physical_device.getFeatures2(&supported_features2);
+	graphics.mesh_shader_enabled = mesh_extension && supported_mesh.meshShader;
+
+	vk::PhysicalDeviceSubgroupSizeControlProperties subgroup_size_control {};
+
+	vk::PhysicalDeviceVulkan11Properties properties11 {};
+	properties11.pNext = &subgroup_size_control;
+
+	vk::PhysicalDeviceProperties2 properties2 {};
+	properties2.pNext = &properties11;
+
+	if (graphics.mesh_shader_enabled) {
+		subgroup_size_control.pNext = &graphics.mesh_shader_properties;
+	}
+	physical_device.getProperties2(&properties2);
+
+	graphics.subgroup_size                 = properties11.subgroupSize;
+	graphics.min_subgroup_size             = subgroup_size_control.minSubgroupSize;
+	graphics.max_subgroup_size             = subgroup_size_control.maxSubgroupSize;
+	graphics.required_subgroup_size_stages = subgroup_size_control.requiredSubgroupSizeStages;
+	graphics.compute_subgroup_size_control_enabled =
+	    supported_features13.subgroupSizeControl == VK_TRUE &&
+	    (graphics.required_subgroup_size_stages & vk::ShaderStageFlagBits::eCompute) &&
+	    subgroup_size_control.minSubgroupSize <= 64 &&
+	    subgroup_size_control.maxSubgroupSize >= 64;
+
+	LOGF("Vulkan subgroup: default=%u min=%u max=%u stages=0x%08x size_control=%s wave64=%s\n",
+	     graphics.subgroup_size, graphics.min_subgroup_size, graphics.max_subgroup_size,
+	     static_cast<vk::ShaderStageFlags::MaskType>(graphics.required_subgroup_size_stages),
+	     graphics.compute_subgroup_size_control_enabled ? "true" : "false",
+	     graphics.SupportsComputeWave64() ? "true" : "false");
+	graphics.provoking_vertex_last_enabled = provoking_extension && provoking_vertex.provokingVertexLast;
 	graphics.attachment_feedback_loop_enabled =
 	    feedback_extensions && feedback_layout.attachmentFeedbackLoopLayout &&
 	    feedback_dynamic.attachmentFeedbackLoopDynamicState;
 	LOGF("Vulkan depth feedback support: %s\n",
 	     graphics.attachment_feedback_loop_enabled ? "true" : "false");
-	graphics.mesh_shader_enabled = mesh_extension && supported_mesh.meshShader;
 	if (graphics.mesh_shader_enabled) {
-		vk::PhysicalDeviceProperties2 properties {};
-		properties.pNext = &graphics.mesh_shader_properties;
-		physical_device.getProperties2(&properties);
 		LOGF("Vulkan MeshEXT: invocations=%u vertices=%u primitives=%u shared=%u\n",
 		     graphics.mesh_shader_properties.maxMeshWorkGroupInvocations,
 		     graphics.mesh_shader_properties.maxMeshOutputVertices,
 		     graphics.mesh_shader_properties.maxMeshOutputPrimitives,
 		     graphics.mesh_shader_properties.maxMeshSharedMemorySize);
 	}
-	const auto required_features12 = WindowContext::RequiredVulkan12Features();
-	const auto required_features13 = WindowContext::RequiredVulkan13Features();
-	EXIT_NOT_IMPLEMENTED(required_features12.samplerMirrorClampToEdge == VK_TRUE &&
-	                     supported_features12.samplerMirrorClampToEdge != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(required_features12.timelineSemaphore == VK_TRUE &&
-	                     supported_features12.timelineSemaphore != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(required_features13.dynamicRendering == VK_TRUE &&
-	                     supported_features13.dynamicRendering != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(required_features13.synchronization2 == VK_TRUE &&
-	                     supported_features13.synchronization2 != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features2.features.sampleRateShading != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features2.features.depthBiasClamp != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features2.features.shaderClipDistance != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features2.features.shaderCullDistance != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features2.features.largePoints != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features2.features.multiViewport != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features2.features.fillModeNonSolid != VK_TRUE);
+	// VulkanFindPhysicalDevice already checked the required creation features. These two
+	// requirements are specific to this creation path and are not part of device selection.
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.shaderInt64 != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.vertexPipelineStoresAndAtomics != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(required_features12.shaderOutputLayer == VK_TRUE &&
-	                     supported_features12.shaderOutputLayer != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(required_features12.shaderOutputViewportIndex == VK_TRUE &&
-	                     supported_features12.shaderOutputViewportIndex != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(required_features12.bufferDeviceAddress == VK_TRUE &&
-	                     supported_features12.bufferDeviceAddress != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(required_features12.shaderBufferInt64Atomics == VK_TRUE &&
-	                     supported_features12.shaderBufferInt64Atomics != VK_TRUE);
-#if !defined(__APPLE__)
-	EXIT_NOT_IMPLEMENTED(supported_fragment_barycentric.fragmentShaderBarycentric != VK_TRUE);
-#endif
 	vk::PhysicalDeviceFeatures device_features {};
 	device_features.fragmentStoresAndAtomics = VK_TRUE;
 	device_features.samplerAnisotropy        = VK_TRUE;
@@ -703,13 +629,10 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	device_features.shaderInt64 = VK_TRUE;
 
 	vk::PhysicalDeviceRobustness2FeaturesEXT robustness2 {};
-	robustness2.sType = vk::StructureType::ePhysicalDeviceRobustness2FeaturesEXT;
 #if defined(__APPLE__)
 	robustness2.pNext = &features12;
 #else
 	vk::PhysicalDeviceFragmentShaderBarycentricFeaturesKHR fragment_barycentric {};
-	fragment_barycentric.sType =
-	    vk::StructureType::ePhysicalDeviceFragmentShaderBarycentricFeaturesKHR;
 	fragment_barycentric.pNext                     = &features12;
 	fragment_barycentric.fragmentShaderBarycentric = VK_TRUE;
 	robustness2.pNext                              = &fragment_barycentric;
@@ -720,11 +643,7 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 		robustness2.nullDescriptor      = supported_robustness2.nullDescriptor;
 	}
 
-	const bool subgroup_size_control_enabled =
-	    graphics.compute_subgroup_size_control_enabled &&
-	    supported_features13.subgroupSizeControl == VK_TRUE;
-
-	auto features13 = required_features13;
+	auto features13 = WindowContext::RequiredVulkan13Features();
 #if defined(__APPLE__)
 	features13.pNext = robustness2_ext_enabled ? static_cast<void*>(&robustness2)
 	                                           : static_cast<void*>(&features12);
@@ -733,7 +652,8 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	                                           : static_cast<void*>(&fragment_barycentric);
 #endif
 	features13.robustImageAccess   = supported_features13.robustImageAccess;
-	features13.subgroupSizeControl = subgroup_size_control_enabled ? VK_TRUE : VK_FALSE;
+	features13.subgroupSizeControl =
+	    graphics.compute_subgroup_size_control_enabled ? VK_TRUE : VK_FALSE;
 
 	LOGF("Vulkan robustness: robustImageAccess=%s robustImageAccess2=%s\n",
 	     features13.robustImageAccess == VK_TRUE ? "true" : "false",
@@ -743,13 +663,16 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	vk::PhysicalDeviceMeshShaderFeaturesEXT mesh_features {};
 	mesh_features.pNext                 = &features13;
 	mesh_features.meshShader            = graphics.mesh_shader_enabled;
-	create_info.sType                   = vk::StructureType::eDeviceCreateInfo;
 	feedback_dynamic.pNext =
 	    mesh_extension ? static_cast<void*>(&mesh_features) : static_cast<void*>(&features13);
 	create_info.pNext = graphics.attachment_feedback_loop_enabled
 	                        ? static_cast<void*>(&feedback_layout)
 	                        : feedback_dynamic.pNext;
-	create_info.flags                   = {};
+	if (graphics.provoking_vertex_last_enabled) {
+		provoking_vertex.pNext = const_cast<void*>(create_info.pNext);
+		provoking_vertex.transformFeedbackPreservesProvokingVertex = VK_FALSE;
+		create_info.pNext = &provoking_vertex;
+	}
 	create_info.pQueueCreateInfos       = &queue_create_info;
 	create_info.queueCreateInfoCount    = 1;
 	create_info.enabledExtensionCount   = static_cast<uint32_t>(device_extensions.size());
@@ -779,8 +702,6 @@ static void VulkanGetExtensions(SDL_Window* window, VulkanExtensions& r) {
 
 	r.required_extensions =
 	    std::vector<const char*>(required_extensions_count); // @suppress("Ambiguous problem")
-	std::memset(r.required_extensions.data(), 0,
-	            sizeof(const char*) * r.required_extensions.size());
 
 	sdl_result = SDL_Vulkan_GetInstanceExtensions(window, &required_extensions_count,
 	                                              r.required_extensions.data());
@@ -972,8 +893,6 @@ void WindowContext::CreateVulkan() {
 	VulkanCheckInstanceVersion();
 
 	vk::ApplicationInfo app_info {};
-	app_info.sType              = vk::StructureType::eApplicationInfo;
-	app_info.pNext              = nullptr;
 	app_info.pApplicationName   = "Kyty";
 	app_info.applicationVersion = 1;
 	app_info.pEngineName        = "Kyty";
@@ -1002,17 +921,11 @@ void WindowContext::CreateVulkan() {
 	}
 
 	vk::ValidationFeaturesEXT validation_features {};
-	validation_features.sType                          = vk::StructureType::eValidationFeaturesEXT;
-	validation_features.pNext                          = nullptr;
 	validation_features.enabledValidationFeatureCount  = enabled_features_count;
 	validation_features.pEnabledValidationFeatures     = enabled_features;
-	validation_features.disabledValidationFeatureCount = 0;
-	validation_features.pDisabledValidationFeatures    = nullptr;
 
 	vk::DebugUtilsMessengerCreateInfoEXT dbg_create_info {};
-	dbg_create_info.sType           = vk::StructureType::eDebugUtilsMessengerCreateInfoEXT;
 	dbg_create_info.pNext           = &validation_features;
-	dbg_create_info.flags           = {};
 	dbg_create_info.messageSeverity = vk::DebugUtilsMessageSeverityFlagBitsEXT::eVerbose |
 	                                  vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo |
 	                                  vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning |
@@ -1021,12 +934,9 @@ void WindowContext::CreateVulkan() {
 	                                  vk::DebugUtilsMessageTypeFlagBitsEXT::eValidation |
 	                                  vk::DebugUtilsMessageTypeFlagBitsEXT::ePerformance;
 	dbg_create_info.pfnUserCallback = VulkanDebugMessengerCallback;
-	dbg_create_info.pUserData       = nullptr;
 
 	vk::InstanceCreateInfo inst_info {};
-	inst_info.sType                   = vk::StructureType::eInstanceCreateInfo;
 	inst_info.pNext                   = (r.enable_validation_layers ? &dbg_create_info : nullptr);
-	inst_info.flags                   = {};
 #if defined(__APPLE__)
 	// MoltenVK requires VK_KHR_portability_enumeration + flag to surface
 	// portability devices. Without this, enumeratePhysicalDevices hides the
@@ -1093,10 +1003,8 @@ void WindowContext::CreateVulkan() {
 	}
 #endif
 
-	uint32_t queue_family = static_cast<uint32_t>(-1);
-
 	VulkanFindPhysicalDevice(graphic_ctx.instance, surface, device_extensions, surface_capabilities,
-	                         graphic_ctx.physical_device, queue_family);
+	                         graphic_ctx.physical_device, graphic_ctx.queue_family);
 
 	if (graphic_ctx.physical_device == nullptr) {
 		EXIT("Could not find suitable device");
@@ -1104,9 +1012,6 @@ void WindowContext::CreateVulkan() {
 
 	vk::PhysicalDevicePushDescriptorProperties push_descriptor_properties {};
 	vk::PhysicalDeviceProperties2              physical_device_properties {};
-	push_descriptor_properties.sType = vk::StructureType::ePhysicalDevicePushDescriptorProperties;
-	push_descriptor_properties.pNext = nullptr;
-	physical_device_properties.sType = vk::StructureType::ePhysicalDeviceProperties2;
 	physical_device_properties.pNext = &push_descriptor_properties;
 	graphic_ctx.physical_device.getProperties2(&physical_device_properties);
 	graphic_ctx.physical_device_properties = physical_device_properties.properties;
@@ -1116,6 +1021,18 @@ void WindowContext::CreateVulkan() {
 
 	LOGF("Select device: %s\n", device_properties.deviceName.data());
 
+	const vk::PhysicalDeviceImageFormatInfo2 block_texel_view_info {
+	    .format = vk::Format::eBc1RgbaUnormBlock,
+	    .type = vk::ImageType::e2D,
+	    .tiling = vk::ImageTiling::eOptimal,
+	    .usage = vk::ImageUsageFlagBits::eSampled,
+	    .flags = vk::ImageCreateFlagBits::eBlockTexelViewCompatible,
+	};
+	const auto block_texel_view_props =
+	    graphic_ctx.physical_device.getImageFormatProperties2(block_texel_view_info);
+	graphic_ctx.supports_block_texel_view = block_texel_view_props.result == vk::Result::eSuccess;
+	LOGF("Block Texel View support: %s\n", graphic_ctx.supports_block_texel_view ? "Yes" : "No");
+
 	{
 		auto available_extensions = EnumerateVulkan<vk::ExtensionProperties>(
 		    "vkEnumerateDeviceExtensionProperties",
@@ -1129,6 +1046,7 @@ void WindowContext::CreateVulkan() {
 			graphic_ctx.memory_budget_ext_enabled = true;
 		}
 		for (const auto* extension: {VK_EXT_ROBUSTNESS_2_EXTENSION_NAME,
+		                             VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME,
 		                             VK_EXT_MESH_SHADER_EXTENSION_NAME,
 		                             VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME}) {
 			if (HasExtension(available_extensions, extension)) {
@@ -1142,16 +1060,12 @@ void WindowContext::CreateVulkan() {
 		}
 	}
 
-	VulkanInitSubgroupSizeControl(graphic_ctx.physical_device, graphic_ctx);
-
-	graphic_ctx.device = VulkanCreateDevice(graphic_ctx.physical_device, r, queue_family,
-	                                        device_extensions, graphic_ctx);
+	graphic_ctx.device = VulkanCreateDevice(graphic_ctx, device_extensions);
 	if (graphic_ctx.device == nullptr) {
 		EXIT("Could not create device");
 	}
 	VULKAN_HPP_DEFAULT_DISPATCHER.init(graphic_ctx.device);
-	graphic_ctx.queue_family = queue_family;
-	graphic_ctx.device.getQueue(queue_family, 0, &graphic_ctx.queue);
+	graphic_ctx.device.getQueue(graphic_ctx.queue_family, 0, &graphic_ctx.queue);
 	EXIT_IF(graphic_ctx.queue == nullptr);
 
 	if (!graphic_ctx.CreateAllocator()) {
@@ -1159,7 +1073,7 @@ void WindowContext::CreateVulkan() {
 	}
 
 	render_context = std::make_unique<RenderContext>(graphic_ctx);
-	LibKernel::Memory::InstallGpuResources(&render_context->GetGpuResources());
+	LibKernel::Memory::InstallGpuResources(render_context.get());
 	presenter = std::make_unique<Presenter>(*this);
 }
 

src/graphics/presentation/window/window.cpp

@@ -30,7 +30,6 @@
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
-#include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/systemOverlay.h"
@@ -1017,7 +1016,7 @@ void WindowContext::UpdateTitle() {
 
 	const auto* device_name = graphic_ctx.GetPhysicalDeviceProperties().deviceName.data();
 	auto text = fmt::format(
-	    "[{} | {}] {}{}{}{}{}{}[{}] [{}], frame: {}, fps: {:f}", KYTY_BUILD_LABEL, build_type,
+	    "[{} | {}] {}{}{}{}{}{}[{}] [{}], frame: {}, fps: {:.0f}", KYTY_BUILD_LABEL, build_type,
 	    (has_title ? title : ""), (has_title ? ", " : ""), (has_title_id ? title_id : ""),
 	    (has_title_id ? ", " : ""), (has_app_ver ? app_ver : ""), (has_app_ver ? " " : ""),
 	    device_name, processor_name, frame_num, current_fps);

src/graphics/rt/hardware.cpp

@@ -1,77 +0,0 @@
-#include "graphics/rt/hardware.h"
-
-#include "common/assert.h"
-#include "common/logging/log.h"
-
-#include <algorithm>
-#include <cstring>
-
-namespace Libs::Graphics {
-namespace {
-
-bool ExtensionAvailable(const std::vector<vk::ExtensionProperties>& available, const char* name) {
-	return std::ranges::any_of(available, [name](const auto& extension) {
-		return std::strcmp(extension.extensionName, name) == 0;
-	});
-}
-
-void AppendExtensionIfMissing(std::vector<const char*>& extensions, const char* name) {
-	if (std::ranges::none_of(extensions, [name](const char* extension) {
-		    return std::strcmp(extension, name) == 0;
-	    })) {
-		extensions.push_back(name);
-	}
-}
-
-} // namespace
-
-void GraphicContext::AppendHardwareRayTracingDeviceExtensions(
-    const std::vector<vk::ExtensionProperties>& available_extensions,
-    std::vector<const char*>&                   device_extensions) {
-
-	const char* required[] = {
-	    VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME,
-	    VK_KHR_DEFERRED_HOST_OPERATIONS_EXTENSION_NAME,
-	    VK_KHR_ACCELERATION_STRUCTURE_EXTENSION_NAME,
-	    VK_KHR_RAY_QUERY_EXTENSION_NAME,
-	};
-	for (const auto* extension: required) {
-		if (!ExtensionAvailable(available_extensions, extension)) {
-			LOGF("Vulkan RT: extension %s is unavailable; hardware ray tracing disabled\n",
-			     extension);
-			rt_extensions_enabled = false;
-			return;
-		}
-	}
-	for (const auto* extension: required) {
-		AppendExtensionIfMissing(device_extensions, extension);
-	}
-	if (ExtensionAvailable(available_extensions, VK_KHR_SPIRV_1_4_EXTENSION_NAME)) {
-		AppendExtensionIfMissing(device_extensions, VK_KHR_SPIRV_1_4_EXTENSION_NAME);
-	}
-	if (ExtensionAvailable(available_extensions, VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME)) {
-		AppendExtensionIfMissing(device_extensions, VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME);
-	}
-
-	rt_extensions_enabled = true;
-	LOGF("Vulkan RT: enabling hardware ray query extensions\n");
-}
-
-void GraphicContext::LoadHardwareRayTracingFunctions() const {
-	EXIT_IF(device == nullptr);
-	if (!rt_extensions_enabled) {
-		return;
-	}
-
-	const auto& dispatcher = VULKAN_HPP_DEFAULT_DISPATCHER;
-	if (dispatcher.vkGetBufferDeviceAddressKHR == nullptr ||
-	    dispatcher.vkCreateAccelerationStructureKHR == nullptr ||
-	    dispatcher.vkDestroyAccelerationStructureKHR == nullptr ||
-	    dispatcher.vkGetAccelerationStructureBuildSizesKHR == nullptr ||
-	    dispatcher.vkCmdBuildAccelerationStructuresKHR == nullptr ||
-	    dispatcher.vkGetAccelerationStructureDeviceAddressKHR == nullptr) {
-		EXIT("Vulkan RT: failed to load required device functions\n");
-	}
-}
-
-} // namespace Libs::Graphics

src/graphics/rt/hardware.h

@@ -1,10 +0,0 @@
-#ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_RT_HARDWARE_H_
-#define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_RT_HARDWARE_H_
-
-#include "graphics/host_gpu/graphicContext.h"
-
-#include <vector>
-
-namespace Libs::Graphics {} // namespace Libs::Graphics
-
-#endif /* EMULATOR_INCLUDE_EMULATOR_GRAPHICS_RT_HARDWARE_H_ */

src/graphics/shader/recompiler/BufferFormat.h

@@ -238,28 +238,13 @@ constexpr BufferFormatInfo GetFormatInfo(Prospero::BufferFormat format) {
 	}
 }
 
-constexpr bool IsKnownFormat(Prospero::BufferFormat format) {
-	return GetFormatInfo(format).type != ComponentType::Unknown;
-}
-
-constexpr uint32_t GetFormatComponentCount(Prospero::BufferFormat format) {
-	return GetFormatInfo(format).component_count;
-}
-
-constexpr uint32_t GetFormatComponentByteOffset(Prospero::BufferFormat format, uint32_t component) {
-	const auto info = GetFormatInfo(format);
+constexpr uint32_t GetFormatComponentByteOffset(const BufferFormatInfo& info, uint32_t component) {
 	if (component >= info.component_count) {
 		return 0;
 	}
 	return info.packed_bitfield ? 0u : info.component_bit_offset[component] / 8u;
 }
 
-constexpr bool CanUseTypedBufferLoad(Prospero::BufferFormat format) {
-	const auto info = GetFormatInfo(format);
-	return info.type != ComponentType::Unknown && !info.packed_bitfield &&
-	       info.component_bits[0] == 32u;
-}
-
 } // namespace Libs::Graphics::ShaderRecompiler::Format
 
 #endif /* EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_BUFFERFORMAT_H_ */

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -1,4 +1,5 @@
 #include "graphics/shader/recompiler/ShaderRecompiler.h"
+#include "graphics/shader/recompiler/Tessellation.h"
 
 #include "common/assert.h"
 #include "common/logging/log.h"
@@ -47,14 +48,16 @@ const char* StageName(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Compute: return "CS";
 		case ShaderType::Vertex: return "VS";
+		case ShaderType::Local: return "LS";
+		case ShaderType::TessellationControl: return "HS";
+		case ShaderType::TessellationEvaluation: return "TES";
 		case ShaderType::Mesh: return "MS";
 		case ShaderType::Pixel: return "PS";
 		default: return "unknown";
 	}
 }
 
-void LogDispatcherFallback(const CompileOptions& options, const CFG::Graph& cfg, const char* phase,
-                           const std::string& reason) {
+void LogDispatcherFallback(const CompileOptions& options, const CFG::Graph& cfg, const char* phase) {
 	const auto* block        = cfg.FindBlock(cfg.failure_block);
 	const auto  start        = block != nullptr ? block->start_pc : UINT32_MAX;
 	const auto  end          = block != nullptr ? block->end_pc : UINT32_MAX;
@@ -68,7 +71,7 @@ void LogDispatcherFallback(const CompileOptions& options, const CFG::Graph& cfg,
 	     CFG::FailureKindToString(cfg.failure_kind).c_str(), cfg.failure_block, start, end,
 	     static_cast<uint64_t>(predecessors), static_cast<uint64_t>(successors),
 	     static_cast<uint64_t>(cfg.blocks.size()), static_cast<uint64_t>(cfg.natural_loops.size()),
-	     static_cast<uint64_t>(cfg.back_edges.size()), reason.c_str());
+	     static_cast<uint64_t>(cfg.back_edges.size()), cfg.unsupported_reason.c_str());
 }
 
 enum class EmbeddedFetchValueType {
@@ -77,19 +80,13 @@ enum class EmbeddedFetchValueType {
 	AttribTable,
 	Attrib,
 	BufferTable,
-	Buffer,
-	Index
+	Buffer
 };
 
 struct EmbeddedFetchSgprInfo {
 	EmbeddedFetchValueType type      = EmbeddedFetchValueType::Unknown;
 	int                    attrib_id = 0;
 	uint32_t               value     = 0;
-	std::vector<uint32_t>  prolog_loads;
-};
-
-struct EmbeddedFetchVgprInfo {
-	EmbeddedFetchValueType type = EmbeddedFetchValueType::Unknown;
 };
 
 using EmbeddedFetchVectorLanes = std::map<uint64_t, EmbeddedFetchSgprInfo>;
@@ -108,7 +105,6 @@ void ClearEmbeddedFetchVectorLanes(EmbeddedFetchVectorLanes* lanes, uint32_t reg
 	lanes->erase(first, last);
 }
 
-using EmbeddedFetchLoad = Frontend::EmbeddedFetchLoad;
 using EmbeddedFetchData = Frontend::EmbeddedFetchPlan;
 
 bool IsDecodedSgpr(const Decoder::Operand& op) {
@@ -136,20 +132,6 @@ uint32_t EmbeddedFetchDstSize(const Decoder::Instruction& inst) {
 	return inst.opcode == Decoder::Opcode::V_MAD_U64_U32 ? 2u : DecodedDstSize(inst);
 }
 
-bool EmbeddedFetchHasBranch(Decoder::Opcode opcode) {
-	switch (opcode) {
-		case Decoder::Opcode::S_SETPC_B64:
-		case Decoder::Opcode::S_BRANCH:
-		case Decoder::Opcode::S_CBRANCH_SCC0:
-		case Decoder::Opcode::S_CBRANCH_SCC1:
-		case Decoder::Opcode::S_CBRANCH_VCCZ:
-		case Decoder::Opcode::S_CBRANCH_VCCNZ:
-		case Decoder::Opcode::S_CBRANCH_EXECZ:
-		case Decoder::Opcode::S_CBRANCH_EXECNZ: return true;
-		default: return false;
-	}
-}
-
 void ClearEmbeddedFetchSgprs(std::array<EmbeddedFetchSgprInfo, 108>& sgprs,
                              const Decoder::Operand& dst, uint32_t size) {
 	if (!IsDecodedSgpr(dst)) {
@@ -232,6 +214,8 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
                                             const ShaderVertexInputInfo* input_info,
                                             uint32_t user_data_base, uint32_t user_data_count,
                                             uint32_t wave_size) {
+	const uint32_t    vertex_index_reg   = input_info->logical_stage == ShaderType::Local ? 2u : 5u;
+	const uint32_t    instance_index_reg = input_info->logical_stage == ShaderType::Local ? 5u : 8u;
 	EmbeddedFetchData data;
 	data.loads.reserve(input_info->resources_num);
 	int32_t vertex_offset_candidate   = -1;
@@ -244,11 +228,14 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 	const int buffer_reg = input_info->fetch_buffer_reg + shift_regs;
 
 	std::array<EmbeddedFetchSgprInfo, 108> sgprs {};
-	std::array<EmbeddedFetchVgprInfo, 256> vgprs {};
+	std::array<bool, 256>                 vgpr_is_index {};
 	EmbeddedFetchVectorLanes               vector_lanes;
 	const bool                             track_vector_lanes =
 	    std::none_of(decoded.instructions.begin(), decoded.instructions.end(),
-	                 [](const auto& inst) { return EmbeddedFetchHasBranch(inst.opcode); });
+	                 [](const auto& inst) {
+		                 return Decoder::IsDirectBranch(inst.opcode) ||
+		                        inst.opcode == Decoder::Opcode::S_SETPC_B64;
+	                 });
 
 	if (attrib_reg >= 0 && attrib_reg < static_cast<int>(sgprs.size())) {
 		sgprs[attrib_reg].type = EmbeddedFetchValueType::AttribTable;
@@ -269,16 +256,17 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 		// corresponding direct-draw offsets before fetching.
 		const bool vertex_index_accumulator =
 		    IsDecodedVgpr(inst.dst) &&
-		    (inst.dst.reg == 0 || (user_data_base == 8 && inst.dst.reg == 5));
+		    (inst.dst.reg == 0 || (user_data_base == 8 && inst.dst.reg == vertex_index_reg));
 		const bool instance_index_accumulator =
-		    IsDecodedVgpr(inst.dst) && (inst.dst.reg == (user_data_base == 8 ? 8u : 3u));
+		    IsDecodedVgpr(inst.dst) &&
+		    (inst.dst.reg == (user_data_base == 8 ? instance_index_reg : 3u));
 		uint32_t   sad_zero = 0;
 		const bool index_offset_add =
-		    (vertex_index_accumulator || instance_index_accumulator) &&
-		    IsDecodedSgpr(inst.src0) &&
+		    (vertex_index_accumulator || instance_index_accumulator) && IsDecodedSgpr(inst.src0) &&
 		    ((inst.opcode == Decoder::Opcode::V_ADD_I32 && IsDecodedVgpr(inst.src1) &&
 		      inst.src1.reg == inst.dst.reg) ||
-		     (user_data_base == 8 && (inst.dst.reg == 5 || inst.dst.reg == 8) &&
+		     (user_data_base == 8 &&
+		      (inst.dst.reg == vertex_index_reg || inst.dst.reg == instance_index_reg) &&
 		      inst.opcode == Decoder::Opcode::V_SAD_U32 && IsDecodedVgpr(inst.src2) &&
 		      inst.src2.reg == inst.dst.reg &&
 		      TryDecodedOperandConstant(sgprs, inst.src1, sad_zero) && sad_zero == 0));
@@ -299,8 +287,8 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 		switch (inst.opcode) {
 			case Decoder::Opcode::V_WRITELANE_B32: {
 				uint32_t lane = 0;
-				if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgprs.size()) {
-					vgprs[inst.dst.reg] = {};
+				if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgpr_is_index.size()) {
+					vgpr_is_index[inst.dst.reg] = false;
 				}
 				if (track_vector_lanes && IsDecodedVgpr(inst.dst) && IsDecodedSgpr(inst.src0) &&
 				    DecodedSgprReg(inst.src0) < sgprs.size() &&
@@ -337,7 +325,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 						auto& dst = sgprs[DecodedSgprReg(inst.dst)];
 						dst.type  = EmbeddedFetchValueType::Constant;
 						dst.value = value;
-						dst.prolog_loads.clear();
 					} else {
 						ClearEmbeddedFetchSgprs(sgprs, inst.dst, 1);
 					}
@@ -348,7 +335,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 					auto& dst = sgprs[DecodedSgprReg(inst.dst)];
 					dst.type  = EmbeddedFetchValueType::Constant;
 					dst.value = inst.src0.value;
-					dst.prolog_loads.clear();
 				}
 				break;
 			default:
@@ -365,7 +351,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 								auto& dst        = sgprs[register_id + i];
 								dst.type         = EmbeddedFetchValueType::Attrib;
 								dst.attrib_id    = index + static_cast<int>(i);
-								dst.prolog_loads = {inst.pc};
 							}
 						} else {
 							ClearEmbeddedFetchSgprs(sgprs, inst.dst, DecodedDstSize(inst));
@@ -383,7 +368,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 								dst.type  = EmbeddedFetchValueType::Buffer;
 								dst.attrib_id =
 								    BufferTableAttribFromOffset(raw_offset, static_cast<int>(i));
-								dst.prolog_loads = {inst.pc};
 							}
 						} else if (IsDecodedSgpr(inst.src1) &&
 						           DecodedSgprReg(inst.src1) < sgprs.size() &&
@@ -395,8 +379,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 								auto& dst        = sgprs[register_id + i];
 								dst.type         = EmbeddedFetchValueType::Buffer;
 								dst.attrib_id    = sgprs[DecodedSgprReg(inst.src1)].attrib_id;
-								dst.prolog_loads = sgprs[DecodedSgprReg(inst.src1)].prolog_loads;
-								dst.prolog_loads.push_back(inst.pc);
 							}
 						} else {
 							ClearEmbeddedFetchSgprs(sgprs, inst.dst, DecodedDstSize(inst));
@@ -405,13 +387,13 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 						ClearEmbeddedFetchSgprs(sgprs, inst.dst, DecodedDstSize(inst));
 					}
 				} else if (inst.opcode == Decoder::Opcode::V_CNDMASK_B32) {
-					if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgprs.size()) {
+					if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgpr_is_index.size()) {
 						ClearEmbeddedFetchVectorLanes(&vector_lanes, inst.dst.reg);
 					}
-					if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgprs.size() &&
-					    IsDecodedVgpr(inst.src0) && inst.src0.reg == 8 &&
-					    IsDecodedVgpr(inst.src1) && inst.src1.reg == 5) {
-						vgprs[inst.dst.reg].type = EmbeddedFetchValueType::Index;
+					if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgpr_is_index.size() &&
+					    IsDecodedVgpr(inst.src0) && inst.src0.reg == instance_index_reg &&
+					    IsDecodedVgpr(inst.src1) && inst.src1.reg == vertex_index_reg) {
+						vgpr_is_index[inst.dst.reg] = true;
 					}
 				} else if (IsEmbeddedFetchAttribPropagationAlu(inst)) {
 					if (IsDecodedSgpr(inst.dst) && IsDecodedSgpr(inst.src0) &&
@@ -435,22 +417,16 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 									break;
 								default: dst.value = src0 + src1; break;
 							}
-							dst.prolog_loads.clear();
 						} else {
 							ClearEmbeddedFetchSgprs(sgprs, inst.dst, 1);
 						}
 					}
 				} else if (IsEmbeddedFetchBufferLoad(inst)) {
-					if (IsDecodedVgpr(inst.src0) && inst.src0.reg < vgprs.size() &&
-					    vgprs[inst.src0.reg].type == EmbeddedFetchValueType::Index &&
+					if (IsDecodedVgpr(inst.src0) && inst.src0.reg < vgpr_is_index.size() &&
+					    vgpr_is_index[inst.src0.reg] &&
 					    IsDecodedSgpr(inst.src1) && DecodedSgprReg(inst.src1) < sgprs.size() &&
 					    sgprs[DecodedSgprReg(inst.src1)].type == EmbeddedFetchValueType::Buffer) {
-						const auto&       buffer = sgprs[DecodedSgprReg(inst.src1)];
-						EmbeddedFetchLoad load;
-						load.pc           = inst.pc;
-						load.attrib_id    = buffer.attrib_id;
-						load.components   = DecodedDstSize(inst);
-						load.prolog_loads = buffer.prolog_loads;
+						const auto& buffer = sgprs[DecodedSgprReg(inst.src1)];
 						if (data.loads.empty()) {
 							if (!vertex_offset_conflict) {
 								data.vertex_offset_sgpr = vertex_offset_candidate;
@@ -459,7 +435,10 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 								data.instance_offset_sgpr = instance_offset_candidate;
 							}
 						}
-						data.loads.push_back(load);
+						auto& load        = data.loads.emplace_back();
+						load.pc           = inst.pc;
+						load.attrib_id    = buffer.attrib_id;
+						load.components   = DecodedDstSize(inst);
 					}
 				}
 				break;
@@ -467,7 +446,8 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 		if (inst.opcode == Decoder::Opcode::V_MOVRELD_B32) {
 			vector_lanes.clear();
 		} else if (inst.opcode != Decoder::Opcode::V_WRITELANE_B32 && IsDecodedVgpr(inst.dst)) {
-			for (uint32_t i = 0; i < EmbeddedFetchDstSize(inst) && inst.dst.reg + i < vgprs.size();
+			for (uint32_t i = 0;
+			     i < EmbeddedFetchDstSize(inst) && inst.dst.reg + i < vgpr_is_index.size();
 			     i++) {
 				ClearEmbeddedFetchVectorLanes(&vector_lanes, inst.dst.reg + i);
 			}
@@ -480,22 +460,9 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 Decoder::Program DecodeFusedProgram(std::span<const uint32_t> front, std::span<const uint32_t> back,
                                     std::vector<uint32_t>& joined_code) {
 	EXIT_IF(back.empty());
-	Decoder::Program result;
-	uint32_t         front_words = 0;
-	while (front_words < front.size()) {
-		auto& inst = result.instructions.emplace_back();
-		Decoder::DecodeInstruction(front, front_words, inst);
-		front_words += inst.word_count;
-		if (inst.opcode == Decoder::Opcode::S_SETPC_B64) {
-			EXIT_NOT_IMPLEMENTED(inst.src0.kind != Decoder::OperandKind::Sgpr ||
-			                     inst.src0.reg != 6u);
-			break;
-		}
-		EXIT_NOT_IMPLEMENTED(inst.opcode == Decoder::Opcode::S_ENDPGM);
-	}
-	EXIT_IF(result.instructions.empty() ||
-	        result.instructions.back().opcode != Decoder::Opcode::S_SETPC_B64);
-	joined_code.assign(front.begin(), front.begin() + front_words);
+	auto       result      = Decoder::DecodeFrontProgram(front);
+	const auto front_words = static_cast<uint32_t>(result.code.size());
+	joined_code.assign(result.code.begin(), result.code.end());
 	joined_code.insert(joined_code.end(), back.begin(), back.end());
 	// The merged-stage ABI passes the back shader in s[6:7]. Give that handoff an
 	// ordinary CFG edge, retaining both bodies in one register and LDS lifetime.
@@ -523,7 +490,9 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		EXIT("shader recompiler input is empty\n");
 	}
 	if (options.stage != ShaderType::Compute && options.stage != ShaderType::Vertex &&
-	    options.stage != ShaderType::Pixel && options.stage != ShaderType::Mesh) {
+	    options.stage != ShaderType::Pixel && options.stage != ShaderType::Mesh &&
+	    options.stage != ShaderType::Local && options.stage != ShaderType::TessellationControl &&
+	    options.stage != ShaderType::TessellationEvaluation) {
 		EXIT("shader recompiler received unsupported stage %u\n",
 		     static_cast<unsigned>(options.stage));
 	}
@@ -541,8 +510,14 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 
 	Decoder::Program decoded;
 	std::vector<uint32_t> joined_code;
-	if (options.stage == ShaderType::Mesh) {
+	if (!options.back_code.empty()) {
 		decoded = DecodeFusedProgram(code, options.back_code, joined_code);
+	} else if (options.stage == ShaderType::Local) {
+		decoded = Decoder::DecodeFrontProgram(code);
+		// The separately compiled hull half runs in the next Vulkan stage.
+		auto& handoff     = decoded.instructions.back();
+		handoff.opcode    = Decoder::Opcode::S_ENDPGM;
+		handoff.src_count = 0;
 	} else {
 		Decoder::DecodeProgram(code, decoded);
 	}
@@ -567,27 +542,13 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
 	     static_cast<uint64_t>(cfg.blocks.size()), static_cast<uint64_t>(cfg.natural_loops.size()),
 	     static_cast<uint64_t>(cfg.back_edges.size()), phase_ms());
-	bool        dispatcher_fallback = false;
-	std::string dispatcher_reason;
 	if (cfg.irreducible) {
-		dispatcher_fallback = true;
-		dispatcher_reason   = cfg.unsupported_reason;
-		LogDispatcherFallback(options, cfg, "build", dispatcher_reason);
+		LogDispatcherFallback(options, cfg, "build");
 	} else {
-		const auto unstructured_cfg = cfg;
 		LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " CFG Structurize\n",
 		     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
 		if (!CFG::Structurize(cfg)) {
-			dispatcher_fallback      = true;
-			dispatcher_reason        = cfg.unsupported_reason;
-			const auto failure_kind  = cfg.failure_kind;
-			const auto failure_block = cfg.failure_block;
-			LogDispatcherFallback(options, cfg, "structurize", dispatcher_reason);
-			cfg                    = unstructured_cfg;
-			cfg.unsupported        = true;
-			cfg.failure_kind       = failure_kind;
-			cfg.failure_block      = failure_block;
-			cfg.unsupported_reason = dispatcher_reason;
+			LogDispatcherFallback(options, cfg, "structurize");
 		} else {
 			LOGF("%s structured CFG success: blocks=%" PRIu64 "\n", GetDumpLabel(options),
 			     static_cast<uint64_t>(cfg.blocks.size()));
@@ -599,45 +560,25 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		     static_cast<uint64_t>(cfg.natural_loops.size()), phase_ms());
 	}
 
-	const ShaderVertexInputInfo*  vertex  = nullptr;
-	const ShaderPixelInputInfo*   pixel   = nullptr;
-	const ShaderComputeInputInfo* compute = nullptr;
-	switch (options.stage) {
-		case ShaderType::Vertex:
-		case ShaderType::Mesh: vertex = options.input_info.vertex; break;
-		case ShaderType::Pixel:
-			pixel = options.input_info.pixel;
-			break;
-		case ShaderType::Compute:
-			compute = options.input_info.compute;
-			break;
-		default: break;
-	}
 	EmbeddedFetchData embedded_fetch;
-	if (options.stage == ShaderType::Vertex && vertex->fetch_embedded) {
-		embedded_fetch = DetectEmbeddedVertexFetch(decoded, vertex, options.user_data_base,
-		                                           static_cast<uint32_t>(options.user_data.size()),
-		                                           options.wave_size);
+	if ((options.stage == ShaderType::Vertex || options.stage == ShaderType::Local) &&
+	    options.input_info.vertex != nullptr && options.input_info.vertex->fetch_embedded) {
+		embedded_fetch = DetectEmbeddedVertexFetch(
+		    decoded, options.input_info.vertex, options.user_data_base,
+		    static_cast<uint32_t>(options.user_data.size()), options.wave_size);
 		if (!embedded_fetch.loads.empty()) {
 			LOGF("%s embedded vertex fetch plan: detected=%" PRIu64 "\n", GetDumpLabel(options),
 			     static_cast<uint64_t>(embedded_fetch.loads.size()));
 		}
 	}
 	Frontend::TranslateOptions translate_options {
-	    .stage               = options.stage,
-	    .wave_size           = options.wave_size,
-	    .shader_hash         = options.shader_hash,
-	    .user_data_base      = options.user_data_base,
-	    .user_data_count     = static_cast<uint32_t>(options.user_data.size()),
-	    .scratch_dwords      = options.scratch_dwords,
-	    .dispatcher_fallback = dispatcher_fallback,
-	    .cfg_failure_kind    = cfg.failure_kind,
-	    .fallback_reason     = dispatcher_reason.empty() ? cfg.unsupported_reason
-	                                                    : dispatcher_reason,
-	    .vertex              = vertex,
-	    .pixel               = pixel,
-	    .compute             = compute,
-	    .embedded_fetch      = embedded_fetch.loads.empty() ? nullptr : &embedded_fetch,
+	    .stage            = options.stage,
+	    .wave_size        = options.wave_size,
+	    .shader_hash      = options.shader_hash,
+	    .user_data_base   = options.user_data_base,
+	    .user_data_count  = static_cast<uint32_t>(options.user_data.size()),
+	    .input_info       = options.input_info,
+	    .embedded_fetch   = embedded_fetch.loads.empty() ? nullptr : &embedded_fetch,
 	};
 	LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " IR TranslateProgram\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
@@ -660,15 +601,11 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		IR::RemoveIdentities(ir.blocks);
 		IR::EliminateDeadCode(ir.blocks);
 	}
+	LowerTessellationMemory(ir, options);
 	IR::BuildSrtPlan(ir);
 	IR::EliminateDeadCode(ir.blocks);
 	IR::TrackResources(ir);
 	IR::EliminateDeadCode(ir.blocks);
-	if (options.stage == ShaderType::Vertex) {
-		ir.info.vertex_offset_sgpr   = embedded_fetch.vertex_offset_sgpr;
-		ir.info.instance_offset_sgpr = embedded_fetch.instance_offset_sgpr;
-	}
-
 	TranslateResult result;
 	result.program = std::move(ir);
 	if (options.dump_ir) {
@@ -687,24 +624,8 @@ CompileResult CompileProgram(TranslateResult translated, const CompileOptions& o
 	IR::RemoveIdentities(ir.blocks);
 	IR::EliminateDeadCode(ir.blocks);
 
-	const ShaderVertexInputInfo*  vertex  = nullptr;
-	const ShaderPixelInputInfo*   pixel   = nullptr;
-	const ShaderComputeInputInfo* compute = nullptr;
-	switch (options.stage) {
-		case ShaderType::Vertex:
-		case ShaderType::Mesh: vertex = options.input_info.vertex; break;
-		case ShaderType::Pixel: pixel = options.input_info.pixel; break;
-		case ShaderType::Compute: compute = options.input_info.compute; break;
-		default: EXIT("invalid shader stage\n");
-	}
-
-	IR::ShaderInfoOptions info_options;
-	info_options.vertex  = vertex;
-	info_options.pixel   = pixel;
-	info_options.compute = compute;
-	IR::CollectShaderInfo(ir, info_options);
+	IR::CollectShaderInfo(ir, options.input_info);
 	IR::AllocateBindings(ir, push_data_start_dword);
-	Spirv::AnalyzeProgramRequirements(ir);
 	std::string ir_dump;
 	if (options.dump_ir) {
 		ir_dump = MakeIrDump(translated.cfg_dump, ir);
@@ -714,11 +635,11 @@ CompileResult CompileProgram(TranslateResult translated, const CompileOptions& o
 	}
 
 	LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " SPIR-V EmitProgram\n",
-	     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
+	     GetDumpLabel(options), StageName(ir.stage), ir.shader_hash);
 	auto spirv = Spirv::EmitProgram(ir, options.input_info);
 	LOGF("%s phase end: stage=%s hash=0x%016" PRIx64 " SPIR-V EmitProgram words=%" PRIu64
 	     " elapsed_ms=%" PRIu64 "\n",
-	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
+	     GetDumpLabel(options), StageName(ir.stage), ir.shader_hash,
 	     static_cast<uint64_t>(spirv.size()),
 	     static_cast<uint64_t>(std::chrono::duration_cast<std::chrono::milliseconds>(
 	                               std::chrono::steady_clock::now() - emit_begin)

src/graphics/shader/recompiler/ShaderRecompiler.h

@@ -15,7 +15,6 @@ struct CompileOptions {
 	ShaderType                  stage           = ShaderType::Compute;
 	uint32_t                    wave_size       = 64;
 	uint32_t                    user_data_base  = 0;
-	uint32_t                    scratch_dwords  = 0;
 	uint64_t                    shader_hash     = 0;
 	bool                        dump_ir                    = true;
 	bool                        early_dump                 = false;

src/graphics/shader/recompiler/Tessellation.cpp

@@ -0,0 +1,330 @@
+#include "graphics/shader/recompiler/Tessellation.h"
+
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "graphics/shader/recompiler/ShaderRecompiler.h"
+#include "graphics/shader/recompiler/frontend/decode/ShaderDecoder.h"
+#include "graphics/shader/recompiler/ir/ShaderIR.h"
+#include "graphics/shader/recompiler/ir/passes/ConstantPropagation.h"
+#include "graphics/shader/recompiler/ir/passes/DeadCodeElimination.h"
+
+#include <algorithm>
+#include <array>
+#include <unordered_map>
+#include <utility>
+
+namespace Libs::Graphics::ShaderRecompiler {
+namespace {
+
+struct TessellationAddress {
+	enum class Kind { Unknown, Affine, PackedControlPoint };
+	Kind     kind        = Kind::Unknown;
+	uint32_t coefficient = 0;
+	uint32_t constant    = 0;
+};
+
+uint32_t ReflectTessellationStride(const Decoder::Program& program, bool local,
+                                   uint32_t control_points, uint32_t input_stride) {
+	using namespace Decoder;
+	using Address = TessellationAddress;
+	using Kind    = Address::Kind;
+	std::array<Address, IR::NumVectorRegs> registers {};
+	registers[local ? 3u : 1u] =
+	    local ? Address {Kind::Affine, 1u, 0u} : Address {Kind::PackedControlPoint};
+	const auto constant = [](uint32_t value) { return Address {Kind::Affine, 0u, value}; };
+	const auto read     = [&](const Operand& operand) -> Address {
+		if (operand.absolute || operand.negate || operand.sdwa_sext || operand.op_sel ||
+		    operand.op_sel_hi || operand.negate_hi || operand.dpp)
+			return {};
+		Address value;
+		if (operand.kind == OperandKind::IntegerInlineConstant ||
+		    operand.kind == OperandKind::LiteralConstant) {
+			value = constant(operand.value);
+		} else if (operand.kind == OperandKind::Vgpr) {
+			value = registers.at(operand.reg);
+		}
+		// The native packed HS ID's low byte is the patch ordinal. The logical
+		// interface addresses one patch, as in the frontend's v1 initialization.
+		if (value.kind == Kind::PackedControlPoint && operand.sdwa_sel == 0u) return constant(0u);
+		return operand.sdwa_sel == 6u ? value : Address {};
+	};
+	const auto affine = [](uint64_t coefficient, uint64_t offset) {
+		return coefficient <= UINT32_MAX && offset <= UINT32_MAX
+		           ? Address {Kind::Affine, static_cast<uint32_t>(coefficient),
+		                      static_cast<uint32_t>(offset)}
+		           : Address {};
+	};
+	const auto add = [&](Address lhs, Address rhs) {
+		return lhs.kind == Kind::Affine && rhs.kind == Kind::Affine
+		           ? affine(uint64_t {lhs.coefficient} + rhs.coefficient,
+		                    uint64_t {lhs.constant} + rhs.constant)
+		           : Address {};
+	};
+	const auto multiply = [&](Address lhs, Address rhs) {
+		if ((lhs.kind == Kind::Affine && lhs.coefficient == 0u && lhs.constant == 0u) ||
+		    (rhs.kind == Kind::Affine && rhs.coefficient == 0u && rhs.constant == 0u))
+			return constant(0u);
+		if (lhs.kind != Kind::Affine || rhs.kind != Kind::Affine) return Address {};
+		if (rhs.coefficient != 0u) std::swap(lhs, rhs);
+		return rhs.coefficient == 0u ? affine(uint64_t {lhs.coefficient} * rhs.constant,
+		                                      uint64_t {lhs.constant} * rhs.constant)
+		                             : Address {};
+	};
+	const auto low24 = [&](Address value) {
+		if (value.kind == Kind::Affine && value.coefficient == 0u)
+			return constant(value.constant & 0xffffffu);
+		return value.kind == Kind::Affine &&
+		               uint64_t {value.coefficient} * (control_points - 1u) + value.constant <=
+		                   0xffffffu
+		           ? value
+		           : Address {};
+	};
+	uint32_t stride = 0;
+	for (const auto& inst: program.instructions) {
+		// Stage exits preserve the active path's definitions. An internal join
+		// would require merging register definitions.
+		EXIT_NOT_IMPLEMENTED(IsDirectBranch(inst.opcode) &&
+		                     inst.branch_target != program.instructions.back().pc);
+		const bool local_store =
+		    inst.opcode == Opcode::DS_WRITE_B32 || inst.opcode == Opcode::DS_WRITE2_B32;
+		const bool buffer_store = inst.opcode == Opcode::BUFFER_STORE_DWORD ||
+		                          inst.opcode == Opcode::BUFFER_STORE_DWORDX2 ||
+		                          inst.opcode == Opcode::BUFFER_STORE_DWORDX3 ||
+		                          inst.opcode == Opcode::BUFFER_STORE_DWORDX4;
+		const bool control_store =
+		    !local && buffer_store && inst.src2.kind == OperandKind::Sgpr && inst.src2.reg == 2u;
+		const bool control_read =
+		    !local && (inst.opcode == Opcode::DS_READ_B32 || inst.opcode == Opcode::DS_READ2_B32);
+		if ((local && local_store) || control_store || control_read) {
+			const auto address = read(inst.src0);
+			if (address.kind != Kind::Affine) {
+				EXIT("%s tessellation address is not affine at pc 0x%08x\n", local ? "LS" : "HS",
+				     inst.pc);
+			}
+			if (address.coefficient != 0u) {
+				const auto expected = control_read ? input_stride : stride;
+				EXIT_NOT_IMPLEMENTED((address.coefficient & 3u) != 0u ||
+				                     (expected != 0u && expected != address.coefficient));
+				if (!control_read) stride = address.coefficient;
+			}
+		}
+		// Store vdata is a source, despite occupying the decoder's dst field.
+		if (local_store || buffer_store || inst.dst.kind != OperandKind::Vgpr) continue;
+		const auto lhs   = read(inst.src0);
+		const auto rhs   = read(inst.src1);
+		const auto third = read(inst.src2);
+		Address    value;
+		switch (inst.opcode) {
+			case Opcode::V_BFE_U32:
+				if (lhs.kind == Kind::PackedControlPoint && rhs.kind == Kind::Affine &&
+				    rhs.coefficient == 0u && third.kind == Kind::Affine &&
+				    third.coefficient == 0u) {
+					if (rhs.constant == 8u && third.constant == 5u) value = {Kind::Affine, 1u, 0u};
+					if (rhs.constant == 0u && third.constant == 8u) value = constant(0u);
+				}
+				break;
+			case Opcode::V_MUL_U32_U24: value = multiply(low24(lhs), low24(rhs)); break;
+			case Opcode::V_MAD_U32_U24: value = add(multiply(low24(lhs), low24(rhs)), third); break;
+			case Opcode::V_LSHL_ADD_U32:
+				if (rhs.kind == Kind::Affine && rhs.coefficient == 0u && rhs.constant < 32u)
+					value = add(multiply(lhs, constant(1u << rhs.constant)), third);
+				break;
+			case Opcode::V_LSHLREV_B32:
+				if (lhs.kind == Kind::Affine && lhs.coefficient == 0u && lhs.constant < 32u)
+					value = multiply(rhs, constant(1u << lhs.constant));
+				break;
+			case Opcode::V_SUB_NC_U32:
+				if (lhs.kind == Kind::Affine && rhs.kind == Kind::Affine &&
+				    lhs.coefficient >= rhs.coefficient && lhs.constant >= rhs.constant)
+					value = affine(lhs.coefficient - rhs.coefficient, lhs.constant - rhs.constant);
+				break;
+			default: break;
+		}
+		if (inst.dst.sdwa_sel != 6u || inst.dst.op_sel || inst.dst.omod || inst.dst.clamp ||
+		    inst.dst.dpp)
+			value = {};
+		for (uint32_t index = 0;
+		     index < std::max(inst.data_dwords, 1u) && inst.dst.reg + index < registers.size(); index++) {
+			registers[inst.dst.reg + index] = {};
+		}
+		registers.at(inst.dst.reg) = value;
+	}
+	EXIT_NOT_IMPLEMENTED(stride == 0u);
+	return stride;
+}
+
+const IR::Inst* TessellationBufferBase(const IR::Inst& inst) {
+	if (IR::BufferAccessOf(inst.GetOpcode()) == IR::BufferAccess::None || inst.NumArgs() <= 3u)
+		return nullptr;
+	const auto* base = inst.Arg(3).Resolve().TryInstruction();
+	return base != nullptr && base->GetOpcode() == IR::ValueOpcode::TessellationBase ? base
+	                                                                                 : nullptr;
+}
+
+IR::Value ActiveAddress(IR::Block& block, IR::Block::iterator before, IR::Value value,
+                        IR::Value predicate, std::unordered_map<IR::Inst*, IR::Value>& resolved) {
+	using namespace IR;
+	value        = value.Resolve();
+	auto* source = value.TryInstruction();
+	if (source == nullptr) return value;
+	if (const auto found = resolved.find(source); found != resolved.end()) return found->second;
+	if (source->GetOpcode() == ValueOpcode::SelectU32) {
+		return source->Arg(0).Resolve() == predicate.Resolve()
+		           ? ActiveAddress(block, before, source->Arg(1), predicate, resolved)
+		           : value;
+	}
+	if (source->GetOpcode() != ValueOpcode::IAdd32 && source->GetOpcode() != ValueOpcode::ISub32)
+		return value;
+	const auto lhs = ActiveAddress(block, before, source->Arg(0), predicate, resolved);
+	const auto rhs = ActiveAddress(block, before, source->Arg(1), predicate, resolved);
+	if (lhs != source->Arg(0).Resolve() || rhs != source->Arg(1).Resolve()) {
+		auto copy = block.PrependNewInst(before, source->GetOpcode(), {lhs, rhs},
+		                                 source->Flags<uint64_t>());
+		value     = Value(&*copy);
+	}
+	resolved.emplace(source, value);
+	return value;
+}
+
+} // namespace
+
+void AnalyzeTessellationPrograms(std::span<const uint32_t> local, std::span<const uint32_t> control,
+                                 ShaderTessellationInputInfo& info) {
+	const auto       local_program = Decoder::DecodeFrontProgram(local);
+	Decoder::Program control_program;
+	Decoder::DecodeProgram(control, control_program);
+	info.ls_stride = ReflectTessellationStride(local_program, true, info.input_control_points, 0u);
+	info.hs_stride = ReflectTessellationStride(control_program, false, info.output_control_points,
+	                                           info.ls_stride);
+	LOGF("Tessellation interface: input_cp=%u output_cp=%u ls_stride=%u hs_stride=%u\n",
+	     info.input_control_points, info.output_control_points, info.ls_stride, info.hs_stride);
+}
+
+void LowerTessellationMemory(IR::Program& program, const CompileOptions& options) {
+	using namespace IR;
+	if (options.stage != ShaderType::Local && options.stage != ShaderType::TessellationControl &&
+	    options.stage != ShaderType::TessellationEvaluation) {
+		return;
+	}
+	const auto& tess = options.input_info.vertex->tess;
+	// Ring addresses can reuse data VGPRs. Their inactive values are irrelevant to
+	// a store guarded by the same EXEC predicate, but must remain intact elsewhere.
+	for (auto* block: program.blocks) {
+		for (auto it = block->begin(); it != block->end(); ++it) {
+			if (TessellationBufferBase(*it) == nullptr) continue;
+			std::unordered_map<Inst*, Value> resolved;
+			it->SetArg(
+			    2, ActiveAddress(*block, it, it->Arg(2), it->Arg(it->NumArgs() - 1u), resolved));
+		}
+	}
+	ConstantPropagationPass(program.blocks);
+	uint32_t reads = 0, writes = 0, factors = 0;
+	for (auto* block: program.blocks) {
+		for (auto it = block->begin(); it != block->end(); ++it) {
+			auto&      inst   = *it;
+			const auto shared = SharedAccessOf(inst.GetOpcode());
+			const auto buffer = BufferAccessOf(inst.GetOpcode());
+			if (shared == SharedAccess::None && buffer == BufferAccess::None) {
+				continue;
+			}
+			const auto&           memory = program.memory_info.at(inst.Flags<MemoryFlags>().index);
+			bool                  write  = false;
+			uint32_t              components = 0;
+			TessellationAttribute kind;
+			Value                 address, predicate;
+			if (shared != SharedAccess::None && memory.kind == ResourceKind::Lds) {
+				write = shared == SharedAccess::Write;
+				EXIT_NOT_IMPLEMENTED(memory.data_bits != 32u ||
+				                     (options.stage == ShaderType::Local
+				                          ? !write
+				                          : options.stage != ShaderType::TessellationControl ||
+				                                shared != SharedAccess::Read));
+				kind       = write ? TessellationAttribute::LocalOutput
+				                   : TessellationAttribute::ControlInput;
+				components = SharedComponentCount(inst.GetOpcode());
+				address    = inst.Arg(0);
+				predicate  = inst.Arg(inst.NumArgs() - 1u);
+			} else if (buffer != BufferAccess::None && memory.kind == ResourceKind::Buffer) {
+				const auto* base = TessellationBufferBase(inst);
+				if (base == nullptr) {
+					continue;
+				}
+				EXIT_NOT_IMPLEMENTED(memory.data_bits != 32u || memory.formatted || memory.idxen ||
+				                     !memory.offen || buffer == BufferAccess::Atomic);
+				write      = buffer == BufferAccess::Write;
+				components = BufferComponentCount(inst.GetOpcode());
+				address    = inst.Arg(2).Resolve();
+				predicate  = inst.Arg(inst.NumArgs() - 1u);
+				if (base->Arg(0).U32() == 1u) {
+					EXIT_NOT_IMPLEMENTED(!write ||
+					                     options.stage != ShaderType::TessellationControl);
+					kind = TessellationAttribute::Factor;
+					factors += components;
+				} else {
+					kind = write ? TessellationAttribute::ControlOutput
+					             : TessellationAttribute::EvaluationInput;
+					EXIT_NOT_IMPLEMENTED(write
+					                         ? options.stage != ShaderType::TessellationControl
+					                         : options.stage != ShaderType::TessellationEvaluation);
+					if (address.IsImmediate() &&
+					    address.U32() >= tess.hs_stride * tess.output_control_points) {
+						EXIT_NOT_IMPLEMENTED(!write);
+						kind = TessellationAttribute::PatchOutput;
+					}
+				}
+			} else {
+				continue;
+			}
+			const auto emit = [&](ValueOpcode opcode, std::initializer_list<Value> args) {
+				return Value(&*block->PrependNewInst(it, opcode, args));
+			};
+			std::array<Value, 4> values;
+			for (uint32_t component = 0; component < components; component++) {
+				const auto offset = memory.offset + 4u * component;
+				const auto byte_address =
+				    offset == 0u ? address : emit(ValueOpcode::IAdd32, {address, Value(offset)});
+				if (write) {
+					Value data = shared != SharedAccess::None ? inst.Arg(component + 1u)
+					             : components == 1u
+					                 ? inst.Arg(4)
+					                 : emit(components == 2u   ? ValueOpcode::CompositeExtractU32x2
+					                        : components == 3u ? ValueOpcode::CompositeExtractU32x3
+					                                           : ValueOpcode::CompositeExtractU32x4,
+					                        {inst.Arg(4), Value(component)});
+					emit(ValueOpcode::SetTessellationAttribute,
+					     {Value(static_cast<uint32_t>(kind)), byte_address, data, predicate});
+					writes++;
+				} else {
+					values[component] =
+					    emit(ValueOpcode::GetTessellationAttribute,
+					         {Value(static_cast<uint32_t>(kind)), byte_address, predicate});
+					reads++;
+				}
+			}
+			if (!write) {
+				Value replacement = values[0];
+				if (components == 2u)
+					replacement =
+					    emit(ValueOpcode::CompositeConstructU32x2, {values[0], values[1]});
+				if (components == 3u)
+					replacement = emit(ValueOpcode::CompositeConstructU32x3,
+					                   {values[0], values[1], values[2]});
+				if (components == 4u)
+					replacement = emit(ValueOpcode::CompositeConstructU32x4,
+					                   {values[0], values[1], values[2], values[3]});
+				inst.ReplaceUsesWith(replacement);
+			}
+			inst.Invalidate();
+		}
+	}
+	ConstantPropagationPass(program.blocks);
+	RemoveIdentities(program.blocks);
+	EliminateDeadCode(program.blocks);
+	LOGF("%s tessellation lowering: reads=%u writes=%u factors=%u\n",
+	     options.stage == ShaderType::Local                 ? "LS"
+	     : options.stage == ShaderType::TessellationControl ? "HS"
+	                                                       : "TES",
+	     reads, writes, factors);
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler

src/graphics/shader/recompiler/Tessellation.h

@@ -0,0 +1,21 @@
+#pragma once
+
+#include <cstdint>
+#include <span>
+
+namespace Libs::Graphics {
+struct ShaderTessellationInputInfo;
+}
+
+namespace Libs::Graphics::ShaderRecompiler {
+
+struct CompileOptions;
+namespace IR {
+struct Program;
+}
+
+void AnalyzeTessellationPrograms(std::span<const uint32_t> local, std::span<const uint32_t> control,
+                                 ShaderTessellationInputInfo& info);
+void LowerTessellationMemory(IR::Program& program, const CompileOptions& options);
+
+} // namespace Libs::Graphics::ShaderRecompiler

src/graphics/shader/recompiler/backend/spirv/SpirvBuilder.cpp

@@ -1,6 +1,6 @@
 #include "graphics/shader/recompiler/backend/spirv/SpirvBuilder.h"
 
-#include "common/debug.h"
+#include "common/assert.h"
 
 #include <algorithm>
 #include <cstring>
@@ -15,9 +15,9 @@ static void AppendInstructionWords(std::vector<uint32_t>& section, const uint32_
 	if (words_num == 0) {
 		return;
 	}
-	const auto opcode     = words[0];
-	const auto word_count = static_cast<uint32_t>(words_num);
-	section.push_back((word_count << 16u) | opcode);
+	const uint32_t opcode     = words[0];
+	const auto     word_count = static_cast<uint32_t>(words_num);
+	section.push_back((word_count << spv::WordCountShift) | opcode);
 	section.insert(section.end(), words + 1, words + words_num);
 }
 
@@ -36,15 +36,17 @@ void Builder::RequireVersion(uint32_t version) {
 	m_version = std::max(m_version, version);
 }
 
-void Builder::RequireCapability(uint32_t capability) {
+void Builder::RequireCapability(spv::Capability capability) {
 	if (m_required_capabilities.insert(capability).second) {
-		AddCapability({capability});
+		AppendInstruction(m_capabilities, spv::OpCapability, capability);
 	}
 }
 
 void Builder::RequireExtension(const char* name) {
 	if (m_required_extensions.emplace(name).second) {
-		AddExtension(name);
+		std::vector<uint32_t> operands;
+		AppendString(operands, name);
+		AppendInstruction(m_extensions, spv::OpExtension, operands);
 	}
 }
 
@@ -54,41 +56,31 @@ uint32_t Builder::Import(const char* name) {
 	}
 	const auto id = AllocateId();
 	m_import_ids.emplace(name, id);
-	AddExtInstImport(id, name);
+	std::vector<uint32_t> operands = {id};
+	AppendString(operands, name);
+	AppendInstruction(m_ext_inst_imports, spv::OpExtInstImport, operands);
 	return id;
 }
 
-uint32_t Builder::Type(uint32_t opcode, std::initializer_list<uint32_t> operands) {
-	return Type(opcode, std::vector<uint32_t>(operands));
-}
-
-uint32_t Builder::Type(uint32_t opcode, const std::vector<uint32_t>& operands) {
-	std::vector<uint32_t> key;
-	key.reserve(operands.size() + 2u);
-	key.push_back(opcode);
-	key.push_back(static_cast<uint32_t>(operands.size()));
-	key.insert(key.end(), operands.begin(), operands.end());
+uint32_t Builder::DeclareType(spv::Op opcode, std::vector<uint32_t> key) {
 	if (const auto it = m_declaration_ids.find(key); it != m_declaration_ids.end()) {
 		return it->second;
 	}
 	const auto id = AllocateId();
+	AppendInstruction(m_declarations, opcode, id, std::span<const uint32_t>(key).subspan(2));
 	m_declaration_ids.emplace(std::move(key), id);
-	std::vector<uint32_t> words {opcode, id};
-	words.insert(words.end(), operands.begin(), operands.end());
-	AppendInstructionWords(m_declarations, words.data(), words.size());
 	return id;
 }
 
-uint32_t Builder::DecoratedType(uint32_t opcode, std::initializer_list<uint32_t> operands,
-                                std::initializer_list<TypeAnnotation> annotations) {
+uint32_t Builder::DeclareDecoratedType(spv::Op opcode, std::vector<uint32_t> key,
+                                       std::initializer_list<TypeAnnotation> annotations) {
 	if (annotations.size() == 0) {
-		return Type(opcode, operands);
+		return DeclareType(opcode, std::move(key));
 	}
-	std::vector<uint32_t> key {opcode, static_cast<uint32_t>(operands.size())};
-	key.insert(key.end(), operands.begin(), operands.end());
+	const auto operand_count = key[1];
 	key.push_back(static_cast<uint32_t>(annotations.size()));
 	for (const auto& annotation: annotations) {
-		key.push_back(annotation.opcode);
+		AppendOperand(key, annotation.opcode);
 		key.push_back(static_cast<uint32_t>(annotation.operands.size()));
 		key.insert(key.end(), annotation.operands.begin(), annotation.operands.end());
 	}
@@ -96,48 +88,35 @@ uint32_t Builder::DecoratedType(uint32_t opcode, std::initializer_list<uint32_t>
 		return it->second;
 	}
 	const auto id = AllocateId();
+	AppendInstruction(m_declarations, opcode, id,
+	                  std::span<const uint32_t>(key).subspan(2, operand_count));
 	m_declaration_ids.emplace(std::move(key), id);
-	std::vector<uint32_t> words {opcode, id};
-	words.insert(words.end(), operands.begin(), operands.end());
-	AppendInstructionWords(m_declarations, words.data(), words.size());
 	for (const auto& annotation: annotations) {
-		words = {annotation.opcode, id};
-		words.insert(words.end(), annotation.operands.begin(), annotation.operands.end());
-		AppendInstructionWords(m_annotations, words.data(), words.size());
+		AppendInstruction(m_annotations, annotation.opcode, id, annotation.operands);
 	}
 	return id;
 }
 
-uint32_t Builder::Constant(uint32_t opcode, uint32_t type,
-                           std::initializer_list<uint32_t> operands) {
-	return Constant(opcode, type, std::vector<uint32_t>(operands));
-}
-
-uint32_t Builder::Constant(uint32_t opcode, uint32_t type, const std::vector<uint32_t>& operands) {
-	std::vector<uint32_t> key;
-	key.reserve(operands.size() + 2u);
-	key.push_back(opcode);
-	key.push_back(type);
-	key.insert(key.end(), operands.begin(), operands.end());
+uint32_t Builder::DeclareConstant(spv::Op opcode, std::vector<uint32_t> key) {
 	if (const auto it = m_declaration_ids.find(key); it != m_declaration_ids.end()) {
 		return it->second;
 	}
 	const auto id = AllocateId();
+	AppendInstruction(m_declarations, opcode, key[1], id,
+	                  std::span<const uint32_t>(key).subspan(2));
 	m_declaration_ids.emplace(std::move(key), id);
-	std::vector<uint32_t> words {opcode, type, id};
-	words.insert(words.end(), operands.begin(), operands.end());
-	AppendInstructionWords(m_declarations, words.data(), words.size());
 	return id;
 }
 
-uint32_t Builder::DefineGlobalVariable(uint32_t pointer_type, uint32_t storage_class) {
+uint32_t Builder::DefineGlobalVariable(uint32_t pointer_type, spv::StorageClass storage_class) {
 	const auto id = AllocateId();
 	DefineGlobalVariable(id, pointer_type, storage_class);
 	return id;
 }
 
-void Builder::DefineGlobalVariable(uint32_t id, uint32_t pointer_type, uint32_t storage_class) {
-	AddType({59u, pointer_type, id, storage_class});
+void Builder::DefineGlobalVariable(uint32_t id, uint32_t pointer_type,
+                                   spv::StorageClass storage_class) {
+	AppendInstruction(m_declarations, spv::OpVariable, pointer_type, id, storage_class);
 }
 
 void Builder::AppendString(std::vector<uint32_t>& words, const char* text) {
@@ -156,49 +135,20 @@ void Builder::AppendString(std::vector<uint32_t>& words, const char* text) {
 	}
 }
 
-void Builder::AppendInstruction(std::vector<uint32_t>& section, uint32_t opcode,
-                                const std::vector<uint32_t>& operands) {
-	const uint32_t word_count = static_cast<uint32_t>(operands.size() + 1u);
-	section.push_back((word_count << 16u) | opcode);
-	section.insert(section.end(), operands.begin(), operands.end());
+void Builder::AddMemoryModel(spv::AddressingModel addressing_model, spv::MemoryModel memory_model) {
+	AppendInstruction(m_memory_model, spv::OpMemoryModel, addressing_model, memory_model);
 }
 
-void Builder::AppendInstruction(std::vector<uint32_t>& section, uint32_t opcode,
-                                std::initializer_list<uint32_t> operands) {
-	const uint32_t word_count = static_cast<uint32_t>(operands.size() + 1u);
-	section.push_back((word_count << 16u) | opcode);
-	section.insert(section.end(), operands.begin(), operands.end());
-}
-
-void Builder::AddCapability(std::initializer_list<uint32_t> operands) {
-	AppendInstruction(m_capabilities, 17u, operands);
-}
-
-void Builder::AddExtension(const char* name) {
+void Builder::AddEntryPoint(spv::ExecutionModel execution_model, uint32_t entry_point,
+                            const char* name, const std::vector<uint32_t>& interfaces) {
 	std::vector<uint32_t> operands;
-	AppendString(operands, name);
-	AppendInstruction(m_extensions, 10u, operands);
-}
-
-void Builder::AddExtInstImport(uint32_t id, const char* name) {
-	std::vector<uint32_t> operands = {id};
-	AppendString(operands, name);
-	AppendInstruction(m_ext_inst_imports, 11u, operands);
-}
-
-void Builder::AddMemoryModel(std::initializer_list<uint32_t> operands) {
-	AppendInstruction(m_memory_model, 14u, operands);
-}
-
-void Builder::AddEntryPoint(uint32_t execution_model, uint32_t entry_point, const char* name,
-                            const std::vector<uint32_t>& interfaces) {
-	std::vector<uint32_t> operands = {execution_model, entry_point};
+	AppendOperands(operands, execution_model, entry_point);
 	AppendString(operands, name);
 	operands.insert(operands.end(), interfaces.begin(), interfaces.end());
 	if (m_version >= 0x00010400u) {
 		for (size_t offset = 0; offset < m_declarations.size();) {
-			const auto count = m_declarations[offset] >> 16u;
-			if ((m_declarations[offset] & 0xffffu) == 59u) {
+			const auto count = m_declarations[offset] >> spv::WordCountShift;
+			if ((m_declarations[offset] & spv::OpCodeMask) == spv::OpVariable) {
 				const auto id = m_declarations[offset + 2u];
 				if (std::find(interfaces.begin(), interfaces.end(), id) == interfaces.end()) {
 					operands.push_back(id);
@@ -207,46 +157,32 @@ void Builder::AddEntryPoint(uint32_t execution_model, uint32_t entry_point, cons
 			offset += count;
 		}
 	}
-	AppendInstruction(m_entry_points, 15u, operands);
-}
-
-void Builder::AddExecutionMode(std::initializer_list<uint32_t> operands) {
-	AppendInstruction(m_execution_modes, 16u, operands);
+	AppendInstruction(m_entry_points, spv::OpEntryPoint, operands);
 }
 
 void Builder::AddName(uint32_t target, const char* name) {
 	std::vector<uint32_t> operands = {target};
 	AppendString(operands, name);
-	AppendInstruction(m_debug, 5u, operands);
+	AppendInstruction(m_debug, spv::OpName, operands);
 }
 
-void Builder::AddAnnotation(std::initializer_list<uint32_t> words) {
-	AppendInstructionWords(m_annotations, words.begin(), words.size());
-}
-
-void Builder::AddType(std::initializer_list<uint32_t> words) {
-	AppendInstructionWords(m_declarations, words.begin(), words.size());
-}
-
-void Builder::AddFunction(std::initializer_list<uint32_t> words) {
-	AppendInstructionWords(m_functions, words.begin(), words.size());
-}
-
-void Builder::AddFunction(const std::vector<uint32_t>& words) {
+void Builder::AddFunction(std::span<const uint32_t> words) {
 	AppendInstructionWords(m_functions, words.data(), words.size());
 }
 
 DeferredPhi Builder::AddDeferredPhi(uint32_t type, uint32_t result, size_t incoming_count) {
-	std::vector<uint32_t> words {245u, type, result};
+	std::vector<uint32_t> words {spv::OpPhi, type, result};
 	words.resize(words.size() + incoming_count * 2u);
-	const DeferredPhi phi {m_functions.size(), incoming_count};
+	const DeferredPhi phi {m_functions.size()};
 	AddFunction(words);
 	m_unpatched_phi_incomings += incoming_count;
 	return phi;
 }
 
 void Builder::PatchDeferredPhi(DeferredPhi phi, size_t incoming, uint32_t value, uint32_t parent) {
-	EXIT_IF(incoming >= phi.incoming_count || value == 0 || parent == 0);
+	const auto incoming_count =
+	    ((m_functions.at(phi.word_offset) >> spv::WordCountShift) - 3u) / 2u;
+	EXIT_IF(incoming >= incoming_count || value == 0 || parent == 0);
 	const auto value_word  = phi.word_offset + 3u + incoming * 2u;
 	const auto parent_word = value_word + 1u;
 	EXIT_IF(m_functions.at(value_word) != 0 || m_functions.at(parent_word) != 0);
@@ -264,7 +200,7 @@ std::vector<uint32_t> Builder::Build() const {
 	               m_debug.size() + m_annotations.size() + m_declarations.size() +
 	               m_functions.size());
 
-	module.push_back(0x07230203u);
+	module.push_back(spv::MagicNumber);
 	module.push_back(m_version);
 	module.push_back(0u);
 	module.push_back(m_next_id);