shader: fix incorrect per-lane VCCZ/EXECZ reads

* Fix opening mount root directories

* Allow empty APR path prefixes

---------

Co-authored-by: nmzik <Nmzik@mail.ru>

CMakeLists.txt

@@ -126,7 +126,7 @@ if (KYTY_LINKER STREQUAL LD AND NOT LINUX)
 	set(KYTY_LD_OPTIONS "-Wl,--image-base=0x100000000000")
 endif()
 
-project(Kyty${KYTY_PROJECT_NAME}${CMAKE_BUILD_TYPE}${KYTY_COMPILER} VERSION 0.2.2)
+project(Kyty${KYTY_PROJECT_NAME}${CMAKE_BUILD_TYPE}${KYTY_COMPILER} VERSION 0.3.0)
 
 include_directories(
 	${KYTY_THIRD_PARTY_DIR}/gtest/include
@@ -431,16 +431,15 @@ add_executable(shader_vertex_metadata_tests EXCLUDE_FROM_ALL
 )
 target_include_directories(shader_vertex_metadata_tests PRIVATE ${inc_headers})
 
-add_executable(shader_stage_runtime_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/ShaderStageRuntimeTests.cpp"
+add_executable(resource_materialization_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/ResourceMaterializationTests.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/gpu_format.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/shaderStageRuntime.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/SrtWalker.cpp"
 )
-target_link_libraries(shader_stage_runtime_tests fmt::fmt)
-target_include_directories(shader_stage_runtime_tests PRIVATE ${inc_headers})
+target_link_libraries(resource_materialization_tests fmt::fmt)
+target_include_directories(resource_materialization_tests PRIVATE ${inc_headers})
 
 add_executable(resource_tracking_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/ResourceTrackingTests.cpp"
@@ -542,7 +541,7 @@ if(BUILD_TESTING)
 	add_test(NAME page_manager COMMAND $<TARGET_FILE:page_manager_tests>)
 	add_test(NAME bit_array COMMAND $<TARGET_FILE:bit_array_tests>)
 	add_test(NAME shader_vertex_metadata COMMAND $<TARGET_FILE:shader_vertex_metadata_tests>)
-	add_test(NAME shader_stage_runtime COMMAND $<TARGET_FILE:shader_stage_runtime_tests>)
+	add_test(NAME resource_materialization COMMAND $<TARGET_FILE:resource_materialization_tests>)
 	add_test(NAME resource_tracking COMMAND $<TARGET_FILE:resource_tracking_tests>)
 	add_test(NAME event_queue_lifetime COMMAND $<TARGET_FILE:event_queue_lifetime_tests>)
 	add_test(NAME kernel_file_system COMMAND $<TARGET_FILE:kernel_file_system_tests>)
@@ -568,6 +567,8 @@ 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
@@ -598,7 +599,7 @@ if(BUILD_TESTING)
 		page_manager_tests
 		bit_array_tests
 		shader_vertex_metadata_tests
-		shader_stage_runtime_tests
+		resource_materialization_tests
 		resource_tracking_tests
 		event_queue_lifetime_tests
 		kernel_file_system_tests

src/common/assert.cpp

@@ -59,6 +59,7 @@ int DbgExitHandler(const char* file, int line, fmt::text_style style, std::strin
 }
 
 void DbgExit(int status) {
+	Subsystems::EmergencyShutdownActive();
 	std::fflush(nullptr);
 	std::_Exit(status);
 }

src/common/assert.h

@@ -29,11 +29,7 @@ void DbgExit(int status);
 
 } // namespace Common
 
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS || KYTY_PLATFORM == KYTY_PLATFORM_LINUX
 #define EXIT_HALT() (Common::DbgExit(321), 1)
-#else
-#define EXIT_HALT() (std::_Exit(321), 1)
-#endif
 
 #ifndef KYTY_FINAL
 #define EXIT_IF(x)                                                                                 \

src/common/file.cpp

@@ -69,7 +69,7 @@ File::File(): m_p(std::make_unique<FilePrivate>()) {
 }
 
 File::~File() {
-	EXIT_IF(m_p->f != nullptr);
+	Close();
 }
 
 File::File(const std::filesystem::path& name): m_p(std::make_unique<FilePrivate>()) {

src/common/hostException.cpp

@@ -5,14 +5,17 @@
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #include <windows.h> // IWYU pragma: keep
-#elif defined(__APPLE__)
+#else
+#include <algorithm>
 #include <csignal>
+#include <cstdlib>
+#include <initializer_list>
+#include <unistd.h>
+#if defined(__APPLE__)
 #include <sys/ucontext.h>
 #else
-#include <csignal>
-#include <initializer_list>
 #include <ucontext.h> // IWYU pragma: keep
-#include <unistd.h>
+#endif
 #endif
 
 // IWYU pragma: no_include <errhandlingapi.h>
@@ -32,6 +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 {
@@ -253,7 +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
@@ -274,8 +322,7 @@ bool InstallHandler(Handler handler) {
 	struct sigaction action {};
 	action.sa_sigaction = SignalHandler;
 	sigemptyset(&action.sa_mask);
-	// Fault resolution needs the normal thread stack.
-	action.sa_flags = SA_SIGINFO | SA_RESTART;
+	action.sa_flags = SA_SIGINFO | SA_RESTART | SA_ONSTACK;
 
 	for (const int signal_number: {SIGSEGV, SIGBUS, SIGILL}) {
 		if (::sigaction(signal_number, &action, nullptr) != 0) {

src/common/hostException.h

@@ -39,6 +39,10 @@ using Handler = bool (*)(const ExceptionInfo&);
 
 bool InstallHandler(Handler handler);
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_LINUX
+bool InitializeThreadSignalStack();
+#endif
+
 } // namespace Common::HostException
 
 #endif /* KYTY_COMMON_HOST_EXCEPTION_H_ */

src/common/stringUtils.h

@@ -316,10 +316,12 @@ inline std::string SafeCsv(std::string_view text) {
 	return std::string(text);
 }
 
-inline std::string Utf16ToUtf8(const char16_t* utf16) {
-	std::u16string                                                    input(utf16);
+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(input);
+	return convert.to_bytes(utf16.data(), utf16.data() + utf16.size());
 }
 
 inline ByteBuffer HexToBin(std::string_view text) {

src/graphics/guest_gpu/command_processor/commandProcessor.h

@@ -73,13 +73,9 @@ public:
 	void SetDrawIndirectArgsBaseAddress(uint64_t draw_indirect_args_base_addr);
 	void SetDispatchIndirectArgsBaseAddress(uint64_t dispatch_indirect_args_base_addr);
 	void SetNumInstances(uint32_t num_instances);
-	void DrawIndex(uint32_t index_count, const void* index_addr, uint32_t flags, uint32_t type,
-	               uint32_t instance_count = 0, const void* object_ids = nullptr,
-	               uint32_t render_target_slice_offset = 0, int32_t vertex_offset_add = 0,
-	               uint32_t first_instance = 0);
-	void DrawIndexOffset(uint32_t index_offset, uint32_t index_count, uint32_t flags);
-	void DrawIndexAuto(uint32_t index_count, uint32_t flags,
-	                   uint32_t render_target_slice_offset = 0);
+	void DrawIndex(DrawIndexArgs args);
+	void DrawIndexOffset(uint32_t index_offset, uint32_t index_count);
+	void DrawIndexAuto(DrawAutoArgs args);
 	void DrawIndirect(uint32_t data_offset, uint32_t draw_initiator, bool indexed);
 	void DrawIndirectMulti(uint32_t data_offset, uint32_t max_count_or_count,
 	                       const volatile uint32_t* count_addr, uint32_t stride_in_bytes,
@@ -149,10 +145,6 @@ private:
 	                      uint32_t interrupt_context_id);
 	void ProcessPm4(Pm4Execution& execution, size_t stop_depth);
 	void SuspendPm4();
-	void SubmitNonIndexedDraw(uint32_t vertex_count, uint32_t flags,
-	                          uint32_t render_target_slice_offset, uint32_t first_vertex,
-	                          uint32_t first_instance);
-
 	CommandScheduler&   GetScheduler() const { return m_renderer.GetCommandScheduler(); }
 	CommandBuffer&      CurrentBuffer() { return GetScheduler().Current(); }
 	void                CheckBuffer() const { GetScheduler().CheckActive(); }

src/graphics/guest_gpu/command_processor/pm4Dispatch.cpp

@@ -24,7 +24,6 @@ constexpr auto MakeContextDispatchTable() {
 	g_hw_ctx_func[Pm4::DB_DEPTH_CLEAR]               = HwCtxSetDepthClear;
 	g_hw_ctx_func[Pm4::PA_SC_SCREEN_SCISSOR_TL]      = HwCtxSetScreenScissor;
 	g_hw_ctx_func[Pm4::DB_STENCIL_INFO]              = HwCtxSetStencilInfo;
-	g_hw_ctx_func[Pm4::DB_SHADING_RATE_ENCODING]     = HwCtxSetDepthShadingRateEncoding;
 	g_hw_ctx_func[Pm4::PA_SU_HARDWARE_SCREEN_OFFSET] = HwCtxSetHardwareScreenOffset;
 	g_hw_ctx_func[Pm4::PA_SC_WINDOW_OFFSET]          = HwCtxSetWindowOffset;
 	g_hw_ctx_func[Pm4::PA_SC_WINDOW_SCISSOR_TL]      = HwCtxSetWindowScissor;
@@ -140,13 +139,13 @@ constexpr auto MakeShaderDispatchTable() {
 	g_hw_sh_func[Pm4::COMPUTE_TMPRING_SIZE]            = HwShSetCsRegisters;
 	g_hw_sh_func[Pm4::COMPUTE_PGM_RSRC3]               = HwShSetCsRegisters;
 	g_hw_sh_func[Pm4::COMPUTE_PACE_ID]                 = HwShSetCsRegisters;
-	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_PS]  = HwShIgnoreRegisters;
-	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS]  = HwShIgnoreRegisters;
-	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS] = HwShIgnoreRegisters;
-	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_GS] = HwShIgnoreRegisters;
-	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_HS]  = HwShIgnoreRegisters;
-	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS] = HwShIgnoreRegisters;
-	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS] = HwShIgnoreRegisters;
+	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_PS]  = HwShSetRegisters;
+	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS]  = HwShSetRegisters;
+	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS] = HwShSetRegisters;
+	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_GS] = HwShSetRegisters;
+	g_hw_sh_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_HS]  = HwShSetRegisters;
+	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS] = HwShSetRegisters;
+	g_hw_sh_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS] = HwShSetRegisters;
 
 	return g_hw_sh_func;
 }

src/graphics/guest_gpu/command_processor/pm4Dispatch.h

@@ -45,8 +45,6 @@ uint32_t HwCtxSetStencilClear(CommandProcessor&, uint32_t, uint32_t, const uint3
 uint32_t HwCtxSetDepthClear(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetScreenScissor(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetStencilInfo(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
-uint32_t HwCtxSetDepthShadingRateEncoding(CommandProcessor&, uint32_t, uint32_t,
-                                          const uint32_t*, uint32_t);
 uint32_t HwCtxSetHardwareScreenOffset(CommandProcessor&, uint32_t, uint32_t, const uint32_t*,
                                       uint32_t);
 uint32_t HwCtxSetWindowOffset(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
@@ -100,7 +98,7 @@ uint32_t HwShSetCsUserSgpr(CommandProcessor&, uint32_t, uint32_t, const uint32_t
 uint32_t HwShIgnoreUserAccumulator(CommandProcessor&, uint32_t, uint32_t, const uint32_t*,
                                    uint32_t);
 uint32_t HwShSetCsRegisters(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
-uint32_t HwShIgnoreRegisters(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
+uint32_t HwShSetRegisters(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwUcSetPrimitiveType(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwUcSetIndexType(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwUcSetObjectId(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -599,17 +599,6 @@ static void HwCtxSetDepthShadingRateEncodingRegister(CommandProcessor& cp, uint3
 	ctx.SetDepthRenderTarget(target);
 }
 
-KYTY_HW_CTX_PARSER(HwCtxSetDepthShadingRateEncoding) {
-	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
-
-	EXIT_NOT_IMPLEMENTED(num_values != 1);
-	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_SHADING_RATE_ENCODING);
-
-	HwCtxSetDepthShadingRateEncodingRegister(cp, buffer[0]);
-
-	return num_values;
-}
-
 static HW::EqaaControl ParseEqaaControl(uint32_t value) {
 	HW::EqaaControl r;
 
@@ -991,7 +980,7 @@ static void HwShIgnoreComputeRegister([[maybe_unused]] uint32_t cmd_offset,
 static void HwShIgnoreShaderRegister([[maybe_unused]] uint32_t cmd_offset,
                                      [[maybe_unused]] uint32_t value) {}
 
-KYTY_HW_SH_PARSER(HwShIgnoreRegisters) {
+KYTY_HW_SH_PARSER(HwShSetRegisters) {
 	const auto reg_num = (cmd_id >> 16u) & 0x3fffu;
 	EXIT_NOT_IMPLEMENTED(reg_num == 0);
 
@@ -1618,15 +1607,15 @@ KYTY_CP_OP_PARSER(CpOpDrawIndex) {
 		auto*    index_addr =
 		    reinterpret_cast<void*>(buffer[1] | (static_cast<uint64_t>(buffer[2]) << 32u));
 		uint32_t max_instance_count = buffer[3];
-		auto*    object_ids =
-		    reinterpret_cast<void*>(buffer[4] | (static_cast<uint64_t>(buffer[5]) << 32u));
-		uint32_t instance_count = buffer[6];
-		uint32_t flags          = buffer[7];
+		uint32_t instance_count     = buffer[6];
+		uint32_t flags              = buffer[7];
 
 		EXIT_NOT_IMPLEMENTED(instance_count > max_instance_count);
 		EXIT_NOT_IMPLEMENTED((flags & ~0xa0u) != 0);
 
-		cp.DrawIndex(index_count, index_addr, 0, 1, instance_count, object_ids);
+		cp.DrawIndex({.index_count    = index_count,
+		              .index_addr     = index_addr,
+		              .instance_count = instance_count});
 
 		return 8;
 	}
@@ -1641,7 +1630,7 @@ KYTY_CP_OP_PARSER(CpOpDrawIndex) {
 		EXIT_NOT_IMPLEMENTED(index_count > max_index_count);
 		EXIT_NOT_IMPLEMENTED((flags & ~0x20u) != 0);
 
-		cp.DrawIndex(index_count, index_addr, 0, 1);
+		cp.DrawIndex({.index_count = index_count, .index_addr = index_addr});
 
 		return 5;
 	}
@@ -1700,7 +1689,7 @@ KYTY_CP_OP_PARSER(CpOpDrawIndexOffset) {
 	EXIT_NOT_IMPLEMENTED(index_count > max_index_size);
 	EXIT_NOT_IMPLEMENTED((flags & ~0x20u) != 0);
 
-	cp.DrawIndexOffset(index_offset, index_count, 0);
+	cp.DrawIndexOffset(index_offset, index_count);
 
 	return 4;
 }
@@ -1715,7 +1704,7 @@ KYTY_CP_OP_PARSER(CpOpDrawIndexAuto) {
 
 	EXIT_NOT_IMPLEMENTED((flags & ~0x22u) != 0);
 
-	cp.DrawIndexAuto(index_count, 0);
+	cp.DrawIndexAuto({.vertex_count = index_count});
 
 	return 2;
 }
@@ -3640,12 +3629,13 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
-	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HwShIgnoreShaderRegister(cmd_offset, value);
-	};
-	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HwShIgnoreShaderRegister(cmd_offset, value);
-	};
+	for (uint32_t offset = Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS;
+	     offset <= Pm4::SPI_SHADER_USER_DATA_ADDR_HI_GS; offset++) {
+		g_hw_sh_indirect_func[offset] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+			cp.GetShCtx().SetGsUserDataAddress(cmd_offset - Pm4::SPI_SHADER_USER_DATA_ADDR_LO_GS,
+			                                   value);
+		};
+	}
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_CHKSUM_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};

src/graphics/guest_gpu/gpu_format.cpp

@@ -15,6 +15,7 @@ struct FormatInfo {
 	uint32_t           render_target_bytes_per_element;
 	bool               sampled_texture;
 	bool               uint_texture;
+	bool               sint_texture = false;
 };
 
 constexpr FormatInfo kFormatInfo[] = {
@@ -24,19 +25,19 @@ constexpr FormatInfo kFormatInfo[] = {
 	{BufferFormat::k16UNorm, 2, 0, 2, true, false},
 	{BufferFormat::k16SNorm, 2, 0, 2, true, false},
 	{BufferFormat::k16UInt, 2, 0, 2, true, true},
-	{BufferFormat::k16SInt, 2, 0, 2, false, false},
+	{BufferFormat::k16SInt, 2, 0, 2, true, false, true},
 	{BufferFormat::k16Float, 2, 0, 2, true, false},
 	{BufferFormat::k8_8UNorm, 2, 0, 2, true, false},
 	{BufferFormat::k8_8SNorm, 2, 0, 2, true, false},
 	{BufferFormat::k8_8UInt, 2, 0, 2, true, true},
-	{BufferFormat::k8_8SInt, 2, 0, 2, false, false},
+	{BufferFormat::k8_8SInt, 2, 0, 2, true, false, true},
 	{BufferFormat::k32UInt, 4, 0, 4, true, true},
-	{BufferFormat::k32SInt, 4, 0, 4, false, false},
+	{BufferFormat::k32SInt, 4, 0, 4, true, false, true},
 	{BufferFormat::k32Float, 4, 0, 4, true, false},
 	{BufferFormat::k16_16UNorm, 4, 0, 4, true, false},
 	{BufferFormat::k16_16SNorm, 4, 0, 4, true, false},
 	{BufferFormat::k16_16UInt, 4, 0, 4, true, true},
-	{BufferFormat::k16_16SInt, 4, 0, 4, false, false},
+	{BufferFormat::k16_16SInt, 4, 0, 4, true, false, true},
 	{BufferFormat::k16_16Float, 4, 0, 4, true, false},
 	{BufferFormat::k11_11_10UInt, 4, 0, 4, true, true},
 	{BufferFormat::k11_11_10Float, 4, 0, 4, true, false},
@@ -45,20 +46,20 @@ constexpr FormatInfo kFormatInfo[] = {
 	{BufferFormat::k8_8_8_8UNorm, 4, 0, 4, true, false},
 	{BufferFormat::k8_8_8_8SNorm, 4, 0, 4, true, false},
 	{BufferFormat::k8_8_8_8UInt, 4, 0, 4, true, true},
-	{BufferFormat::k8_8_8_8SInt, 4, 0, 4, false, false},
+	{BufferFormat::k8_8_8_8SInt, 4, 0, 4, true, false, true},
 	{BufferFormat::k32_32UInt, 8, 0, 8, true, true},
-	{BufferFormat::k32_32SInt, 8, 0, 8, false, false},
+	{BufferFormat::k32_32SInt, 8, 0, 8, true, false, true},
 	{BufferFormat::k32_32Float, 8, 0, 8, true, false},
 	{BufferFormat::k16_16_16_16UNorm, 8, 0, 8, true, false},
 	{BufferFormat::k16_16_16_16SNorm, 8, 0, 8, true, false},
 	{BufferFormat::k16_16_16_16UInt, 8, 0, 8, true, true},
-	{BufferFormat::k16_16_16_16SInt, 8, 0, 8, false, false},
+	{BufferFormat::k16_16_16_16SInt, 8, 0, 8, true, false, true},
 	{BufferFormat::k16_16_16_16Float, 8, 0, 8, true, false},
 	{BufferFormat::k32_32_32UInt, 12, 0, 12, true, true},
-	{BufferFormat::k32_32_32SInt, 12, 0, 12, false, false},
+	{BufferFormat::k32_32_32SInt, 12, 0, 12, true, false, true},
 	{BufferFormat::k32_32_32Float, 12, 0, 12, true, false},
 	{BufferFormat::k32_32_32_32UInt, 16, 0, 16, true, true},
-	{BufferFormat::k32_32_32_32SInt, 16, 0, 16, false, false},
+	{BufferFormat::k32_32_32_32SInt, 16, 0, 16, true, false, true},
 	{BufferFormat::k32_32_32_32Float, 16, 0, 16, true, false},
 	{BufferFormat::k8Srgb, 1, 0, 0, true, false},
 	{BufferFormat::k8_8Srgb, 2, 0, 0, true, false},
@@ -218,14 +219,15 @@ bool IsFmaskTextureFormat(BufferFormat format) {
 	return format >= BufferFormat::kFmask8_S2_F1 && format <= BufferFormat::kFmask64_S16_F8;
 }
 
-bool IsSampledTextureFormat(BufferFormat format) {
+TextureNumericClass SampledTextureNumericClass(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
-	return info != nullptr && info->sampled_texture;
-}
-
-bool IsUintTextureFormat(BufferFormat format) {
-	const auto* info = FindFormatInfo(format);
-	return info != nullptr && info->uint_texture;
+	if (info == nullptr || !info->sampled_texture) {
+		return TextureNumericClass::Unsupported;
+	}
+	if (info->sint_texture) {
+		return TextureNumericClass::Sint;
+	}
+	return info->uint_texture ? TextureNumericClass::Uint : TextureNumericClass::Float;
 }
 
 BufferFormat RemapTextureFormat(BufferFormat format) {

src/graphics/guest_gpu/gpu_format.h

@@ -37,13 +37,14 @@ struct RenderTargetFormatEncoding {
 	}
 };
 
+enum class TextureNumericClass { Unsupported, Float, Uint, Sint };
+
 RenderTargetFormatEncoding ResolveRenderTargetFormat(ChannelLayout layout, ChannelType type);
 uint32_t                   NumBytesPerElement(BufferFormat format);
 uint32_t                   BlockCompressedBytesPerBlock(BufferFormat format);
 uint32_t                   RenderTargetBytesPerElement(BufferFormat format);
 bool                       IsFmaskTextureFormat(BufferFormat format);
-bool                       IsSampledTextureFormat(BufferFormat format);
-bool                       IsUintTextureFormat(BufferFormat format);
+TextureNumericClass        SampledTextureNumericClass(BufferFormat format);
 BufferFormat               RemapTextureFormat(BufferFormat format);
 
 } // namespace Libs::Graphics::Prospero

src/graphics/guest_gpu/graphicsRun.cpp

@@ -13,7 +13,6 @@
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/sync.h"
-#include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window.h"
 #include "graphics/shader/shader.h"
@@ -40,6 +39,21 @@ static thread_local bool              g_gpu_mutex_owned   = false;
 static thread_local bool              g_gpu_thread        = false;
 static thread_local GuestGpu*         g_gpu_state         = nullptr;
 
+struct DrawIndirectArgs {
+	uint32_t vertex_count_per_instance;
+	uint32_t instance_count;
+	uint32_t start_vertex_location;
+	uint32_t start_instance_location;
+};
+
+struct DrawIndexedIndirectArgs {
+	uint32_t index_count_per_instance;
+	uint32_t instance_count;
+	uint32_t start_index_location;
+	uint32_t base_vertex_location;
+	uint32_t start_instance_location;
+};
+
 class GpuMutexLock final {
 public:
 	explicit GpuMutexLock(Common::Mutex& mutex): m_mutex(mutex) {
@@ -546,10 +560,6 @@ void GuestGpu::ThreadRun(void* data) {
 
 bool GuestGpu::Process(Submission& submission) {
 	const bool first_slice = !submission.started;
-	if (first_slice && RenderDocCaptureRequested()) {
-		Common::LockGuard render_lock(m_renderer.GetMutex());
-		RenderDocStartCapture();
-	}
 	auto& cp = GetProcessor(submission.queue_id);
 
 	if (first_slice && submission.reset_processor) {
@@ -847,36 +857,21 @@ void CommandProcessor::SetPredication(uint32_t condition, uint32_t op, uint32_t
 	}
 }
 
-void CommandProcessor::DrawIndex(uint32_t index_count, const void* index_addr, uint32_t flags,
-                                 uint32_t type, uint32_t instance_count, const void* object_ids,
-                                 uint32_t render_target_slice_offset, int32_t vertex_offset_add,
-                                 uint32_t first_instance) {
+void CommandProcessor::DrawIndex(DrawIndexArgs args) {
 	CheckBuffer();
 
-	if (instance_count == 0) {
-		instance_count = m_num_instances;
+	args.index_type_and_size = m_index_type_and_size;
+	if (args.instance_count == 0) {
+		args.instance_count = m_num_instances;
 	}
-	if (object_ids != nullptr) {
-		LOGF("\t draw indexed multi-instanced objectIds = 0x%016" PRIx64 "\n",
-		     reinterpret_cast<uint64_t>(object_ids));
+	if (args.base_vertex != 0 || args.first_instance != 0) {
+		LOGF("\t draw indexed offsets: base_vertex = %" PRId32 ", first_instance = %" PRIu32 "\n",
+		     args.base_vertex, args.first_instance);
 	}
-	if (render_target_slice_offset != 0) {
-		LOGF("\t draw render target slice offset = %" PRIu32 "\n", render_target_slice_offset);
-	}
-	if (vertex_offset_add != 0 || first_instance != 0) {
-		LOGF("\t draw indexed offsets: vertex_offset_add = %" PRId32 ", first_instance = %" PRIu32
-		     "\n",
-		     vertex_offset_add, first_instance);
-	}
-	m_renderer.GetRenderExecutor().DrawIndex(
-	    m_submit_id, CurrentBuffer(), m_index_type_and_size, index_count, index_addr, flags, type,
-	    instance_count, render_target_slice_offset, vertex_offset_add, first_instance);
+	m_renderer.GetRenderExecutor().DrawIndex(m_submit_id, CurrentBuffer(), args);
 }
 
-void CommandProcessor::DrawIndexOffset(uint32_t index_offset, uint32_t index_count,
-                                       uint32_t flags) {
-	CheckBuffer();
-
+void CommandProcessor::DrawIndexOffset(uint32_t index_offset, uint32_t index_count) {
 	uint64_t index_size = 0;
 	switch (m_index_type_and_size) {
 		case 0: index_size = 2; break;
@@ -888,25 +883,10 @@ void CommandProcessor::DrawIndexOffset(uint32_t index_offset, uint32_t index_cou
 	auto* index_addr = reinterpret_cast<const void*>(
 	    m_index_base_addr + static_cast<uint64_t>(index_offset) * index_size);
 
-	m_renderer.GetRenderExecutor().DrawIndex(m_submit_id, CurrentBuffer(), m_index_type_and_size,
-	                                         index_count, index_addr, flags, 1, m_num_instances);
+	DrawIndex({.index_count = index_count, .index_addr = index_addr});
 }
 
 void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiator, bool indexed) {
-	struct DrawIndirectArgs {
-		uint32_t vertex_count_per_instance;
-		uint32_t instance_count;
-		uint32_t start_vertex_location;
-		uint32_t start_instance_location;
-	};
-	struct DrawIndexedIndirectArgs {
-		uint32_t index_count_per_instance;
-		uint32_t instance_count;
-		uint32_t start_index_location;
-		uint32_t base_vertex_location;
-		uint32_t start_instance_location;
-	};
-
 	EXIT_NOT_IMPLEMENTED((draw_initiator & ~0x20u) != 2u);
 	EXIT_NOT_IMPLEMENTED(m_draw_indirect_args_base_addr == 0);
 
@@ -928,8 +908,11 @@ void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiato
 			}
 		}
 		m_num_instances = args.instance_count;
-		SubmitNonIndexedDraw(args.vertex_count_per_instance, 0, 0, args.start_vertex_location,
-		                     args.start_instance_location);
+		DrawIndexAuto({.vertex_count   = args.vertex_count_per_instance,
+		               .instance_count = args.instance_count,
+		               .first_vertex   = args.start_vertex_location,
+		               .first_instance = args.start_instance_location,
+		               .offset_source  = DrawOffsetSource::IndirectArgs});
 		return;
 	}
 
@@ -970,28 +953,18 @@ void CommandProcessor::DrawIndirect(uint32_t data_offset, uint32_t draw_initiato
 	}
 
 	m_num_instances = args.instance_count;
-	DrawIndex(index_count, index_addr, 0, 1, args.instance_count, nullptr, 0,
-	          static_cast<int32_t>(args.base_vertex_location), args.start_instance_location);
+	DrawIndex({.index_count    = index_count,
+	           .index_addr     = index_addr,
+	           .instance_count = args.instance_count,
+	           .base_vertex    = static_cast<int32_t>(args.base_vertex_location),
+	           .first_instance = args.start_instance_location,
+	           .offset_source  = DrawOffsetSource::IndirectArgs});
 }
 
 void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_count_or_count,
                                          const volatile uint32_t* count_addr,
                                          uint32_t stride_in_bytes, uint32_t draw_initiator,
                                          bool indexed) {
-	struct DrawIndirectArgs {
-		uint32_t vertex_count_per_instance;
-		uint32_t instance_count;
-		uint32_t start_vertex_location;
-		uint32_t start_instance_location;
-	};
-	struct DrawIndexedIndirectArgs {
-		uint32_t index_count_per_instance;
-		uint32_t instance_count;
-		uint32_t start_index_location;
-		uint32_t base_vertex_location;
-		uint32_t start_instance_location;
-	};
-
 	EXIT_NOT_IMPLEMENTED((draw_initiator & ~0x20u) != 2u);
 	EXIT_NOT_IMPLEMENTED(m_draw_indirect_args_base_addr == 0);
 
@@ -1028,8 +1001,11 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 				}
 			}
 			m_num_instances = args->instance_count;
-			SubmitNonIndexedDraw(args->vertex_count_per_instance, 0, 0, args->start_vertex_location,
-			                     args->start_instance_location);
+			DrawIndexAuto({.vertex_count   = args->vertex_count_per_instance,
+			               .instance_count = args->instance_count,
+			               .first_vertex   = args->start_vertex_location,
+			               .first_instance = args->start_instance_location,
+			               .offset_source  = DrawOffsetSource::IndirectArgs});
 			continue;
 		}
 
@@ -1071,8 +1047,12 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 		}
 
 		m_num_instances = args->instance_count;
-		DrawIndex(index_count, index_addr, 0, 1, args->instance_count, nullptr, 0,
-		          static_cast<int32_t>(args->base_vertex_location), args->start_instance_location);
+		DrawIndex({.index_count    = index_count,
+		           .index_addr     = index_addr,
+		           .instance_count = args->instance_count,
+		           .base_vertex    = static_cast<int32_t>(args->base_vertex_location),
+		           .first_instance = args->start_instance_location,
+		           .offset_source  = DrawOffsetSource::IndirectArgs});
 	}
 }
 
@@ -1163,19 +1143,13 @@ void CommandProcessor::DispatchIndirect(uint32_t data_offset, uint32_t mode) {
 	DispatchDirect(args->thread_group_x, args->thread_group_y, args->thread_group_z, mode);
 }
 
-void CommandProcessor::DrawIndexAuto(uint32_t index_count, uint32_t flags,
-                                     uint32_t render_target_slice_offset) {
-	SubmitNonIndexedDraw(index_count, flags, render_target_slice_offset, 0, 0);
-}
-
-void CommandProcessor::SubmitNonIndexedDraw(uint32_t vertex_count, uint32_t flags,
-                                            uint32_t render_target_slice_offset,
-                                            uint32_t first_vertex, uint32_t first_instance) {
+void CommandProcessor::DrawIndexAuto(DrawAutoArgs args) {
 	CheckBuffer();
 
-	m_renderer.GetRenderExecutor().DrawAuto(m_submit_id, CurrentBuffer(), vertex_count, flags,
-	                                        render_target_slice_offset, m_num_instances,
-	                                        first_vertex, first_instance);
+	if (args.instance_count == 0) {
+		args.instance_count = m_num_instances;
+	}
+	m_renderer.GetRenderExecutor().DrawAuto(m_submit_id, CurrentBuffer(), args);
 }
 
 void CommandProcessor::WaitFlipDone(uint32_t video_out_handle, uint32_t display_buffer_index) {

src/graphics/guest_gpu/hardwareContext.h

@@ -362,16 +362,16 @@ struct AaConfig {
 struct Viewport {
 	float zmin                                  = 0.0f;
 	float zmax                                  = 0.0f;
-	float xscale                                = 0.0f;
+	float xscale                                = 1.0f;
 	float xoffset                               = 0.0f;
-	float yscale                                = 0.0f;
+	float yscale                                = 1.0f;
 	float yoffset                               = 0.0f;
-	float zscale                                = 0.0f;
+	float zscale                                = 1.0f;
 	float zoffset                               = 0.0f;
 	int   viewport_scissor_left                 = 0;
 	int   viewport_scissor_top                  = 0;
-	int   viewport_scissor_right                = 0;
-	int   viewport_scissor_bottom               = 0;
+	int   viewport_scissor_right                = 16384;
+	int   viewport_scissor_bottom               = 16384;
 	bool  viewport_scissor_window_offset_enable = false;
 };
 
@@ -380,26 +380,26 @@ struct ScreenViewport {
 	uint32_t transform_control                    = 1087;
 	int      screen_scissor_left                  = 0;
 	int      screen_scissor_top                   = 0;
-	int      screen_scissor_right                 = 0;
-	int      screen_scissor_bottom                = 0;
+	int      screen_scissor_right                 = 16384;
+	int      screen_scissor_bottom                = 16384;
 	int      window_scissor_left                  = 0;
 	int      window_scissor_top                   = 0;
-	int      window_scissor_right                 = 0;
-	int      window_scissor_bottom                = 0;
+	int      window_scissor_right                 = 16384;
+	int      window_scissor_bottom                = 16384;
 	bool     window_scissor_window_offset_enable  = false;
 	int      generic_scissor_left                 = 0;
 	int      generic_scissor_top                  = 0;
-	int      generic_scissor_right                = 0;
-	int      generic_scissor_bottom               = 0;
+	int      generic_scissor_right                = 16384;
+	int      generic_scissor_bottom               = 16384;
 	bool     generic_scissor_window_offset_enable = false;
 	int      window_offset_x                      = 0;
 	int      window_offset_y                      = 0;
 	uint32_t hw_offset_x                          = 0;
 	uint32_t hw_offset_y                          = 0;
-	float    guard_band_horz_clip                 = 0.0f;
-	float    guard_band_vert_clip                 = 0.0f;
-	float    guard_band_horz_discard              = 0.0f;
-	float    guard_band_vert_discard              = 0.0f;
+	float    guard_band_horz_clip                 = 1.0f;
+	float    guard_band_vert_clip                 = 1.0f;
+	float    guard_band_horz_discard              = 1.0f;
+	float    guard_band_vert_discard              = 1.0f;
 	uint16_t clip_rect_rule                       = 0xffffu;
 	int      clip_rect_left[4]                    = {};
 	int      clip_rect_top[4]                     = {};
@@ -520,7 +520,8 @@ struct GsShaderResource2 {
 };
 
 struct GsStageRegisters {
-	uint64_t          data_addr = 0;
+	uint64_t          data_addr      = 0;
+	uint64_t          user_data_addr = 0;
 	GsShaderResource1 rsrc1;
 	GsShaderResource2 rsrc2;
 };
@@ -976,6 +977,12 @@ public:
 	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);
+	}
 	void SetGsShaderResource1(const GsShaderResource1& rsrc1) { m_vs.gs_regs.rsrc1 = rsrc1; }
 	void SetGsShaderResource2(const GsShaderResource2& rsrc2) { m_vs.gs_regs.rsrc2 = rsrc2; }
 

src/graphics/guest_gpu/tile.cpp

@@ -1221,6 +1221,31 @@ bool TileGetRenderTargetSize(uint32_t width, uint32_t height, uint32_t pitch,
 	return true;
 }
 
+bool TileGetDccSize(uint32_t width, uint32_t height, uint32_t slices,
+                    uint32_t bytes_per_element, uint32_t levels, Prospero::TileMode tile,
+                    TileSizeAlign& total_size, uint32_t num_fragments_log2) {
+	total_size = {};
+	if (width == 0 || height == 0 || slices == 0 || levels != 1 || num_fragments_log2 != 0 ||
+	    !std::has_single_bit(bytes_per_element) || bytes_per_element > 16 ||
+	    (tile != Prospero::TileMode::kRenderTarget && tile != Prospero::TileMode::kDepth)) {
+		return false;
+	}
+	// Gen5 color metadata uses 4 KiB blocks, with one byte per 256 bytes of color data.
+	// Metadata stays thin for volume surfaces, so every depth slice has its own block raster.
+	const uint32_t coverage_bits = 20u - std::countr_zero(bytes_per_element);
+	const uint32_t block_width   = 1u << ((coverage_bits + 1u) / 2u);
+	const uint32_t block_height  = 1u << (coverage_bits / 2u);
+	const uint64_t blocks_x = (static_cast<uint64_t>(width) + block_width - 1u) / block_width;
+	const uint64_t blocks_y = (static_cast<uint64_t>(height) + block_height - 1u) / block_height;
+	const uint64_t blocks   = blocks_x * blocks_y;
+	if (blocks > UINT32_MAX / 4096u / slices) {
+		return false;
+	}
+	total_size.size  = static_cast<uint32_t>(blocks * slices * 4096u);
+	total_size.align = 4096;
+	return true;
+}
+
 bool TileGetRenderTargetMipLayout(uint32_t width, uint32_t height, uint32_t pitch,
                                   uint32_t bytes_per_element, uint32_t levels,
                                   TileSizeAlign& total_size, TileSizeOffset* level_sizes,

src/graphics/guest_gpu/tile.h

@@ -122,6 +122,9 @@ uint32_t TileGetDepthPitch(uint32_t width, uint32_t bytes_per_element,
 bool     TileGetRenderTargetSize(uint32_t width, uint32_t height, uint32_t pitch,
                                  uint32_t bytes_per_element, TileSizeAlign& total_size,
                                  uint32_t num_fragments_log2 = 0);
+bool     TileGetDccSize(uint32_t width, uint32_t height, uint32_t slices,
+                        uint32_t bytes_per_element, uint32_t levels, Prospero::TileMode tile,
+                        TileSizeAlign& total_size, uint32_t num_fragments_log2 = 0);
 bool     TileGetRenderTargetMipLayout(uint32_t width, uint32_t height, uint32_t pitch,
                                       uint32_t bytes_per_element, uint32_t levels,
                                       TileSizeAlign& total_size, TileSizeOffset* level_sizes,

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;
 
@@ -31,8 +29,12 @@ struct GraphicContext {
 	bool                               memory_budget_ext_enabled             = false;
 	bool                               rt_extensions_enabled                 = false;
 	bool                               compute_subgroup_size_control_enabled = false;
-	bool                               compute_wave64_supported              = false;
 	bool                               sample_rate_shading_enabled           = false;
+	bool                               attachment_feedback_loop_enabled      = false;
+	bool                               provoking_vertex_last_enabled         = false;
+	bool                               supports_block_texel_view              = false;
+	bool                                      mesh_shader_enabled                   = false;
+	vk::PhysicalDeviceMeshShaderPropertiesEXT mesh_shader_properties                = {};
 	uint32_t                           subgroup_size                         = 0;
 	uint32_t                           min_subgroup_size                     = 0;
 	uint32_t                           max_subgroup_size                     = 0;
@@ -80,6 +82,10 @@ struct GraphicContext {
 		return it->second.first;
 	}
 
+	[[nodiscard]] bool SupportsComputeWave64() const noexcept {
+		return subgroup_size == 64u || compute_subgroup_size_control_enabled;
+	}
+
 	[[nodiscard]] vk::DeviceSize StorageMinAlignment() const {
 		const auto alignment = physical_device_properties.limits.minStorageBufferOffsetAlignment;
 		return alignment != 0 ? alignment : 1;
@@ -91,11 +97,8 @@ struct GraphicContext {
 	[[nodiscard]] bool CanReportMemoryUsage() const noexcept { return memory_budget_ext_enabled; }
 	[[nodiscard]] uint64_t GetDeviceMemoryUsage() const;
 	[[nodiscard]] uint64_t GetTotalMemoryBudget() const;
-	void                   CreateBuffer(uint64_t size, VulkanBuffer& buffer);
 	[[nodiscard]] bool     CreateImage(const vk::ImageCreateInfo& info, VulkanImage& image);
 	void                   DeleteImage(VulkanImage& image);
-	void                   MapMemory(VulkanMemory& memory, void*& data);
-	void                   UnmapMemory(VulkanMemory& memory);
 	void                   AppendHardwareRayTracingDeviceExtensions(
 	    const std::vector<vk::ExtensionProperties>& available_extensions,
 	    std::vector<const char*>&                   device_extensions);
@@ -114,18 +117,6 @@ private:
 	    m_image_format_properties;
 };
 
-struct VulkanMemory {
-	vk::MemoryRequirements  requirements       = {};
-	vk::MemoryPropertyFlags property           = {};
-	vk::MemoryPropertyFlags preferred_property = {};
-	vk::DeviceMemory        memory             = nullptr;
-	VmaAllocation           allocation         = nullptr;
-	VmaAllocationInfo       allocation_info    = {};
-	vk::DeviceSize          offset             = 0;
-	uint32_t                type               = 0;
-	uint64_t                unique_id          = 0;
-};
-
 struct VulkanImageState {
 	vk::PipelineStageFlags2 pl_stage    = vk::PipelineStageFlagBits2::eAllCommands;
 	vk::AccessFlags2        access_mask = vk::AccessFlagBits2::eNone;
@@ -139,7 +130,6 @@ 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;
@@ -148,15 +138,10 @@ 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

@@ -23,10 +23,7 @@ void MemoryTracker::ValidateGpuDirtyPages(const RangeSet& dirty, uint64_t vaddr,
 		EXIT("MemoryTracker: invalid dirty-page validation range\n");
 	}
 	for (auto page = vaddr; page < vaddr + size; page += TRACKER_PAGE_SIZE) {
-		bool found = false;
-		dirty.ForEachIntersection(page, TRACKER_PAGE_SIZE,
-		                          [&found](RangeSet::Range) { found = true; });
-		if (!found) {
+		if (!dirty.Intersects(page, TRACKER_PAGE_SIZE)) {
 			EXIT("MemoryTracker: GPU-dirty tracker page has no dirty bytes, operation=%s "
 			     "addr=0x%016" PRIx64 "\n",
 			     operation, page);
@@ -40,9 +37,7 @@ void MemoryTracker::ValidateGpuDirtyOwnership(const RangeSet& dirty, uint64_t va
 	const auto begin = vaddr & ~(TRACKER_PAGE_SIZE - 1);
 	const auto end   = (vaddr + size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
 	for (auto page = begin; page < end; page += TRACKER_PAGE_SIZE) {
-		bool has_dirty_bytes = false;
-		dirty.ForEachIntersection(page, TRACKER_PAGE_SIZE,
-		                          [&has_dirty_bytes](RangeSet::Range) { has_dirty_bytes = true; });
+		const bool has_dirty_bytes = dirty.Intersects(page, TRACKER_PAGE_SIZE);
 		if (IsRegionGpuModified(page, TRACKER_PAGE_SIZE) != has_dirty_bytes) {
 			EXIT("MemoryTracker: tracker and byte ownership disagree, operation=%s "
 			     "addr=0x%016" PRIx64 "\n",
@@ -113,7 +108,8 @@ void MemoryTracker::UnmarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	});
 }
 
-void MemoryTracker::UntrackMemoryImpl(uint64_t vaddr, uint64_t size) {
+void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
+	CheckNotInUploadCallback();
 	std::vector<RegionManager*> managers;
 	managers.reserve((vaddr % TRACKER_REGION_SIZE + size + TRACKER_REGION_SIZE - 1) /
 	                 TRACKER_REGION_SIZE);
@@ -134,12 +130,6 @@ void MemoryTracker::UntrackMemoryImpl(uint64_t vaddr, uint64_t size) {
 	Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
 	});
-	locks.clear();
-}
-
-void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
-	CheckNotInUploadCallback();
-	UntrackMemoryImpl(vaddr, size);
 }
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/memoryTracker.h

@@ -34,7 +34,6 @@ public:
 	void InvalidateRegion(uint64_t vaddr, uint64_t size, Flush&& on_flush) noexcept {
 		static_assert(std::is_invocable_v<Flush&>);
 		CheckNotInUploadCallback();
-		ValidateRange(vaddr, size);
 
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 			const bool should_flush = [&] {
@@ -171,7 +170,6 @@ private:
 	}
 
 	static void    ValidateRange(uint64_t vaddr, uint64_t size);
-	void           UntrackMemoryImpl(uint64_t vaddr, uint64_t size);
 	RegionManager* GetOrCreateRegion(uint64_t index);
 
 	std::unique_ptr<std::atomic<RegionManager*>[]> m_regions;

src/graphics/host_gpu/pageManager.cpp

@@ -371,8 +371,4 @@ template void PageManager::UpdatePageWatchersForRegion<true, false>(uint64_t, Re
 template void PageManager::UpdatePageWatchersForRegion<false, true>(uint64_t, RegionBits&);
 template void PageManager::UpdatePageWatchersForRegion<false, false>(uint64_t, RegionBits&);
 
-void PageManager::OnGpuMap(uint64_t, uint64_t) {}
-
-void PageManager::OnGpuUnmap(uint64_t, uint64_t) {}
-
 } // namespace Libs::Graphics

src/graphics/host_gpu/pageManager.h

@@ -24,8 +24,6 @@ public:
 	void UpdatePageWatchers(uint64_t vaddr, uint64_t size);
 	template <bool track, bool is_read = false>
 	void UpdatePageWatchersForRegion(uint64_t base_addr, RegionBits& mask);
-	void OnGpuMap(uint64_t vaddr, uint64_t size);
-	void OnGpuUnmap(uint64_t vaddr, uint64_t size);
 
 private:
 	struct Impl;

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

@@ -13,7 +13,6 @@
 #include "kernel/memory.h"
 
 #include <algorithm>
-#include <array>
 #include <cinttypes>
 #include <cstring>
 #include <memory>
@@ -193,17 +192,16 @@ void BufferCache::DownloadBufferMemory(std::span<const DownloadCopy> copies) {
 
 BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
                          PageManager& page_manager, TextureCache& texture_cache)
-	: m_graphics(graphics), m_scheduler(scheduler),
-	  m_fault_manager(graphics, scheduler, *this, CACHING_PAGEBITS, CACHING_NUMPAGES),
-	  m_gds_buffer(graphics, scheduler, MemoryUsage::Stream, 0, AllFlags, GdsBufferSize),
-	  m_bda_pagetable_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 0, AllFlags,
-	                         BDA_PAGETABLE_SIZE),
-	  m_memory_tracker(page_manager),
-	  m_staging_buffer(graphics, scheduler, MemoryUsage::Upload, 512 * MiB),
-	  m_stream_buffer(graphics, scheduler, MemoryUsage::Stream, 64 * MiB),
-	  m_download_buffer(graphics, scheduler, MemoryUsage::Download, 32 * MiB),
-	  m_device_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 128 * MiB),
-	  m_texture_cache(texture_cache) {
+    : m_graphics(graphics), m_scheduler(scheduler), m_fault_manager(graphics, scheduler, *this),
+      m_gds_buffer(graphics, scheduler, MemoryUsage::Stream, 0, AllFlags, GdsBufferSize),
+      m_bda_pagetable_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 0, AllFlags,
+                             BDA_PAGETABLE_SIZE),
+      m_memory_tracker(page_manager),
+      m_staging_buffer(graphics, scheduler, MemoryUsage::Upload, 512 * MiB),
+      m_stream_buffer(graphics, scheduler, MemoryUsage::Stream, 64 * MiB),
+      m_download_buffer(graphics, scheduler, MemoryUsage::Download, 32 * MiB),
+      m_device_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 128 * MiB),
+      m_texture_cache(texture_cache) {
 	std::memset(m_gds_buffer.Mapped().data(), 0, static_cast<size_t>(m_gds_buffer.Size()));
 	m_gds_buffer.Flush(0, m_gds_buffer.Size());
 	SetVulkanObjectNameF(m_graphics.device, m_bda_pagetable_buffer.Handle(),
@@ -259,51 +257,36 @@ void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size, bool is_write) {
 }
 
 void BufferCache::ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write) {
-	// CPU invalidation reaches this point only for a GPU-owned tracker page. Resolve the exact
-	// Buffer owner on the GPU thread so the cache index remains single-thread-owned.
 	if (is_write && !IsRegionRegistered(vaddr, size)) {
 		return;
 	}
+	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(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>(
-	    vaddr, size,
+	    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)) {
-			    for (uint64_t copied = 0; copied < range.size;) {
-				    const auto copy_address = range.address + copied;
-				    const auto* owner = m_page_table.Find(copy_address >> PageTable::kPageBits);
-				    if (owner == nullptr || !*owner) {
-					    EXIT("BufferCache: invalidation readback has no buffer owner\n");
-				    }
-				    auto& buffer = m_slot_buffers[*owner];
-				    if (!buffer.IsInBounds(copy_address, 1)) {
-					    EXIT("BufferCache: invalidation readback is outside its buffer owner\n");
-				    }
-				    const auto copy_size = std::min(
-				        range.size - copied, buffer.CpuAddress() + buffer.Size() - copy_address);
-				    copies.push_back(
-				        {&buffer, buffer.Offset(copy_address), copy_address, copy_size});
-				    copied += copy_size;
-			    }
-		    }
+	    [&](uint64_t address, uint64_t bytes) noexcept {
+		    m_gpu_modified_ranges.ForEachIntersection(address, bytes, [&](RangeSet::Range range) {
+			    copies.push_back(
+			        {&buffer, buffer.Offset(range.address), range.address, range.size});
+		    });
 	    });
-	if (copies.empty()) {
-		if (!is_write) {
-			return;
-		}
-		// A preceding read fault can consume the last GPU-owned copy after this write invalidation
-		// has already chosen to flush. Complete the CPU ownership handoff even though this callback
-		// no longer has bytes to download.
-		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
-		return;
+	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);
 	}
-	DownloadBufferMemory(copies);
-	// The enumeration above covered whole dirty pages and every exact interval on them.
-	m_memory_tracker.UnmarkRegionAsGpuModified(vaddr, size);
 	if (is_write) {
 		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
 	}
@@ -326,38 +309,81 @@ BufferId BufferCache::FindBuffer(uint64_t vaddr, uint64_t size) {
 	return CreateBuffer(vaddr, size);
 }
 
-BufferId BufferCache::CreateBuffer(uint64_t vaddr, uint64_t size) {
-	auto& command = m_scheduler.Current();
-	EXIT_IF(command.IsInvalid());
-	auto       begin = vaddr & ~(CACHING_PAGESIZE - 1);
-	auto       end   = (vaddr + size + CACHING_PAGESIZE - 1) & ~(CACHING_PAGESIZE - 1);
-	auto       first = m_buffers.lower_bound(begin);
-	if (first != m_buffers.begin()) {
-		const auto previous = std::prev(first);
-		if (const auto& buffer = m_slot_buffers[previous->second];
-		    buffer.CpuAddress() + buffer.Size() > begin) {
-			first = previous;
+BufferCache::OverlapResult BufferCache::ResolveOverlaps(uint64_t vaddr, uint64_t size) {
+	static constexpr int      StreamLeapThreshold = 16;
+	static constexpr uint64_t StreamLeapSize      = CACHING_PAGESIZE * 128;
+
+	auto       begin      = vaddr;
+	auto       end        = vaddr + size;
+	const auto find_first = [&](uint64_t address) {
+		auto first = m_buffers.lower_bound(address);
+		if (first != m_buffers.begin()) {
+			const auto  previous = std::prev(first);
+			const auto& buffer   = m_slot_buffers[previous->second];
+			if (buffer.CpuAddress() + buffer.Size() > address) {
+				first = previous;
+			}
+		}
+		return first;
+	};
+	auto first           = find_first(begin);
+	auto last            = first;
+	int  stream_score    = 0;
+	bool has_stream_leap = false;
+	for (; last != m_buffers.end() && last->first < end; ++last) {
+		const auto& buffer        = m_slot_buffers[last->second];
+		const auto  buffer_begin  = buffer.CpuAddress();
+		const auto  buffer_end    = buffer_begin + buffer.Size();
+		const bool  expands_left  = buffer_begin < begin;
+		const bool  expands_right = buffer_end > end;
+		begin                     = std::min(begin, buffer_begin);
+		end                       = std::max(end, buffer_end);
+		if (!has_stream_leap && (stream_score += buffer.StreamScore()) > StreamLeapThreshold) {
+			has_stream_leap = true;
+			if (expands_right) {
+				end += std::min(StreamLeapSize, PageTable::kAddressSpaceSize - end);
+			}
+			if (expands_left) {
+				const auto minimum = CACHING_PAGESIZE * 2;
+				if (begin > minimum) {
+					begin -= std::min(StreamLeapSize, begin - minimum);
+				}
+				first = find_first(begin);
+				begin = std::min(begin, first->first);
+			}
 		}
 	}
-	auto last = first;
-	for (; last != m_buffers.end() && last->first < end; ++last) {
-		const auto& buffer = m_slot_buffers[last->second];
-		begin              = std::min(begin, buffer.CpuAddress());
-		end                = std::max(end, buffer.CpuAddress() + buffer.Size());
+	return {first, last, begin, end, has_stream_leap};
+}
+
+void BufferCache::JoinOverlap(BufferId new_id, BufferId overlap_id, bool accumulate_stream_score) {
+	auto& new_buffer = m_slot_buffers[new_id];
+	auto& overlap    = m_slot_buffers[overlap_id];
+	if (accumulate_stream_score) {
+		new_buffer.IncreaseStreamScore(overlap.StreamScore() + 1);
 	}
+	new_buffer.CopyFrom(m_scheduler.Current(), overlap, 0,
+	                    overlap.CpuAddress() - new_buffer.CpuAddress(), overlap.Size());
+	DeleteBuffer(overlap_id);
+}
+
+BufferId BufferCache::CreateBuffer(uint64_t vaddr, uint64_t size) {
+	EXIT_IF(m_scheduler.Current().IsInvalid());
+	const auto end = (vaddr + size + CACHING_PAGESIZE - 1) & ~(CACHING_PAGESIZE - 1);
+	vaddr &= ~(CACHING_PAGESIZE - 1);
+	size               = end - vaddr;
+	const auto overlap = ResolveOverlaps(vaddr, size);
 
 	const auto id = m_slot_buffers.insert(
-	    m_graphics, m_scheduler, MemoryUsage::DeviceLocal, begin,
-	    AllFlags | vk::BufferUsageFlagBits::eShaderDeviceAddress, end - begin);
-	auto&      buffer = m_slot_buffers[id];
+	    m_graphics, m_scheduler, MemoryUsage::DeviceLocal, overlap.begin,
+	    AllFlags | vk::BufferUsageFlagBits::eShaderDeviceAddress, overlap.end - overlap.begin);
+	const auto& buffer = m_slot_buffers[id];
 	SetVulkanObjectNameF(m_graphics.device, buffer.Handle(),
-	                     "Kyty.GameBuffer[guest=0x{:016x} size=0x{:x}]", begin, end - begin);
-	for (auto overlap = first; overlap != last;) {
-		const auto current = overlap++;
-		const auto old_id  = current->second;
-		const auto& old    = m_slot_buffers[old_id];
-		buffer.CopyFrom(command, old, 0, old.CpuAddress() - begin, old.Size());
-		DeleteBuffer(old_id);
+	                     "Kyty.GameBuffer[guest=0x{:016x} size=0x{:x}]", overlap.begin,
+	                     overlap.end - overlap.begin);
+	for (auto it = overlap.first; it != overlap.last;) {
+		const auto old_id = (it++)->second;
+		JoinOverlap(id, old_id, !overlap.has_stream_leap);
 	}
 	Register(id);
 	return id;
@@ -380,7 +406,6 @@ bool BufferCache::SynchronizeBuffer(Buffer& buffer, uint64_t vaddr, uint64_t siz
 		command.EndRendering();
 		const auto native = command.Handle();
 		vk::BufferMemoryBarrier before {};
-		before.sType         = vk::StructureType::eBufferMemoryBarrier;
 		before.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite |
 		                       vk::AccessFlagBits::eTransferRead |
 		                       vk::AccessFlagBits::eTransferWrite;
@@ -475,37 +500,18 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBufferForImage(uint64_t vaddr, u
 	if (!GuestRange {vaddr, size}.Valid()) {
 		EXIT("BufferCache: invalid image source\n");
 	}
-	auto find_owner = [&]() -> Buffer* {
-		const auto* owner = m_page_table.Find(vaddr >> PageTable::kPageBits);
-		if (owner == nullptr || !*owner) {
-			return nullptr;
-		}
+	const auto* owner = m_page_table.Find(vaddr >> PageTable::kPageBits);
+	if (owner != nullptr && *owner) {
 		auto& buffer = m_slot_buffers[*owner];
-		return buffer.IsInBounds(vaddr, size) ? &buffer : nullptr;
-	};
-
-	{
-		const bool cpu_modified            = m_memory_tracker.IsRegionCpuModified(vaddr, size);
-		const bool gpu_modified            = m_memory_tracker.IsRegionGpuModified(vaddr, size);
-		const bool has_dirty_buffer_source = m_gpu_modified_ranges.Intersects(vaddr, size);
-		m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size,
-		                                           "image source");
-
-		auto* owner = find_owner();
-		if (has_dirty_buffer_source && owner == nullptr) {
-			if (!IsRegionRegistered(vaddr, size)) {
-				EXIT("BufferCache: GPU-dirty image source has no native buffer\n");
-			}
-			owner = &m_slot_buffers[FindBuffer(vaddr, size)];
-		}
-		if (owner != nullptr && !cpu_modified && (!gpu_modified || has_dirty_buffer_source)) {
-			TouchBuffer(*owner);
-			return {owner, owner->Offset(vaddr)};
-		}
-		if (has_dirty_buffer_source && owner == nullptr) {
-			EXIT("BufferCache: GPU-dirty image source could not resolve its native owner\n");
+		if (buffer.IsInBounds(vaddr, size)) {
+			TouchBuffer(buffer);
+			(void)SynchronizeBuffer(buffer, vaddr, size, false, false);
+			return {&buffer, buffer.Offset(vaddr)};
 		}
 	}
+	if (IsRegionGpuModified(vaddr, size)) {
+		return ObtainBuffer(vaddr, size, false, false);
+	}
 
 	auto [staging, stage_offset] = m_staging_buffer.Map(size, 16);
 	if (staging == nullptr || (!Libs::LibKernel::Memory::TryReadBacking(vaddr, staging, size) &&
@@ -513,52 +519,7 @@ std::pair<Buffer*, uint64_t> BufferCache::ObtainBufferForImage(uint64_t vaddr, u
 		EXIT("BufferCache: failed to read mapped guest image backing\n");
 	}
 	m_staging_buffer.Commit();
-
-	const bool has_dirty_buffer_source = m_gpu_modified_ranges.Intersects(vaddr, size);
-	auto*      owner                   = find_owner();
-	if (has_dirty_buffer_source && owner == nullptr) {
-		EXIT("BufferCache: GPU-dirty image source lost its native owner\n");
-	}
-	if (owner == nullptr ||
-	    (m_memory_tracker.IsRegionGpuModified(vaddr, size) && !has_dirty_buffer_source)) {
-		return {&m_staging_buffer, stage_offset};
-	}
-
-	TouchBuffer(*owner);
-	std::vector<std::pair<uint64_t, uint64_t>> uploads;
-	m_memory_tracker.ForEachUploadRange(
-	    vaddr, size, false,
-	    [&](uint64_t address, uint64_t upload_size) noexcept {
-		    uploads.emplace_back(address, upload_size);
-	    },
-	    [&]() noexcept {
-		    for (const auto& [address, upload_size]: uploads) {
-			    owner->CopyFrom(m_scheduler.Current(), m_staging_buffer,
-			                    stage_offset + address - vaddr, owner->Offset(address), upload_size,
-			                    vk::AccessFlagBits::eHostWrite);
-		    }
-	    });
-	return {owner, owner->Offset(vaddr)};
-}
-
-void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data) {
-	if (vaddr == 0 || data.empty() || data.size() > UINT64_MAX - vaddr) {
-		EXIT("BufferCache: invalid host DMA write\n");
-	}
-	Libs::LibKernel::Memory::WriteBacking(vaddr, data.data(), data.size());
-
-	const auto end = vaddr + data.size();
-	for (const auto& [address, id]: m_buffers) {
-		auto&      buffer     = m_slot_buffers[id];
-		const auto buffer_end = address + buffer.Size();
-		const auto begin      = std::max(vaddr, address);
-		const auto range_end  = std::min(end, buffer_end);
-		if (begin >= range_end) {
-			continue;
-		}
-		WriteDataBuffer(buffer, begin, data.data() + begin - vaddr, range_end - begin);
-		TouchBuffer(buffer);
-	}
+	return {&m_staging_buffer, stage_offset};
 }
 
 void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool is_gds) {
@@ -576,23 +537,11 @@ void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool
 		EXIT("BufferCache: invalid fill memory address\n");
 	}
 	(void)m_texture_cache.ClearMeta(vaddr);
-	{
-		const auto region = m_texture_cache.QueryRegion(vaddr, size);
-		if (!HasGpuDirtyBytes(vaddr, size) && !region.gpu_image_bytes) {
-			if (region.image_bytes) {
-				m_texture_cache.InvalidateMemory(vaddr, size);
-			}
-			std::array<uint32_t, 4096> values;
-			values.fill(value);
-			const std::span<const uint8_t> bytes {reinterpret_cast<const uint8_t*>(values.data()),
-			                                      sizeof(values)};
-			for (uint64_t offset = 0; offset < size;) {
-				const auto chunk = std::min<uint64_t>(size - offset, bytes.size());
-				WriteHostMemory(vaddr + offset, bytes.first(chunk));
-				offset += chunk;
-			}
-			return;
-		}
+	if (!IsRegionGpuModified(vaddr, size)) {
+		// Access the guest mapping so write faults invalidate cached buffers and images.
+		auto* destination = reinterpret_cast<uint32_t*>(vaddr);
+		std::fill(destination, destination + size / sizeof(uint32_t), value);
+		return;
 	}
 
 	m_texture_cache.InvalidateMemoryFromGPU(vaddr, size);
@@ -615,29 +564,11 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 		     " size=0x%016" PRIx64 " src_gds=%d dst_gds=%d\n",
 		     src_vaddr, dst_vaddr, size, static_cast<int>(src_gds), static_cast<int>(dst_gds));
 	}
-	if (src_memory || dst_memory) {
-		const auto src_region =
-		    src_memory ? m_texture_cache.QueryRegion(src_vaddr, size) : TextureCache::RegionInfo {};
-		const auto dst_region =
-		    dst_memory ? m_texture_cache.QueryRegion(dst_vaddr, size) : TextureCache::RegionInfo {};
-		if (src_memory && dst_memory && !HasGpuDirtyBytes(src_vaddr, size) &&
-		    !HasGpuDirtyBytes(dst_vaddr, size) && !src_region.gpu_image_bytes &&
-		    !dst_region.gpu_image_bytes) {
-			if (dst_region.image_bytes) {
-				m_texture_cache.InvalidateMemory(dst_vaddr, size);
-			}
-			std::array<uint8_t, 64 * 1024> bytes;
-			for (uint64_t offset = 0; offset < size;) {
-				const auto chunk = std::min<uint64_t>(size - offset, bytes.size());
-				if (!Libs::LibKernel::Memory::TryReadBacking(src_vaddr + offset, bytes.data(),
-				                                             chunk)) {
-					EXIT("BufferCache: host DMA source has no direct backing\n");
-				}
-				WriteHostMemory(dst_vaddr + offset, std::span {bytes}.first(chunk));
-				offset += chunk;
-			}
-			return;
-		}
+	if (src_memory && dst_memory && !IsRegionGpuModified(dst_vaddr, size) &&
+	    !IsRegionGpuModified(src_vaddr, size) && !m_texture_cache.FindImageFromRange(src_vaddr, size)) {
+		std::memcpy(reinterpret_cast<void*>(dst_vaddr), reinterpret_cast<const void*>(src_vaddr),
+		            size);
+		return;
 	}
 
 	auto& command = m_scheduler.Current();

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

@@ -74,6 +74,15 @@ public:
 private:
 	friend struct BufferCacheTestAccess;
 
+	using BufferMap = std::map<uint64_t, BufferId>;
+	struct OverlapResult {
+		BufferMap::iterator first;
+		BufferMap::iterator last;
+		uint64_t            begin;
+		uint64_t            end;
+		bool                has_stream_leap;
+	};
+
 	struct DownloadCopy;
 	using PageTable = MultiLevelPageTable<BufferId, CACHING_PAGEBITS, 40, 16>;
 	static_assert(CACHING_PAGESIZE == (uint64_t {1} << PageTable::kPageBits));
@@ -84,6 +93,8 @@ private:
 	[[nodiscard]] static std::pair<uint64_t, uint64_t> DownloadEnvelope(const DownloadCopy& copy);
 	void WriteDataBuffer(Buffer& buffer, uint64_t address, const void* source, uint64_t size);
 	void TouchBuffer(const Buffer& buffer);
+	[[nodiscard]] OverlapResult ResolveOverlaps(uint64_t vaddr, uint64_t size);
+	void JoinOverlap(BufferId new_id, BufferId overlap_id, bool accumulate_stream_score);
 	[[nodiscard]] BufferId CreateBuffer(uint64_t vaddr, uint64_t size);
 	void                   Register(BufferId id);
 	void Unregister(BufferId id);
@@ -96,7 +107,6 @@ private:
 	                                      uint64_t total_size);
 	[[nodiscard]] bool SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uint64_t size);
 	void DownloadBufferMemory(std::span<const DownloadCopy> copies);
-	void WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data);
 	void ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write);
 
 	GraphicContext&                                   m_graphics;
@@ -106,7 +116,7 @@ private:
 	Buffer                                            m_bda_pagetable_buffer;
 	Common::SlotVector<Buffer>                        m_slot_buffers;
 	Common::LeastRecentlyUsedCache<BufferId, uint64_t> m_lru_cache;
-	std::map<uint64_t, BufferId>                      m_buffers;
+	BufferMap                                         m_buffers;
 	PageTable                                         m_page_table;
 	RangeSet                                          m_gpu_modified_ranges;
 	MemoryTracker                                     m_memory_tracker;

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

@@ -23,13 +23,10 @@ constexpr size_t PageFaultAreaSize = MaxPageFaults * sizeof(uint64_t);
 } // namespace
 
 FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler,
-                           BufferCache& buffer_cache, uint32_t caching_pagebits,
-                           uint64_t caching_num_pages)
+                           BufferCache& buffer_cache)
     : m_graphics(graphics), m_scheduler(scheduler), m_buffer_cache(buffer_cache),
-      m_caching_pagesize(uint64_t {1} << caching_pagebits),
-      m_caching_num_pages(caching_num_pages), m_fault_buffer_size(caching_num_pages / 8),
       m_fault_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 0, AllFlags,
-                     m_fault_buffer_size),
+                     BufferCache::CACHING_NUMPAGES / 8),
       m_download_buffer(graphics, scheduler, MemoryUsage::Download, 0, AllFlags,
                         MaxPendingFaults * PageFaultAreaSize) {
 	SetVulkanObjectNameF(m_graphics.device, m_fault_buffer.Handle(), "Fault Buffer");
@@ -39,7 +36,6 @@ FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler
 	    {1, vk::DescriptorType::eStorageBuffer, 1, vk::ShaderStageFlagBits::eCompute, nullptr},
 	};
 	vk::DescriptorSetLayoutCreateInfo layout_info {};
-	layout_info.sType        = vk::StructureType::eDescriptorSetLayoutCreateInfo;
 	layout_info.flags        = vk::DescriptorSetLayoutCreateFlagBits::ePushDescriptorKHR;
 	layout_info.bindingCount = std::size(bindings);
 	layout_info.pBindings    = bindings;
@@ -49,7 +45,6 @@ FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler
 	    "create fault-buffer descriptor layout");
 
 	vk::ShaderModuleCreateInfo module_info {};
-	module_info.sType    = vk::StructureType::eShaderModuleCreateInfo;
 	module_info.codeSize = std::size(FAULT_BUFFER_PROCESS_SPV) * sizeof(uint32_t);
 	module_info.pCode    = FAULT_BUFFER_PROCESS_SPV;
 	vk::ShaderModule module = nullptr;
@@ -57,7 +52,6 @@ FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler
 	                     "create fault-buffer shader module");
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
-	pipeline_layout_info.sType          = vk::StructureType::ePipelineLayoutCreateInfo;
 	pipeline_layout_info.setLayoutCount = 1;
 	pipeline_layout_info.pSetLayouts    = &m_fault_process_desc_layout;
 	RequireVulkanSuccess(
@@ -66,12 +60,10 @@ FaultManager::FaultManager(GraphicContext& graphics, CommandScheduler& scheduler
 	    "create fault-buffer pipeline layout");
 
 	vk::PipelineShaderStageCreateInfo stage {};
-	stage.sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
 	stage.stage  = vk::ShaderStageFlagBits::eCompute;
 	stage.module = module;
 	stage.pName  = "main";
 	vk::ComputePipelineCreateInfo pipeline_info {};
-	pipeline_info.sType  = vk::StructureType::eComputePipelineCreateInfo;
 	pipeline_info.stage  = stage;
 	pipeline_info.layout = m_fault_process_pipeline_layout;
 	const auto result = m_graphics.device.createComputePipelines(
@@ -99,14 +91,13 @@ void FaultManager::ProcessFaultBuffer() {
 	m_download_buffer.Flush(offset, PageFaultAreaSize);
 
 	vk::BufferMemoryBarrier2 pre_barrier {};
-	pre_barrier.sType         = vk::StructureType::eBufferMemoryBarrier2;
 	pre_barrier.srcStageMask  = vk::PipelineStageFlagBits2::eAllCommands;
 	pre_barrier.srcAccessMask = vk::AccessFlagBits2::eShaderWrite;
 	pre_barrier.dstStageMask  = vk::PipelineStageFlagBits2::eComputeShader;
 	pre_barrier.dstAccessMask = vk::AccessFlagBits2::eShaderRead;
 	pre_barrier.buffer        = m_fault_buffer.Handle();
 	pre_barrier.offset        = 0;
-	pre_barrier.size          = m_fault_buffer_size;
+	pre_barrier.size           = m_fault_buffer.Size();
 	auto post_barrier         = pre_barrier;
 	post_barrier.srcStageMask  = vk::PipelineStageFlagBits2::eComputeShader;
 	post_barrier.srcAccessMask = vk::AccessFlagBits2::eShaderWrite;
@@ -114,12 +105,11 @@ void FaultManager::ProcessFaultBuffer() {
 	post_barrier.dstAccessMask = vk::AccessFlagBits2::eShaderWrite;
 
 	const vk::DescriptorBufferInfo infos[] {
-	    {m_fault_buffer.Handle(), 0, m_fault_buffer_size},
+	    {m_fault_buffer.Handle(), 0, m_fault_buffer.Size()},
 	    {m_download_buffer.Handle(), offset, PageFaultAreaSize},
 	};
 	std::array<vk::WriteDescriptorSet, 2> writes {};
 	for (uint32_t index = 0; index < writes.size(); ++index) {
-		writes[index].sType           = vk::StructureType::eWriteDescriptorSet;
 		writes[index].dstBinding      = index;
 		writes[index].descriptorCount = 1;
 		writes[index].descriptorType  = vk::DescriptorType::eStorageBuffer;
@@ -129,7 +119,6 @@ void FaultManager::ProcessFaultBuffer() {
 	m_scheduler.EndRendering();
 	auto command = m_scheduler.Current().Handle();
 	vk::DependencyInfo dependency {};
-	dependency.sType                    = vk::StructureType::eDependencyInfo;
 	dependency.dependencyFlags          = vk::DependencyFlagBits::eByRegion;
 	dependency.bufferMemoryBarrierCount = 1;
 	dependency.pBufferMemoryBarriers    = &pre_barrier;
@@ -137,7 +126,7 @@ void FaultManager::ProcessFaultBuffer() {
 	command.bindPipeline(vk::PipelineBindPoint::eCompute, m_fault_process_pipeline);
 	command.pushDescriptorSetKHR(vk::PipelineBindPoint::eCompute,
 	                             m_fault_process_pipeline_layout, 0, writes);
-	const auto num_threads    = m_caching_num_pages / 32;
+	const auto num_threads    = BufferCache::CACHING_NUMPAGES / 32;
 	const auto num_workgroups = (num_threads + 63) / 64;
 	command.dispatch(static_cast<uint32_t>(num_workgroups), 1, 1);
 	dependency.pBufferMemoryBarriers = &post_barrier;
@@ -146,14 +135,14 @@ void FaultManager::ProcessFaultBuffer() {
 	const auto area = m_current_area;
 	m_scheduler.DeferOperation([this, mapped, offset, area] {
 		m_download_buffer.Invalidate(offset, PageFaultAreaSize);
-		m_fault_ranges.Clear();
+		RangeSet    fault_ranges;
 		const auto* faults = std::bit_cast<const uint64_t*>(mapped);
 		const auto  count  = static_cast<uint32_t>(faults[0]);
 		for (uint32_t index = 1; index <= count; ++index) {
-			m_fault_ranges.Add(faults[index], m_caching_pagesize);
+			fault_ranges.Add(faults[index], BufferCache::CACHING_PAGESIZE);
 			LOGF("Accessed non-GPU cached memory at 0x%016" PRIx64 "\n", faults[index]);
 		}
-		m_fault_ranges.ForEach([this](uint64_t start, uint64_t end) {
+		fault_ranges.ForEach([this](uint64_t start, uint64_t end) {
 			EXIT_IF(end - start > std::numeric_limits<uint32_t>::max());
 			(void)m_buffer_cache.FindBuffer(start, end - start);
 		});

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

@@ -2,7 +2,6 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_CACHE_FAULTMANAGER_H_
 
 #include "common/abi.h"
-#include "graphics/host_gpu/rangeSet.h"
 #include "graphics/host_gpu/renderer/cache/streamBuffer.h"
 
 #include <array>
@@ -16,8 +15,7 @@ class FaultManager {
 	static constexpr size_t MaxPendingFaults = 8;
 
 public:
-	FaultManager(GraphicContext& graphics, CommandScheduler& scheduler, BufferCache& buffer_cache,
-	             uint32_t caching_pagebits, uint64_t caching_num_pages);
+	FaultManager(GraphicContext& graphics, CommandScheduler& scheduler, BufferCache& buffer_cache);
 	~FaultManager();
 	KYTY_CLASS_NO_COPY(FaultManager);
 
@@ -28,10 +26,6 @@ private:
 	GraphicContext&                            m_graphics;
 	CommandScheduler&                          m_scheduler;
 	BufferCache&                               m_buffer_cache;
-	RangeSet                                   m_fault_ranges;
-	uint64_t                                   m_caching_pagesize;
-	uint64_t                                   m_caching_num_pages;
-	uint64_t                                   m_fault_buffer_size;
 	Buffer                                     m_fault_buffer;
 	Buffer                                     m_download_buffer;
 	std::array<uint64_t, MaxPendingFaults>      m_fault_areas {};

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

@@ -12,7 +12,9 @@ GpuResourceManager::GpuResourceManager(GraphicContext& graphics, CommandSchedule
 GpuResourceManager::~GpuResourceManager() = default;
 
 bool GpuResourceManager::HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept {
-	constexpr uint64_t fault_size = 8;
+	// 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;
 	}
@@ -47,7 +49,6 @@ void GpuResourceManager::MapMemory(uint64_t vaddr, uint64_t size) {
 		std::lock_guard lock(m_mapped_ranges_mutex);
 		m_mapped_ranges.Add(vaddr, size);
 	}
-	m_page_manager.OnGpuMap(vaddr, size);
 }
 
 void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
@@ -64,7 +65,6 @@ void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		}
 		m_buffer_cache.InvalidateMemory(vaddr, size);
 		m_texture_cache.UnmapMemory(vaddr, size);
-		m_page_manager.OnGpuUnmap(vaddr, size);
 		std::lock_guard lock(m_mapped_ranges_mutex);
 		m_mapped_ranges.Subtract(vaddr, size);
 	};

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

@@ -256,17 +256,6 @@ private:
 	size_type                              m_inline_size = 0;
 };
 
-// Removes only the requested owner. Other owners in the same page entry remain intact.
-template <typename Container, typename Value>
-[[nodiscard]] bool EraseExact(Container& owners, const Value& owner) {
-	const auto it = std::find(owners.begin(), owners.end(), owner);
-	if (it == owners.end()) {
-		return false;
-	}
-	owners.erase(it);
-	return true;
-}
-
 } // namespace Libs::Graphics
 
 #endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_MULTILEVELPAGETABLE_H_

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

@@ -116,9 +116,6 @@ vk::Sampler SamplerCache::GetSampler(const ShaderSamplerResource& r) {
 		default: EXIT("unknown border color: %d", static_cast<int>(r.BorderColorType()));
 	}
 
-	sampler_info.sType     = vk::StructureType::eSamplerCreateInfo;
-	sampler_info.pNext     = nullptr;
-	sampler_info.flags     = {};
 	sampler_info.magFilter = to_vk_filter(mag_filter);
 	sampler_info.minFilter = to_vk_filter(min_filter);
 	sampler_info.mipmapMode =

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

@@ -4,7 +4,6 @@
 #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>
@@ -64,14 +63,13 @@ constexpr size_t WATCHES_RESERVE_CHUNK   = 0x1000;
 Buffer::Buffer(GraphicContext& graphics, CommandScheduler& scheduler, MemoryUsage usage,
                uint64_t cpu_address, vk::BufferUsageFlags flags, uint64_t size)
     : m_graphics(&graphics), m_scheduler(&scheduler), m_usage(usage), m_cpu_address(cpu_address),
-      m_size(size), m_buffer(std::make_unique<VulkanBuffer>()) {
+      m_size(size) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(graphics.allocator == nullptr || size == 0);
 
 	vk::BufferCreateInfo buffer_info {};
 	buffer_info.size        = size;
 	buffer_info.usage       = flags;
-	buffer_info.sharingMode = vk::SharingMode::eExclusive;
 
 	const bool with_bda = bool(flags & vk::BufferUsageFlagBits::eShaderDeviceAddress);
 	const VmaAllocationCreateFlags bda_flag =
@@ -88,93 +86,76 @@ 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.allocation_info = allocation_result;
-	m_buffer->memory.memory          = allocation_result.deviceMemory;
-	m_buffer->memory.offset          = allocation_result.offset;
-	m_buffer->memory.type            = allocation_result.memoryType;
-	m_buffer->memory.unique_id       = VulkanNextMemoryUniqueId();
-	graphics.device.getBufferMemoryRequirements(m_buffer->buffer, &m_buffer->memory.requirements);
-	if (static_cast<bool>(flags & vk::BufferUsageFlagBits::eShaderDeviceAddress)) {
+	m_buffer = native_buffer;
+	if (with_bda) {
 		vk::BufferDeviceAddressInfo address_info {};
-		address_info.sType  = vk::StructureType::eBufferDeviceAddressInfo;
-		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);
-	m_is_coherent             = (properties & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT) != 0;
+	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;
-}
-
 vk::DeviceAddress Buffer::BufferDeviceAddress() const noexcept {
 	EXIT_IF(m_device_address == 0);
 	return m_device_address;
 }
 
 bool Buffer::IsInBounds(uint64_t address, uint64_t size) const noexcept {
-	return address >= m_cpu_address && size <= m_size && address - m_cpu_address <= m_size - size;
+	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 > m_size || size > m_size - offset);
+	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 > m_size || size > m_size - offset);
-	if (!m_is_coherent && size != 0) {
+	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);
 	}
 }
 
 void Buffer::Invalidate(uint64_t offset, uint64_t size) {
-	EXIT_IF(m_usage != MemoryUsage::Download || offset > m_size || size > m_size - offset);
-	if (!m_is_coherent && size != 0) {
+	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);
 	}
 }
 
 vk::BufferMemoryBarrier Buffer::Barrier(uint64_t offset, uint64_t size, vk::AccessFlags source,
                                         vk::AccessFlags destination) const {
-	if (Handle() == nullptr || size == 0 || offset > m_size || size > m_size - offset) {
+	if (Handle() == nullptr || size == 0 || offset > Size() || size > Size() - offset) {
 		EXIT("Buffer: invalid DMA barrier, handle=%p offset=0x%016" PRIx64 " size=0x%016" PRIx64
 		     " capacity=0x%016" PRIx64 "\n",
-		     static_cast<const void*>(Handle()), offset, size, m_size);
+		     static_cast<const void*>(Handle()), offset, size, Size());
 	}
 	vk::BufferMemoryBarrier barrier {};
-	barrier.sType               = vk::StructureType::eBufferMemoryBarrier;
 	barrier.srcAccessMask       = source;
 	barrier.dstAccessMask       = destination;
 	barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
@@ -189,8 +170,8 @@ void Buffer::CopyFrom(CommandBuffer& command, const Buffer& source, uint64_t sou
                       uint64_t destination_offset, uint64_t size, vk::AccessFlags source_before,
                       vk::AccessFlags destination_before, vk::AccessFlags source_after,
                       vk::AccessFlags destination_after) {
-	if (size == 0 || source_offset > source.m_size || size > source.m_size - source_offset ||
-	    destination_offset > m_size || size > m_size - destination_offset) {
+	if (size == 0 || source_offset > source.Size() || size > source.Size() - source_offset ||
+	    destination_offset > Size() || size > Size() - destination_offset) {
 		EXIT("Buffer: invalid copy range\n");
 	}
 	if (source.Handle() == Handle() && source_offset < destination_offset + size &&
@@ -247,10 +228,8 @@ void Buffer::Fill(uint64_t offset, uint64_t size, uint32_t value) {
 
 StreamBuffer::StreamBuffer(GraphicContext& graphics, CommandScheduler& scheduler, MemoryUsage usage,
                            uint64_t size)
-    : Buffer(graphics, scheduler, usage, 0, AllFlags, size) {
-	ReserveWatches(m_current_watches, WATCHES_INITIAL_RESERVE);
-	ReserveWatches(m_previous_watches, WATCHES_INITIAL_RESERVE);
-}
+    : Buffer(graphics, scheduler, usage, 0, AllFlags, size),
+      m_current_watches(WATCHES_INITIAL_RESERVE), m_previous_watches(WATCHES_INITIAL_RESERVE) {}
 
 bool StreamBuffer::NormalizeReservation(bool coherent, uint64_t atom, uint64_t& size,
                                         uint64_t& alignment) {
@@ -315,10 +294,8 @@ std::pair<uint8_t*, uint64_t> StreamBuffer::Map(uint64_t size, uint64_t alignmen
 }
 
 void StreamBuffer::Commit() {
-	if (!IsCoherent() && Usage() != MemoryUsage::Download && m_mapped_size != 0) {
-		const auto result = vmaFlushAllocation(
-		    Graphics().allocator, NativeBuffer().memory.allocation, m_offset, m_mapped_size);
-		EXIT_NOT_IMPLEMENTED(static_cast<vk::Result>(result) != vk::Result::eSuccess);
+	if (Usage() != MemoryUsage::Download && m_mapped_size != 0) {
+		Flush(m_offset, m_mapped_size);
 	}
 
 	m_offset += m_mapped_size;
@@ -328,7 +305,7 @@ void StreamBuffer::Commit() {
 		return;
 	}
 	if (m_current_watch_cursor + 1 >= m_current_watches.size()) {
-		ReserveWatches(m_current_watches, WATCHES_RESERVE_CHUNK);
+		m_current_watches.resize(m_current_watches.size() + WATCHES_RESERVE_CHUNK);
 	}
 	auto& watch       = m_current_watches[m_current_watch_cursor++];
 	watch.upper_bound = m_offset;
@@ -344,10 +321,6 @@ uint64_t StreamBuffer::Copy(const void* source, uint64_t size, uint64_t alignmen
 	return offset;
 }
 
-void StreamBuffer::ReserveWatches(std::vector<Watch>& watches, size_t grow_size) {
-	watches.resize(watches.size() + grow_size);
-}
-
 bool StreamBuffer::WaitPendingOperations(const std::vector<Watch>& watches,
                                          std::optional<size_t>     invalidation_mark,
                                          uint64_t requested_upper_bound, bool allow_wait,

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]] 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 { return m_is_coherent; }
+	[[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;
@@ -53,6 +53,8 @@ public:
 		return address - m_cpu_address;
 	}
 	[[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);
@@ -68,13 +70,13 @@ public:
 	void Fill(uint64_t offset, uint64_t size, uint32_t value);
 
 	// BufferCache state lives directly on the resource.
-	bool   is_deleted = false;
-	size_t lru_id     = 0;
+	bool   is_deleted   = false;
+	int    stream_score = 0;
+	size_t lru_id       = 0;
 
 protected:
 	[[nodiscard]] GraphicContext&   Graphics() const noexcept { return *m_graphics; }
 	[[nodiscard]] CommandScheduler& Scheduler() const noexcept { return *m_scheduler; }
-	[[nodiscard]] VulkanBuffer&     NativeBuffer() noexcept { return *m_buffer; }
 
 private:
 	[[nodiscard]] vk::BufferMemoryBarrier Barrier(uint64_t offset, uint64_t size,
@@ -85,11 +87,12 @@ private:
 	CommandScheduler*             m_scheduler   = nullptr;
 	MemoryUsage                   m_usage       = MemoryUsage::DeviceLocal;
 	uint64_t                      m_cpu_address = 0;
-	uint64_t                      m_size        = 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;
-	bool                          m_is_coherent = false;
 };
 
 class StreamBuffer final: public Buffer {
@@ -110,7 +113,6 @@ private:
 		uint64_t upper_bound = 0;
 	};
 
-	void                      ReserveWatches(std::vector<Watch>& watches, size_t grow_size);
 	[[nodiscard]] static bool NormalizeReservation(bool coherent, uint64_t atom, uint64_t& size,
 	                                               uint64_t& alignment);
 	[[nodiscard]] bool        WaitPendingOperations(const std::vector<Watch>& watches,

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

@@ -37,12 +37,6 @@ public:
 		BindingType   type = BindingType::Texture;
 	};
 
-	struct RegionInfo {
-		bool image_pages     = false;
-		bool image_bytes     = false;
-		bool gpu_image_bytes = false;
-	};
-
 	TextureCache(GraphicContext& graphics, CommandScheduler& scheduler, PageManager& page_manager,
 	             BufferCache& buffer_cache);
 	~TextureCache();
@@ -66,15 +60,14 @@ public:
 	                                        uint32_t packed_clear);
 	void               InvalidateMemory(uint64_t address, uint64_t size);
 	void               InvalidateMemoryFromGPU(uint64_t address, uint64_t size);
-	[[nodiscard]] RegionInfo QueryRegion(uint64_t address, uint64_t size);
+	[[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 ClearMeta(uint64_t address);
-	// Returns true when registered DCC absorbed the fill and the caller may skip the dispatch.
-	// False may still record PendingDcc state, but the guest dispatch must execute.
-	[[nodiscard]] bool TryConsumeDccFill(uint64_t address, uint64_t size, uint32_t fill_value);
+	// 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);
@@ -83,12 +76,12 @@ public:
 
 private:
 	enum class TransferDirection { Upload, Download };
-	struct ColorTransferPlan;
+	struct TextureTransferPlan;
 	struct DownloadPlan;
 
 	struct MetaDataInfo {
 		// A guest metadata-fill dispatch may initialize DCC before its render target is bound.
-		// PendingDcc retains that exact fill until FindRenderTarget classifies the address,
+		// 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.
@@ -137,16 +130,20 @@ private:
 	[[nodiscard]] ImageId       ResolveDepthOverlap(const ImageInfo& requested, BindingType binding,
 	                                                ImageId cached);
 	[[nodiscard]] ImageId       ExpandImage(const ImageInfo& info, ImageId source);
-	void                        RefreshImage(ImageId id, const ImageDesc& desc);
-	void                        InitializeImage(ImageId id, const ImageDesc& desc);
-	[[nodiscard]] ColorTransferPlan BuildColorTransfer(const Image& image, BindingType binding,
-	                                                   TransferDirection direction) const;
-	[[nodiscard]] DownloadPlan      BuildDownload(const Image& image) const;
-	void UploadImage(Image& image, const ImageDesc& desc, Buffer& source, uint64_t source_offset);
+	void                        RefreshImage(ImageId id);
+	void                        PrepareDccClear(ImageId id, const ImageDesc& desc);
+	void                        InitializeImage(ImageId id);
+	[[nodiscard]] TextureTransferPlan
+	BuildTextureTransfer(const Image& image, BindingType binding, TransferDirection direction) const;
+	[[nodiscard]] DownloadPlan 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 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 PrepareImageCopy(Image& image);
 	void RefreshCopySource(ImageId id);
 	[[nodiscard]] bool CopyD16(Image& destination, Image& source);

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -4,6 +4,7 @@
 #include "common/logging/log.h"
 #include "common/profiler.h"
 #include "graphics/guest_gpu/gpu_defs.h"
+#include "graphics/guest_gpu/gpu_format.h"
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
@@ -21,47 +22,40 @@ namespace Libs::Graphics {
 
 static std::atomic<uint32_t> g_render_color_log_count = 0;
 
-static void ResolveDccClearInfo(RenderColorInfo& info, vk::Format format, bool has_dcc,
-                                uint32_t packed_clear) {
-	// Register-backed DCC clears use the target's packed clear value. Decode one-word guest
-	// formats here; unsupported encodings remain tracked without unsafe materialization.
-	info.metadata_clear_supported =
-	    has_dcc && DecodePackedColorClear(format, packed_clear, info.color_clear_value);
-	switch (format) {
-		case vk::Format::eR8G8B8A8Unorm:
-		case vk::Format::eR8G8B8A8Srgb:
-		case vk::Format::eB8G8R8A8Unorm:
-		case vk::Format::eB8G8R8A8Srgb:
-		case vk::Format::eA2B10G10R10UnormPack32:
-		case vk::Format::eA2R10G10B10UnormPack32:
-			info.metadata_fixed_clear_supported = has_dcc;
-			break;
-		default: info.metadata_fixed_clear_supported = false; break;
+static bool DccAlphaOnMsb(const HW::ColorInfo& info) {
+	switch (info.format) {
+		case Prospero::ChannelLayout::k10_10_10_2:
+		case Prospero::ChannelLayout::k10_10_10_2Float:
+		case Prospero::ChannelLayout::k5_5_5_1: return true;
+		case Prospero::ChannelLayout::k2_10_10_10:
+		case Prospero::ChannelLayout::k1_5_5_5: return false;
+		default: break;
 	}
-	if (!info.metadata_clear_supported) {
-		info.color_clear_value = {};
+	const auto components =
+	    Prospero::ResolveRenderTargetFormat(info.format, info.channel_type).components;
+	if (components == 1) {
+		return info.channel_order != Prospero::ChannelOrder::kStandard;
 	}
+	return components == 3 || info.channel_order == Prospero::ChannelOrder::kStandard ||
+	       info.channel_order == Prospero::ChannelOrder::kAlt;
 }
 
 // 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) {
 		mask = 0x0f;
 	}
 
-	r.target_slot    = rt_slot;
-	r.export_mapping = {};
+	r             = {};
+	r.target_slot = rt_slot;
 
 	if (rt.base.addr == 0 || mask == 0) {
 		if (graphics_debug_dump_enabled()) {
@@ -76,23 +70,6 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer&
 			}
 		}
 
-		// No color output
-		r.type                           = RenderColorType::NoColorOutput;
-		r.desc                           = {};
-		r.base_addr                      = 0;
-		r.image_id                       = {};
-		r.image_view                     = nullptr;
-		r.format                         = vk::Format::eUndefined;
-		r.extent                         = {};
-		r.base_mip_level                 = 0;
-		r.base_array_layer               = 0;
-		r.buffer_size                    = 0;
-		r.samples                        = 1;
-		r.export_mapping                 = {};
-		r.color_clear_enable             = false;
-		r.metadata_clear_supported       = false;
-		r.metadata_fixed_clear_supported = false;
-		r.color_clear_value              = {};
 		return;
 	}
 	const auto samples = render_sample_count(rt.attrib.num_fragments);
@@ -100,50 +77,11 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer&
 		EXIT("unsupported render-target sample configuration: samples=%u fragments=%u\n",
 		     rt.attrib.num_samples, rt.attrib.num_fragments);
 	}
-	const auto view = ResolveTargetViewInfo(
-	    rt.view.base_array_slice_index, rt.view.last_array_slice_index, render_target_slice_offset);
-	switch (view.type) {
-		case TargetViewType::Image2D:
-		case TargetViewType::Image2DArray: break;
-		case TargetViewType::Unsupported:
-			EXIT("invalid render-target view: base=%u last=%u draw_offset=%u\n",
-			     rt.view.base_array_slice_index, rt.view.last_array_slice_index,
-			     render_target_slice_offset);
-	}
-	r.base_array_layer    = view.base_layer;
 	const uint32_t levels = rt.attrib2.num_mip_levels + 1u;
 	if (levels == 0 || levels > 16 || rt.view.current_mip_level >= levels) {
 		EXIT("unsupported render-target mip range: current=%u levels=%u\n",
 		     rt.view.current_mip_level, levels);
 	}
-	if (graphics_debug_dump_enabled()) {
-		static std::atomic_uint log_count = 0;
-		const auto              log_id    = log_count.fetch_add(1, std::memory_order_relaxed);
-		if (log_id < 128) {
-			LOGF("RenderColorTarget: inspect slot=%" PRIu32 " base=0x%010" PRIx64
-			     " mask=0x%01" PRIx32 " attrib2_width=%" PRIu32 " attrib2_height=%" PRIu32
-			     " attrib3_tile=0x%08" PRIx32 " attrib3_dim=0x%08" PRIx32 " fmt=0x%08" PRIx32
-			     " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32 "\n",
-			     rt_slot, rt.base.addr, mask, rt.attrib2.width, rt.attrib2.height,
-			     static_cast<uint32_t>(rt.attrib3.tile_mode), rt.attrib3.dimension,
-			     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
-			     static_cast<uint32_t>(rt.info.channel_order));
-		}
-	}
-
-	// Color-control state selects the color-buffer operation and logical blend operation.
-	// The normal copy operation is a regular color write, not an attachment clear.
-	// Nonlinear clear values are still stored as normalized components.
-	// Fast color clears are metadata driven and must be handled explicitly when
-	// that metadata path is implemented; render-pass load must preserve contents.
-	r.color_clear_enable = false;
-	r.color_clear_value  = {};
-
-	uint32_t   width  = 0;
-	uint32_t   height = 0;
-	uint32_t   pitch  = 0;
-	uint64_t   size   = 0;
-	bool       tile   = false;
 	static constexpr std::array image_types {Prospero::ImageType::kColor1D,
 	                                         Prospero::ImageType::kColor2D,
 	                                         Prospero::ImageType::kColor3D};
@@ -165,7 +103,48 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer&
 	if (volume && samples != 1) {
 		EXIT("multisampled 3D render targets are unsupported\n");
 	}
-	const uint32_t depth        = volume ? rt.attrib3.depth + 1u : 1u;
+	const uint32_t depth = volume ? rt.attrib3.depth + 1u : 1u;
+	// For volumes, CB_COLOR_VIEW bounds exported slices; ATTRIB3 defines storage depth.
+	// The host attachment contains only the selected slices that exist in this mip.
+	const uint32_t last_layer = volume
+	                                ? std::min(rt.view.last_array_slice_index,
+	                                           std::max(depth >> rt.view.current_mip_level, 1u) - 1u)
+	                                : rt.view.last_array_slice_index;
+	const auto view = ResolveTargetViewInfo(
+	    rt.view.base_array_slice_index, last_layer, render_target_slice_offset);
+	switch (view.type) {
+		case TargetViewType::Image2D:
+		case TargetViewType::Image2DArray: break;
+		case TargetViewType::Unsupported:
+			EXIT("invalid render-target view: base=%u last=%u draw_offset=%u\n",
+			     rt.view.base_array_slice_index, rt.view.last_array_slice_index,
+			     render_target_slice_offset);
+	}
+	if (graphics_debug_dump_enabled()) {
+		static std::atomic_uint log_count = 0;
+		const auto              log_id    = log_count.fetch_add(1, std::memory_order_relaxed);
+		if (log_id < 128) {
+			LOGF("RenderColorTarget: inspect slot=%" PRIu32 " base=0x%010" PRIx64
+			     " mask=0x%01" PRIx32 " attrib2_width=%" PRIu32 " attrib2_height=%" PRIu32
+			     " attrib3_tile=0x%08" PRIx32 " attrib3_dim=0x%08" PRIx32 " fmt=0x%08" PRIx32
+			     " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32 "\n",
+			     rt_slot, rt.base.addr, mask, rt.attrib2.width, rt.attrib2.height,
+			     static_cast<uint32_t>(rt.attrib3.tile_mode), rt.attrib3.dimension,
+			     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
+			     static_cast<uint32_t>(rt.info.channel_order));
+		}
+	}
+
+	// Color-control state selects the color-buffer operation and logical blend operation.
+	// The normal copy operation is a regular color write, not an attachment clear.
+	// Nonlinear clear values are still stored as normalized components.
+	// Fast color clears are metadata driven and must be handled explicitly when
+	// that metadata path is implemented; render-pass load must preserve contents.
+	uint32_t   width  = 0;
+	uint32_t   height = 0;
+	uint32_t   pitch  = 0;
+	uint64_t   size   = 0;
+	bool       tile   = false;
 	const bool     standard4    = rt.attrib3.tile_mode == Prospero::TileMode::kStandard4KB;
 	const bool     standard64   = rt.attrib3.tile_mode == Prospero::TileMode::kStandard64KB;
 	const bool     depth_tile   = rt.attrib3.tile_mode == Prospero::TileMode::kDepth;
@@ -224,19 +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) {
-			pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
-		} else if (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] {};
@@ -305,12 +279,6 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer&
 
 	const vk::Extent2D view_extent = {std::max(width >> rt.view.current_mip_level, 1u),
 	                                  std::max(height >> rt.view.current_mip_level, 1u)};
-	const uint32_t     view_depth  = std::max(depth >> rt.view.current_mip_level, 1u);
-	if (volume &&
-	    (view.base_layer >= view_depth || view.layer_count > view_depth - view.base_layer)) {
-		EXIT("3D render-target view exceeds mip depth: base=%u count=%u depth=%u mip=%u\n",
-		     view.base_layer, view.layer_count, view_depth, rt.view.current_mip_level);
-	}
 
 	auto decision_log_id = g_render_color_log_count.fetch_add(1);
 	if (decision_log_id < 128) {
@@ -337,10 +305,14 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer&
 	desc.info.tile_mode       = rt.attrib3.tile_mode;
 	const bool has_dcc        = rt.info.dcc_compression_enable && rt.dcc_addr.addr != 0;
 	if (has_dcc) {
-		// DCC uses a separate metadata allocation. Carry its address through ImageInfo so
-		// TextureCache can associate metadata fills observed before target registration.
-		desc.info.metadata.kind          = ImageMetadataKind::Dcc;
-		desc.info.metadata.range.address = rt.dcc_addr.addr;
+		TileSizeAlign metadata_size {};
+		(void)TileGetDccSize(width, height, volume ? depth : view.image_layers, bytes_per_element,
+		                     levels, rt.attrib3.tile_mode, metadata_size, rt.attrib.num_fragments);
+		desc.info.metadata.kind                     = ImageMetadataKind::Dcc;
+		desc.info.metadata.range                    = {rt.dcc_addr.addr, metadata_size.size};
+		desc.info.metadata.dcc_clear_word           = rt.clear_word0.word0;
+		desc.info.metadata.dcc_clear_register_valid = true;
+		desc.info.metadata.dcc_alpha_msb            = DccAlphaOnMsb(rt.info);
 	}
 	for (uint32_t level = 0; level < levels; level++) {
 		if (volume) {
@@ -382,18 +354,10 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer&
 	desc.view_info.usage       = vk::ImageUsageFlagBits::eColorAttachment;
 	auto& texture_cache        = m_context.GetTextureCache();
 	r.desc                     = std::move(desc);
+	r.guest_mip_level          = rt.view.current_mip_level;
+	r.guest_array_layer        = view.base_layer;
 	r.image_id                 = texture_cache.FindImage(r.desc, exact_format);
-	r.type                     = RenderColorType::RenderTexture;
-	r.base_addr                = rt.base.addr;
-	r.image_view               = nullptr;
-	r.format                   = r.desc.view_info.format;
-	r.extent                   = view_extent;
-	r.base_mip_level           = rt.view.current_mip_level;
-	r.buffer_size              = backing_size;
-	r.samples                  = samples;
 	r.export_mapping           = target_format.export_mapping;
-	r.color_clear_enable       = false;
-	ResolveDccClearInfo(r, target_format.format, has_dcc, rt.clear_word0.word0);
 	BindRenderTarget(r.image_id);
 }
 

src/graphics/host_gpu/renderer/colorRenderTarget.h

@@ -6,33 +6,24 @@
 #include "graphics/host_gpu/renderer/renderTarget.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 
+#include <algorithm>
 #include <cstdint>
 
 namespace Libs::Graphics {
 
-enum class RenderColorType {
-	NoColorOutput,
-	RenderTexture,
-};
-
 struct RenderColorInfo {
-	RenderColorType                 type = RenderColorType::NoColorOutput;
+	// Discovery keeps guest image information but can remap the view into a larger cache image.
 	TextureCache::ImageDesc         desc;
 	ImageId                         image_id;
-	vk::ImageView                   image_view       = nullptr;
-	vk::Format                      format           = vk::Format::eUndefined;
-	vk::Extent2D                    extent           = {};
-	uint32_t                        base_mip_level   = 0;
-	uint32_t                        base_array_layer = 0;
-	uint64_t                        base_addr        = 0;
-	uint64_t                        buffer_size      = 0;
 	uint32_t                        target_slot      = 0;
-	uint32_t                        samples          = 1;
+	uint32_t                        guest_mip_level   = 0;
+	uint32_t                        guest_array_layer = 0;
 	Prospero::ColorComponentMapping export_mapping;
-	bool                            color_clear_enable             = false;
-	bool                            metadata_clear_supported       = false;
-	bool                            metadata_fixed_clear_supported = false;
-	vk::ClearColorValue             color_clear_value {};
+
+	[[nodiscard]] vk::Extent2D Extent() const {
+		return {std::max(desc.info.extent.width >> guest_mip_level, 1u),
+		        std::max(desc.info.extent.height >> guest_mip_level, 1u)};
+	}
 };
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/renderer/commandScheduler.cpp

@@ -19,11 +19,11 @@ void ReportVulkanFatal(const char* what, vk::Result result, uint64_t tick, uint3
                        uint32_t arg3, uint64_t arg4) {
 	LOGF("%s failed: %s (%d), tick=%" PRIu64 " debug_op=%u debug_submit=%" PRIu64
 	     " args=%u,%u,%u,%u,0x%016" PRIx64 "\n",
-	     what, VulkanToString(result).c_str(), static_cast<int>(result), tick, debug_op,
+	     what, vk::to_string(result).c_str(), static_cast<int>(result), tick, debug_op,
 	     debug_submit, arg0, arg1, arg2, arg3, arg4);
 	std::printf("%s failed: %s (%d), tick=%" PRIu64 " debug_op=%u debug_submit=%" PRIu64
 	            " args=%u,%u,%u,%u,0x%016" PRIx64 "\n",
-	            what, VulkanToString(result).c_str(), static_cast<int>(result), tick, debug_op,
+	            what, vk::to_string(result).c_str(), static_cast<int>(result), tick, debug_op,
 	            debug_submit, arg0, arg1, arg2, arg3, arg4);
 	std::fflush(stdout);
 }
@@ -34,7 +34,6 @@ CommandScheduler::CommandPool::CommandPool(GraphicContext& graphics, MasterSemap
     : m_graphics(graphics), m_master(master) {
 	EXIT_IF(graphics.queue_family == static_cast<uint32_t>(-1));
 	vk::CommandPoolCreateInfo create {};
-	create.sType            = vk::StructureType::eCommandPoolCreateInfo;
 	create.queueFamilyIndex = graphics.queue_family;
 	create.flags            = vk::CommandPoolCreateFlagBits::eTransient |
 	                          vk::CommandPoolCreateFlagBits::eResetCommandBuffer;
@@ -52,7 +51,6 @@ size_t CommandScheduler::CommandPool::Grow() {
 	m_buffers.resize(first + GrowStep);
 
 	vk::CommandBufferAllocateInfo allocate {};
-	allocate.sType              = vk::StructureType::eCommandBufferAllocateInfo;
 	allocate.commandPool        = m_pool;
 	allocate.level              = vk::CommandBufferLevel::ePrimary;
 	allocate.commandBufferCount = static_cast<uint32_t>(GrowStep);
@@ -87,10 +85,7 @@ vk::CommandBuffer CommandScheduler::CommandPool::Commit() {
 		m_ticks[*found] = m_master.CurrentTick();
 	}
 
-	m_hint = *found + 1;
-	if (m_hint == m_ticks.size()) {
-		m_hint = 0;
-	}
+	m_hint = (*found + 1) % m_ticks.size();
 	return m_buffers[*found];
 }
 
@@ -153,14 +148,10 @@ void CommandScheduler::Begin(HW::Context& registers, HW::UserConfig& user_config
 		std::lock_guard lock(m_operation_mutex);
 		EXIT_IF(m_operation_state != OperationState::Open);
 	}
-	m_registers   = &registers;
-	m_user_config = &user_config;
-	m_shaders     = &shaders;
+	m_command.Bind(registers, user_config, shaders);
 
 	if (m_command.IsInvalid()) {
 		BeginNext();
-	} else {
-		BindCurrent();
 	}
 }
 
@@ -217,13 +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;
 		{
@@ -351,11 +336,6 @@ CommandBuffer& CommandScheduler::Current() {
 	return m_command;
 }
 
-void CommandScheduler::BindCurrent() {
-	EXIT_IF(m_registers == nullptr || m_user_config == nullptr || m_shaders == nullptr);
-	m_command.Bind(*m_registers, *m_user_config, *m_shaders);
-}
-
 CommandBuffer& CommandScheduler::BeginCommand() {
 	EXIT_IF(!m_command.IsInvalid());
 	m_command.m_buffer = m_command_pool.Commit();
@@ -381,14 +361,12 @@ uint64_t CommandScheduler::Submit(SubmitInfo submit) {
 		submit.AddSignal(m_master.Handle(), tick);
 
 		vk::TimelineSemaphoreSubmitInfo timeline_info {};
-		timeline_info.sType                     = vk::StructureType::eTimelineSemaphoreSubmitInfo;
 		timeline_info.waitSemaphoreValueCount   = submit.num_wait_semaphores;
 		timeline_info.pWaitSemaphoreValues      = submit.wait_ticks.data();
 		timeline_info.signalSemaphoreValueCount = submit.num_signal_semaphores;
 		timeline_info.pSignalSemaphoreValues    = submit.signal_ticks.data();
 
 		vk::SubmitInfo submit_info {};
-		submit_info.sType                = vk::StructureType::eSubmitInfo;
 		submit_info.pNext                = &timeline_info;
 		submit_info.waitSemaphoreCount   = submit.num_wait_semaphores;
 		submit_info.pWaitSemaphores      = submit.wait_semaphores.data();
@@ -414,8 +392,7 @@ uint64_t CommandScheduler::Submit(SubmitInfo submit) {
 }
 
 void CommandScheduler::BeginNext() {
-	EXIT_IF(!m_command.IsInvalid());
-	BindCurrent();
+	CheckActive();
 	BeginCommand();
 }
 

src/graphics/host_gpu/renderer/commandScheduler.h

@@ -42,7 +42,7 @@ public:
 	void                      DeferPriorityOperation(Common::UniqueFunction<void>&& operation);
 	[[nodiscard]] static bool InDeferredOperation() noexcept;
 
-	[[nodiscard]] bool             Active() const noexcept { return m_registers != nullptr; }
+	[[nodiscard]] bool Active() const noexcept { return m_command.m_registers != nullptr; }
 	void                           CheckActive() const;
 	CommandBuffer&                 Current();
 	[[nodiscard]] uint64_t         CurrentTick() const noexcept { return m_master.CurrentTick(); }
@@ -80,9 +80,7 @@ private:
 		uint64_t                     tick = 0;
 	};
 
-	void BindCurrent();
 	void BeginNext();
-	void PopPendingOperations(bool refresh_gpu_tick);
 	void PriorityOperationsThread(std::stop_token stop);
 	void RunOperation(Common::UniqueFunction<void>&& operation);
 
@@ -99,9 +97,6 @@ private:
 	bool                         m_priority_active      = false;
 	uint64_t                     m_priority_active_tick = 0;
 	OperationState               m_operation_state      = OperationState::Open;
-	HW::Context*                 m_registers            = nullptr;
-	HW::UserConfig*              m_user_config          = nullptr;
-	HW::Shader*                  m_shaders              = nullptr;
 };
 
 } // namespace Libs::Graphics

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>
@@ -34,10 +33,7 @@ void CommandBuffer::Begin() {
 	auto buffer = Handle();
 
 	vk::CommandBufferBeginInfo begin_info {};
-	begin_info.sType            = vk::StructureType::eCommandBufferBeginInfo;
-	begin_info.pNext            = nullptr;
-	begin_info.flags            = vk::CommandBufferUsageFlagBits::eOneTimeSubmit;
-	begin_info.pInheritanceInfo = nullptr;
+	begin_info.flags = vk::CommandBufferUsageFlagBits::eOneTimeSubmit;
 
 	auto result = buffer.begin(&begin_info);
 
@@ -75,7 +71,6 @@ void CommandBuffer::BeginRendering(const RenderState& state) const {
 	std::array<vk::RenderingAttachmentInfo, RENDER_COLOR_ATTACHMENTS_MAX> colors {};
 	for (uint32_t i = 0; i < state.num_color_attachments; i++) {
 		const auto& attachment = state.color_attachments[i];
-		colors[i].sType        = vk::StructureType::eRenderingAttachmentInfo;
 		colors[i].imageView    = attachment.image_view;
 		colors[i].imageLayout  = attachment.image_layout;
 		colors[i].loadOp =
@@ -86,7 +81,6 @@ void CommandBuffer::BeginRendering(const RenderState& state) const {
 
 	const auto&                 depth_stencil = state.depth_stencil_attachment;
 	vk::RenderingAttachmentInfo depth {};
-	depth.sType       = vk::StructureType::eRenderingAttachmentInfo;
 	depth.imageView   = depth_stencil.image_view;
 	depth.imageLayout = depth_stencil.image_layout;
 	depth.loadOp =
@@ -95,7 +89,6 @@ void CommandBuffer::BeginRendering(const RenderState& state) const {
 	depth.clearValue.depthStencil.depth = std::bit_cast<float>(depth_stencil.clear_value[0]);
 
 	vk::RenderingAttachmentInfo stencil {};
-	stencil.sType       = vk::StructureType::eRenderingAttachmentInfo;
 	stencil.imageView   = depth_stencil.image_view;
 	stencil.imageLayout = depth_stencil.image_layout;
 	stencil.loadOp =
@@ -104,7 +97,6 @@ void CommandBuffer::BeginRendering(const RenderState& state) const {
 	stencil.clearValue.depthStencil.stencil = depth_stencil.clear_value[1];
 
 	vk::RenderingInfo rendering {};
-	rendering.sType                = vk::StructureType::eRenderingInfo;
 	rendering.renderArea.extent    = {state.width, state.height};
 	rendering.layerCount           = state.num_layers;
 	rendering.colorAttachmentCount = state.num_color_attachments;

src/graphics/host_gpu/renderer/debug.cpp

@@ -17,8 +17,6 @@
 
 namespace Libs::Graphics {
 
-static std::atomic<uint32_t> g_scissor_default_log_count = 0;
-
 uint32_t render_target_mask_slot(uint32_t mask, uint32_t slot) {
 	return (mask >> (slot * 4u)) & 0x0fu;
 }
@@ -387,81 +385,6 @@ static void ZPrint(const char* func, const HW::DepthRenderTarget& z) {
 	     z.size.valid ? "true" : "false");
 }
 
-// NOLINTNEXTLINE(readability-function-cognitive-complexity)
-static void ZCheck(const HW::DepthRenderTarget& z, const HW::DepthControl& dc,
-                   const HW::RenderControl& rc) {
-	const bool depth_active = dc.z_enable || dc.z_write_enable || dc.depth_bounds_enable ||
-	                          rc.depth_clear_enable || rc.copy_depth_to_color;
-	const bool stencil_active =
-	    dc.stencil_enable || rc.stencil_clear_enable || rc.copy_stencil_to_color;
-	if (!depth_active && !stencil_active) {
-		return;
-	}
-
-	EXIT_NOT_IMPLEMENTED(rc.copy_depth_to_color || rc.copy_stencil_to_color);
-	EXIT_NOT_IMPLEMENTED(!z.z_info.HasValidTextureCompatibility());
-	EXIT_NOT_IMPLEMENTED(!z.stencil_info.HasValidTextureCompatibility());
-	if (z.z_info.format == Prospero::DepthFormat::kInvalid) {
-		EXIT_NOT_IMPLEMENTED(z.z_info.format != Prospero::DepthFormat::kInvalid);
-		EXIT_NOT_IMPLEMENTED(z.z_info.num_samples != 0);
-		EXIT_NOT_IMPLEMENTED(z.z_info.htile_acceleration != false);
-		EXIT_NOT_IMPLEMENTED(z.z_info.expclear_enabled != false);
-		EXIT_NOT_IMPLEMENTED(z.z_info.partially_resident != false);
-		EXIT_NOT_IMPLEMENTED(z.z_info.max_mip_level != 0);
-	} else {
-		EXIT_NOT_IMPLEMENTED(z.z_info.format != Prospero::DepthFormat::kZ16 &&
-		                     z.z_info.format != Prospero::DepthFormat::kZ32F);
-		if (z.z_info.num_samples != 0x00000000) {
-			static bool logged = false;
-			if (!logged) {
-				LOGF("DepthTarget: using native num_samples=0x%08" PRIx32 "\n",
-				     z.z_info.num_samples);
-				logged = true;
-			}
-		}
-		EXIT_NOT_IMPLEMENTED(z.z_info.expclear_enabled != false);
-		EXIT_NOT_IMPLEMENTED(z.z_info.partially_resident != false);
-		EXIT_NOT_IMPLEMENTED(z.z_info.max_mip_level != 0);
-	}
-
-	if (z.stencil_info.format == Prospero::StencilFormat::kInvalid) {
-		EXIT_NOT_IMPLEMENTED(z.stencil_info.expclear_enabled != false);
-		EXIT_NOT_IMPLEMENTED(z.stencil_info.partially_resident != false);
-	} else {
-		EXIT_NOT_IMPLEMENTED(z.stencil_info.format != Prospero::StencilFormat::k8UInt);
-		EXIT_NOT_IMPLEMENTED(z.stencil_info.expclear_enabled != false);
-		EXIT_NOT_IMPLEMENTED(z.stencil_info.partially_resident != false);
-	}
-
-	if (z.z_info.format != Prospero::DepthFormat::kInvalid ||
-	    z.stencil_info.format != Prospero::StencilFormat::kInvalid) {
-		EXIT_NOT_IMPLEMENTED(z.shading_rate_encoding != 0);
-
-		if (z.depth_view.current_mip_level != 0x00000000) {
-			static std::atomic<uint32_t> log_count {0};
-			if (log_count.fetch_add(1, std::memory_order_relaxed) < 16) {
-				LOGF("DepthTarget: temporary: ignoring PS5 current mip level=0x%08" PRIx32 "\n",
-				     z.depth_view.current_mip_level);
-			}
-		}
-		if (z.depth_view.depth_write_disable || z.depth_view.stencil_write_disable) {
-			static std::atomic<uint32_t> log_count {0};
-			if (log_count.fetch_add(1, std::memory_order_relaxed) < 16) {
-				LOGF("DepthTarget: honoring write disable depth=%s, stencil=%s\n",
-				     z.depth_view.depth_write_disable ? "true" : "false",
-				     z.depth_view.stencil_write_disable ? "true" : "false");
-			}
-		}
-		EXIT_NOT_IMPLEMENTED(!z.depth_view.depth_write_disable &&
-		                     z.z_read_base_addr != z.z_write_base_addr);
-		EXIT_NOT_IMPLEMENTED(!z.depth_view.stencil_write_disable &&
-		                     z.stencil_read_base_addr != z.stencil_write_base_addr);
-		EXIT_NOT_IMPLEMENTED(!z.depth_view.depth_write_disable && z.z_write_base_addr == 0);
-		// EXIT_NOT_IMPLEMENTED(z.htile_data_base_addr == 0);
-		EXIT_NOT_IMPLEMENTED(!z.size.valid);
-	}
-}
-
 static void ClipPrint(const char* func, const HW::ClipControl& c) {
 	LOGF("%s\n", func);
 
@@ -546,16 +469,6 @@ static void McCheck(const HW::ModeControl& c) {
 	// EXIT_NOT_IMPLEMENTED(c.cull_front != false);
 	// EXIT_NOT_IMPLEMENTED(c.cull_back != false);
 	// EXIT_NOT_IMPLEMENTED(c.face != false);
-	if (c.poly_mode != 0) {
-		static bool logged = false;
-		if (!logged) {
-			LOGF("\t temporary: PA_SU_SC_MODE_CNTL.POLY_MODE is not fully implemented; continuing "
-			     "with filled polygons\n");
-			logged = true;
-		}
-	}
-	EXIT_NOT_IMPLEMENTED(c.polymode_front_ptype != 0 && c.polymode_front_ptype != 2);
-	EXIT_NOT_IMPLEMENTED(c.polymode_back_ptype != 0 && c.polymode_back_ptype != 2);
 	if (c.vtx_window_offset_enable) {
 		static bool logged = false;
 		if (!logged) {
@@ -565,7 +478,6 @@ 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);
 }
 
@@ -845,15 +757,6 @@ static void VpCheck(const HW::ScreenViewport& vp, const HW::ScanModeControl& smc
 	// EXIT_NOT_IMPLEMENTED(smc.vport_scissor_enable);
 	EXIT_NOT_IMPLEMENTED(smc.line_stipple_enable);
 
-	if (vp.viewports[0].zmin > 0.000000 || vp.viewports[0].zmax != 1.000000) {
-		static bool logged = false;
-		if (!logged) {
-			LOGF("\t warning: non-default viewport depth clamp zmin = %f, zmax = %f; using "
-			     "viewport scale/offset for Vulkan depth range\n",
-			     vp.viewports[0].zmin, vp.viewports[0].zmax);
-			logged = true;
-		}
-	}
 	// EXIT_NOT_IMPLEMENTED(vp.viewports[0].xscale != 960.000000);
 	// EXIT_NOT_IMPLEMENTED(vp.viewports[0].xoffset != 960.000000);
 	// EXIT_NOT_IMPLEMENTED(vp.viewports[0].yscale != -540.000000);
@@ -873,7 +776,7 @@ static void VpCheck(const HW::ScreenViewport& vp, const HW::ScanModeControl& smc
 	// EXIT_NOT_IMPLEMENTED(fabsf(vp.guard_band_horz_clip - 33.133327f) > 0.001f);
 	// EXIT_NOT_IMPLEMENTED(fabsf(vp.guard_band_vert_clip - 59.629623f) > 0.001f);
 
-	if (vp.guard_band_horz_discard != 0.0f || vp.guard_band_vert_discard != 0.0f) {
+	if (vp.guard_band_horz_discard != 1.0f || vp.guard_band_vert_discard != 1.0f) {
 		static std::atomic<uint32_t> log_count {0};
 		if (log_count.fetch_add(1) < 16) {
 			LOGF("\t warning: unsupported PS5 guard band discard = %f, %f, continuing\n",
@@ -893,10 +796,6 @@ static bool ScissorRectValid(const ScissorRect& r) {
 	return r.right > r.left && r.bottom > r.top;
 }
 
-static bool ScissorRectSet(const ScissorRect& r) {
-	return r.left != 0 || r.top != 0 || r.right != 0 || r.bottom != 0;
-}
-
 static ScissorRect ScissorRectOffset(ScissorRect r, int x, int y) {
 	r.left += x;
 	r.right += x;
@@ -906,8 +805,8 @@ static ScissorRect ScissorRectOffset(ScissorRect r, int x, int y) {
 }
 
 static ScissorRect ScissorRectIntersect(const ScissorRect& a, const ScissorRect& b) {
-	return {a.left > b.left ? a.left : b.left, a.top > b.top ? a.top : b.top,
-	        a.right < b.right ? a.right : b.right, a.bottom < b.bottom ? a.bottom : b.bottom};
+	return {std::max(a.left, b.left), std::max(a.top, b.top),
+	        std::min(a.right, b.right), std::min(a.bottom, b.bottom)};
 }
 
 static ScissorRect ScissorRectClamp(ScissorRect r, uint32_t width, uint32_t height) {
@@ -954,49 +853,26 @@ static bool ScissorClipRuleToIntersectionMask(uint16_t rule, uint8_t* mask) {
 }
 
 ScissorRect calc_final_scissor(const HW::ScreenViewport& vp, const HW::ScanModeControl& smc,
-                               vk::Extent2D extent) {
-	ScissorRect screen {vp.screen_scissor_left, vp.screen_scissor_top, vp.screen_scissor_right,
-	                    vp.screen_scissor_bottom};
-	ScissorRect final = screen;
-
-	if (!ScissorRectSet(screen)) {
-		final = {0, 0, static_cast<int>(extent.width), static_cast<int>(extent.height)};
-
-		auto log_id = g_scissor_default_log_count.fetch_add(1);
-		if (log_id < 32) {
-			LOGF("temporary: default unset screen scissor to framebuffer extent %ux%u\n",
-			     extent.width, extent.height);
+                               vk::Extent2D extent, uint32_t viewport_index) {
+	EXIT_IF(viewport_index >= std::size(vp.viewports));
+	ScissorRect final {vp.screen_scissor_left, vp.screen_scissor_top, vp.screen_scissor_right,
+	                   vp.screen_scissor_bottom};
+	const auto intersect = [&](ScissorRect rect, bool window_offset) {
+		if (window_offset) {
+			rect = ScissorRectOffset(rect, vp.window_offset_x, vp.window_offset_y);
 		}
-	}
+		final = ScissorRectIntersect(final, rect);
+	};
+	intersect({vp.window_scissor_left, vp.window_scissor_top, vp.window_scissor_right,
+	           vp.window_scissor_bottom}, vp.window_scissor_window_offset_enable);
+	intersect({vp.generic_scissor_left, vp.generic_scissor_top, vp.generic_scissor_right,
+	           vp.generic_scissor_bottom}, vp.generic_scissor_window_offset_enable);
 
-	ScissorRect window {vp.window_scissor_left, vp.window_scissor_top, vp.window_scissor_right,
-	                    vp.window_scissor_bottom};
-	if (ScissorRectSet(window)) {
-		if (vp.window_scissor_window_offset_enable) {
-			window = ScissorRectOffset(window, vp.window_offset_x, vp.window_offset_y);
-		}
-		final = ScissorRectIntersect(final, window);
-	}
-
-	ScissorRect generic {vp.generic_scissor_left, vp.generic_scissor_top, vp.generic_scissor_right,
-	                     vp.generic_scissor_bottom};
-	if (ScissorRectSet(generic)) {
-		if (vp.generic_scissor_window_offset_enable) {
-			generic = ScissorRectOffset(generic, vp.window_offset_x, vp.window_offset_y);
-		}
-		final = ScissorRectIntersect(final, generic);
-	}
-
-	const auto& viewport = vp.viewports[0];
-	ScissorRect viewport_scissor {viewport.viewport_scissor_left, viewport.viewport_scissor_top,
-	                              viewport.viewport_scissor_right,
-	                              viewport.viewport_scissor_bottom};
-	if (smc.vport_scissor_enable && ScissorRectSet(viewport_scissor)) {
-		if (viewport.viewport_scissor_window_offset_enable) {
-			viewport_scissor =
-			    ScissorRectOffset(viewport_scissor, vp.window_offset_x, vp.window_offset_y);
-		}
-		final = ScissorRectIntersect(final, viewport_scissor);
+	const auto& viewport = vp.viewports[viewport_index];
+	if (smc.vport_scissor_enable) {
+		intersect({viewport.viewport_scissor_left, viewport.viewport_scissor_top,
+		           viewport.viewport_scissor_right, viewport.viewport_scissor_bottom},
+		          viewport.viewport_scissor_window_offset_enable);
 	}
 
 	if (vp.clip_rect_rule == 0) {
@@ -1009,12 +885,8 @@ ScissorRect calc_final_scissor(const HW::ScreenViewport& vp, const HW::ScanModeC
 					continue;
 				}
 
-				ScissorRect clip {vp.clip_rect_left[i], vp.clip_rect_top[i], vp.clip_rect_right[i],
-				                  vp.clip_rect_bottom[i]};
-				if (vp.clip_rect_window_offset_enable[i]) {
-					clip = ScissorRectOffset(clip, vp.window_offset_x, vp.window_offset_y);
-				}
-				final = ScissorRectIntersect(final, clip);
+				intersect({vp.clip_rect_left[i], vp.clip_rect_top[i], vp.clip_rect_right[i],
+				           vp.clip_rect_bottom[i]}, vp.clip_rect_window_offset_enable[i]);
 			}
 		} else {
 			static std::atomic<uint32_t> log_count {0};
@@ -1035,7 +907,6 @@ void hw_check(const CommandBuffer& buffer) {
 	const auto& bc      = hw.GetBlendControl(rt_slot);
 	const auto& bclr    = hw.GetBlendColor();
 	const auto& vp      = hw.GetScreenViewport();
-	const auto& z       = hw.GetDepthRenderTarget();
 	const auto& c       = hw.GetClipControl();
 	const auto& rc      = hw.GetRenderControl();
 	const auto& d       = hw.GetDepthControl();
@@ -1061,8 +932,6 @@ void hw_check(const CommandBuffer& buffer) {
 	RtCheck(rt);
 	log_phase("vp");
 	VpCheck(vp, smc);
-	log_phase("z");
-	ZCheck(z, d, rc);
 	log_phase("clip");
 	ClipCheck(c);
 	log_phase("rc");

src/graphics/host_gpu/renderer/debug.h

@@ -39,7 +39,7 @@ void                     hw_print(const CommandBuffer& buffer);
 void                     hw_check(const CommandBuffer& buffer);
 void                     LogDrawPhase(const char* draw_name, const char* phase);
 ScissorRect calc_final_scissor(const HW::ScreenViewport& vp, const HW::ScanModeControl& smc,
-                               vk::Extent2D extent);
+                               vk::Extent2D extent, uint32_t viewport_index);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -97,10 +97,8 @@ static bool UsesStencilOpValue(uint8_t fail, uint8_t pass, uint8_t depth_fail) {
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer& buffer,
-                                              RenderDepthInfo& r) {
+void RenderExecutor::ResolveRenderDepthTarget(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();
@@ -108,7 +106,7 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 	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.z_write_enable || dc.depth_bounds_enable ||
+	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);
@@ -144,6 +142,8 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 		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);
 	const auto samples   = render_sample_count(z.z_info.num_samples);
 	if (samples == 0) {
 		DepthFatal("unsupported depth fragment count: %u", z.z_info.num_samples);
@@ -162,7 +162,7 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 	    rc.copy_sample != 0 || z.z_info.expclear_enabled ||
 	    z.stencil_info.expclear_enabled || z.z_info.partially_resident ||
 	    z.stencil_info.partially_resident || z.z_info.max_mip_level != 0 ||
-	    z.depth_view.current_mip_level != 0 || z.shading_rate_encoding != 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 ||
@@ -201,10 +201,10 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 		DepthFatal("unsupported depth/stencil format pair");
 	}
 	const auto ideal_format = DepthAttachmentFormat(*policy, has_stencil);
-	r.format = ResolveHostDepthAttachmentFormat(buffer, *policy, has_stencil, samples);
-	if (r.format == vk::Format::eUndefined) {
+	const auto format = ResolveHostDepthAttachmentFormat(buffer, *policy, has_stencil, samples);
+	if (format == vk::Format::eUndefined) {
 		DepthFatal("no host depth/stencil format supports required usage for %s",
-		           VulkanToString(ideal_format).c_str());
+		           vk::to_string(ideal_format).c_str());
 	}
 	const auto     guest_format = policy->guest_format;
 	const uint32_t bytes        = policy->bytes_per_element;
@@ -232,22 +232,12 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 	    (has_htile && !GuestRange {z.htile_data_base_addr, htile_backing_size}.Valid())) {
 		DepthFatal("layered depth backing range is invalid");
 	}
-	r.htile                   = has_htile;
-	r.width                   = width;
-	r.height                  = height;
-	r.samples                 = samples;
-	r.depth_buffer_size       = depth_backing_size;
-	r.depth_buffer_vaddr      = z.z_read_base_addr;
-	r.stencil_buffer_size     = has_stencil ? stencil_backing_size : 0;
-	r.stencil_buffer_vaddr    = has_stencil ? z.stencil_read_base_addr : 0;
-	r.htile_buffer_size       = has_htile ? htile_backing_size : 0;
-	r.htile_buffer_vaddr      = has_htile ? z.htile_data_base_addr : 0;
 	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      = dc.z_write_enable && !z.depth_view.depth_write_disable;
+	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;
@@ -292,18 +282,12 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 			r.stencil_dynamic_back = r.stencil_dynamic_front;
 		}
 	}
-	r.vaddr_num = has_stencil ? 2 : 1;
-	r.vaddr[0]  = r.depth_buffer_vaddr;
-	r.size[0]   = r.depth_buffer_size;
-	if (has_stencil) {
-		r.vaddr[1] = r.stencil_buffer_vaddr;
-		r.size[1]  = r.stencil_buffer_size;
-	}
 	TextureCache::ImageDesc desc {};
 	desc.type                 = TextureCache::BindingType::DepthTarget;
-	desc.info.data            = {r.depth_buffer_vaddr, r.depth_buffer_size};
-	desc.info.stencil         = {r.stencil_buffer_vaddr, r.stencil_buffer_size};
-	desc.info.pixel_format    = r.format;
+	desc.info.data            = {z.z_read_base_addr, depth_backing_size};
+	desc.info.stencil =
+	    has_stencil ? GuestRange {z.stencil_read_base_addr, stencil_backing_size} : GuestRange {};
+	desc.info.pixel_format    = format;
 	desc.info.guest_format    = guest_format;
 	desc.info.type            = Prospero::ImageType::kColor2D;
 	desc.info.extent          = {width, height, 1};
@@ -312,15 +296,16 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 	desc.info.bytes_per_block = bytes;
 	desc.info.samples         = samples;
 	desc.info.tile_mode       = Prospero::TileMode::kDepth;
-	desc.info.mip_layout[0]   = {0, r.depth_buffer_size, pitch, height};
-	desc.info.metadata.range  = {r.htile_buffer_vaddr, r.htile_buffer_size};
+	desc.info.mip_layout[0]   = {0, depth_backing_size, pitch, height};
+	desc.info.metadata.range =
+	    has_htile ? GuestRange {z.htile_data_base_addr, htile_backing_size} : GuestRange {};
 	desc.info.metadata.kind   = has_htile ? ImageMetadataKind::Htile : ImageMetadataKind::None;
 	desc.info.metadata.stencil_compressed =
 	    has_stencil && has_htile && !z.stencil_info.htile_stencil_disabled;
-	desc.view_info.format = r.format;
+	desc.view_info.format = format;
 	desc.view_info.type =
 	    view.layer_count == 1 ? vk::ImageViewType::e2D : vk::ImageViewType::e2DArray;
-	desc.view_info.aspect      = ImageViewOps::DepthAspectMask(r.format);
+	desc.view_info.aspect      = ImageViewOps::DepthAspectMask(format);
 	desc.view_info.base_level  = 0;
 	desc.view_info.level_count = 1;
 	desc.view_info.base_layer  = view.base_layer;
@@ -329,11 +314,11 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, CommandBuffer&
 	r.desc                     = std::move(desc);
 	auto& cache                = m_context.GetTextureCache();
 	r.image_id                 = cache.FindImage(r.desc);
-	r.image_view               = nullptr;
 	BindRenderTarget(r.image_id);
 }
 
 vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
+	const auto format = desc.view_info.format;
 	if (format == vk::Format::eUndefined) {
 		return {};
 	}
@@ -341,7 +326,7 @@ vk::ImageAspectFlags RenderDepthInfo::AttachmentWriteAspects() const {
 	const auto           available = ImageViewOps::DepthAspectMask(format);
 	vk::ImageAspectFlags writes {};
 	if ((available & vk::ImageAspectFlagBits::eDepth) &&
-	    (depth_load_clear_enable || (depth_test_enable && depth_write_enable))) {
+	    (depth_load_clear_enable || depth_write_enable)) {
 		writes |= vk::ImageAspectFlagBits::eDepth;
 	}
 	if (!(available & vk::ImageAspectFlagBits::eStencil)) {

src/graphics/host_gpu/renderer/depthRenderTarget.h

@@ -19,26 +19,13 @@ inline constexpr bool depth_htile_stencil_acceleration_compatible(bool has_stenc
 }
 
 struct RenderDepthInfo {
+	// Discovery keeps guest image information but can remap the view into a larger cache image.
 	TextureCache::ImageDesc     desc;
-	vk::Format                  format                   = vk::Format::eUndefined;
-	uint32_t                    width                    = 0;
-	uint32_t                    height                   = 0;
-	uint32_t                    samples                  = 1;
-	bool                        htile                    = false;
-	uint64_t                    depth_buffer_size        = 0;
-	uint64_t                    depth_buffer_vaddr       = 0;
-	uint64_t                    depth_tile_swizzle       = 0;
-	uint64_t                    stencil_buffer_size      = 0;
-	uint64_t                    stencil_buffer_vaddr     = 0;
-	uint64_t                    stencil_tile_swizzle     = 0;
-	uint64_t                    htile_buffer_size        = 0;
-	uint64_t                    htile_buffer_vaddr       = 0;
-	uint64_t                    htile_tile_swizzle       = 0;
 	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.
 	bool                        depth_write_enable       = false;
 	vk::CompareOp               depth_compare_op         = vk::CompareOp::eNever;
 	bool                        depth_bounds_test_enable = false;
@@ -52,24 +39,17 @@ struct RenderDepthInfo {
 	PipelineStencilDynamicState stencil_dynamic_front;
 	PipelineStencilDynamicState stencil_dynamic_back;
 	ImageId                     image_id;
-	vk::ImageView               image_view = nullptr;
-	uint64_t                    vaddr[3]   = {};
-	uint64_t                    size[3]    = {};
-	int                         vaddr_num  = 0;
 
 	[[nodiscard]] vk::ImageAspectFlags AttachmentWriteAspects() const;
 };
 
-inline bool depth_attachment_read_only(const RenderDepthInfo& depth) {
-	return !depth.AttachmentWriteAspects();
-}
-
 inline vk::ImageLayout depth_attachment_layout(const RenderDepthInfo& depth) {
-	const auto available     = ImageViewOps::DepthAspectMask(depth.format);
+	const auto available     = ImageViewOps::DepthAspectMask(depth.desc.view_info.format);
 	const auto writes        = depth.AttachmentWriteAspects();
 	const bool has_depth     = static_cast<bool>(available & vk::ImageAspectFlagBits::eDepth);
 	const bool has_stencil   = static_cast<bool>(available & vk::ImageAspectFlagBits::eStencil);
-	const bool depth_write   = static_cast<bool>(writes & vk::ImageAspectFlagBits::eDepth);
+	// LoadOp clears are separate from guest depth writes.
+	const bool depth_write   = depth.depth_write_enable;
 	const bool stencil_write = static_cast<bool>(writes & vk::ImageAspectFlagBits::eStencil);
 	if (!has_stencil) {
 		return depth_write ? vk::ImageLayout::eDepthAttachmentOptimal

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

@@ -23,7 +23,6 @@ BlitHelper::BlitHelper(GraphicContext& graphics, CommandScheduler& scheduler)
 	texture_binding.stageFlags      = vk::ShaderStageFlagBits::eFragment;
 
 	vk::DescriptorSetLayoutCreateInfo descriptor_info {};
-	descriptor_info.sType        = vk::StructureType::eDescriptorSetLayoutCreateInfo;
 	descriptor_info.flags        = vk::DescriptorSetLayoutCreateFlagBits::ePushDescriptorKHR;
 	descriptor_info.bindingCount = 1;
 	descriptor_info.pBindings    = &texture_binding;
@@ -32,7 +31,6 @@ BlitHelper::BlitHelper(GraphicContext& graphics, CommandScheduler& scheduler)
 	                     "create BlitHelper descriptor layout");
 
 	vk::PipelineLayoutCreateInfo layout_info {};
-	layout_info.sType          = vk::StructureType::ePipelineLayoutCreateInfo;
 	layout_info.setLayoutCount = 1;
 	layout_info.pSetLayouts    = &m_descriptor_layout;
 	RequireVulkanSuccess(
@@ -65,7 +63,6 @@ BlitHelper::~BlitHelper() {
 vk::ShaderModule BlitHelper::CreateShader(const uint32_t* code, size_t words) const {
 	EXIT_IF(code == nullptr || words == 0);
 	vk::ShaderModuleCreateInfo create {};
-	create.sType            = vk::StructureType::eShaderModuleCreateInfo;
 	create.codeSize         = words * sizeof(uint32_t);
 	create.pCode            = code;
 	vk::ShaderModule module = nullptr;
@@ -84,51 +81,37 @@ vk::Pipeline BlitHelper::GetPipeline(PipelineKey key) {
 	EXIT_IF(samples == vk::SampleCountFlagBits {} || key.format == vk::Format::eUndefined);
 
 	std::array<vk::PipelineShaderStageCreateInfo, 2> stages {};
-	stages[0].sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
 	stages[0].stage  = vk::ShaderStageFlagBits::eVertex;
 	stages[0].module = m_vertex_shader;
 	stages[0].pName  = "main";
-	stages[1].sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
 	stages[1].stage  = vk::ShaderStageFlagBits::eFragment;
 	stages[1].module = m_fragment_shader;
 	stages[1].pName  = "main";
 
 	vk::PipelineVertexInputStateCreateInfo vertex_input {};
-	vertex_input.sType = vk::StructureType::ePipelineVertexInputStateCreateInfo;
 	vk::PipelineInputAssemblyStateCreateInfo input_assembly {};
-	input_assembly.sType    = vk::StructureType::ePipelineInputAssemblyStateCreateInfo;
 	input_assembly.topology = vk::PrimitiveTopology::eTriangleList;
 	vk::PipelineViewportStateCreateInfo viewport {};
-	viewport.sType         = vk::StructureType::ePipelineViewportStateCreateInfo;
 	viewport.viewportCount = 1;
 	viewport.scissorCount  = 1;
 	vk::PipelineRasterizationStateCreateInfo rasterization {};
-	rasterization.sType       = vk::StructureType::ePipelineRasterizationStateCreateInfo;
-	rasterization.polygonMode = vk::PolygonMode::eFill;
-	rasterization.cullMode    = vk::CullModeFlagBits::eNone;
-	rasterization.lineWidth   = 1.0f;
+	rasterization.lineWidth = 1.0f;
 	vk::PipelineMultisampleStateCreateInfo multisample {};
-	multisample.sType                = vk::StructureType::ePipelineMultisampleStateCreateInfo;
 	multisample.rasterizationSamples = samples;
 	vk::PipelineDepthStencilStateCreateInfo depth {};
-	depth.sType            = vk::StructureType::ePipelineDepthStencilStateCreateInfo;
 	depth.depthTestEnable  = VK_TRUE;
 	depth.depthWriteEnable = VK_TRUE;
 	depth.depthCompareOp   = vk::CompareOp::eAlways;
 	vk::PipelineColorBlendStateCreateInfo color_blend {};
-	color_blend.sType = vk::StructureType::ePipelineColorBlendStateCreateInfo;
 	const std::array dynamic_states {vk::DynamicState::eViewport, vk::DynamicState::eScissor};
 	vk::PipelineDynamicStateCreateInfo dynamic {};
-	dynamic.sType             = vk::StructureType::ePipelineDynamicStateCreateInfo;
 	dynamic.dynamicStateCount = static_cast<uint32_t>(dynamic_states.size());
 	dynamic.pDynamicStates    = dynamic_states.data();
 
 	vk::PipelineRenderingCreateInfo rendering {};
-	rendering.sType                 = vk::StructureType::ePipelineRenderingCreateInfo;
 	rendering.depthAttachmentFormat = key.format;
 
 	vk::GraphicsPipelineCreateInfo create {};
-	create.sType               = vk::StructureType::eGraphicsPipelineCreateInfo;
 	create.pNext               = &rendering;
 	create.stageCount          = static_cast<uint32_t>(stages.size());
 	create.pStages             = stages.data();
@@ -186,7 +169,6 @@ void BlitHelper::ReinterpretColorAsMsDepth(Image& source, Image& destination) {
 	                    command);
 
 	vk::RenderingAttachmentInfo depth_attachment {};
-	depth_attachment.sType                   = vk::StructureType::eRenderingAttachmentInfo;
 	depth_attachment.imageView               = destination_view;
 	depth_attachment.imageLayout             = ColorToMsDepthLayout;
 	depth_attachment.loadOp                  = vk::AttachmentLoadOp::eClear;
@@ -194,7 +176,6 @@ void BlitHelper::ReinterpretColorAsMsDepth(Image& source, Image& destination) {
 	depth_attachment.clearValue.depthStencil = {0.0f, 0};
 
 	vk::RenderingInfo rendering {};
-	rendering.sType             = vk::StructureType::eRenderingInfo;
 	rendering.renderArea.extent = {destination_info.extent.width, destination_info.extent.height};
 	rendering.layerCount        = 1;
 	rendering.pDepthAttachment  = &depth_attachment;
@@ -204,7 +185,6 @@ void BlitHelper::ReinterpretColorAsMsDepth(Image& source, Image& destination) {
 	descriptor_image.imageView   = source_view;
 	descriptor_image.imageLayout = vk::ImageLayout::eShaderReadOnlyOptimal;
 	vk::WriteDescriptorSet descriptor_write {};
-	descriptor_write.sType           = vk::StructureType::eWriteDescriptorSet;
 	descriptor_write.dstBinding      = 0;
 	descriptor_write.descriptorCount = 1;
 	descriptor_write.descriptorType  = vk::DescriptorType::eSampledImage;

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)) {
@@ -54,6 +59,9 @@ namespace {
 	}
 	if (DepthAspectTransferFormat(info.pixel_format) != vk::Format::eUndefined) {
 		usage |= vk::ImageUsageFlagBits::eDepthStencilAttachment;
+		if (graphics.attachment_feedback_loop_enabled && (usage & vk::ImageUsageFlagBits::eSampled)) {
+			usage |= vk::ImageUsageFlagBits::eAttachmentFeedbackLoopEXT;
+		}
 		return usage;
 	}
 	if (HasFormatFeature(properties, vk::FormatFeatureFlagBits::eColorAttachment)) {
@@ -202,9 +210,7 @@ void Image::Transit(vk::ImageLayout destination_layout, vk::AccessFlags2 destina
 	if (barriers.empty()) {
 		return;
 	}
-	if (m_scheduler != nullptr) {
-		m_scheduler->EndRendering();
-	}
+	m_scheduler.EndRendering();
 	vk::DependencyInfo dependency {};
 	dependency.imageMemoryBarrierCount = static_cast<uint32_t>(barriers.size());
 	dependency.pImageMemoryBarriers    = barriers.data();
@@ -213,8 +219,8 @@ void Image::Transit(vk::ImageLayout destination_layout, vk::AccessFlags2 destina
 
 void Image::Upload(std::span<const vk::BufferImageCopy> copies, vk::Buffer buffer, uint64_t offset,
                    uint64_t size) {
-	EXIT_IF(m_scheduler == nullptr || copies.empty() || buffer == nullptr || size == 0);
-	m_scheduler->EndRendering();
+	EXIT_IF(copies.empty() || buffer == nullptr || size == 0);
+	m_scheduler.EndRendering();
 	vk::BufferMemoryBarrier2 buffer_barrier {};
 	buffer_barrier.srcStageMask        = vk::PipelineStageFlagBits2::eAllCommands;
 	buffer_barrier.srcAccessMask       = vk::AccessFlagBits2::eMemoryWrite;
@@ -234,7 +240,7 @@ void Image::Upload(std::span<const vk::BufferImageCopy> copies, vk::Buffer buffe
 	dependency.pBufferMemoryBarriers    = &buffer_barrier;
 	dependency.imageMemoryBarrierCount  = static_cast<uint32_t>(image_barriers.size());
 	dependency.pImageMemoryBarriers     = image_barriers.data();
-	auto command                        = m_scheduler->Current().Handle();
+	auto command                        = m_scheduler.Current().Handle();
 	command.pipelineBarrier2(dependency);
 	command.copyBufferToImage(buffer, backing.image, vk::ImageLayout::eTransferDstOptimal,
 	                          static_cast<uint32_t>(copies.size()), copies.data());
@@ -252,8 +258,8 @@ void Image::Upload(std::span<const vk::BufferImageCopy> copies, vk::Buffer buffe
 
 void Image::Download(std::span<const vk::BufferImageCopy> copies, vk::Buffer buffer,
                      uint64_t offset, uint64_t size) {
-	EXIT_IF(m_scheduler == nullptr || copies.empty() || buffer == nullptr || size == 0);
-	m_scheduler->EndRendering();
+	EXIT_IF(copies.empty() || buffer == nullptr || size == 0);
+	m_scheduler.EndRendering();
 	vk::BufferMemoryBarrier2 buffer_barrier {};
 	buffer_barrier.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands;
 	buffer_barrier.srcAccessMask =
@@ -274,7 +280,7 @@ void Image::Download(std::span<const vk::BufferImageCopy> copies, vk::Buffer buf
 	dependency.pBufferMemoryBarriers    = &buffer_barrier;
 	dependency.imageMemoryBarrierCount  = static_cast<uint32_t>(image_barriers.size());
 	dependency.pImageMemoryBarriers     = image_barriers.data();
-	auto command                        = m_scheduler->Current().Handle();
+	auto command                        = m_scheduler.Current().Handle();
 	command.pipelineBarrier2(dependency);
 	command.copyImageToBuffer(backing.image, vk::ImageLayout::eTransferSrcOptimal, buffer,
 	                          static_cast<uint32_t>(copies.size()), copies.data());
@@ -311,8 +317,8 @@ std::pair<uint32_t, uint32_t> Image::SanitizeCopyLayers(const Image& source,
 }
 
 void Image::CopyImage(Image& source) {
-	EXIT_IF(m_scheduler == nullptr || source.backing.samples != backing.samples);
-	m_scheduler->EndRendering();
+	EXIT_IF(source.backing.samples != backing.samples);
+	m_scheduler.EndRendering();
 	const uint32_t levels     = std::min(source.backing.mip_levels, backing.mip_levels);
 	const uint32_t base_depth = backing.image_type == vk::ImageType::e3D
 	                                ? backing.extent.depth
@@ -351,7 +357,7 @@ void Image::CopyImage(Image& source) {
 	if (copies.empty()) {
 		return;
 	}
-	auto command = m_scheduler->Current().Handle();
+	auto command = m_scheduler.Current().Handle();
 	source.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead, {},
 	               command);
 	Transit(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite, {}, command);
@@ -364,8 +370,7 @@ void Image::CopyImage(Image& source) {
 
 void Image::Resolve(Image& source, const ImageSubresourceRange& source_range,
                     const ImageSubresourceRange& destination_range) {
-	EXIT_IF(m_scheduler == nullptr || backing.samples != 1 ||
-	        source.backing.image_type != vk::ImageType::e2D ||
+	EXIT_IF(backing.samples != 1 || source.backing.image_type != vk::ImageType::e2D ||
 	        backing.image_type != vk::ImageType::e2D || source_range.level_count != 1 ||
 	        destination_range.level_count != 1 ||
 	        source_range.base_level >= source.backing.mip_levels ||
@@ -393,8 +398,8 @@ void Image::Resolve(Image& source, const ImageSubresourceRange& source_range,
 	resolved_destination_range.layer_count = layers;
 	const vk::Extent3D resolve_extent {info.extent.width, info.extent.height, 1};
 
-	m_scheduler->EndRendering();
-	auto command = m_scheduler->Current().Handle();
+	m_scheduler.EndRendering();
+	auto command = m_scheduler.Current().Handle();
 	source.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead,
 	               resolved_source_range, command);
 	Transit(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite,
@@ -430,9 +435,8 @@ uint32_t Image::CopyRows(uint64_t row_size, uint32_t rows, uint64_t capacity) no
 }
 
 void Image::CopyImageWithBuffer(Image& source, Buffer& buffer) {
-	EXIT_IF(m_scheduler == nullptr || buffer.Handle() == nullptr || source.backing.samples != 1 ||
-	        backing.samples != 1);
-	m_scheduler->EndRendering();
+	EXIT_IF(buffer.Handle() == nullptr || source.backing.samples != 1 || backing.samples != 1);
+	m_scheduler.EndRendering();
 	const uint32_t levels = std::min(source.backing.mip_levels, backing.mip_levels);
 	const auto     source_aspect =
 	    FullAspectMask(source.backing.format) & ~vk::ImageAspectFlagBits::eStencil;
@@ -462,7 +466,7 @@ void Image::CopyImageWithBuffer(Image& source, Buffer& buffer) {
 	dependency.dependencyFlags          = vk::DependencyFlagBits::eByRegion;
 	dependency.bufferMemoryBarrierCount = 1;
 	dependency.pBufferMemoryBarriers    = &barrier;
-	auto command                        = m_scheduler->Current().Handle();
+	auto command                        = m_scheduler.Current().Handle();
 	source.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead, {},
 	               command);
 	Transit(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite, {}, command);
@@ -522,9 +526,9 @@ void Image::CopyImageWithBuffer(Image& source, Buffer& buffer) {
 }
 
 void Image::CopyMip(Image& source, uint32_t mip, uint32_t layer) {
-	EXIT_IF(m_scheduler == nullptr || source.backing.samples != backing.samples ||
-	        mip >= backing.mip_levels || layer >= backing.layers);
-	m_scheduler->EndRendering();
+	EXIT_IF(source.backing.samples != backing.samples || mip >= backing.mip_levels ||
+	        layer >= backing.layers);
+	m_scheduler.EndRendering();
 	const auto width  = std::max(backing.extent.width >> mip, 1u);
 	const auto height = std::max(backing.extent.height >> mip, 1u);
 	const auto depth  = std::max(backing.extent.depth >> mip, 1u);
@@ -544,7 +548,7 @@ void Image::CopyMip(Image& source, uint32_t mip, uint32_t layer) {
 		copy.dstSubresource = {aspect, mip, layer, destination_layers};
 		copy.extent         = {width, height, depth};
 	}
-	auto command = m_scheduler->Current().Handle();
+	auto command = m_scheduler.Current().Handle();
 	Transit(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite, {}, command);
 	source.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead, {},
 	               command);
@@ -648,7 +652,7 @@ Prospero::BufferFormat RenderTargetTransferFormat(uint32_t bytes_per_element) {
 } // namespace ImageOps
 
 Image::Image(GraphicContext& graphics, CommandScheduler& scheduler, const ImageInfo& image_info)
-    : info(image_info), m_graphics(&graphics), m_scheduler(&scheduler) {
+    : info(image_info), m_graphics(graphics), m_scheduler(scheduler) {
 	KYTY_PROFILER_FUNCTION();
 	ImageOps::Validate(info);
 	m_cpu_dirty =
@@ -657,29 +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.sType         = vk::StructureType::eImageCreateInfo;
-	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.sharingMode   = vk::SharingMode::eExclusive;
-	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,
@@ -690,10 +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,
@@ -721,16 +712,13 @@ uint64_t Image::HashGuestEdges() const {
 
 Image::~Image() {
 	KYTY_PROFILER_FUNCTION();
-	if (m_graphics == nullptr) {
-		return;
-	}
 	for (const auto& cached: views) {
 		if (cached.view != nullptr) {
-			m_graphics->device.destroyImageView(cached.view, nullptr);
+			m_graphics.device.destroyImageView(cached.view, nullptr);
 		}
 	}
 	if (backing.image != nullptr) {
-		m_graphics->DeleteImage(backing);
+		m_graphics.DeleteImage(backing);
 	}
 }
 

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

@@ -35,11 +35,13 @@ struct ImageUsage {
 };
 
 struct ImageBinding {
-	bool is_bound      = false;
-	bool is_target     = false;
-	bool needs_rebind  = false;
-	bool force_general = false;
-	bool shader_write  = false;
+	vk::ImageLayout  attachment_layout = vk::ImageLayout::eUndefined;
+	vk::AccessFlags2 attachment_access;
+	bool             is_bound      = false;
+	bool             is_target     = false;
+	bool             needs_rebind  = false;
+	bool             force_general = false;
+	bool             shader_write  = false;
 };
 
 class Image final {
@@ -49,7 +51,6 @@ public:
 	KYTY_CLASS_NO_COPY(Image);
 
 	[[nodiscard]] vk::ImageView FindView(const ImageViewInfo& view_info);
-	void                        AssociateDepth(ImageId image_id) { depth_id = image_id; }
 	using Barriers = std::vector<vk::ImageMemoryBarrier2>;
 	[[nodiscard]] Barriers GetBarriers(vk::ImageLayout                      destination_layout,
 	                                   vk::AccessFlags2                     destination_access,
@@ -127,9 +128,6 @@ public:
 		return pages ? ImagePageRangesOverlap(info.data.address, info.data.size, address, size)
 		             : ImageRangeOverlaps(info.data.address, info.data.size, address, size);
 	}
-	[[nodiscard]] bool GpuOverlaps(uint64_t address, uint64_t size) const noexcept {
-		return IsGpuModified() && Overlaps(address, size);
-	}
 	[[nodiscard]] bool SafeToDownload() const noexcept {
 		return IsGpuModified() && !IsBufferModified() && !IsCpuDirty();
 	}
@@ -161,8 +159,8 @@ private:
 	[[nodiscard]] static std::pair<uint32_t, uint32_t>
 	SanitizeCopyLayers(const Image& source, const Image& destination, uint32_t depth);
 
-	GraphicContext*   m_graphics         = nullptr;
-	CommandScheduler* m_scheduler        = nullptr;
+	GraphicContext&   m_graphics;
+	CommandScheduler& m_scheduler;
 	uint64_t          m_maybe_cpu_hash   = 0;
 	bool              m_cpu_dirty        = false;
 	bool              m_maybe_cpu_dirty  = false;

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

@@ -23,8 +23,11 @@ struct ImageMetadataInfo {
 	GuestRange          range;
 	ImageMetadataKind   kind               = ImageMetadataKind::None;
 	uint32_t            control            = 0;
+	uint32_t            dcc_clear_word           = 0;
 	VideoOutCompression compression        = VideoOutCompression::Uncompressed;
 	bool                stencil_compressed = false;
+	bool                dcc_clear_register_valid = false;
+	bool                dcc_alpha_msb            = true;
 };
 
 struct ImageSubresources {
@@ -44,6 +47,7 @@ struct ImageSubresourceRange {
 struct ImageMipInfo {
 	uint64_t offset                                 = 0;
 	uint64_t size                                   = 0;
+	// Padded dimensions in storage elements (compressed blocks for BC formats).
 	uint32_t pitch                                  = 0;
 	uint32_t height                                 = 0;
 	auto     operator<=>(const ImageMipInfo&) const = default;

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

@@ -21,10 +21,6 @@ namespace {
 	}
 }
 
-[[nodiscard]] bool IsCompatibleViewFormat(vk::Format image_format, vk::Format view_format) {
-	return ImageViewOps::FormatsCompatible(image_format, view_format);
-}
-
 [[nodiscard]] bool IsStencilViewFormat(vk::Format format) {
 	switch (format) {
 		case vk::Format::eS8Uint:
@@ -327,7 +323,7 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 	}
 	normalized.usage = is_storage ? vk::ImageUsageFlagBits::eStorage : vk::ImageUsageFlags {};
 	const bool format_compatible = normalized.format != vk::Format::eUndefined &&
-	                               IsCompatibleViewFormat(image.format, normalized.format);
+	                               ImageViewOps::FormatsCompatible(image.format, normalized.format);
 	const bool slice_view =
 	    image.image_type == vk::ImageType::e3D && (normalized.type == vk::ImageViewType::e2D ||
 	                                               normalized.type == vk::ImageViewType::e2DArray);
@@ -362,13 +358,11 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 	}
 
 	vk::ImageViewUsageCreateInfo usage {};
-	usage.sType = vk::StructureType::eImageViewUsageCreateInfo;
 	usage.usage = image.usage;
 	if (!is_storage) {
 		usage.usage &= ~vk::ImageUsageFlagBits::eStorage;
 	}
 	vk::ImageViewCreateInfo create {};
-	create.sType                           = vk::StructureType::eImageViewCreateInfo;
 	create.pNext                           = &usage;
 	create.image                           = image.image;
 	create.viewType                        = normalized.type;
@@ -381,7 +375,7 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 	create.subresourceRange.layerCount     = normalized.layer_count;
 
 	vk::ImageView view   = nullptr;
-	const auto    result = m_graphics->device.createImageView(&create, nullptr, &view);
+	const auto    result = m_graphics.device.createImageView(&create, nullptr, &view);
 	if (result != vk::Result::eSuccess || view == nullptr) {
 		EXIT("failed to create image view: result=%d image_format=%d view_format=%d type=%d "
 		     "aspect=0x%x mip=%u+%u layer=%u+%u usage=0x%x\n",

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

@@ -92,29 +92,18 @@ SelectSampledColorView(vk::Format image_format, vk::Format view_format, uint32_t
 	}
 }
 
-[[nodiscard]] inline uint32_t
-SelectSampledDepthView(vk::Format image_format, vk::Format view_format, uint32_t swizzle) noexcept {
-	if (IsSupportedSampledDepthView(image_format, view_format, swizzle)) {
-		return swizzle;
-	}
-	EXIT("unsupported sampled depth image view: image_format=%d view_format=%d swizzle=0x%03x\n",
-	     static_cast<int>(image_format), static_cast<int>(view_format), swizzle);
-}
-
 [[nodiscard]] inline bool
 IsSupportedSampledDepthResource(const ShaderRecompiler::IR::ImageResource& resource) noexcept {
-	if (resource.kind != ShaderRecompiler::IR::ResourceKind::Image &&
-	    resource.kind != ShaderRecompiler::IR::ResourceKind::ImageUint) {
+	if (resource.resource_class != ShaderRecompiler::IR::ImageResourceClass::Sampled) {
 		return false;
 	}
-	if (resource.kind == ShaderRecompiler::IR::ResourceKind::ImageUint && resource.depth_compare) {
+	if (resource.numeric_class == Prospero::TextureNumericClass::Unsupported) {
 		return false;
 	}
-	return (resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D ||
-	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray ||
-	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaa ||
-	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaaArray) &&
-	       resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None && resource.read &&
+	if (resource.numeric_class != Prospero::TextureNumericClass::Float && resource.depth_compare) {
+		return false;
+	}
+	return resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None && resource.read &&
 	       !resource.written && !resource.atomic;
 }
 
@@ -128,8 +117,9 @@ inline void ValidateStorageColorView(vk::Format image_format, vk::Format view_fo
 
 [[nodiscard]] inline bool
 IsSupportedStorageImageResource(const ShaderRecompiler::IR::ImageResource& resource) noexcept {
-	return (resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImage ||
-	        resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint) &&
+	return resource.resource_class == ShaderRecompiler::IR::ImageResourceClass::Storage &&
+	       (resource.numeric_class == Prospero::TextureNumericClass::Float ||
+	        resource.numeric_class == Prospero::TextureNumericClass::Uint) &&
 	       (resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim1D ||
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim1DArray ||
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D ||
@@ -141,19 +131,19 @@ IsSupportedStorageImageResource(const ShaderRecompiler::IR::ImageResource& resou
 	         resource.mip_count != 0u)) &&
 	       resource.written &&
 	       (!resource.atomic ||
-	        (resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint &&
-	         resource.read)) &&
+	        (resource.numeric_class == Prospero::TextureNumericClass::Uint && resource.read)) &&
 	       !resource.depth_compare;
 }
 
 inline void
 ValidateStorageImageResource(const ShaderRecompiler::IR::ImageResource& resource) noexcept {
 	if (!IsSupportedStorageImageResource(resource)) {
-		EXIT("unsupported storage color image resource: kind=%u dimension=%u mip=%u "
+		EXIT("unsupported storage color image resource: class=%u numeric=%u dimension=%u mip=%u "
 		     "read=%d written=%d atomic=%d depth_compare=%d\n",
-		     static_cast<uint32_t>(resource.kind), static_cast<uint32_t>(resource.dimension),
-		     static_cast<uint32_t>(resource.mip_mode), resource.read, resource.written,
-		     resource.atomic, resource.depth_compare);
+		     static_cast<uint32_t>(resource.resource_class),
+		     static_cast<uint32_t>(resource.numeric_class),
+		     static_cast<uint32_t>(resource.dimension), static_cast<uint32_t>(resource.mip_mode),
+		     resource.read, resource.written, resource.atomic, resource.depth_compare);
 	}
 }
 

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

@@ -5,6 +5,7 @@
 #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>
@@ -104,18 +105,14 @@ vk::ComponentSwizzle TextureGetComponentSwizzle(uint8_t 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)));
+	components.r = TextureGetComponentSwizzle(GetDstSel(swizzle, 0));
+	components.g = TextureGetComponentSwizzle(GetDstSel(swizzle, 1));
+	components.b = TextureGetComponentSwizzle(GetDstSel(swizzle, 2));
+	components.a = TextureGetComponentSwizzle(GetDstSel(swizzle, 3));
 	return components;
 }
 

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

@@ -40,7 +40,6 @@ TileManager::TileManager(GraphicContext& graphics, CommandScheduler& scheduler,
 	               nullptr};
 
 	vk::DescriptorSetLayoutCreateInfo descriptor_info {};
-	descriptor_info.sType        = vk::StructureType::eDescriptorSetLayoutCreateInfo;
 	descriptor_info.flags        = vk::DescriptorSetLayoutCreateFlagBits::ePushDescriptorKHR;
 	descriptor_info.bindingCount = static_cast<uint32_t>(bindings.size());
 	descriptor_info.pBindings    = bindings.data();
@@ -50,7 +49,6 @@ TileManager::TileManager(GraphicContext& graphics, CommandScheduler& scheduler,
 
 	const vk::PushConstantRange  push_range {vk::ShaderStageFlagBits::eCompute, 0, sizeof(Push)};
 	vk::PipelineLayoutCreateInfo layout_info {};
-	layout_info.sType                  = vk::StructureType::ePipelineLayoutCreateInfo;
 	layout_info.setLayoutCount         = 1;
 	layout_info.pSetLayouts            = &m_descriptor_layout;
 	layout_info.pushConstantRangeCount = 1;
@@ -92,11 +90,9 @@ TileManager::~TileManager() {
 TileManager::Scratch TileManager::AllocateScratch(uint64_t size) {
 	EXIT_IF(size == 0);
 	vk::BufferCreateInfo create {};
-	create.sType = vk::StructureType::eBufferCreateInfo;
 	create.size  = size;
 	create.usage = vk::BufferUsageFlagBits::eStorageBuffer | vk::BufferUsageFlagBits::eTransferSrc |
 	               vk::BufferUsageFlagBits::eTransferDst;
-	create.sharingMode = vk::SharingMode::eExclusive;
 
 	VmaAllocationCreateInfo allocate {};
 	allocate.usage       = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE;
@@ -264,20 +260,17 @@ vk::Pipeline TileManager::GetPipeline(uint32_t slot) {
 	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.sType       = vk::StructureType::eShaderModuleCreateInfo;
 	module_info.codeSize    = shaders[family_index].words * sizeof(uint32_t);
 	module_info.pCode       = shaders[family_index].code;
 	vk::ShaderModule module = nullptr;
 	RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
 	                     "create TileManager shader module");
 	vk::PipelineShaderStageCreateInfo stage {};
-	stage.sType               = vk::StructureType::ePipelineShaderStageCreateInfo;
 	stage.stage               = vk::ShaderStageFlagBits::eCompute;
 	stage.module              = module;
 	stage.pName               = "main";
 	stage.pSpecializationInfo = &specialization;
 	vk::ComputePipelineCreateInfo create {};
-	create.sType  = vk::StructureType::eComputePipelineCreateInfo;
 	create.stage  = stage;
 	create.layout = m_pipeline_layout;
 	const auto result =
@@ -287,7 +280,7 @@ vk::Pipeline TileManager::GetPipeline(uint32_t slot) {
 	return m_pipelines[slot];
 }
 
-void TileManager::Record(bool tile, vk::Buffer source, uint64_t source_offset,
+void TileManager::Record(vk::Buffer source, uint64_t source_offset,
                          uint64_t source_capacity, vk::Buffer target, uint64_t target_offset,
                          uint64_t target_capacity, std::span<Dispatch> dispatches,
                          bool clear_target) {
@@ -307,7 +300,6 @@ void TileManager::Record(bool tile, vk::Buffer source, uint64_t source_offset,
 	m_scheduler.EndRendering();
 	auto                    command = m_scheduler.Current().Handle();
 	vk::BufferMemoryBarrier barriers[3] {};
-	barriers[0].sType         = vk::StructureType::eBufferMemoryBarrier;
 	barriers[0].srcAccessMask = vk::AccessFlagBits::eMemoryWrite | vk::AccessFlagBits::eHostWrite;
 	barriers[0].dstAccessMask = vk::AccessFlagBits::eShaderRead;
 	barriers[0].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
@@ -316,7 +308,6 @@ void TileManager::Record(bool tile, vk::Buffer source, uint64_t source_offset,
 	barriers[0].offset              = source_offset;
 	barriers[0].size                = source_capacity;
 	barriers[1]                     = barriers[0];
-	barriers[1].srcAccessMask = vk::AccessFlagBits::eMemoryWrite | vk::AccessFlagBits::eHostWrite;
 	barriers[1].dstAccessMask =
 	    clear_target ? vk::AccessFlagBits::eTransferWrite
 	                 : vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
@@ -352,7 +343,6 @@ void TileManager::Record(bool tile, vk::Buffer source, uint64_t source_offset,
 		const vk::DescriptorBufferInfo        infos[] {source_info, target_info, params_info};
 		std::array<vk::WriteDescriptorSet, 3> writes {};
 		for (uint32_t index = 0; index < writes.size(); index++) {
-			writes[index].sType           = vk::StructureType::eWriteDescriptorSet;
 			writes[index].dstBinding      = index;
 			writes[index].descriptorCount = 1;
 			writes[index].descriptorType  = index == 2 ? vk::DescriptorType::eUniformBuffer
@@ -384,7 +374,7 @@ TileManager::Result TileManager::Detile(vk::Buffer tiled, uint64_t tiled_offset,
 	Prepare(false, tiled_capacity, linear_capacity, infos, source_base, 0, dispatches);
 	auto scratch = AllocateScratch((linear_capacity + 3u) & ~uint64_t {3});
 	DeferDestroy(scratch);
-	Record(false, tiled, tiled_offset, tiled_capacity, scratch.buffer, 0, scratch.size, dispatches,
+	Record(tiled, tiled_offset, tiled_capacity, scratch.buffer, 0, scratch.size, dispatches,
 	       true);
 	return {scratch.buffer, 0, linear_capacity};
 }
@@ -399,7 +389,7 @@ void TileManager::Tile(vk::Buffer linear, uint64_t linear_offset, uint64_t linea
 	const uint64_t        target_base = tiled_offset & (descriptor_alignment - 1);
 	std::vector<Dispatch> dispatches;
 	Prepare(true, tiled_capacity, linear_capacity, infos, source_base, target_base, dispatches);
-	Record(true, linear, linear_offset, linear_capacity, tiled, tiled_offset, tiled_capacity,
+	Record(linear, linear_offset, linear_capacity, tiled, tiled_offset, tiled_capacity,
 	       dispatches, false);
 }
 
@@ -423,7 +413,7 @@ void TileManager::TileImage(Image& image, std::span<const vk::BufferImageCopy> r
 	if (transform == ColorTransform::SwapBgra16) {
 		source = SwapBgra16(source);
 	}
-	Record(true, source.buffer, source.offset, linear_capacity, tiled, tiled_offset, tiled_capacity,
+	Record(source.buffer, source.offset, linear_capacity, tiled, tiled_offset, tiled_capacity,
 	       dispatches, false);
 }
 
@@ -482,20 +472,17 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 			words = std::size(GPU_TILER_DEMOTE_D16_SPV);
 		}
 		vk::ShaderModuleCreateInfo module_info {};
-		module_info.sType       = vk::StructureType::eShaderModuleCreateInfo;
 		module_info.codeSize    = words * sizeof(uint32_t);
 		module_info.pCode       = code;
 		vk::ShaderModule module = nullptr;
 		RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
 		                     "create D16 conversion shader module");
 		vk::PipelineShaderStageCreateInfo stage {};
-		stage.sType               = vk::StructureType::ePipelineShaderStageCreateInfo;
 		stage.stage               = vk::ShaderStageFlagBits::eCompute;
 		stage.module              = module;
 		stage.pName               = "main";
 		stage.pSpecializationInfo = &specialization;
 		vk::ComputePipelineCreateInfo create {};
-		create.sType  = vk::StructureType::eComputePipelineCreateInfo;
 		create.stage  = stage;
 		create.layout = m_pipeline_layout;
 		const auto result =
@@ -532,7 +519,6 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 	m_scheduler.EndRendering();
 	auto                    command = m_scheduler.Current().Handle();
 	vk::BufferMemoryBarrier barriers[2] {};
-	barriers[0].sType               = vk::StructureType::eBufferMemoryBarrier;
 	barriers[0].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
 	barriers[0].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
 	barriers[0].buffer              = source.buffer;
@@ -542,7 +528,6 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 	                            vk::AccessFlagBits::eTransferWrite |
 	                            vk::AccessFlagBits::eShaderWrite;
 	barriers[0].dstAccessMask = vk::AccessFlagBits::eShaderRead;
-	barriers[1].sType         = vk::StructureType::eBufferMemoryBarrier;
 	barriers[1].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
 	barriers[1].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
 	barriers[1].buffer              = target.buffer;
@@ -596,7 +581,6 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 			};
 			std::array<vk::WriteDescriptorSet, 2> writes {};
 			for (uint32_t index = 0; index < writes.size(); index++) {
-				writes[index].sType           = vk::StructureType::eWriteDescriptorSet;
 				writes[index].dstBinding      = index;
 				writes[index].descriptorCount = 1;
 				writes[index].descriptorType  = vk::DescriptorType::eStorageBuffer;
@@ -627,19 +611,16 @@ 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.sType       = vk::StructureType::eShaderModuleCreateInfo;
 		module_info.codeSize    = std::size(GPU_TILER_SWAP_BGRA16_SPV) * sizeof(uint32_t);
 		module_info.pCode       = GPU_TILER_SWAP_BGRA16_SPV;
 		vk::ShaderModule module = nullptr;
 		RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
 		                     "create BGRA16 swap shader module");
 		vk::PipelineShaderStageCreateInfo stage {};
-		stage.sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
 		stage.stage  = vk::ShaderStageFlagBits::eCompute;
 		stage.module = module;
 		stage.pName  = "main";
 		vk::ComputePipelineCreateInfo create {};
-		create.sType  = vk::StructureType::eComputePipelineCreateInfo;
 		create.stage  = stage;
 		create.layout = m_pipeline_layout;
 		const auto result =
@@ -658,7 +639,6 @@ void TileManager::SwapBgra16(Result input, Result output, uint32_t pixels) {
 	};
 	std::array<vk::WriteDescriptorSet, 2> writes {};
 	for (uint32_t index = 0; index < writes.size(); index++) {
-		writes[index].sType           = vk::StructureType::eWriteDescriptorSet;
 		writes[index].dstBinding      = index;
 		writes[index].descriptorCount = 1;
 		writes[index].descriptorType  = vk::DescriptorType::eStorageBuffer;
@@ -666,7 +646,6 @@ void TileManager::SwapBgra16(Result input, Result output, uint32_t pixels) {
 	}
 	vk::BufferMemoryBarrier barriers[2] {};
 	for (uint32_t index = 0; index < 2; index++) {
-		barriers[index].sType               = vk::StructureType::eBufferMemoryBarrier;
 		barriers[index].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
 		barriers[index].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
 		barriers[index].buffer              = infos[index].buffer;

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

@@ -125,7 +125,7 @@ private:
 	void Prepare(bool tile, uint64_t tiled_capacity, uint64_t linear_capacity,
 	             std::span<const GpuTileInfo> infos, uint64_t source_base, uint64_t target_base,
 	             std::vector<Dispatch>& dispatches);
-	void Record(bool tile, vk::Buffer source, uint64_t source_offset, uint64_t source_capacity,
+	void Record(vk::Buffer source, uint64_t source_offset, uint64_t source_capacity,
 	            vk::Buffer target, uint64_t target_offset, uint64_t target_capacity,
 	            std::span<Dispatch> dispatches, bool clear_target);
 	[[nodiscard]] vk::Pipeline GetPipeline(uint32_t slot);

src/graphics/host_gpu/renderer/masterSemaphore.cpp

@@ -7,12 +7,10 @@ namespace Libs::Graphics {
 
 MasterSemaphore::MasterSemaphore(GraphicContext& graphics): m_graphics(graphics) {
 	vk::SemaphoreTypeCreateInfo type_info {};
-	type_info.sType         = vk::StructureType::eSemaphoreTypeCreateInfo;
 	type_info.semaphoreType = vk::SemaphoreType::eTimeline;
 	type_info.initialValue  = 0;
 
 	vk::SemaphoreCreateInfo create_info {};
-	create_info.sType = vk::StructureType::eSemaphoreCreateInfo;
 	create_info.pNext = &type_info;
 
 	const auto result = m_graphics.device.createSemaphore(&create_info, nullptr, &m_semaphore);
@@ -47,7 +45,6 @@ void MasterSemaphore::Wait(uint64_t tick) {
 	}
 
 	vk::SemaphoreWaitInfo wait_info {};
-	wait_info.sType          = vk::StructureType::eSemaphoreWaitInfo;
 	wait_info.semaphoreCount = 1;
 	wait_info.pSemaphores    = &m_semaphore;
 	wait_info.pValues        = &tick;

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

@@ -63,7 +63,6 @@ bool DescriptorHeap::Allocate(vk::DescriptorSetLayout layout, Batch& batch) {
 	layouts.fill(layout);
 
 	vk::DescriptorSetAllocateInfo allocate {};
-	allocate.sType          = vk::StructureType::eDescriptorSetAllocateInfo;
 	allocate.descriptorPool = m_current_pool;
 	allocate.pSetLayouts    = layouts.data();
 
@@ -85,8 +84,6 @@ bool DescriptorHeap::Allocate(vk::DescriptorSetLayout layout, Batch& batch) {
 
 void DescriptorHeap::CreatePool() {
 	vk::DescriptorPoolCreateInfo create {};
-	create.sType         = vk::StructureType::eDescriptorPoolCreateInfo;
-	create.flags         = {};
 	create.maxSets       = DescriptorHeapCount;
 	create.poolSizeCount = static_cast<uint32_t>(DescriptorPoolSizes.size());
 	create.pPoolSizes    = DescriptorPoolSizes.data();

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

@@ -16,15 +16,7 @@
 
 namespace Libs::Graphics {
 
-namespace ShaderRecompiler::IR {
-struct ResourceSnapshot;
-}
-
-struct BufferView {
-	vk::Buffer     buffer = nullptr;
-	vk::DeviceSize offset = 0;
-	vk::DeviceSize range  = VK_WHOLE_SIZE;
-};
+struct ShaderStageRuntime;
 
 struct TextureBinding {
 	ImageId                    image_id;
@@ -34,23 +26,25 @@ struct TextureBinding {
 	std::vector<vk::ImageView> mip_views;
 };
 
-struct NativeDescriptors {
-	std::vector<BufferView>     buffers;
-	std::vector<TextureBinding> images;
-	std::vector<vk::Sampler>    samplers;
-	BufferView                  gds;
-	BufferView                  flattened_srt;
-	BufferView                  user_data;
-};
-
 struct PreparedBindings {
-	const ShaderRecompiler::IR::Program*          program  = nullptr;
-	const ShaderRecompiler::IR::ResourceSnapshot* snapshot = nullptr;
-	NativeDescriptors                             resources;
-	std::vector<BufferId>                         buffer_ids;
-	std::vector<uint32_t>                         flattened_srt;
-	std::vector<uint32_t>                         user_data;
-	bool                                          committed = false;
+	struct BufferSource {
+		uint64_t address = 0;
+		uint64_t size    = 0;
+		BufferId id;
+	};
+
+	// 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;
+	std::vector<vk::DescriptorBufferInfo> buffers;
+	std::vector<TextureBinding>           images;
+	std::vector<vk::Sampler>              samplers;
+	vk::DescriptorBufferInfo              gds {nullptr, 0, VK_WHOLE_SIZE};
+	vk::DescriptorBufferInfo              flattened_srt;
+	vk::DescriptorBufferInfo              shader_data_buffer;
+	std::vector<uint32_t>                 shader_data;
 };
 
 [[nodiscard]] vk::DescriptorType
@@ -70,20 +64,8 @@ template <typename T>
 	return result;
 }
 
-struct TargetTextureViewInfo {
-	vk::ImageViewType type        = static_cast<vk::ImageViewType>(VK_IMAGE_VIEW_TYPE_MAX_ENUM);
-	uint32_t          base_layer  = 0;
-	uint32_t          layer_count = 0;
-};
-
-[[nodiscard]] TargetTextureViewInfo
-ResolveTargetTextureView(const ShaderRecompiler::IR::ImageResource& resource,
-                         Prospero::ImageType type, uint32_t base_layer, uint32_t image_layers);
-
-[[nodiscard]] bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor,
-                                                    const Image& image, bool r128 = false);
 [[nodiscard]] bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor,
-                                                   const Image& image, bool r128 = false);
+                                                   bool r128 = false);
 [[nodiscard]] bool
 IsSupportedSampledVideoOutView(const ShaderRecompiler::IR::ImageResource& resource,
                                const ShaderTextureResource& descriptor, const Image& image);

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

@@ -1,6 +1,7 @@
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 
 #include "common/assert.h"
+#include "common/emulatorConfig.h"
 #include "common/file.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
@@ -11,28 +12,54 @@
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
+#include "graphics/shader/recompiler/ShaderRecompiler.h"
 #include "graphics/shader/shaderCompiler.h"
+#include "kernel/memory.h"
 #include "kytyGitVersion.h"
 #include "loader/systemContent.h"
 
 #include <algorithm>
+#include <array>
 #include <atomic>
 #include <cctype>
 #include <cstdio>
 #include <cstring>
+#include <fmt/format.h>
 #include <limits>
 #include <span>
+#include <spirv-tools/libspirv.hpp>
 #include <string_view>
+#include <tuple>
 #include <utility>
 #include <vector>
-
-#include <fmt/format.h>
 #include <xxhash.h>
 
 namespace Libs::Graphics {
 
 namespace {
 
+vk::PolygonMode ResolvePolygonMode(const HW::ModeControl& mode, bool cull_front, bool cull_back) {
+	// CxPrimitiveSetup::PolygonMode disables both per-face modes when it is zero.
+	if (mode.poly_mode == 0) {
+		return vk::PolygonMode::eFill;
+	}
+	EXIT_NOT_IMPLEMENTED(mode.poly_mode != 1);
+	if (cull_front && cull_back) {
+		return vk::PolygonMode::eFill;
+	}
+	if (!cull_front && !cull_back && mode.polymode_front_ptype != mode.polymode_back_ptype) {
+		EXIT("Pipeline: different polygon modes for two visible faces are unsupported\n");
+	}
+	// Vulkan has one polygon mode. A culled face does not constrain that mode.
+	const auto polygon_mode = cull_front ? mode.polymode_back_ptype : mode.polymode_front_ptype;
+	switch (polygon_mode) {
+		case 0: return vk::PolygonMode::ePoint;
+		case 1: return vk::PolygonMode::eLine;
+		case 2: return vk::PolygonMode::eFill;
+		default: EXIT("Pipeline: invalid polygon mode %u\n", polygon_mode);
+	}
+}
+
 std::string DriverCacheSignature(const vk::PhysicalDeviceProperties& properties) {
 	constexpr char hex[] = "0123456789abcdef";
 	std::string    uuid(VK_UUID_SIZE * 2, '0');
@@ -70,97 +97,304 @@ void PipelineCacheLog(fmt::format_string<Args...> format, Args&&... args) {
 	Log::Flush();
 }
 
-void NormalizeStaticParamsForDynamicState(PipelineStaticParameters& static_params) {
-	static_params.viewport_scale[0]  = 0.5f;
-	static_params.viewport_scale[1]  = 0.5f;
-	static_params.viewport_scale[2]  = 1.0f;
-	static_params.viewport_offset[0] = 0.5f;
-	static_params.viewport_offset[1] = 0.5f;
-	static_params.viewport_offset[2] = 0.0f;
+bool ReadShaderGuestMemory(void*, uint64_t address, uint32_t* value) {
+	return value != nullptr &&
+	       Libs::LibKernel::Memory::TryReadGpuCleanBacking(address, value, sizeof(*value));
+}
 
-	static_params.scissor_ltrb[0] = 0;
-	static_params.scissor_ltrb[1] = 0;
-	static_params.scissor_ltrb[2] = 1;
-	static_params.scissor_ltrb[3] = 1;
+void DumpShaderSpirv(const char* stage_name, uint64_t shader_hash,
+                     const std::vector<uint32_t>& spirv) {
+	if (!Config::GraphicsDebugDumpEnabled()) {
+		return;
+	}
+	static std::atomic_int id = 0;
+	const auto path = Config::GetShaderLogFolder() / fmt::format("{:04d}_new_shader_{}_{:016x}.spv",
+	                                                             id++, stage_name, shader_hash);
+	Common::File::CreateDirectories(path.parent_path());
+	Common::File file(path);
+	if (file.IsInvalid()) {
+		const auto path_text = Common::PathToString(path);
+		LOGF_COLOR(Log::Color::BrightRed, "Can't create file: %s\n", path_text.c_str());
+		return;
+	}
+	file.Write(spirv.data(), spirv.size() * sizeof(uint32_t));
+}
+
+void DumpShaderOriginal(const char* stage_name, uint64_t shader_hash,
+                        std::span<const uint32_t> code, const std::string& decoded_dump) {
+	if (!Config::GraphicsDebugDumpEnabled()) {
+		return;
+	}
+	EXIT_IF(code.empty());
+	static std::atomic_int id = 0;
+	const auto base = Config::GetShaderLogFolder() / "original" /
+	                  fmt::format("{:04d}_new_shader_{}_{:016x}", id++, stage_name, shader_hash);
+	Common::File::CreateDirectories(base.parent_path());
+	for (const auto& [suffix, data, size]: {
+	         std::tuple {".bin", static_cast<const void*>(code.data()), code.size_bytes()},
+	         std::tuple {".rdna2", static_cast<const void*>(decoded_dump.data()),
+	                     decoded_dump.size()},
+	     }) {
+		if (size == 0) {
+			continue;
+		}
+		auto path = base;
+		path += suffix;
+		Common::File file(path);
+		if (file.IsInvalid()) {
+			const auto path_text = Common::PathToString(path);
+			LOGF_COLOR(Log::Color::BrightRed, "Can't create file: %s\n", path_text.c_str());
+		} else {
+			file.Write(data, size);
+		}
+	}
+}
+
+bool ValidateShaderSpirv(const char* label, uint64_t shader_hash,
+                         const std::vector<uint32_t>& spirv) {
+	if (!Config::ShaderValidationEnabled()) {
+		return true;
+	}
+	spvtools::SpirvTools tools(SPV_ENV_VULKAN_1_3);
+	std::string          messages;
+	tools.SetMessageConsumer([&messages](spv_message_level_t, const char*,
+	                                     const spv_position_t& position, const char* message) {
+		messages += fmt::format("{}: {} ({}) {}\n", static_cast<int>(position.line),
+		                        static_cast<int>(position.column), static_cast<int>(position.index),
+		                        message);
+	});
+	if (tools.Validate(spirv)) {
+		return true;
+	}
+	spvtools::SpirvTools disassembler(SPV_ENV_VULKAN_1_2);
+	std::string          text;
+	disassembler.Disassemble(spirv, &text,
+	                         static_cast<uint32_t>(SPV_BINARY_TO_TEXT_OPTION_NO_HEADER) |
+	                             static_cast<uint32_t>(SPV_BINARY_TO_TEXT_OPTION_FRIENDLY_NAMES) |
+	                             static_cast<uint32_t>(SPV_BINARY_TO_TEXT_OPTION_COMMENT) |
+	                             static_cast<uint32_t>(SPV_BINARY_TO_TEXT_OPTION_INDENT) |
+	                             static_cast<uint32_t>(SPV_BINARY_TO_TEXT_OPTION_COLOR));
+	LOGF_COLOR(Log::Color::BrightRed, "%s SPIR-V validation failed hash=0x%016" PRIx64 ":\n%s",
+	           label, shader_hash, messages.c_str());
+	LOGF("%s\n", text.c_str());
+	return false;
 }
 
 } // namespace
 
 struct PipelineCache::ProgramCache {
-	struct SourceKey {
-		ShaderType stage = ShaderType::Unknown;
-		uint64_t   hash  = 0;
+	struct ProgramKey {
+		ShaderType            stage           = ShaderType::Unknown;
+		uint64_t              hash            = 0;
+		uint32_t              user_data_count = 0;
+		uint32_t              code_size       = 0;
+		std::vector<uint32_t> static_state;
 
-		bool operator==(const SourceKey&) const = default;
+		bool operator==(const ProgramKey&) const = default;
 	};
 
 	struct Permutation {
-		std::vector<uint32_t>                               static_key;
-		std::shared_ptr<const ShaderRecompiler::IR::Program> program;
-		ShaderProgram                                       handle;
+		ShaderRecompiler::IR::ResourceSpecialization specialization;
+		ShaderRecompiler::IR::CompiledShaderInfo     program;
+		ShaderProgram                                handle;
 	};
 
-	struct SourceKeyHash {
-		std::size_t operator()(const SourceKey& key) const {
+	struct SourceEntry {
+		explicit SourceEntry(ShaderRecompiler::IR::ResourcePlan plan)
+		    : resource_plan(std::move(plan)) {
+			permutations.reserve(8);
+		}
+
+		ShaderRecompiler::IR::ResourcePlan resource_plan;
+		std::vector<Permutation>           permutations;
+	};
+
+	struct ProgramKeyHash {
+		std::size_t operator()(const ProgramKey& key) const {
 			std::size_t hash = static_cast<std::size_t>(key.stage);
 			PipelineKeyHash::Mix(hash, static_cast<std::size_t>(key.hash));
 			if constexpr (sizeof(std::size_t) < sizeof(uint64_t)) {
 				PipelineKeyHash::Mix(hash, static_cast<std::size_t>(key.hash >> 32u));
 			}
+			PipelineKeyHash::Mix(hash, key.user_data_count);
+			PipelineKeyHash::Mix(hash, key.code_size);
+			PipelineKeyHash::Mix(hash, key.static_state.size());
+			// Bucket same-shape static variants by source. ProgramKey equality performs the one
+			// exact state comparison needed on a stable hit without hashing up to 429 words first.
 			return hash;
 		}
 	};
 
-	static uint64_t MixId(uint64_t hash, uint64_t value) {
-		return hash ^ (value + 0x9e3779b97f4a7c15ull + (hash << 6u) + (hash >> 2u));
+	static constexpr std::size_t MaxStaticKeyWords = 13 + ShaderVertexInputInfo::RES_MAX * 13;
+
+	Permutation CompilePermutation(const ShaderParams&                          params,
+	                               const ShaderRecompiler::CompileOptions&      options,
+	                               ShaderRecompiler::TranslateResult            translated,
+	                               ShaderRecompiler::IR::ResourceSpecialization specialization,
+	                               uint32_t push_data_start_dword) {
+		const char* stage_name = nullptr;
+		switch (options.stage) {
+			case ShaderType::Vertex: stage_name = "vs"; break;
+			case ShaderType::Mesh: stage_name = "ms"; break;
+			case ShaderType::Pixel: stage_name = "ps"; break;
+			case ShaderType::Compute: stage_name = "cs"; break;
+			default: EXIT("invalid pipeline shader stage\n");
+		}
+		auto result = ShaderRecompiler::CompileProgram(std::move(translated), options,
+		                                               specialization, push_data_start_dword);
+		DumpShaderOriginal(stage_name, options.shader_hash, params.code, result.decoded_dump);
+		if (!ValidateShaderSpirv(options.dump_label, options.shader_hash, result.spirv)) {
+			DumpShaderSpirv(stage_name, options.shader_hash, result.spirv);
+			EXIT("%s failed hash=0x%016" PRIx64 ": SPIR-V validation failed\n", options.dump_label,
+			     options.shader_hash);
+		}
+		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");
+		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);
+		}
+		return {
+		    .specialization = std::move(specialization),
+		    .program        = std::move(result.program).TakeCompiledInfo(),
+		    .handle         = {.id = ++next_shader_id, .module = module},
+		};
 	}
 
-	static constexpr std::size_t MaxStaticKeyWords =
-	    12 + ShaderVertexInputInfo::RES_MAX * 17;
-
 	template <typename InputInfo>
-	ShaderProgram Get(const ShaderParams& params, InputInfo& input_info) {
-		constexpr ShaderType stage = [] {
-			if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
-				return ShaderType::Vertex;
-			} else if constexpr (std::is_same_v<InputInfo, ShaderPixelInputInfo>) {
-				return ShaderType::Pixel;
-			} else {
-				static_assert(std::is_same_v<InputInfo, ShaderComputeInputInfo>);
-				return ShaderType::Compute;
-			}
-		}();
+	ShaderProgram Get(const ShaderParams& params, InputInfo& input_info,
+	                  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;
+		} else if constexpr (std::is_same_v<InputInfo, ShaderPixelInputInfo>) {
+			stage = ShaderType::Pixel;
+		} else {
+			static_assert(std::is_same_v<InputInfo, ShaderComputeInputInfo>);
+			stage = ShaderType::Compute;
+		}
 
-		BuildStageStaticKey(input_info, key_scratch);
-		auto& permutations = programs[{stage, params.hash}];
-		for (const auto& permutation: permutations) {
-			if (permutation.static_key == key_scratch &&
-			    MaterializeProgram(permutation.program, params, input_info)) {
-				return permutation.handle;
+		lookup_key.stage           = stage;
+		lookup_key.hash            = params.hash;
+		lookup_key.user_data_count = static_cast<uint32_t>(params.user_data.size());
+		lookup_key.code_size       = static_cast<uint32_t>(params.code.size());
+		BuildStageStaticKey(input_info, lookup_key.static_state);
+		auto                                         entry = programs.find(lookup_key);
+		ShaderRecompiler::IR::ResourceSnapshot       resources;
+		ShaderRecompiler::IR::ResourceSpecialization specialization;
+		const ShaderRecompiler::IR::SrtRuntime       runtime {
+		    .user_data                  = params.user_data,
+		    .shader_base                = params.Base(),
+		    .read_specialization_memory = ReadShaderGuestMemory,
+		};
+		if (entry != programs.end()) {
+			EXIT_IF(!ShaderRecompiler::IR::MaterializeResources(
+			    entry->second.resource_plan, runtime, resources, specialization));
+			if (const auto permutation = std::ranges::find_if(
+			        entry->second.permutations, [&](const Permutation& candidate) {
+				        const auto& layout = candidate.program.bindings;
+				        return layout.push_data_start_dword ==
+				                   ShaderRecompiler::IR::PushData::StartFor(
+				                       push_data_cursor, layout.ShaderDataDwords()) &&
+				               candidate.specialization == specialization;
+			        });
+			    permutation != entry->second.permutations.end()) {
+				input_info.stage = {.program   = &permutation->program,
+				                    .resources = std::move(resources)};
+				permutation->program.bindings.AdvancePushData(push_data_cursor);
+				return permutation->handle;
 			}
 		}
 
-		const auto module = CompileProgram(device, params, input_info);
-		EXIT_IF(module == nullptr || !input_info.stage);
-		uint64_t id = MixId(MixId(params.hash, static_cast<uint64_t>(stage)), permutations.size());
-		if (id == 0) {
-			id = 1;
+		ShaderStageInputInfo stage_input {};
+		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
+			stage_input.vertex = &input_info;
+		} else if constexpr (std::is_same_v<InputInfo, ShaderPixelInputInfo>) {
+			stage_input.pixel = &input_info;
+		} else {
+			stage_input.compute = &input_info;
 		}
-		const ShaderProgram handle {.id = id, .module = module};
-		permutations.push_back({key_scratch, input_info.stage.program, handle});
+		const char* label = nullptr;
+		switch (stage) {
+			case ShaderType::Vertex: label = "ShaderRecompiler VS"; break;
+			case ShaderType::Mesh: label = "ShaderRecompiler MS"; break;
+			case ShaderType::Pixel: label = "ShaderRecompiler PS"; break;
+			case ShaderType::Compute: label = "ShaderRecompiler CS"; break;
+			default: EXIT("invalid pipeline shader stage\n");
+		}
+		ShaderRecompiler::CompileOptions options;
+		options.stage       = stage;
+		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.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) {
+				options.user_data_base = 0;
+				options.wave_size      = input_info.mesh.wave_size;
+				options.scratch_dwords = input_info.mesh.scratch_size_dwords;
+			}
+		} else if constexpr (std::is_same_v<InputInfo, ShaderComputeInputInfo>) {
+			options.wave_size = input_info.wave_size;
+		}
+		auto translated = ShaderRecompiler::TranslateProgram(params.code, options);
+		if (entry == programs.end()) {
+			auto resource_plan = ShaderRecompiler::IR::ExtractResourcePlan(translated.program);
+			EXIT_IF(!ShaderRecompiler::IR::MaterializeResources(resource_plan, runtime, resources,
+			                                                    specialization));
+			entry = programs.try_emplace(lookup_key, std::move(resource_plan)).first;
+		}
+		entry->second.permutations.push_back(CompilePermutation(
+		    params, options, std::move(translated), std::move(specialization), push_data_cursor));
+		const auto& permutation = entry->second.permutations.back();
+		input_info.stage = {.program = &permutation.program, .resources = std::move(resources)};
+		permutation.program.bindings.AdvancePushData(push_data_cursor);
 
-		std::printf("Num compiled %u shaders\n", ++num_compiled);
-		return handle;
+		std::array<size_t, static_cast<size_t>(ShaderType::Mesh) + 1> counts {};
+		for (const auto& [key, source]: programs) {
+			counts[static_cast<size_t>(key.stage)] += source.permutations.size();
+		}
+		// Guest geometry shaders are compiled through the host mesh stage.
+		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu\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)]);
+		return permutation.handle;
 	}
 
 	explicit ProgramCache(vk::Device device): device(device) {
-		key_scratch.reserve(MaxStaticKeyWords);
+		lookup_key.static_state.reserve(MaxStaticKeyWords);
+	}
+	~ProgramCache() {
+		for (const auto& [key, entry]: programs) {
+			(void)key;
+			for (const auto& permutation: entry.permutations) {
+				device.destroyShaderModule(permutation.handle.module, nullptr);
+			}
+		}
 	}
 
-	std::unordered_map<SourceKey, std::vector<Permutation>, SourceKeyHash> programs;
-	std::vector<uint32_t> key_scratch;
-	vk::Device device;
-	uint32_t   num_compiled = 0;
+	std::unordered_map<ProgramKey, SourceEntry, ProgramKeyHash> programs;
+	ProgramKey                                                  lookup_key;
+	vk::Device                                                  device;
+	uint64_t                                                    next_shader_id = 0;
 };
 
 PipelineCache::PipelineCache(GraphicContext& graphics)
@@ -181,12 +415,6 @@ PipelineCache::~PipelineCache() {
 	};
 	destroy(m_graphics_pipelines);
 	destroy(m_compute_pipelines);
-	for (const auto& [key, permutations]: m_program_cache->programs) {
-		(void)key;
-		for (const auto& permutation: permutations) {
-			m_graphics.device.destroyShaderModule(permutation.handle.module, nullptr);
-		}
-	}
 	if (m_driver_cache != nullptr) {
 		m_graphics.device.destroyPipelineCache(m_driver_cache, nullptr);
 	}
@@ -212,7 +440,7 @@ void PipelineCache::InitializeDriverCache() {
 		return;
 	}
 
-	m_driver_cache_path = std::filesystem::path("_PipelineCache") / (title_id + ".bin");
+	m_driver_cache_path     = std::filesystem::path("_PipelineCache") / (title_id + ".bin");
 	const auto path         = Common::PathToString(m_driver_cache_path);
 	const bool cache_exists = Common::File::IsFileExisting(m_driver_cache_path);
 	if (cache_exists) {
@@ -224,7 +452,7 @@ void PipelineCache::InitializeDriverCache() {
 	if (cache_exists) {
 		Common::File file(m_driver_cache_path, Common::File::Mode::Read);
 		const auto   file_size = file.IsInvalid() ? 0 : file.Size();
-		const auto signature = DriverCacheSignature(m_graphics.GetPhysicalDeviceProperties());
+		const auto   signature = DriverCacheSignature(m_graphics.GetPhysicalDeviceProperties());
 		if (file_size >= signature.size() + sizeof(uint64_t) &&
 		    file_size <= std::numeric_limits<uint32_t>::max()) {
 			std::string cached_signature(signature.size(), '\0');
@@ -236,7 +464,8 @@ void PipelineCache::InitializeDriverCache() {
 			file.Read(cached_signature.data(), static_cast<uint32_t>(cached_signature.size()),
 			          &signature_read);
 			file.Read(&payload_hash, sizeof(payload_hash), &hash_read);
-			file.Read(initial_data.data(), static_cast<uint32_t>(initial_data.size()), &payload_read);
+			file.Read(initial_data.data(), static_cast<uint32_t>(initial_data.size()),
+			          &payload_read);
 			file.Close();
 			if (signature_read != cached_signature.size() || hash_read != sizeof(payload_hash) ||
 			    payload_read != initial_data.size() || cached_signature != signature ||
@@ -248,26 +477,24 @@ void PipelineCache::InitializeDriverCache() {
 			}
 		} else {
 			file.Close();
-			PipelineCacheLog("Vulkan pipeline cache: invalidating {} (invalid file size)",
-			                 path);
+			PipelineCacheLog("Vulkan pipeline cache: invalidating {} (invalid file size)", path);
 		}
 	}
 
 	vk::PipelineCacheCreateInfo create {};
-	create.sType           = vk::StructureType::ePipelineCacheCreateInfo;
 	create.initialDataSize = initial_data.size();
 	create.pInitialData    = initial_data.empty() ? nullptr : initial_data.data();
 	auto result = m_graphics.device.createPipelineCache(&create, nullptr, &m_driver_cache);
 	if (result != vk::Result::eSuccess && !initial_data.empty()) {
 		PipelineCacheLog("Vulkan pipeline cache: driver rejected {} ({}); starting empty", path,
-		                 VulkanToString(result));
+		                 vk::to_string(result));
 		initial_data.clear();
 		create.initialDataSize = 0;
 		create.pInitialData    = nullptr;
 		result = m_graphics.device.createPipelineCache(&create, nullptr, &m_driver_cache);
 	}
 	if (result != vk::Result::eSuccess) {
-		PipelineCacheLog("Vulkan pipeline cache: disabled ({})", VulkanToString(result));
+		PipelineCacheLog("Vulkan pipeline cache: disabled ({})", vk::to_string(result));
 		m_driver_cache = nullptr;
 		return;
 	}
@@ -304,11 +531,11 @@ void PipelineCache::Save() {
 	if (result != vk::Result::eSuccess || size == 0 ||
 	    size > std::numeric_limits<uint32_t>::max()) {
 		PipelineCacheLog("Vulkan pipeline cache: save failed ({}, {} bytes)",
-		                 VulkanToString(result), size);
+		                 vk::to_string(result), size);
 		return;
 	}
 	payload.resize(size);
-	auto prefix = DriverCacheSignature(m_graphics.GetPhysicalDeviceProperties());
+	auto       prefix       = DriverCacheSignature(m_graphics.GetPhysicalDeviceProperties());
 	const auto payload_hash = XXH3_64bits(payload.data(), payload.size());
 	prefix.append(reinterpret_cast<const char*>(&payload_hash), sizeof(payload_hash));
 	if (!Common::File::CreateDirectories(m_driver_cache_path.parent_path())) {
@@ -338,15 +565,39 @@ void PipelineCache::Save() {
 	m_driver_cache = nullptr;
 }
 
-ShaderProgram PipelineCache::GetVertexProgram(const HW::VertexShaderInfo& regs,
-                                              const HW::ShaderRegisters&  sh,
-                                              const HW::Context&          context,
-                                              ShaderVertexInputInfo&      input_info) {
-	const auto params = PrepareProgram(regs, sh, input_info);
+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;
+	if (mesh_active) {
+		EXIT_NOT_IMPLEMENTED(!m_graphics.mesh_shader_enabled);
+		auto& mesh              = vertex_info.mesh;
+		mesh.host_subgroup_size = m_graphics.subgroup_size;
+		const auto& limits      = m_graphics.mesh_shader_properties;
+		const auto  logical_threads =
+		    mesh.threads_num[0] * mesh.threads_num[1] * mesh.threads_num[2];
+		const auto host_threads = ((logical_threads + mesh.wave_size - 1u) / mesh.wave_size) *
+		                          std::min(mesh.host_subgroup_size, mesh.wave_size);
+		if (host_threads > limits.maxMeshWorkGroupInvocations ||
+		    host_threads > limits.maxMeshWorkGroupSize[0] ||
+		    mesh.max_vertices > limits.maxMeshOutputVertices ||
+		    mesh.max_primitives > limits.maxMeshOutputPrimitives ||
+		    mesh.lds_size_dwords * sizeof(uint32_t) > limits.maxMeshSharedMemorySize) {
+			EXIT("mesh shader exceeds host limits: threads=%u vertices=%u primitives=%u LDS=%u\n",
+			     host_threads, mesh.max_vertices, mesh.max_primitives, mesh.lds_size_dwords);
+		}
+	}
+	ShaderParams pixel_params;
+	if (pixel_active) {
+		pixel_params = PrepareProgram(pixel_regs, sh, target_export_mapping, pixel_info);
+	}
 	if (context.GetClipControl().clip_disable) {
 		const auto& viewport = context.GetScreenViewport().viewports[0];
 		const auto& limits   = m_graphics.GetPhysicalDeviceProperties().limits;
-		auto&       clip     = input_info.clip_space;
+		auto&       clip     = vertex_info.clip_space;
 		clip.scale[0]        = viewport.xscale;
 		clip.scale[1]        = viewport.yscale;
 		clip.offset[0]       = viewport.xoffset;
@@ -358,33 +609,31 @@ ShaderProgram PipelineCache::GetVertexProgram(const HW::VertexShaderInfo& regs,
 		clip.enabled = true;
 	}
 	Common::LockGuard lock(m_mutex);
-	return m_program_cache->Get(params, input_info);
-}
-
-ShaderProgram PipelineCache::GetPixelProgram(
-    const HW::PixelShaderInfo& regs, const HW::ShaderRegisters& sh,
-    const ShaderVertexInputInfo&                        vertex_info,
-    std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
-    ShaderPixelInputInfo&                               input_info) {
-	const auto params = PrepareProgram(regs, sh, vertex_info, target_export_mapping, input_info);
-	Common::LockGuard lock(m_mutex);
-	return m_program_cache->Get(params, input_info);
+	uint32_t          push_data_cursor =
+	    mesh_active ? ShaderRecompiler::IR::PushData::MeshDrawDwordCount : 0;
+	GraphicsPrograms  result;
+	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);
+	return result;
 }
 
 ShaderProgram PipelineCache::GetComputeProgram(const HW::ComputeShaderInfo& regs,
                                                const HW::ShaderRegisters&   sh,
                                                ShaderComputeInputInfo&      input_info) {
-	input_info.needs_lds_barriers = !m_graphics.compute_wave64_supported;
-	const auto params             = PrepareProgram(regs, sh, input_info);
+	input_info.host_subgroup_size = m_graphics.SupportsComputeWave64() ? 64u : 32u;
+	const auto        params      = PrepareProgram(regs, sh, input_info);
 	Common::LockGuard lock(m_mutex);
-	return m_program_cache->Get(params, input_info);
+	uint32_t          push_data_cursor = 0;
+	return m_program_cache->Get(params, input_info, push_data_cursor);
 }
 
 bool PipelineStaticParameters::operator==(const PipelineStaticParameters& other) const noexcept {
 	return std::memcmp(this, &other, sizeof(*this)) == 0;
 }
 
-PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
+PipelineCache::Pipeline& PipelineCache::CreateGraphicsPipeline(
     std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
     const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
     const ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology,
@@ -401,51 +650,45 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 	Common::LockGuard lock(m_mutex);
 	auto&             ctx = command.GetRegisters();
 
-	const HW::BlendColor& bclr                                     = ctx.GetBlendColor();
-	uint32_t              color_mask[RENDER_COLOR_ATTACHMENTS_MAX] = {};
-	for (uint32_t i = 0; i < color_count; i++) {
-		color_mask[i] =
-		    (colors[i].image_id ? colors[i].export_mapping.ApplyMask(render_target_mask_slot(
-		                              ctx.GetRenderTargetMask(), colors[i].target_slot))
-		                        : 0);
-	}
 	const HW::ModeControl& mc = ctx.GetModeControl();
 
 	const auto vs_id = vertex_program.id;
 	const auto ps_id = ps_active ? pixel_program.id : 0;
 
-	PipelineStaticParameters static_params {};
-	GraphicsPipeline         p {};
-	p.ps_shader_id = ps_id;
-	p.vs_shader_id = vs_id;
-
-	static_params.color_count = color_count;
-	PipelineRenderingState rendering {};
+	GraphicsPipelineKey key {};
+	key.vs_shader_id            = vs_id;
+	key.ps_shader_id            = ps_id;
+	auto& static_params         = key.static_params;
+	auto& rendering             = key.rendering;
 	rendering.color_count       = color_count;
 	uint32_t attachment_samples = 0;
 	for (uint32_t i = 0; i < color_count; i++) {
-		EXIT_IF(!colors[i].image_id || colors[i].format == vk::Format::eUndefined);
-		rendering.color_formats[i] = colors[i].format;
+		EXIT_IF(!colors[i].image_id || colors[i].desc.view_info.format == vk::Format::eUndefined);
+		static_params.color_mask[i] = colors[i].export_mapping.ApplyMask(
+		    render_target_mask_slot(ctx.GetRenderTargetMask(), colors[i].target_slot));
+		rendering.color_formats[i] = colors[i].desc.view_info.format;
 		if (attachment_samples == 0) {
-			attachment_samples = colors[i].samples;
-		} else if (attachment_samples != colors[i].samples) {
+			attachment_samples = colors[i].desc.info.samples;
+		} else if (attachment_samples != colors[i].desc.info.samples) {
 			EXIT("mixed color attachment sample counts are unsupported: %u and %u\n",
-			     attachment_samples, colors[i].samples);
+			     attachment_samples, colors[i].desc.info.samples);
 		}
 	}
 	const bool with_depth =
-	    depth.format != vk::Format::eUndefined && static_cast<bool>(depth.image_id);
+	    depth.desc.view_info.format != vk::Format::eUndefined && static_cast<bool>(depth.image_id);
 	if (with_depth) {
-		const auto aspects = ImageViewOps::DepthAspectMask(depth.format);
-		rendering.depth_format =
-		    aspects & vk::ImageAspectFlagBits::eDepth ? depth.format : vk::Format::eUndefined;
-		rendering.stencil_format =
-		    aspects & vk::ImageAspectFlagBits::eStencil ? depth.format : vk::Format::eUndefined;
+		const auto aspects       = ImageViewOps::DepthAspectMask(depth.desc.view_info.format);
+		rendering.depth_format   = aspects & vk::ImageAspectFlagBits::eDepth
+		                               ? depth.desc.view_info.format
+		                               : vk::Format::eUndefined;
+		rendering.stencil_format = aspects & vk::ImageAspectFlagBits::eStencil
+		                               ? depth.desc.view_info.format
+		                               : vk::Format::eUndefined;
 		if (attachment_samples == 0) {
-			attachment_samples = depth.samples;
-		} else if (attachment_samples != depth.samples) {
+			attachment_samples = depth.desc.info.samples;
+		} else if (attachment_samples != depth.desc.info.samples) {
 			EXIT("mixed color/depth sample counts are unsupported: %u and %u\n", attachment_samples,
-			     depth.samples);
+			     depth.desc.info.samples);
 		}
 	}
 	if (color_count == 0 && !with_depth) {
@@ -475,23 +718,19 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 	if (static_params.sample_shading_enable && !m_graphics.sample_rate_shading_enabled) {
 		EXIT("Pipeline: sample-rate shading is required but unsupported by the host\n");
 	}
-	static_params.with_depth         = with_depth;
-	static_params.depth_test_enable  = depth.depth_test_enable;
-	static_params.depth_write_enable = (depth.depth_write_enable && !depth.depth_clear_enable);
-	static_params.depth_compare_op   = depth.depth_compare_op;
 	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;
-	for (uint32_t i = 0; i < RENDER_COLOR_ATTACHMENTS_MAX; i++) {
-		static_params.color_mask[i] = color_mask[i];
-	}
 	const bool rect_list     = topology == vk::PrimitiveTopology::ePatchList;
 	static_params.cull_back  = !rect_list && mc.cull_back;
 	static_params.cull_front = !rect_list && mc.cull_front;
 	static_params.face       = mc.face;
+	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);
@@ -506,18 +745,32 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 		static_params.blend_enable[i]         = bc.enable;
 		static_params.blend_bypass[i]         = rt.info.blend_bypass;
 	}
-	static_params.blend_color_red   = bclr.red;
-	static_params.blend_color_green = bclr.green;
-	static_params.blend_color_blue  = bclr.blue;
-	static_params.blend_color_alpha = bclr.alpha;
-
-	NormalizeStaticParamsForDynamicState(static_params);
-
-	GraphicsPipelineKey key {};
-	key.rendering     = rendering;
-	key.vs_shader_id  = p.vs_shader_id;
-	key.ps_shader_id  = p.ps_shader_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 ||
+		        vs_input_info.resources_num < 0 ||
+		        vs_input_info.resources_num > ShaderVertexInputInfo::RES_MAX);
+		key.vertex_input.binding_count   = static_cast<uint8_t>(vs_input_info.buffers_num);
+		key.vertex_input.attribute_count = static_cast<uint8_t>(vs_input_info.resources_num);
+		uint32_t attributes_num          = 0;
+		for (int binding = 0; binding < vs_input_info.buffers_num; binding++) {
+			const auto& buffer = vs_input_info.buffers[binding];
+			EXIT_IF(buffer.attr_num < 0 || buffer.attr_num > ShaderVertexInputBuffer::ATTR_MAX);
+			attributes_num += static_cast<uint32_t>(buffer.attr_num);
+			EXIT_IF(attributes_num > static_cast<uint32_t>(vs_input_info.resources_num));
+			key.vertex_input.bindings[binding] = {.stride   = buffer.stride,
+			                                      .instance = buffer.fetch_index != 0};
+			for (int attribute = 0; attribute < buffer.attr_num; attribute++) {
+				const auto index = buffer.attr_indices[attribute];
+				EXIT_IF(index < 0 || index >= vs_input_info.resources_num);
+				key.vertex_input.attributes[index] = {
+				    .offset  = buffer.attr_offsets[attribute],
+				    .binding = static_cast<uint8_t>(binding),
+				};
+			}
+		}
+		EXIT_IF(attributes_num != static_cast<uint32_t>(vs_input_info.resources_num));
+	}
 
 	if (auto iter = m_graphics_pipelines.find(key); iter != m_graphics_pipelines.end()) {
 		return *iter->second;
@@ -528,16 +781,16 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 		if (ps_active) {
 			ShaderDbgDumpInputInfo(*ps_input_info);
 		}
-		LOGF("PipelineTrace: shader modules VS=%" PRIu64 " module=%p PS=%" PRIu64
-		     " module=%p\n",
+		LOGF("PipelineTrace: shader modules VS=%" PRIu64 " module=%p PS=%" PRIu64 " module=%p\n",
 		     vs_id, static_cast<void*>(vertex_program.module), ps_id,
 		     static_cast<void*>(pixel_program.module));
 	}
 
-	auto cached = std::make_unique<GraphicsPipeline>(p);
+	auto cached = std::make_unique<Pipeline>();
 	LogPipelineTrace("CreatePipelineInternal begin", vs_id, ps_id);
-	CreatePipelineInternal(m_graphics, *cached, rendering, vs_input_info, vertex_program.module,
-	                       ps_input_info, pixel_program.module, static_params, m_driver_cache);
+	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vs_input_info,
+	                       vertex_program, ps_input_info, pixel_program, static_params,
+	                       m_driver_cache);
 	LogPipelineTrace("CreatePipelineInternal done", vs_id, ps_id);
 
 	EXIT_NOT_IMPLEMENTED(cached->pipeline == nullptr);
@@ -549,22 +802,17 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 	return *iter->second;
 }
 
-PipelineCache::ComputePipeline&
-PipelineCache::CreateComputePipeline(ShaderComputeInputInfo& input_info,
-                                     const ShaderProgram&    compute_program) {
+PipelineCache::Pipeline&
+PipelineCache::CreateComputePipeline(const ShaderComputeInputInfo& input_info,
+                                     const ShaderProgram&          compute_program) {
 	KYTY_PROFILER_BLOCK("PipelineCache::CreatePipeline(Compute)", profiler::colors::RedA100);
 
 	EXIT_IF(!compute_program);
 
 	Common::LockGuard lock(m_mutex);
 
-	ComputePipeline p {};
-	p.cs_shader_id = compute_program.id;
-
-	ComputePipelineKey key {};
-	key.cs_shader_id = p.cs_shader_id;
-
-	if (auto iter = m_compute_pipelines.find(key); iter != m_compute_pipelines.end()) {
+	if (auto iter = m_compute_pipelines.find(compute_program.id);
+	    iter != m_compute_pipelines.end()) {
 		return *iter->second;
 	}
 
@@ -572,13 +820,13 @@ PipelineCache::CreateComputePipeline(ShaderComputeInputInfo& input_info,
 		ShaderDbgDumpInputInfo(input_info);
 	}
 
-	auto cached = std::make_unique<ComputePipeline>(p);
+	auto cached = std::make_unique<Pipeline>();
 	CreatePipelineInternal(m_graphics, *cached, input_info, compute_program.module, m_driver_cache);
 
 	EXIT_NOT_IMPLEMENTED(cached->pipeline == nullptr);
 	EXIT_NOT_IMPLEMENTED(cached->pipeline_layout == nullptr);
 
-	auto [iter, inserted] = m_compute_pipelines.emplace(std::move(key), std::move(cached));
+	auto [iter, inserted] = m_compute_pipelines.emplace(compute_program.id, std::move(cached));
 	EXIT_IF(!inserted);
 
 	return *iter->second;

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

@@ -26,36 +26,31 @@ class CommandBuffer;
 namespace HW {
 class Context;
 class Shader;
+class UserConfig;
 struct ComputeShaderInfo;
 } // namespace HW
 
 #pragma pack(push, 1)
 
 struct PipelineStaticParameters {
-	float                      viewport_scale[3]        = {};
-	float                      viewport_offset[3]       = {};
 	bool                       negative_one_to_one      = false;
 	bool                       depth_clip_enable        = true;
-	int                        scissor_ltrb[4]          = {0};
 	vk::PrimitiveTopology      topology                 = vk::PrimitiveTopology::ePointList;
 	bool                       primitive_restart_enable = false;
 	uint32_t                   samples                  = 1;
 	bool                       sample_shading_enable    = false;
-	bool                       with_depth               = false;
-	bool                       depth_test_enable        = false;
-	bool                       depth_write_enable       = false;
-	vk::CompareOp              depth_compare_op         = vk::CompareOp::eNever;
 	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_count                                        = 1;
 	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]       = {};
 	uint8_t                    color_destblend[RENDER_COLOR_ATTACHMENTS_MAX]      = {};
@@ -65,10 +60,6 @@ struct PipelineStaticParameters {
 	bool                       separate_alpha_blend[RENDER_COLOR_ATTACHMENTS_MAX] = {};
 	bool                       blend_enable[RENDER_COLOR_ATTACHMENTS_MAX]         = {};
 	bool                       blend_bypass[RENDER_COLOR_ATTACHMENTS_MAX]         = {};
-	float                      blend_color_red                                    = 0.0f;
-	float                      blend_color_green                                  = 0.0f;
-	float                      blend_color_blue                                   = 0.0f;
-	float                      blend_color_alpha                                  = 0.0f;
 
 	bool operator==(const PipelineStaticParameters& other) const noexcept;
 };
@@ -78,14 +69,7 @@ struct PipelineStaticParameters {
 static_assert(std::is_trivially_copyable_v<PipelineStaticParameters>);
 static_assert(std::is_standard_layout_v<PipelineStaticParameters>);
 static_assert(alignof(PipelineStaticParameters) == 1);
-static_assert(sizeof(PipelineStaticParameters) ==
-              sizeof(float[3]) + sizeof(float[3]) + sizeof(bool) * 2 + sizeof(int[4]) +
-                  sizeof(vk::PrimitiveTopology) + sizeof(bool) + sizeof(uint32_t) +
-                  sizeof(bool) * 4 + sizeof(vk::CompareOp) + sizeof(bool) + sizeof(float) * 2 +
-                  sizeof(bool) + sizeof(PipelineStencilStaticState) * 2 + sizeof(uint32_t) +
-                  sizeof(uint32_t[RENDER_COLOR_ATTACHMENTS_MAX]) + sizeof(bool) * 3 +
-                  sizeof(uint8_t[RENDER_COLOR_ATTACHMENTS_MAX]) * 6 +
-                  sizeof(bool[RENDER_COLOR_ATTACHMENTS_MAX]) * 3 + sizeof(float) * 4);
+static_assert(sizeof(PipelineStaticParameters) == 166);
 
 struct PipelineRenderingState {
 	std::array<vk::Format, RENDER_COLOR_ATTACHMENTS_MAX> color_formats {};
@@ -96,13 +80,31 @@ struct PipelineRenderingState {
 	bool operator==(const PipelineRenderingState&) const = default;
 };
 
+struct PipelineVertexInputState {
+	struct Binding {
+		uint32_t stride                           = 0;
+		bool     instance                         = false;
+		bool     operator==(const Binding&) const = default;
+	};
+	struct Attribute {
+		uint32_t offset                             = 0;
+		uint8_t  binding                            = 0;
+		bool     operator==(const Attribute&) const = default;
+	};
+
+	std::array<Binding, ShaderVertexInputInfo::RES_MAX>   bindings {};
+	std::array<Attribute, ShaderVertexInputInfo::RES_MAX> attributes {};
+	uint8_t                                               binding_count   = 0;
+	uint8_t                                               attribute_count = 0;
+
+	bool operator==(const PipelineVertexInputState&) const = default;
+};
+
 struct ShaderProgram {
 	uint64_t         id     = 0;
 	vk::ShaderModule module = nullptr;
 
-	explicit operator bool() const {
-		return id != 0 && module != nullptr;
-	}
+	explicit operator bool() const { return id != 0 && module != nullptr; }
 };
 
 class PipelineCache {
@@ -119,36 +121,30 @@ public:
 		bool                    uses_push_descriptors = false;
 	};
 
-	struct GraphicsPipeline: Pipeline {
-		uint64_t vs_shader_id = 0;
-		uint64_t ps_shader_id = 0;
+	struct GraphicsPrograms {
+		ShaderProgram vertex;
+		ShaderProgram pixel;
 	};
 
-	struct ComputePipeline: Pipeline {
-		uint64_t cs_shader_id = 0;
-	};
-
-	ShaderProgram GetVertexProgram(const HW::VertexShaderInfo& regs,
-	                               const HW::ShaderRegisters&  sh,
-	                               const HW::Context&          context,
-	                               ShaderVertexInputInfo&      input_info);
-	ShaderProgram
-	GetPixelProgram(const HW::PixelShaderInfo& regs, const HW::ShaderRegisters& sh,
-	                const ShaderVertexInputInfo&                        vertex_info,
-	                std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
-	                ShaderPixelInputInfo&                               input_info);
+	GraphicsPrograms
+	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);
 	ShaderProgram GetComputeProgram(const HW::ComputeShaderInfo& regs,
-	                                const HW::ShaderRegisters& sh,
-	                                ShaderComputeInputInfo& input_info);
+	                                const HW::ShaderRegisters&   sh,
+	                                ShaderComputeInputInfo&      input_info);
 
-	GraphicsPipeline& 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);
-	ComputePipeline& CreateComputePipeline(ShaderComputeInputInfo& input_info,
-	                                       const ShaderProgram&    compute_program);
+	Pipeline&
+	CreateGraphicsPipeline(std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
+	                       const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
+	                       const ShaderPixelInputInfo* ps_input_info,
+	                       vk::PrimitiveTopology topology, bool primitive_restart_enable,
+	                       const ShaderProgram& vertex_program, const ShaderProgram& pixel_program);
+	Pipeline& CreateComputePipeline(const ShaderComputeInputInfo& input_info,
+	                                const ShaderProgram&          compute_program);
 
 private:
 	struct ProgramCache;
@@ -157,19 +153,13 @@ private:
 		PipelineRenderingState   rendering;
 		uint64_t                 vs_shader_id = 0;
 		uint64_t                 ps_shader_id = 0;
+		PipelineVertexInputState vertex_input;
 		PipelineStaticParameters static_params;
 
 		bool operator==(const GraphicsPipelineKey& other) const {
 			return rendering == other.rendering && vs_shader_id == other.vs_shader_id &&
-			       ps_shader_id == other.ps_shader_id && static_params == other.static_params;
-		}
-	};
-
-	struct ComputePipelineKey {
-		uint64_t cs_shader_id = 0;
-
-		bool operator==(const ComputePipelineKey& other) const {
-			return cs_shader_id == other.cs_shader_id;
+			       ps_shader_id == other.ps_shader_id && vertex_input == other.vertex_input &&
+			       static_params == other.static_params;
 		}
 	};
 
@@ -202,28 +192,28 @@ private:
 			PipelineKeyHash::MixRendering(hash, key.rendering);
 			PipelineKeyHash::Mix(hash, key.vs_shader_id);
 			PipelineKeyHash::Mix(hash, key.ps_shader_id);
+			PipelineKeyHash::Mix(hash, key.vertex_input.binding_count);
+			for (uint32_t i = 0; i < key.vertex_input.binding_count; i++) {
+				PipelineKeyHash::Mix(hash, key.vertex_input.bindings[i].stride);
+				PipelineKeyHash::Mix(hash, key.vertex_input.bindings[i].instance);
+			}
+			PipelineKeyHash::Mix(hash, key.vertex_input.attribute_count);
+			for (uint32_t i = 0; i < key.vertex_input.attribute_count; i++) {
+				PipelineKeyHash::Mix(hash, key.vertex_input.attributes[i].offset);
+				PipelineKeyHash::Mix(hash, key.vertex_input.attributes[i].binding);
+			}
 			PipelineKeyHash::MixStaticParams(hash, key.static_params);
 			return hash;
 		}
 	};
 
-	struct ComputePipelineKeyHash {
-		std::size_t operator()(const ComputePipelineKey& key) const {
-			std::size_t hash = 0;
-			PipelineKeyHash::Mix(hash, key.cs_shader_id);
-			return hash;
-		}
-	};
-
-	GraphicContext&                m_graphics;
+	GraphicContext&               m_graphics;
 	std::unique_ptr<ProgramCache> m_program_cache;
 	vk::PipelineCache             m_driver_cache = nullptr;
 	std::filesystem::path         m_driver_cache_path;
-	std::unordered_map<GraphicsPipelineKey, std::unique_ptr<GraphicsPipeline>,
-	                   GraphicsPipelineKeyHash>
-	    m_graphics_pipelines;
-	std::unordered_map<ComputePipelineKey, std::unique_ptr<ComputePipeline>, ComputePipelineKeyHash>
-	              m_compute_pipelines;
+	std::unordered_map<GraphicsPipelineKey, std::unique_ptr<Pipeline>, GraphicsPipelineKeyHash>
+	                                                        m_graphics_pipelines;
+	std::unordered_map<uint64_t, std::unique_ptr<Pipeline>> m_compute_pipelines;
 	Common::Mutex m_mutex;
 
 	void InitializeDriverCache();
@@ -231,14 +221,14 @@ private:
 
 void LogPipelineTrace(const char* phase, uint64_t vertex_program_id, uint64_t pixel_program_id);
 void CreatePipelineInternal(
-    GraphicContext& graphics, PipelineCache::GraphicsPipeline& pipeline,
-    const PipelineRenderingState& rendering, const ShaderVertexInputInfo& vs_input_info,
-    vk::ShaderModule vertex_module, const ShaderPixelInputInfo* ps_input_info,
-	vk::ShaderModule pixel_module, const PipelineStaticParameters& static_params,
-	vk::PipelineCache driver_cache);
-void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::ComputePipeline& pipeline,
+    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
+    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
+    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
+    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache);
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
                             const ShaderComputeInputInfo& input_info,
-	vk::ShaderModule compute_module, vk::PipelineCache driver_cache);
+                            vk::ShaderModule compute_module, vk::PipelineCache driver_cache);
 
 } // namespace Libs::Graphics
 

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

@@ -13,6 +13,9 @@ vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	if (stages & vk::ShaderStageFlagBits::eVertex) {
 		result |= vk::PipelineStageFlagBits::eVertexShader;
 	}
+	if (stages & vk::ShaderStageFlagBits::eMeshEXT) {
+		result |= vk::PipelineStageFlagBits::eMeshShaderEXT;
+	}
 	if (stages & vk::ShaderStageFlagBits::eFragment) {
 		result |= vk::PipelineStageFlagBits::eFragmentShader;
 	}
@@ -25,7 +28,6 @@ vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 
 VulkanMemoryBarrier MakeShaderWriteDependency() {
 	VulkanMemoryBarrier barrier {};
-	barrier.sType         = vk::StructureType::eMemoryBarrier;
 	barrier.srcAccessMask = vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite |
 	                        vk::AccessFlagBits::eVertexAttributeRead |
@@ -38,7 +40,6 @@ VulkanMemoryBarrier MakeShaderWriteDependency() {
 
 VulkanMemoryBarrier MakeShaderAccessDependency() {
 	VulkanMemoryBarrier barrier {};
-	barrier.sType         = vk::StructureType::eMemoryBarrier;
 	barrier.srcAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
 	return barrier;
@@ -46,7 +47,6 @@ VulkanMemoryBarrier MakeShaderAccessDependency() {
 
 VulkanMemoryBarrier MakeShaderWriteHazardDependency() {
 	VulkanMemoryBarrier barrier {};
-	barrier.sType         = vk::StructureType::eMemoryBarrier;
 	barrier.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
 	return barrier;
@@ -56,7 +56,6 @@ vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer) {
 	EXIT_IF(buffer == nullptr);
 
 	vk::BufferMemoryBarrier barrier {};
-	barrier.sType         = vk::StructureType::eBufferMemoryBarrier;
 	barrier.srcAccessMask = vk::AccessFlagBits::eHostWrite | vk::AccessFlagBits::eTransferWrite |
 	                        vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
@@ -68,11 +67,12 @@ vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer) {
 	return barrier;
 }
 
-std::vector<ShaderBufferWriteRange>
-CollectShaderBufferWrites(const ShaderRecompiler::IR::Program&          program,
-                          const ShaderRecompiler::IR::ResourceSnapshot& resources) {
+bool HasShaderBufferWrites(const ShaderStageRuntime& runtime) {
+	EXIT_IF(!runtime);
+	const auto& program   = *runtime.program;
+	const auto& resources = runtime.resources;
 	EXIT_IF(resources.buffers.size() != program.info.buffers.size());
-	std::vector<ShaderBufferWriteRange> writes;
+	bool has_writes = false;
 	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {
 		if (!program.info.buffers[i].written) {
 			continue;
@@ -81,23 +81,11 @@ CollectShaderBufferWrites(const ShaderRecompiler::IR::Program&          program,
 		EXIT_IF(value.dword_count < 4);
 		ShaderBufferResource descriptor;
 		std::memcpy(descriptor.fields, value.dwords.data(), sizeof(descriptor.fields));
-		const auto address = descriptor.Base48();
-		const auto records = static_cast<uint64_t>(descriptor.NumRecords());
-		const auto stride  = static_cast<uint64_t>(descriptor.Stride());
-		if (stride != 0 && records > UINT64_MAX / stride) {
-			EXIT("shader resource barrier buffer footprint overflow\n");
-		}
-		const auto size = stride == 0 ? records : stride * records;
-		if (address != 0 && size != 0) {
-			writes.push_back({address, size});
-		}
+		// A zero stride means byte addressing. For either addressing mode a nonzero record
+		// count is exactly the condition for a nonempty descriptor range.
+		has_writes |= descriptor.Base48() != 0 && descriptor.NumRecords() != 0;
 	}
-	return writes;
-}
-
-bool HasShaderBufferWrites(const ShaderStageRuntime& runtime) {
-	EXIT_IF(!runtime);
-	return !CollectShaderBufferWrites(*runtime.program, *runtime.resources).empty();
+	return has_writes;
 }
 
 void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages) {
@@ -118,13 +106,11 @@ void ShaderWriteHazardBarrier(vk::CommandBuffer      vk_buffer,
 void ShaderWriteBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages) {
 	EXIT_IF(vk_buffer == nullptr || !source_stages);
 	const auto barrier = MakeShaderWriteDependency();
-	vk_buffer.pipelineBarrier(
-	    source_stages,
-	    vk::PipelineStageFlagBits::eComputeShader | vk::PipelineStageFlagBits::eVertexInput |
-	        vk::PipelineStageFlagBits::eVertexShader | vk::PipelineStageFlagBits::eFragmentShader |
-	        vk::PipelineStageFlagBits::eTransfer |
-	        vk::PipelineStageFlagBits::eColorAttachmentOutput,
-	    vk::DependencyFlags {}, 1, &barrier, 0, nullptr, 0, nullptr);
+	vk_buffer.pipelineBarrier(source_stages,
+	                          vk::PipelineStageFlagBits::eComputeShader |
+	                              vk::PipelineStageFlagBits::eAllGraphics |
+	                              vk::PipelineStageFlagBits::eTransfer,
+	                          vk::DependencyFlags {}, 1, &barrier, 0, nullptr, 0, nullptr);
 }
 
 } // namespace Libs::Graphics

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

@@ -4,27 +4,15 @@
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 
-#include <vector>
-
 namespace Libs::Graphics {
 
 struct ShaderStageRuntime;
 
-struct ShaderBufferWriteRange {
-	uint64_t address = 0;
-	uint64_t size    = 0;
-
-	bool operator==(const ShaderBufferWriteRange& other) const = default;
-};
-
 vk::PipelineStageFlags  ShaderPipelineStages(vk::ShaderStageFlags stages);
 VulkanMemoryBarrier     MakeShaderAccessDependency();
 VulkanMemoryBarrier     MakeShaderWriteHazardDependency();
 VulkanMemoryBarrier     MakeShaderWriteDependency();
 vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer);
-std::vector<ShaderBufferWriteRange>
-CollectShaderBufferWrites(const ShaderRecompiler::IR::Program&          program,
-                          const ShaderRecompiler::IR::ResourceSnapshot& resources);
 bool HasShaderBufferWrites(const ShaderStageRuntime& runtime);
 void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages);
 void ShaderWriteHazardBarrier(vk::CommandBuffer      vk_buffer,

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

@@ -103,7 +103,7 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 		size   = 4;
 		if (NarrowInputFormat(format, size, used_components)) {
 			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
-			     VulkanToString(format).c_str(), used_components);
+			     vk::to_string(format).c_str(), used_components);
 		}
 		return;
 	}
@@ -117,7 +117,7 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 		size   = 2;
 		if (NarrowInputFormat(format, size, used_components)) {
 			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
-			     VulkanToString(format).c_str(), used_components);
+			     vk::to_string(format).c_str(), used_components);
 		}
 		return;
 	}
@@ -390,7 +390,7 @@ static vk::BlendOp GetBlendOp(uint32_t op) {
 }
 
 static void AddLayoutBindings(std::vector<vk::DescriptorSetLayoutBinding>& descriptor_bindings,
-                              const ShaderRecompiler::IR::Program& program,
+                              const ShaderRecompiler::IR::CompiledShaderInfo& program,
                               vk::ShaderStageFlagBits              stage) {
 	for (const auto& binding: program.bindings.descriptors) {
 		descriptor_bindings.push_back(
@@ -408,7 +408,6 @@ static void CreateDescriptorLayout(GraphicContext& graphics, PipelineCache::Pipe
 	pipeline.uses_push_descriptors = descriptor_count <= graphics.max_push_descriptors;
 
 	vk::DescriptorSetLayoutCreateInfo create {};
-	create.sType        = vk::StructureType::eDescriptorSetLayoutCreateInfo;
 	create.flags        = pipeline.uses_push_descriptors
 	                          ? vk::DescriptorSetLayoutCreateFlagBits::ePushDescriptorKHR
 	                          : vk::DescriptorSetLayoutCreateFlags {};
@@ -420,21 +419,27 @@ static void CreateDescriptorLayout(GraphicContext& graphics, PipelineCache::Pipe
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
 void CreatePipelineInternal(
-    GraphicContext& graphics, PipelineCache::GraphicsPipeline& pipeline,
-    const PipelineRenderingState& rendering, const ShaderVertexInputInfo& vs_input_info,
-    vk::ShaderModule vertex_module, const ShaderPixelInputInfo* ps_input_info,
-	vk::ShaderModule pixel_module, const PipelineStaticParameters& static_params,
-	vk::PipelineCache driver_cache) {
+    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
+    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
+    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
+    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache) {
 	const bool ps_active = ps_input_info != nullptr;
-	EXIT_IF(vertex_module == nullptr || (ps_active && pixel_module == nullptr));
+	EXIT_IF(!vertex_program || (ps_active && !pixel_program));
+	const bool with_depth = rendering.depth_format != vk::Format::eUndefined ||
+	                        rendering.stencil_format != vk::Format::eUndefined;
+	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 = static_params.topology == vk::PrimitiveTopology::ePatchList;
+	const bool rect_list = !mesh && static_params.topology == vk::PrimitiveTopology::ePatchList;
 
 	vk::ShaderModule tess_control_shader_module = nullptr;
 	vk::ShaderModule tess_eval_shader_module    = nullptr;
 
 	vk::ShaderModuleCreateInfo create_info {};
-	create_info.sType = vk::StructureType::eShaderModuleCreateInfo;
 	vk::Result result {};
 	if (rect_list) {
 		const auto shaders =
@@ -445,7 +450,7 @@ void CreatePipelineInternal(
 		    graphics.device.createShaderModule(&create_info, nullptr, &tess_control_shader_module);
 		if (graphics_debug_dump_enabled()) {
 			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done result=%s module=%p\n",
-			     VulkanToString(result).c_str(), static_cast<void*>(tess_control_shader_module));
+			     vk::to_string(result).c_str(), static_cast<void*>(tess_control_shader_module));
 		}
 		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
@@ -455,7 +460,7 @@ void CreatePipelineInternal(
 		    graphics.device.createShaderModule(&create_info, nullptr, &tess_eval_shader_module);
 		if (graphics_debug_dump_enabled()) {
 			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done result=%s module=%p\n",
-			     VulkanToString(result).c_str(), static_cast<void*>(tess_eval_shader_module));
+			     vk::to_string(result).c_str(), static_cast<void*>(tess_eval_shader_module));
 		}
 		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 	}
@@ -464,32 +469,21 @@ void CreatePipelineInternal(
 	    rect_list && (tess_control_shader_module == nullptr || tess_eval_shader_module == nullptr));
 
 	vk::PipelineShaderStageCreateInfo vert_shader_stage_info {};
-
-	vert_shader_stage_info.sType               = vk::StructureType::ePipelineShaderStageCreateInfo;
-	vert_shader_stage_info.pNext               = nullptr;
-	vert_shader_stage_info.flags               = {};
-	vert_shader_stage_info.stage               = vk::ShaderStageFlagBits::eVertex;
-	vert_shader_stage_info.module              = vertex_module;
-	vert_shader_stage_info.pName               = "main";
-	vert_shader_stage_info.pSpecializationInfo = nullptr;
+	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.sType               = vk::StructureType::ePipelineShaderStageCreateInfo;
-	frag_shader_stage_info.pNext               = nullptr;
-	frag_shader_stage_info.flags               = {};
-	frag_shader_stage_info.stage               = vk::ShaderStageFlagBits::eFragment;
-	frag_shader_stage_info.module              = pixel_module;
-	frag_shader_stage_info.pName               = "main";
-	frag_shader_stage_info.pSpecializationInfo = nullptr;
+	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.sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
 	tess_control_shader_stage_info.stage  = vk::ShaderStageFlagBits::eTessellationControl;
 	tess_control_shader_stage_info.module = tess_control_shader_module;
 	tess_control_shader_stage_info.pName  = "main";
 
 	vk::PipelineShaderStageCreateInfo tess_eval_shader_stage_info {};
-	tess_eval_shader_stage_info.sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
 	tess_eval_shader_stage_info.stage  = vk::ShaderStageFlagBits::eTessellationEvaluation;
 	tess_eval_shader_stage_info.module = tess_eval_shader_module;
 	tess_eval_shader_stage_info.pName  = "main";
@@ -505,182 +499,81 @@ void CreatePipelineInternal(
 		shader_stages[shader_stage_count++] = frag_shader_stage_info;
 	}
 
-	vk::VertexInputAttributeDescription input_attr[ShaderVertexInputInfo::RES_MAX];
-	vk::VertexInputBindingDescription   input_desc[ShaderVertexInputInfo::RES_MAX];
+	vk::VertexInputAttributeDescription input_attr[ShaderVertexInputInfo::RES_MAX] {};
+	vk::VertexInputBindingDescription   input_desc[ShaderVertexInputInfo::RES_MAX] {};
 
-	for (int bi = 0; bi < vs_input_info.buffers_num; bi++) {
-		const auto& b          = vs_input_info.buffers[bi];
-		input_desc[bi].binding = bi;
-		input_desc[bi].stride  = b.stride;
-		input_desc[bi].inputRate =
-		    (b.fetch_index == 0 ? vk::VertexInputRate::eVertex : vk::VertexInputRate::eInstance);
+	for (uint32_t binding = 0; binding < vertex_input.binding_count; binding++) {
+		input_desc[binding].binding   = binding;
+		input_desc[binding].stride    = vertex_input.bindings[binding].stride;
+		input_desc[binding].inputRate = vertex_input.bindings[binding].instance
+		                                    ? vk::VertexInputRate::eInstance
+		                                    : vk::VertexInputRate::eVertex;
+	}
+	for (uint32_t index = 0; index < vertex_input.attribute_count; index++) {
+		input_attr[index].binding  = vertex_input.attributes[index].binding;
+		input_attr[index].location = index;
+		input_attr[index].offset   = vertex_input.attributes[index].offset;
 
-		for (int ai = 0; ai < b.attr_num; ai++) {
-			auto index                 = b.attr_indices[ai];
-			input_attr[index].binding  = bi;
-			input_attr[index].location = index;
-			input_attr[index].offset   = b.attr_offsets[ai];
+		uint32_t   attr_size     = 4;
+		const auto registers_num = vs_input_info.resources_dst[index].registers_num;
+		const auto compiled_components =
+		    vs_input_info.stage.program->info.vertex_fetch_components[index];
+		const auto used_components =
+		    compiled_components > 0 ? static_cast<int>(compiled_components) : registers_num;
+		GetInputFormat(vs_input_info.resources[index], input_attr[index].format, attr_size,
+		               static_cast<uint32_t>(used_components));
 
-			uint32_t attr_size       = 4;
-			auto     registers_num   = vs_input_info.resources_dst[index].registers_num;
-			const auto compiled_components =
-			    vs_input_info.stage.program->info.vertex_fetch_components[index];
-			auto used_components =
-			    compiled_components > 0 ? static_cast<int>(compiled_components) : registers_num;
-			GetInputFormat(vs_input_info.resources[index], input_attr[index].format, attr_size,
-			               static_cast<uint32_t>(used_components));
-
-			if (graphics_debug_dump_enabled()) {
-				static std::atomic_uint log_count = 0;
-				const auto              log_id = log_count.fetch_add(1, std::memory_order_relaxed);
-				if (log_id < 128) {
-					LOGF("VertexInputState[%u]: attr=%u binding=%u offset=%u stride=%u fmt=%d "
-					     "src_fmt=%u dst=v%u regs=%u"
-					     " fetched_components=%u attr_size=%u swizzle=%u,%u,%u,%u\n",
-					     log_id, index, input_attr[index].binding, input_attr[index].offset,
-					     input_desc[bi].stride, static_cast<int>(input_attr[index].format),
-					     static_cast<uint32_t>(vs_input_info.resources[index].Format()),
-					     static_cast<uint32_t>(vs_input_info.resources_dst[index].register_start),
-					     static_cast<uint32_t>(registers_num),
-					     static_cast<uint32_t>(used_components), attr_size,
-					     static_cast<uint32_t>(vs_input_info.resources[index].DstSelX()),
-					     static_cast<uint32_t>(vs_input_info.resources[index].DstSelY()),
-					     static_cast<uint32_t>(vs_input_info.resources[index].DstSelZ()),
-					     static_cast<uint32_t>(vs_input_info.resources[index].DstSelW()));
-				}
-			}
-
-			if (vs_input_info.resources[index].OutOfBounds() != 0) {
-				static bool logged = false;
-				if (!logged) {
-					LOGF("VertexInput: temporary: accepting PS5 out-of-bounds behavior %" PRIu8
-					     "\n",
-					     vs_input_info.resources[index].OutOfBounds());
-					logged = true;
-				}
-			}
-
-			EXIT_NOT_IMPLEMENTED(vs_input_info.resources[index].AddTid());
-			EXIT_NOT_IMPLEMENTED(vs_input_info.resources[index].SwizzleEnabled());
-
-			auto log_unsupported_vertex_swizzle = [index, attr_size, registers_num](
-			                                          uint32_t swizzle, uint32_t expected) {
-				static bool logged = false;
-				if (!logged) {
-					LOGF(
-					    "VertexInput: temporary: accepting unsupported dst swizzle at attr %" PRIu32
-					    " (attr_size=%" PRIu32 ", regs=%" PRIu32 ", swizzle=0x%03" PRIx32
-					    ", expected=0x%03" PRIx32 ")\n",
-					    static_cast<uint32_t>(index), attr_size, registers_num, swizzle, expected);
-					logged = true;
-				}
-			};
-
-			switch (registers_num) {
-				case 1: {
-					auto swizzle = vs_input_info.resources[index].DstSelX();
-					if (swizzle != DstSel(4)) {
-						log_unsupported_vertex_swizzle(swizzle, DstSel(4));
-					}
-					break;
-				}
-				case 2: {
-					auto swizzle  = vs_input_info.resources[index].DstSelXY();
-					auto expected = (attr_size == 1 ? DstSel(4, 0) : DstSel(4, 5));
-					if (swizzle != expected) {
-						log_unsupported_vertex_swizzle(swizzle, expected);
-					}
-					break;
-				}
-				case 3: {
-					auto swizzle  = vs_input_info.resources[index].DstSelXYZ();
-					auto expected = DstSel(4, 5, 6);
-					switch (attr_size) {
-						case 1: expected = DstSel(4, 0, 0); break;
-						case 2: expected = DstSel(4, 5, 0); break;
-						default: break;
-					}
-					if (swizzle != expected) {
-						log_unsupported_vertex_swizzle(swizzle, expected);
-					}
-					break;
-				}
-				case 4: {
-					auto swizzle   = vs_input_info.resources[index].DstSelXYZW();
-					auto expected  = DstSel(4, 5, 6, 7);
-					bool supported = false;
-					switch (attr_size) {
-						case 1:
-							expected  = DstSel(4, 0, 0, 1);
-							supported = (swizzle == expected);
-							break;
-						case 2:
-							expected  = DstSel(4, 5, 0, 1);
-							supported = (swizzle == expected);
-							break;
-						case 3:
-							expected  = DstSel(4, 5, 6, 1);
-							supported = (swizzle == expected || swizzle == DstSel(4, 5, 6, 0));
-							break;
-						default:
-							supported = (swizzle == expected || swizzle == DstSel(4, 5, 6, 1) ||
-							             swizzle == DstSel(4, 5, 6, 0));
-							break;
-					}
-					if (!supported) {
-						log_unsupported_vertex_swizzle(swizzle, expected);
-					}
-					break;
-				}
-				default: EXIT("invalid registers_num");
+		if (graphics_debug_dump_enabled()) {
+			static std::atomic_uint log_count = 0;
+			const auto              log_id    = log_count.fetch_add(1, std::memory_order_relaxed);
+			if (log_id < 128) {
+				LOGF("VertexInputState[%u]: attr=%u binding=%u offset=%u stride=%u fmt=%d "
+				     "src_fmt=%u dst=v%u regs=%u"
+				     " fetched_components=%u attr_size=%u swizzle=%u,%u,%u,%u\n",
+				     log_id, index, input_attr[index].binding, input_attr[index].offset,
+				     input_desc[input_attr[index].binding].stride,
+				     static_cast<int>(input_attr[index].format),
+				     static_cast<uint32_t>(vs_input_info.resources[index].Format()),
+				     static_cast<uint32_t>(vs_input_info.resources_dst[index].register_start),
+				     static_cast<uint32_t>(registers_num), static_cast<uint32_t>(used_components),
+				     attr_size, static_cast<uint32_t>(vs_input_info.resources[index].DstSelX()),
+				     static_cast<uint32_t>(vs_input_info.resources[index].DstSelY()),
+				     static_cast<uint32_t>(vs_input_info.resources[index].DstSelZ()),
+				     static_cast<uint32_t>(vs_input_info.resources[index].DstSelW()));
 			}
 		}
+
+		if (vs_input_info.resources[index].OutOfBounds() != 0) {
+			static bool logged = false;
+			if (!logged) {
+				LOGF("VertexInput: temporary: accepting PS5 out-of-bounds behavior %" PRIu8 "\n",
+				     vs_input_info.resources[index].OutOfBounds());
+				logged = true;
+			}
+		}
+
+		EXIT_NOT_IMPLEMENTED(vs_input_info.resources[index].AddTid());
+		EXIT_NOT_IMPLEMENTED(vs_input_info.resources[index].SwizzleEnabled());
+
+		EXIT_IF(registers_num < 1 || registers_num > 4);
 	}
 
 	vk::PipelineVertexInputStateCreateInfo vertex_input_info {};
-	vertex_input_info.sType = vk::StructureType::ePipelineVertexInputStateCreateInfo;
-	vertex_input_info.pNext = nullptr;
-	vertex_input_info.flags = {};
-	vertex_input_info.vertexBindingDescriptionCount   = vs_input_info.buffers_num;
+	vertex_input_info.vertexBindingDescriptionCount   = vertex_input.binding_count;
 	vertex_input_info.pVertexBindingDescriptions      = input_desc;
-	vertex_input_info.vertexAttributeDescriptionCount = vs_input_info.resources_num;
+	vertex_input_info.vertexAttributeDescriptionCount = vertex_input.attribute_count;
 	vertex_input_info.pVertexAttributeDescriptions    = input_attr;
 
 	vk::PipelineInputAssemblyStateCreateInfo input_assembly {};
-	input_assembly.sType    = vk::StructureType::ePipelineInputAssemblyStateCreateInfo;
-	input_assembly.pNext    = nullptr;
-	input_assembly.flags    = {};
 	input_assembly.topology = static_params.topology;
 	input_assembly.primitiveRestartEnable =
 	    static_params.primitive_restart_enable ? VK_TRUE : VK_FALSE;
 
-	vk::Viewport viewport {};
-	viewport.x        = static_params.viewport_offset[0] - static_params.viewport_scale[0];
-	viewport.y        = static_params.viewport_offset[1] - static_params.viewport_scale[1];
-	viewport.width    = static_params.viewport_scale[0] * 2.0f;
-	viewport.height   = static_params.viewport_scale[1] * 2.0f;
-	viewport.minDepth = static_params.viewport_offset[2];
-	viewport.maxDepth = static_params.viewport_scale[2] + static_params.viewport_offset[2];
-
-	vk::Rect2D scissor {};
-	scissor.offset = {static_params.scissor_ltrb[0], static_params.scissor_ltrb[1]};
-	scissor.extent = {
-	    static_cast<uint32_t>(static_params.scissor_ltrb[2] - static_params.scissor_ltrb[0]),
-	    static_cast<uint32_t>(static_params.scissor_ltrb[3] - static_params.scissor_ltrb[1])};
-
 	vk::PipelineViewportDepthClipControlCreateInfoEXT depth_clip_control {};
-	depth_clip_control.sType = vk::StructureType::ePipelineViewportDepthClipControlCreateInfoEXT;
-	depth_clip_control.pNext = nullptr;
 	depth_clip_control.negativeOneToOne = (static_params.negative_one_to_one ? VK_TRUE : VK_FALSE);
 
 	vk::PipelineViewportStateCreateInfo viewport_state {};
-	viewport_state.sType         = vk::StructureType::ePipelineViewportStateCreateInfo;
-	viewport_state.pNext         = &depth_clip_control;
-	viewport_state.flags         = {};
-	viewport_state.viewportCount = 1;
-	viewport_state.pViewports    = &viewport;
-	viewport_state.scissorCount  = 1;
-	viewport_state.pScissors     = &scissor;
+	viewport_state.pNext = &depth_clip_control;
 
 	vk::CullModeFlags cull_mode = vk::CullModeFlagBits::eNone;
 	if (static_params.cull_back) {
@@ -694,65 +587,38 @@ void CreatePipelineInternal(
 	    (static_params.face ? vk::FrontFace::eClockwise : vk::FrontFace::eCounterClockwise);
 
 	vk::PipelineRasterizationDepthClipStateCreateInfoEXT clip_ext {};
-	clip_ext.sType           = vk::StructureType::ePipelineRasterizationDepthClipStateCreateInfoEXT;
-	clip_ext.pNext           = nullptr;
-	clip_ext.flags           = {};
 	clip_ext.depthClipEnable = static_params.depth_clip_enable ? VK_TRUE : VK_FALSE;
 
 	vk::PipelineRasterizationStateCreateInfo rasterizer {};
-	rasterizer.sType = vk::StructureType::ePipelineRasterizationStateCreateInfo;
 	// MoltenVK lacks VK_EXT_depth_clip_enable; omit the depth-clip struct on macOS and accept
 	// Vulkan's default depth clipping (enabled) instead of the PS5's clamp behavior.
-#if defined(__APPLE__)
-	rasterizer.pNext = nullptr;
-#else
+#if !defined(__APPLE__)
 	rasterizer.pNext = &clip_ext;
 #endif
-	rasterizer.flags                   = {};
-	rasterizer.depthClampEnable        = VK_FALSE;
-	rasterizer.rasterizerDiscardEnable = VK_FALSE;
-	rasterizer.polygonMode             = vk::PolygonMode::eFill;
-	rasterizer.cullMode                = cull_mode;
-	rasterizer.frontFace               = front_face;
-	rasterizer.depthBiasEnable         = VK_FALSE;
-	rasterizer.depthBiasConstantFactor = 0.0f;
-	rasterizer.depthBiasClamp          = 0.0f;
-	rasterizer.depthBiasSlopeFactor    = 0.0f;
-	rasterizer.lineWidth               = 1.0f;
+	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;
+	rasterizer.lineWidth = 1.0f;
 
 	vk::PipelineMultisampleStateCreateInfo multisampling {};
-	multisampling.sType                 = vk::StructureType::ePipelineMultisampleStateCreateInfo;
-	multisampling.pNext                 = nullptr;
-	multisampling.flags                 = {};
-	multisampling.sampleShadingEnable   = static_params.sample_shading_enable ? VK_TRUE : VK_FALSE;
-	multisampling.rasterizationSamples  = vulkan_sample_count(static_params.samples);
-	multisampling.minSampleShading      = 1.0f;
-	multisampling.pSampleMask           = nullptr;
-	multisampling.alphaToCoverageEnable = VK_FALSE;
-	multisampling.alphaToOneEnable      = VK_FALSE;
+	multisampling.sampleShadingEnable  = static_params.sample_shading_enable ? VK_TRUE : VK_FALSE;
+	multisampling.rasterizationSamples = vulkan_sample_count(static_params.samples);
+	multisampling.minSampleShading     = 1.0f;
 
 	vk::PipelineColorBlendAttachmentState color_blend_attachment[RENDER_COLOR_ATTACHMENTS_MAX] = {};
-	for (uint32_t i = 0; i < static_params.color_count; i++) {
-		vk::ColorComponentFlags color_write_mask = {};
+	for (uint32_t i = 0; i < rendering.color_count; i++) {
 		EXIT_NOT_IMPLEMENTED((static_params.color_mask[i] & ~0x0fu) != 0);
-		if ((static_params.color_mask[i] & 0x1u) != 0) {
-			color_write_mask |=
-			    static_cast<vk::ColorComponentFlags>(vk::ColorComponentFlagBits::eR);
-		}
-		if ((static_params.color_mask[i] & 0x2u) != 0) {
-			color_write_mask |=
-			    static_cast<vk::ColorComponentFlags>(vk::ColorComponentFlagBits::eG);
-		}
-		if ((static_params.color_mask[i] & 0x4u) != 0) {
-			color_write_mask |=
-			    static_cast<vk::ColorComponentFlags>(vk::ColorComponentFlagBits::eB);
-		}
-		if ((static_params.color_mask[i] & 0x8u) != 0) {
-			color_write_mask |=
-			    static_cast<vk::ColorComponentFlags>(vk::ColorComponentFlagBits::eA);
-		}
-
-		color_blend_attachment[i].colorWriteMask = color_write_mask;
+		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].srcColorBlendFactor =
@@ -773,54 +639,37 @@ void CreatePipelineInternal(
 	}
 
 	vk::Bool32 color_write_enable[RENDER_COLOR_ATTACHMENTS_MAX] = {};
-	for (uint32_t i = 0; i < static_params.color_count; i++) {
+	for (uint32_t i = 0; i < rendering.color_count; i++) {
 		color_write_enable[i] = VK_TRUE;
 	}
 
 	vk::PipelineColorWriteCreateInfoEXT color_write {};
-	color_write.sType              = vk::StructureType::ePipelineColorWriteCreateInfoEXT;
-	color_write.pNext              = nullptr;
-	color_write.attachmentCount    = static_params.color_count;
+	color_write.attachmentCount    = rendering.color_count;
 	color_write.pColorWriteEnables = color_write_enable;
 
 	vk::PipelineColorBlendStateCreateInfo color_blending {};
-	color_blending.sType = vk::StructureType::ePipelineColorBlendStateCreateInfo;
 	// MoltenVK lacks VK_EXT_color_write_enable; drop the dynamic color-write struct on macOS
 	// and rely on each attachment's static colorWriteMask (all channels enabled by default).
-#if defined(__APPLE__)
-	color_blending.pNext = nullptr;
-#else
+#if !defined(__APPLE__)
 	color_blending.pNext = &color_write;
 #endif
-	color_blending.flags             = {};
-	color_blending.logicOpEnable     = VK_FALSE;
-	color_blending.logicOp           = vk::LogicOp::eCopy;
-	color_blending.attachmentCount   = static_params.color_count;
-	color_blending.pAttachments      = color_blend_attachment;
-	color_blending.blendConstants[0] = static_params.blend_color_red;
-	color_blending.blendConstants[1] = static_params.blend_color_green;
-	color_blending.blendConstants[2] = static_params.blend_color_blue;
-	color_blending.blendConstants[3] = static_params.blend_color_alpha;
+	color_blending.logicOp         = vk::LogicOp::eCopy;
+	color_blending.attachmentCount = rendering.color_count;
+	color_blending.pAttachments    = color_blend_attachment;
 
-	EXIT_IF(!vs_input_info.stage);
 	std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
-	AddLayoutBindings(descriptor_bindings, *vs_input_info.stage.program,
-	                  vk::ShaderStageFlagBits::eVertex);
+	AddLayoutBindings(descriptor_bindings, *vs_input_info.stage.program, vertex_stage);
 	if (ps_active) {
 		EXIT_IF(!ps_input_info->stage);
 		AddLayoutBindings(descriptor_bindings, *ps_input_info->stage.program,
 		                  vk::ShaderStageFlagBits::eFragment);
 	}
 	CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
-	constexpr auto GraphicsStages =
-	    vk::ShaderStageFlagBits::eVertex | vk::ShaderStageFlagBits::eFragment;
-	const vk::PushConstantRange push_constants {
-	    GraphicsStages, 0, ShaderRecompiler::IR::NativePushConstantSize};
+	const auto                  GraphicsStages = vertex_stage | vk::ShaderStageFlagBits::eFragment;
+	const vk::PushConstantRange push_constants {GraphicsStages, 0,
+	                                            ShaderRecompiler::IR::NativePushConstantSize};
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
-	pipeline_layout_info.sType                  = vk::StructureType::ePipelineLayoutCreateInfo;
-	pipeline_layout_info.pNext                  = nullptr;
-	pipeline_layout_info.flags                  = {};
 	pipeline_layout_info.setLayoutCount         = 1;
 	pipeline_layout_info.pSetLayouts            = &pipeline.descriptor_set_layout;
 	pipeline_layout_info.pushConstantRangeCount = 1;
@@ -831,25 +680,19 @@ void CreatePipelineInternal(
 	if (graphics_debug_dump_enabled()) {
 		LOGF("PipelineTrace: vkCreatePipelineLayout begin VS=%" PRIu64 " PS=%" PRIu64
 		     " set_layouts=1 push_constants=%" PRIu32 "\n",
-		     pipeline.vs_shader_id, pipeline.ps_shader_id, 1u);
+		     vertex_program.id, ps_active ? pixel_program.id : 0, 1u);
 	}
 	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",
-		     VulkanToString(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
+		     vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
 	}
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
 	EXIT_NOT_IMPLEMENTED(pipeline.pipeline_layout == nullptr);
 
 	vk::PipelineDepthStencilStateCreateInfo depth_stencil_info {};
-	depth_stencil_info.sType            = vk::StructureType::ePipelineDepthStencilStateCreateInfo;
-	depth_stencil_info.pNext            = nullptr;
-	depth_stencil_info.flags            = {};
-	depth_stencil_info.depthTestEnable  = (static_params.depth_test_enable ? VK_TRUE : VK_FALSE);
-	depth_stencil_info.depthWriteEnable = (static_params.depth_write_enable ? VK_TRUE : VK_FALSE);
-	depth_stencil_info.depthCompareOp   = static_params.depth_compare_op;
 	depth_stencil_info.depthBoundsTestEnable =
 #if defined(__APPLE__)
 	    VK_FALSE; // MoltenVK lacks the depthBounds feature; depth-bounds testing is disabled
@@ -868,84 +711,72 @@ void CreatePipelineInternal(
 	depth_stencil_info.minDepthBounds    = static_params.depth_min_bounds;
 	depth_stencil_info.maxDepthBounds    = static_params.depth_max_bounds;
 
-	const vk::DynamicState dynamic_states[] = {
-	    vk::DynamicState::eViewport,
-	    vk::DynamicState::eScissor,
+	std::vector<vk::DynamicState> dynamic_states {
+	    vk::DynamicState::eViewportWithCount,
+	    vk::DynamicState::eScissorWithCount,
 	    vk::DynamicState::eLineWidth,
+	    vk::DynamicState::eDepthTestEnable,
+	    vk::DynamicState::eDepthWriteEnable,
+	    vk::DynamicState::eDepthCompareOp,
 	    vk::DynamicState::eDepthBiasEnable,
 	    vk::DynamicState::eDepthBias,
 	    vk::DynamicState::eStencilCompareMask,
 	    vk::DynamicState::eStencilReference,
 	    vk::DynamicState::eStencilWriteMask,
-#if !defined(__APPLE__)
-	    // Keep last so depth-only pipelines can omit this dynamic state.
-	    vk::DynamicState::eColorWriteEnableEXT, // unsupported by MoltenVK; static mask instead
-#endif
+	    vk::DynamicState::eBlendConstants,
 	};
-	auto dynamic_states_count =
-	    static_cast<uint32_t>(sizeof(dynamic_states) / sizeof(dynamic_states[0]));
 #if !defined(__APPLE__)
-	if (static_params.color_count == 0) {
-		dynamic_states_count--;
+	if (rendering.color_count != 0) {
+		dynamic_states.push_back(vk::DynamicState::eColorWriteEnableEXT);
 	}
 #endif
+	if (graphics.attachment_feedback_loop_enabled) {
+		dynamic_states.push_back(vk::DynamicState::eAttachmentFeedbackLoopEnableEXT);
+	}
 
 	vk::PipelineDynamicStateCreateInfo dynamic_state {};
-	dynamic_state.sType             = vk::StructureType::ePipelineDynamicStateCreateInfo;
-	dynamic_state.pNext             = nullptr;
-	dynamic_state.flags             = {};
-	dynamic_state.dynamicStateCount = dynamic_states_count;
-	dynamic_state.pDynamicStates    = dynamic_states;
+	dynamic_state.dynamicStateCount = static_cast<uint32_t>(dynamic_states.size());
+	dynamic_state.pDynamicStates    = dynamic_states.data();
 
 	vk::GraphicsPipelineCreateInfo  pipeline_info {};
 	vk::PipelineRenderingCreateInfo rendering_info {};
-	rendering_info.sType                   = vk::StructureType::ePipelineRenderingCreateInfo;
 	rendering_info.colorAttachmentCount    = rendering.color_count;
 	rendering_info.pColorAttachmentFormats = rendering.color_formats.data();
 	rendering_info.depthAttachmentFormat   = rendering.depth_format;
 	rendering_info.stencilAttachmentFormat = rendering.stencil_format;
-	pipeline_info.sType                    = vk::StructureType::eGraphicsPipelineCreateInfo;
 	pipeline_info.pNext                    = &rendering_info;
-	pipeline_info.flags                    = {};
 	pipeline_info.stageCount               = shader_stage_count;
 	pipeline_info.pStages                  = shader_stages;
-	pipeline_info.pVertexInputState        = &vertex_input_info;
-	pipeline_info.pInputAssemblyState      = &input_assembly;
+	pipeline_info.pVertexInputState        = mesh ? nullptr : &vertex_input_info;
+	pipeline_info.pInputAssemblyState      = mesh ? nullptr : &input_assembly;
 	vk::PipelineTessellationStateCreateInfo tessellation_state {};
-	tessellation_state.sType              = vk::StructureType::ePipelineTessellationStateCreateInfo;
 	tessellation_state.patchControlPoints = 3;
 	pipeline_info.pTessellationState      = (rect_list ? &tessellation_state : nullptr);
 	pipeline_info.pViewportState          = &viewport_state;
 	pipeline_info.pRasterizationState     = &rasterizer;
 	pipeline_info.pMultisampleState       = &multisampling;
-	pipeline_info.pDepthStencilState = (static_params.with_depth ? &depth_stencil_info : nullptr);
-	pipeline_info.pColorBlendState   = &color_blending;
-	pipeline_info.pDynamicState      = &dynamic_state;
-	pipeline_info.layout             = pipeline.pipeline_layout;
-	pipeline_info.renderPass         = nullptr;
-	pipeline_info.subpass            = 0;
-	pipeline_info.basePipelineHandle = nullptr;
-	pipeline_info.basePipelineIndex  = -1;
+	pipeline_info.pDepthStencilState      = (with_depth ? &depth_stencil_info : nullptr);
+	pipeline_info.pColorBlendState        = &color_blending;
+	pipeline_info.pDynamicState           = &dynamic_state;
+	pipeline_info.layout                  = pipeline.pipeline_layout;
+	pipeline_info.basePipelineIndex       = -1;
 
 	EXIT_IF(pipeline.pipeline != nullptr);
 
 	if (graphics_debug_dump_enabled()) {
 		LOGF("PipelineTrace: vkCreateGraphicsPipelines begin VS=%" PRIu64 " PS=%" PRIu64
 		     " topology=%" PRIu32 " color_mask=0x%08" PRIx32
-		     " depth=%s blend=%s dyn_states=%" PRIu32 " viewport=%f,%f %fx%f"
-		     " scissor=%d,%d %ux%u\n",
-		     pipeline.vs_shader_id, pipeline.ps_shader_id,
-		     static_cast<uint32_t>(static_params.topology),
-		     static_params.color_mask[0], (static_params.with_depth ? "true" : "false"),
-		     (static_params.blend_enable[0] ? "true" : "false"), dynamic_states_count, viewport.x,
-		     viewport.y, viewport.width, viewport.height, scissor.offset.x, scissor.offset.y,
-		     scissor.extent.width, scissor.extent.height);
+		     " depth=%s blend=%s dyn_states=%" PRIu32 "\n",
+		     vertex_program.id, ps_active ? pixel_program.id : 0,
+		     static_cast<uint32_t>(static_params.topology), static_params.color_mask[0],
+		     (with_depth ? "true" : "false"), (static_params.blend_enable[0] ? "true" : "false"),
+		     dynamic_state.dynamicStateCount);
 	}
 	result = graphics.device.createGraphicsPipelines(driver_cache, 1, &pipeline_info, nullptr,
 	                                                 &pipeline.pipeline);
 	if (graphics_debug_dump_enabled()) {
 		LOGF("PipelineTrace: vkCreateGraphicsPipelines done result=%s pipeline=%p\n",
-		     VulkanToString(result).c_str(), static_cast<void*>(pipeline.pipeline));
+		     vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline));
 	}
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
@@ -960,26 +791,20 @@ void CreatePipelineInternal(
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::ComputePipeline& pipeline,
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
                             const ShaderComputeInputInfo& input_info,
-	vk::ShaderModule compute_module, vk::PipelineCache driver_cache) {
+                            vk::ShaderModule compute_module, vk::PipelineCache driver_cache) {
 	EXIT_IF(compute_module == nullptr);
 
 	vk::PipelineShaderStageCreateInfo                     comp_shader_stage_info {};
 	vk::PipelineShaderStageRequiredSubgroupSizeCreateInfo comp_subgroup_size {};
-	comp_shader_stage_info.sType               = vk::StructureType::ePipelineShaderStageCreateInfo;
-	comp_shader_stage_info.pNext               = nullptr;
-	comp_shader_stage_info.flags               = {};
-	comp_shader_stage_info.stage               = vk::ShaderStageFlagBits::eCompute;
-	comp_shader_stage_info.module              = compute_module;
-	comp_shader_stage_info.pName               = "main";
-	comp_shader_stage_info.pSpecializationInfo = nullptr;
+	comp_shader_stage_info.stage  = vk::ShaderStageFlagBits::eCompute;
+	comp_shader_stage_info.module = compute_module;
+	comp_shader_stage_info.pName  = "main";
 	EXIT_IF(!input_info.stage);
 	const auto wave_size = input_info.stage.program->wave_size;
 	if (graphics.compute_subgroup_size_control_enabled &&
 	    wave_size >= graphics.min_subgroup_size && wave_size <= graphics.max_subgroup_size) {
-		comp_subgroup_size.sType =
-		    vk::StructureType::ePipelineShaderStageRequiredSubgroupSizeCreateInfo;
 		comp_subgroup_size.requiredSubgroupSize = wave_size;
 		comp_shader_stage_info.pNext            = &comp_subgroup_size;
 	}
@@ -992,9 +817,6 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::ComputePipe
 	                                            ShaderRecompiler::IR::NativePushConstantSize};
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
-	pipeline_layout_info.sType                  = vk::StructureType::ePipelineLayoutCreateInfo;
-	pipeline_layout_info.pNext                  = nullptr;
-	pipeline_layout_info.flags                  = {};
 	pipeline_layout_info.setLayoutCount         = 1;
 	pipeline_layout_info.pSetLayouts            = &pipeline.descriptor_set_layout;
 	pipeline_layout_info.pushConstantRangeCount = 1;
@@ -1007,19 +829,15 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::ComputePipe
 	auto result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
 	                                                  &pipeline.pipeline_layout);
 	LOGF("PipelineTrace: vkCreatePipelineLayout CS done result=%s layout=%p\n",
-	     VulkanToString(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
+	     vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
 	EXIT_NOT_IMPLEMENTED(pipeline.pipeline_layout == nullptr);
 
 	vk::ComputePipelineCreateInfo info {};
-	info.sType              = vk::StructureType::eComputePipelineCreateInfo;
-	info.pNext              = nullptr;
-	info.flags              = {};
-	info.stage              = comp_shader_stage_info;
-	info.layout             = pipeline.pipeline_layout;
-	info.basePipelineHandle = nullptr;
-	info.basePipelineIndex  = -1;
+	info.stage             = comp_shader_stage_info;
+	info.layout            = pipeline.pipeline_layout;
+	info.basePipelineIndex = -1;
 
 	EXIT_IF(pipeline.pipeline != nullptr);
 
@@ -1028,7 +846,7 @@ void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::ComputePipe
 	result = graphics.device.createComputePipelines(driver_cache, 1, &info, nullptr,
 	                                                &pipeline.pipeline);
 	LOGF("PipelineTrace: vkCreateComputePipelines done result=%s pipeline=%p\n",
-	     VulkanToString(result).c_str(), static_cast<void*>(pipeline.pipeline));
+	     vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline));
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
 	EXIT_NOT_IMPLEMENTED(pipeline.pipeline == nullptr);

src/graphics/host_gpu/renderer/render.h

@@ -48,6 +48,31 @@ enum class CommandBufferDebugOp : uint32_t {
 	Unknown,
 };
 
+enum class DrawOffsetSource : uint8_t {
+	DrawState,
+	IndirectArgs,
+};
+
+struct DrawIndexArgs {
+	uint32_t         index_count                = 0;
+	const void*      index_addr                 = nullptr;
+	uint32_t         instance_count             = 0;
+	uint32_t         index_type_and_size        = 0;
+	int32_t          base_vertex                = 0;
+	uint32_t         first_instance             = 0;
+	DrawOffsetSource offset_source              = DrawOffsetSource::DrawState;
+	uint32_t         render_target_slice_offset = 0;
+};
+
+struct DrawAutoArgs {
+	uint32_t         vertex_count               = 0;
+	uint32_t         instance_count             = 0;
+	uint32_t         first_vertex               = 0;
+	uint32_t         first_instance             = 0;
+	DrawOffsetSource offset_source              = DrawOffsetSource::DrawState;
+	uint32_t         render_target_slice_offset = 0;
+};
+
 struct SubmitInfo {
 	static constexpr uint32_t MaxSemaphores = 3;
 
@@ -129,13 +154,6 @@ public:
 	explicit RenderExecutor(RenderContext& context): m_context(context) {}
 	KYTY_CLASS_NO_COPY(RenderExecutor);
 
-	void DrawIndex(uint64_t submit_id, CommandBuffer& buffer, uint32_t index_type_and_size,
-	               uint32_t index_count, const void* index_addr, uint32_t flags, uint32_t type,
-	               uint32_t instance_count = 1, uint32_t render_target_slice_offset = 0,
-	               int32_t vertex_offset_add = 0, uint32_t first_instance = 0);
-	void DrawAuto(uint64_t submit_id, CommandBuffer& buffer, uint32_t index_count, uint32_t flags,
-	              uint32_t render_target_slice_offset = 0, uint32_t instance_count = 1,
-	              uint32_t first_vertex = 0, uint32_t first_instance = 0);
 	void DispatchDirect(uint64_t submit_id, CommandBuffer& buffer, uint32_t thread_group_x,
 	                    uint32_t thread_group_y, uint32_t thread_group_z, uint32_t mode);
 
@@ -148,6 +166,9 @@ public:
 	                    std::span<PreparedBindings* const> bindings);
 
 private:
+	void DrawIndex(uint64_t submit_id, CommandBuffer& buffer, const DrawIndexArgs& args);
+	void DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const DrawAutoArgs& args);
+
 	struct GraphicsBindings {
 		PreparedBindings                vertex;
 		std::optional<PreparedBindings> pixel;
@@ -158,13 +179,11 @@ private:
 	[[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 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 PrepareDrawRenderState(CommandBuffer& buffer,
 	                                          const DrawCallInfo& draw,
 	                                          uint32_t            render_target_slice_offset,
 	                                          bool log_setup_phases, DrawRenderState& state);
@@ -174,8 +193,9 @@ private:
 	                         bool primitive_restart_enable, bool log_pipeline_phase,
 	                         bool set_bind_debug, bool set_auto_debug);
 	[[nodiscard]] RenderState AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
-	                                               uint32_t color_count, RenderDepthInfo& depth);
-	[[nodiscard]] bool        ResolveColorTargets(uint64_t submit_id, CommandBuffer& buffer,
+	                                               uint32_t color_count, RenderDepthInfo& depth,
+	                                               const std::optional<PreparedBindings>& pixel = std::nullopt);
+	[[nodiscard]] bool        ResolveColorTargets(CommandBuffer& buffer,
 	                                              uint32_t render_target_slice_offset);
 	void                      BindImage(ImageId id, bool storage);
 	void                      BindRenderTarget(ImageId id);
@@ -183,6 +203,9 @@ private:
 	void                      ResetBindings();
 	[[nodiscard]] bool        TryConsumeComputeMetaClear(const ShaderComputeInputInfo& input,
 	                                                     const CommandBuffer&          buffer);
+	[[nodiscard]] bool TryConsumeComputeImageClear(const ShaderComputeInputInfo& input,
+	                                              CommandBuffer& command, uint32_t group_x,
+	                                              uint32_t group_y, uint32_t group_z, uint32_t mode);
 
 	RenderContext&                        m_context;
 	std::vector<ImageId>                  m_bound_images;
@@ -190,13 +213,12 @@ private:
 	std::vector<vk::DescriptorImageInfo>  m_descriptor_images;
 	std::vector<vk::WriteDescriptorSet>   m_descriptor_writes;
 	std::vector<uint32_t>                 m_image_occurrences;
-	std::array<uint32_t, ShaderRecompiler::IR::NativePushConstantSize / sizeof(uint32_t)>
-	    m_push_constants {};
 
+	friend class CommandProcessor;
 	friend struct RenderExecutorTestAccess;
 };
 
-[[nodiscard]] bool ResolveComputeImageClear(const ShaderComputeInputInfo& input, uint32_t group_x,
+[[nodiscard]] bool ResolveComputeBufferFill(const ShaderComputeInputInfo& input, uint32_t group_x,
                                             uint32_t group_y, uint32_t group_z, uint32_t mode,
                                             ShaderBufferResource& descriptor,
                                             uint32_t& packed_clear, uint64_t& size);

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -25,10 +25,12 @@
 #include "libs/errno.h"
 
 #include <algorithm>
+#include <array>
 #include <atomic>
 #include <cmath>
 #include <cstring>
 #include <limits>
+#include <mutex>
 #include <span>
 #include <unordered_map>
 #include <vector>
@@ -37,16 +39,25 @@ namespace Libs::Graphics {
 static uint64_t BufferDescriptorSize(const ShaderBufferResource& descriptor) {
 	const uint64_t records = descriptor.NumRecords();
 	const uint64_t stride  = descriptor.Stride();
-	if (stride != 0 && records > UINT64_MAX / stride) {
-		EXIT("compute buffer descriptor footprint overflow\n");
-	}
 	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);
+		if (source.Base48() < destination.End() && destination.address < source.Base48() + bytes)
+			return false;
+	}
+	return true;
+}
+
 bool RenderExecutor::TryConsumeComputeMetaClear(const ShaderComputeInputInfo& input,
                                                 const CommandBuffer&          buffer) {
 	const auto& program   = *input.stage.program;
-	const auto& resources = *input.stage.resources;
+	const auto& resources = input.stage.resources;
 	if (resources.buffers.size() != program.info.buffers.size()) {
 		EXIT("compute runtime buffer count does not match shader metadata\n");
 	}
@@ -76,82 +87,115 @@ bool RenderExecutor::TryConsumeComputeMetaClear(const ShaderComputeInputInfo& in
 	return false;
 }
 
-bool ResolveComputeImageClear(const ShaderComputeInputInfo& input, uint32_t group_x,
+bool ResolveComputeBufferFill(const ShaderComputeInputInfo& input, uint32_t group_x,
                               uint32_t group_y, uint32_t group_z, uint32_t mode,
                               ShaderBufferResource& resolved_descriptor, uint32_t& resolved_clear,
                               uint64_t& resolved_size) {
-	const auto& program   = *input.stage.program;
-	const auto& resources = *input.stage.resources;
-	if (program.info.buffers.size() != 1 || resources.buffers.size() != 1 ||
-	    !program.info.images.empty() || !program.info.samplers.empty() || program.info.uses_dma ||
-	    !resources.images.empty() || !resources.samplers.empty()) {
+	const auto& resources = input.stage.resources;
+	const auto& fill      = resources.uniform_fill;
+	if (fill.kind != ShaderRecompiler::IR::UniformFillKind::Buffer) {
 		return false;
 	}
-	const auto& resource   = program.info.buffers.front();
-	const auto& raw        = resources.buffers.front();
-	const auto  descriptor = DecodeNativeDescriptor<ShaderBufferResource>(raw);
-	if (!resource.formatted || !resource.written || resource.read || resource.atomic ||
-	    resource.scalar || resource.max_byte_extent != 16 || descriptor.Stride() != 16 ||
-	    descriptor.Format() != Prospero::BufferFormat::k32_32_32_32UInt ||
+	const auto element_size = fill.words * sizeof(uint32_t);
+	const auto descriptor =
+	    DecodeNativeDescriptor<ShaderBufferResource>(resources.buffers[fill.resource]);
+	constexpr std::array formats {
+	    Prospero::BufferFormat::k32UInt, Prospero::BufferFormat::k32_32UInt,
+	    Prospero::BufferFormat::k32_32_32UInt, Prospero::BufferFormat::k32_32_32_32UInt};
+	if (descriptor.Stride() != element_size || descriptor.Format() != formats[fill.words - 1] ||
 	    descriptor.SwizzleEnabled() || descriptor.IndexStride() != 0 || descriptor.AddTid() ||
-	    resource.packed_stride != descriptor.PackedStride() || raw.dword_count != 4 ||
-	    program.user_data_base != 0 || resources.user_data.size() != 8) {
+	    descriptor.Base48() == 0) {
 		return false;
 	}
-	for (uint32_t i = 0; i < raw.dword_count; i++) {
-		if (raw.dwords[i] != resources.user_data[i]) {
-			return false;
-		}
-	}
-	const uint32_t clear = resources.user_data[4];
-	if (resources.user_data[5] != clear || resources.user_data[6] != clear ||
-	    resources.user_data[7] != clear) {
+	if (input.threads_num[0] == 0 || input.threads_num[0] != fill.group_stride[0] ||
+	    input.threads_num[1] != 1 || input.threads_num[2] != 1 || group_x == 0 || group_y != 1 ||
+	    group_z != 1 || mode != (input.dispatch_thread_dimensions ? 0x61u : 0x41u)) {
 		return false;
 	}
-	const bool full_dispatch =
-	    input.dispatch_thread_dimensions && input.threads_num[0] == 64 &&
-	    input.threads_num[1] == 1 && input.threads_num[2] == 1 && group_x != 0 && group_y == 1 &&
-	    group_z == 1 && input.dispatch_threads_num[0] == group_x &&
-	    input.dispatch_threads_num[1] == 1 && input.dispatch_threads_num[2] == 1 &&
-	    input.group_id[0] && !input.group_id[1] && !input.group_id[2] &&
-	    input.thread_ids_num == 1 && input.wave_size == 64 && !input.tg_size_en && mode == 0x61u &&
-	    group_x % input.threads_num[0] == 0 && descriptor.NumRecords() == group_x;
-	const auto size = BufferDescriptorSize(descriptor);
-	if (!full_dispatch || size == 0) {
+	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);
+	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 ||
+	      input.dispatch_threads_num[1] != 1 || input.dispatch_threads_num[2] != 1))) {
 		return false;
 	}
+	if (!FillSourcesDisjoint(resources.buffers, {descriptor.Base48(), size}, fill.resource))
+		return false;
 	resolved_descriptor = descriptor;
-	resolved_clear      = clear;
+	resolved_clear      = fill.value;
 	resolved_size       = size;
 	return true;
 }
 
-static bool TryConsumeComputeImageClear(const ShaderComputeInputInfo& input, CommandBuffer& command,
-                                        uint32_t group_x, uint32_t group_y, uint32_t group_z,
-                                        uint32_t mode) {
+bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& input,
+                                                CommandBuffer& command, uint32_t group_x,
+                                                uint32_t group_y, uint32_t group_z, uint32_t mode) {
+	const auto& program   = *input.stage.program;
+	const auto& resources = input.stage.resources;
+	const auto& fill      = resources.uniform_fill;
+	auto&       cache     = command.GetContext().GetTextureCache();
+	if (fill.kind == ShaderRecompiler::IR::UniformFillKind::Image) {
+		if (mode != 0x41u || input.dispatch_thread_dimensions || fill.value > 255 ||
+		    input.threads_num[2] != 1)
+			return false;
+		const auto  descriptor = DecodeNativeDescriptor<ShaderTextureResource>(resources.images[0]);
+		const auto& resource   = program.info.images[0];
+		if (descriptor.IsNull() || descriptor.Format() != Prospero::BufferFormat::k8UInt ||
+		    descriptor.Type() != Prospero::ImageType::kColor2DArray || descriptor.MetaCompress() ||
+		    descriptor.WriteCompress() || descriptor.BaseLevel() > descriptor.LastLevel() ||
+		    descriptor.BaseLevel() > descriptor.MaxMip() ||
+		    descriptor.BaseArray5() > descriptor.Depth() || descriptor.DstSelX() != 4)
+			return false;
+		const std::array extents {
+		    std::max(1u, (descriptor.Width5() + 1u) >> descriptor.BaseLevel()),
+		    std::max(1u, (descriptor.Height5() + 1u) >> descriptor.BaseLevel()),
+		    descriptor.Depth() - descriptor.BaseArray5() + 1u};
+		const std::array groups {group_x, group_y, group_z};
+		for (uint32_t axis = 0; axis < 3; ++axis) {
+			const uint64_t threads = input.threads_num[axis];
+			// Guest image writes outside the descriptor dimensions are discarded. Only the
+			// final workgroup may extend beyond the selected image view.
+			if (threads == 0 || threads != fill.group_stride[axis] ||
+			    groups[axis] != (extents[axis] + threads - 1) / threads ||
+			    groups[axis] * threads > UINT32_MAX) return false;
+		}
+		const auto  binding     = ResolveTexture(resource, resources.images[0]);
+		const auto& destination = binding.desc.info.data;
+		if (!FillSourcesDisjoint(resources.buffers, destination)) return false;
+		std::scoped_lock lock {cache.m_lock};
+		const auto&      image = cache.GetImage(binding.image_id);
+		const auto&      view  = binding.desc.view_info;
+		if (image.backing.format != vk::Format::eD32SfloatS8Uint || image.info.samples != 1 ||
+		    image.info.stencil != destination || view.base_level >= image.backing.mip_levels ||
+		    view.base_layer >= image.backing.layers || view.layer_count != extents[2] ||
+		    view.layer_count > image.backing.layers - view.base_layer ||
+		    std::max(1u, image.info.extent.width >> view.base_level) != extents[0] ||
+		    std::max(1u, image.info.extent.height >> view.base_level) != extents[1]) return false;
+		const vk::ImageSubresourceRange range {vk::ImageAspectFlagBits::eStencil, view.base_level,
+		                                       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);
+		return true;
+	}
 	ShaderBufferResource descriptor;
 	uint32_t             packed_clear = 0;
 	uint64_t             size         = 0;
-	if (!ResolveComputeImageClear(input, group_x, group_y, group_z, mode, descriptor, packed_clear,
+	if (!ResolveComputeBufferFill(input, group_x, group_y, group_z, mode, descriptor, packed_clear,
 	                              size)) {
 		return false;
 	}
-	auto& cache = command.GetContext().GetTextureCache();
 	if (!cache.ClearImageFromBuffer(command, descriptor.Base48(), size, packed_clear)) {
-		// Recognized metadata-fill shaders access DCC as an ordinary storage buffer and may run
-		// before the render target is bound. TryConsumeDccFill either consumes registered state
-		// or retains a PendingDcc fill while allowing the dispatch to run.
-		const bool registered_metadata =
-		    cache.TryConsumeDccFill(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: %s metadata clear shader=0x%016" PRIx64
+			LOGF("GraphicsRenderDispatchDirect: metadata fill shader=0x%016" PRIx64
 			     " addr=0x%016" PRIx64 " size=0x%016" PRIx64 " value=0x%08" PRIx32 "\n",
-			     registered_metadata ? "tracked" : "deferred", input.stage.program->shader_hash,
-			     descriptor.Base48(), size, packed_clear);
-		}
-		if (registered_metadata) {
-			return true;
+			     input.stage.program->shader_hash, descriptor.Base48(), size, packed_clear);
 		}
 		return false;
 	}
@@ -224,7 +268,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	const bool     large_workgroup =
 	    (input_info.threads_num[0] * input_info.threads_num[1] * input_info.threads_num[2] >= 512);
 	const auto& program   = *input_info.stage.program;
-	const auto& resources = *input_info.stage.resources;
+	const auto& resources = input_info.stage.resources;
 	if (TryConsumeComputeMetaClear(input_info, buffer)) {
 		ResetBindings();
 		return;
@@ -236,8 +280,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	}
 	const auto sampled_images = std::count_if(
 	    program.info.images.begin(), program.info.images.end(), [](const auto& image) {
-		    return image.kind == ShaderRecompiler::IR::ResourceKind::Image ||
-		           image.kind == ShaderRecompiler::IR::ResourceKind::ImageUint;
+		    return image.resource_class == ShaderRecompiler::IR::ImageResourceClass::Sampled;
 	    });
 	const bool                   has_sampler = !program.info.samplers.empty();
 	static std::atomic<uint32_t> dispatch_log_count {0};
@@ -250,7 +293,10 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		     thread_group_z, mode, input_info.threads_num[0], input_info.threads_num[1],
 		     input_info.threads_num[2], program.info.buffers.size(), program.info.images.size(),
 		     sampled_images, program.info.images.size() - sampled_images,
-		     program.info.samplers.size(), program.bindings.push_constant_size);
+		     program.info.samplers.size(),
+		     program.bindings.UsesPushData()
+		         ? static_cast<uint32_t>(sizeof(ShaderRecompiler::IR::PushData))
+		         : 0u);
 		for (uint32_t i = 0; i < program.info.buffers.size(); i++) {
 			const auto& buffer = program.info.buffers[i];
 			const auto  r      = DecodeNativeDescriptor<ShaderBufferResource>(resources.buffers[i]);
@@ -265,8 +311,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 			LOGF("  CS texture[%u]: source=%u usage=%s sampled=%s addr=0x%010" PRIx64
 			     " type=%u fmt=%u extent=%ux%u depth=%u levels=%u tile=%u\n",
 			     i, image.source, image.written ? "read-write" : "read-only",
-			     (image.kind == ShaderRecompiler::IR::ResourceKind::Image ||
-			      image.kind == ShaderRecompiler::IR::ResourceKind::ImageUint)
+			     image.resource_class == ShaderRecompiler::IR::ImageResourceClass::Sampled
 			         ? "true"
 			         : "false",
 			     r.Base40(), static_cast<uint32_t>(r.Type()), static_cast<uint32_t>(r.Format()),
@@ -347,8 +392,8 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	    std::any_of(program.info.images.begin(), program.info.images.end(),
 	                [](const auto& image) {
 		                return image.written &&
-		                       (image.kind == ShaderRecompiler::IR::ResourceKind::StorageImage ||
-		                        image.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint);
+		                       image.resource_class ==
+		                           ShaderRecompiler::IR::ImageResourceClass::Storage;
 	                }) ||
 	    has_storage_writes;
 	if (has_storage_writes) {

src/graphics/host_gpu/renderer/renderDraw.h

@@ -7,8 +7,9 @@ namespace Libs::Graphics {
 
 struct ShaderVertexInputInfo;
 
-[[nodiscard]] int32_t ResolveVertexOffset(uint32_t                     index_offset,
-                                          const ShaderVertexInputInfo& vs_input_info);
+[[nodiscard]] int32_t  ResolveVertexOffset(uint32_t                     index_offset,
+                                           const ShaderVertexInputInfo& vs_input_info);
+[[nodiscard]] uint32_t ResolveInstanceOffset(const ShaderVertexInputInfo& vs_input_info);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/renderer/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/vma.cpp

@@ -18,45 +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;
-
-void TrackAllocationImpl(const VulkanMemory& memory) {
-	g_memory_stats.allocated[memory.type] += memory.requirements.size;
-	g_memory_stats.count[memory.type]++;
-}
-
-void UntrackAllocationImpl(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]--;
-}
-
-} // namespace
-
-void VulkanTrackAllocation(const VulkanMemory& memory) {
-	TrackAllocationImpl(memory);
-}
-
-void VulkanUntrackAllocation(const VulkanMemory& memory) {
-	UntrackAllocationImpl(memory);
-}
-
 bool GraphicContext::CreateAllocator() {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(instance == nullptr || physical_device == nullptr || device == nullptr ||
@@ -79,7 +46,7 @@ bool GraphicContext::CreateAllocator() {
 
 	const auto result = static_cast<vk::Result>(vmaCreateAllocator(&info, &allocator));
 	if (result != vk::Result::eSuccess) {
-		LOGF("vmaCreateAllocator failed: %s\n", VulkanToString(result).c_str());
+		LOGF("vmaCreateAllocator failed: %s\n", vk::to_string(result).c_str());
 		return false;
 	}
 	return true;
@@ -93,11 +60,6 @@ void GraphicContext::DestroyAllocator() {
 	allocator = nullptr;
 }
 
-uint64_t VulkanNextMemoryUniqueId() {
-	static std::atomic_uint64_t sequence = 0;
-	return ++sequence;
-}
-
 void GraphicContext::LogMemoryBudget() const {
 	if (allocator == nullptr || physical_device == nullptr) {
 		return;
@@ -166,96 +128,44 @@ uint64_t GraphicContext::GetTotalMemoryBudget() const {
 	return std::max(local, available > system_reserve ? available - system_reserve : uint64_t {0});
 }
 
-void GraphicContext::CreateBuffer(uint64_t size, VulkanBuffer& buffer) {
-	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(allocator == nullptr || buffer.buffer != nullptr ||
-	        buffer.memory.allocation != nullptr || size == 0);
-
-	vk::BufferCreateInfo buffer_info {};
-	buffer_info.sType       = vk::StructureType::eBufferCreateInfo;
-	buffer_info.size        = size;
-	buffer_info.usage       = buffer.usage;
-	buffer_info.sharingMode = vk::SharingMode::eExclusive;
-
-	VmaAllocationCreateInfo alloc_info {};
-	alloc_info.requiredFlags =
-	    static_cast<vk::MemoryPropertyFlags::MaskType>(buffer.memory.property);
-	alloc_info.preferredFlags =
-	    static_cast<vk::MemoryPropertyFlags::MaskType>(buffer.memory.preferred_property);
-
-	vk::Buffer::CType native_buffer = VK_NULL_HANDLE;
-	const auto        result        = static_cast<vk::Result>(vmaCreateBuffer(
-	    allocator, static_cast<const vk::BufferCreateInfo::NativeType*>(buffer_info), &alloc_info,
-	    &native_buffer, &buffer.memory.allocation, &buffer.memory.allocation_info));
-	buffer.buffer                   = native_buffer;
-	if (result != vk::Result::eSuccess) {
-		LogMemoryBudget();
-	}
-	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
-
-	device.getBufferMemoryRequirements(buffer.buffer, &buffer.memory.requirements);
-	buffer.memory.type      = buffer.memory.allocation_info.memoryType;
-	buffer.memory.memory    = buffer.memory.allocation_info.deviceMemory;
-	buffer.memory.offset    = buffer.memory.allocation_info.offset;
-	buffer.memory.unique_id = VulkanNextMemoryUniqueId();
-	buffer.buffer_size      = size;
-	VulkanTrackAllocation(buffer.memory);
-}
-
 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;
 
 	vk::Image::CType native_image = VK_NULL_HANDLE;
-	const auto       result       = static_cast<vk::Result>(
+	const auto        result       = static_cast<vk::Result>(
 	    vmaCreateImage(allocator, static_cast<const vk::ImageCreateInfo::NativeType*>(image_info),
-	                   &alloc_info, &native_image, &memory.allocation, &memory.allocation_info));
+	                   &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      = memory.allocation_info.memoryType;
-	memory.memory    = memory.allocation_info.deviceMemory;
-	memory.offset    = memory.allocation_info.offset;
-	memory.unique_id = VulkanNextMemoryUniqueId();
-	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.memory          = nullptr;
-	memory.allocation      = nullptr;
-	memory.allocation_info = {};
-	memory.offset          = 0;
-}
-
-void GraphicContext::MapMemory(VulkanMemory& memory, void*& data) {
-	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(allocator == nullptr || memory.allocation == nullptr);
-	EXIT_NOT_IMPLEMENTED(static_cast<vk::Result>(vmaMapMemory(allocator, memory.allocation,
-	                                                          &data)) != vk::Result::eSuccess);
-}
-
-void GraphicContext::UnmapMemory(VulkanMemory& memory) {
-	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(allocator == nullptr || memory.allocation == nullptr);
-	vmaUnmapMemory(allocator, memory.allocation);
+	vmaDestroyImage(allocator, image.image, image.allocation);
+	image.image      = nullptr;
+	image.allocation = nullptr;
 }
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/vma.h

@@ -1,15 +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 {
-
-uint64_t VulkanNextMemoryUniqueId();
-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

@@ -65,7 +65,6 @@ constexpr FormatMapping kFormatMappings[] = {
     {Prospero::BufferFormat::k5_6_5UNorm, vk::Format::eB5G6R5UnormPack16},
     {Prospero::BufferFormat::k5_5_5_1UNorm, vk::Format::eR5G5B5A1UnormPack16},
     {Prospero::BufferFormat::k4_4_4_4UNorm, vk::Format::eR4G4B4A4UnormPack16},
-    {Prospero::BufferFormat::kFmask8_S4_F4, vk::Format::eR32Sfloat},
     {Prospero::BufferFormat::kBc1UNorm, vk::Format::eBc1RgbaUnormBlock},
     {Prospero::BufferFormat::kBc1Srgb, vk::Format::eBc1RgbaSrgbBlock},
     {Prospero::BufferFormat::kBc2UNorm, vk::Format::eBc2UnormBlock},
@@ -96,22 +95,6 @@ constexpr auto kFormatLookup = MakeFormatLookup();
 
 } // namespace
 
-std::string VulkanToString(vk::Result value) {
-	return vk::to_string(value);
-}
-
-std::string VulkanToString(vk::Format value) {
-	return vk::to_string(value);
-}
-
-std::string VulkanToString(vk::ImageLayout value) {
-	return vk::to_string(value);
-}
-
-std::string VulkanToString(vk::QueueFlags value) {
-	return vk::to_string(value);
-}
-
 vk::Format VulkanFormat(Prospero::BufferFormat guest_format) {
 	const auto index = static_cast<size_t>(guest_format);
 	return index < kFormatLookup.size() ? kFormatLookup[index] : vk::Format::eUndefined;
@@ -119,7 +102,7 @@ vk::Format VulkanFormat(Prospero::BufferFormat guest_format) {
 
 void RequireVulkanSuccess(vk::Result result, const char* operation) {
 	if (result != vk::Result::eSuccess) {
-		EXIT("%s failed: %s (%d)\n", operation, VulkanToString(result).c_str(),
+		EXIT("%s failed: %s (%d)\n", operation, vk::to_string(result).c_str(),
 		     static_cast<int>(result));
 	}
 }

src/graphics/host_gpu/vulkanCommon.h

@@ -25,10 +25,6 @@ namespace Libs::Graphics {
 
 using VulkanMemoryBarrier = vk::MemoryBarrier;
 
-std::string VulkanToString(vk::Result value);
-std::string VulkanToString(vk::Format value);
-std::string VulkanToString(vk::ImageLayout value);
-std::string VulkanToString(vk::QueueFlags value);
 vk::Format  VulkanFormat(Prospero::BufferFormat guest_format);
 void        RequireVulkanSuccess(vk::Result result, const char* operation);
 

src/graphics/presentation/imeOverlay.h

@@ -1,42 +0,0 @@
-#ifndef EMULATOR_SRC_GRAPHICS_PRESENTATION_IMEOVERLAY_H_
-#define EMULATOR_SRC_GRAPHICS_PRESENTATION_IMEOVERLAY_H_
-
-#include "common/common.h"
-#include "graphics/host_gpu/vulkanCommon.h"
-
-#include <memory>
-
-union SDL_Event;
-
-namespace Libs::Graphics {
-
-struct GraphicContext;
-
-struct ImeVisualState {
-	bool     active;
-	uint64_t revision;
-};
-
-void           InitializeImeInput();
-void           ShutdownImeInput();
-bool           ProcessImeInput(const SDL_Event& event);
-ImeVisualState GetImeVisualState() noexcept;
-
-class ImeOverlay final {
-public:
-	explicit ImeOverlay(GraphicContext& graphics);
-	~ImeOverlay();
-	KYTY_CLASS_NO_COPY(ImeOverlay);
-
-	[[nodiscard]] bool PrepareFrame(vk::Extent2D extent, vk::Format format, uint32_t image_count);
-	void               Record(vk::CommandBuffer command, vk::ImageView target);
-	void               ReleaseVulkan();
-
-private:
-	struct Impl;
-	std::unique_ptr<Impl> m_impl;
-};
-
-} // namespace Libs::Graphics
-
-#endif // EMULATOR_SRC_GRAPHICS_PRESENTATION_IMEOVERLAY_H_

src/graphics/presentation/presenter.h

@@ -25,7 +25,7 @@ public:
 	                                               CommandBuffer* producer = nullptr);
 	[[nodiscard]] Frame*         PrepareLastFrame();
 	[[nodiscard]] bool           IsGuestPaused() const noexcept;
-	[[nodiscard]] bool           NeedsImeRefresh() const noexcept;
+	[[nodiscard]] bool           NeedsSystemOverlayRefresh() const noexcept;
 	[[nodiscard]] RenderContext& Renderer() const noexcept;
 	void                         Present(Frame& frame, bool reuse = false);
 	void                         Discard(Frame& frame);

src/graphics/presentation/renderDoc.cpp

@@ -1,6 +1,8 @@
 #include "graphics/presentation/renderDoc.h"
 
 #include "common/logging/log.h"
+#include "graphics/host_gpu/graphicContext.h"
+#include "graphics/host_gpu/renderer/renderContext.h"
 
 #include <array>
 #include <atomic>
@@ -25,13 +27,12 @@ namespace Libs::Graphics {
 enum class RenderDocState : uint32_t {
 	Idle,
 	Requested,
-	Starting,
 	Capturing,
 };
 
 static RENDERDOC_API_1_6_0*        g_api             = nullptr;
 static std::atomic<RenderDocState> g_state           = RenderDocState::Idle;
-static std::atomic_uint32_t        g_captured_flips  = 0;
+static uint32_t                    g_captured_flips  = 0;
 static std::atomic_bool            g_unavailable_log = false;
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
@@ -132,21 +133,7 @@ void RenderDocRequestCapture() {
 	}
 }
 
-bool RenderDocCaptureRequested() {
-	return g_state.load(std::memory_order_acquire) == RenderDocState::Requested;
-}
-
-bool RenderDocCaptureInProgress() {
-	return g_state.load(std::memory_order_acquire) == RenderDocState::Capturing;
-}
-
-void RenderDocStartCapture() {
-	RenderDocState expected = RenderDocState::Requested;
-	if (g_api == nullptr || !g_state.compare_exchange_strong(expected, RenderDocState::Starting,
-	                                                         std::memory_order_acq_rel)) {
-		return;
-	}
-
+static void StartCapture() {
 	if (g_api->IsFrameCapturing() != 0) {
 		g_state.store(RenderDocState::Idle, std::memory_order_release);
 		LOGF("RenderDoc: capture request ignored because a capture is already active\n");
@@ -164,29 +151,32 @@ void RenderDocStartCapture() {
 		LOGF("RenderDoc: capture failed to start\n");
 		return;
 	}
-	g_captured_flips.store(0, std::memory_order_release);
+	g_captured_flips = 0;
 	g_state.store(RenderDocState::Capturing, std::memory_order_release);
 	LOGF("RenderDoc: capture started\n");
 }
 
-void RenderDocEndCapture() {
-	if (g_api == nullptr || !RenderDocCaptureInProgress()) {
+void RenderDocOnGuestFlip(RenderContext& renderer) {
+	const auto state = g_state.load(std::memory_order_acquire);
+	if (g_api == nullptr || state == RenderDocState::Idle) {
 		return;
 	}
-
-	const auto ok = g_api->EndFrameCapture(nullptr, nullptr);
-	g_state.store(RenderDocState::Idle, std::memory_order_release);
-	LOGF(ok != 0 ? "RenderDoc: capture finished\n" : "RenderDoc: capture failed\n");
-}
-
-void RenderDocOnGuestFlip() {
-	if (!RenderDocCaptureInProgress()) {
-		return;
+	if (state == RenderDocState::Capturing) {
+		LOGF("RenderDoc: captured guest flip %u/2\n", ++g_captured_flips);
+		if (g_captured_flips < 2) {
+			return;
+		}
 	}
-	const auto flip = g_captured_flips.fetch_add(1, std::memory_order_acq_rel) + 1;
-	LOGF("RenderDoc: captured guest flip %u/2\n", flip);
-	if (flip >= 2) {
-		RenderDocEndCapture();
+
+	// Capture boundaries follow presentation and exclude concurrent queue access.
+	Common::LockGuard render_lock(renderer.GetMutex());
+	Common::LockGuard queue_lock(renderer.GetGraphics().queue_mutex);
+	if (state == RenderDocState::Requested) {
+		StartCapture();
+	} else {
+		const auto ok = g_api->EndFrameCapture(nullptr, nullptr);
+		g_state.store(RenderDocState::Idle, std::memory_order_release);
+		LOGF(ok != 0 ? "RenderDoc: capture finished\n" : "RenderDoc: capture failed\n");
 	}
 }
 

src/graphics/presentation/renderDoc.h

@@ -3,14 +3,12 @@
 
 namespace Libs::Graphics {
 
+class RenderContext;
+
 void RenderDocInit();
 void RenderDocRequestCapture();
-void RenderDocStartCapture();
-void RenderDocEndCapture();
-void RenderDocOnGuestFlip();
-
-[[nodiscard]] bool RenderDocCaptureRequested();
-[[nodiscard]] bool RenderDocCaptureInProgress();
+// Called by the presentation thread after releasing video-out locks.
+void RenderDocOnGuestFlip(RenderContext& renderer);
 
 } // namespace Libs::Graphics
 

src/graphics/presentation/systemOverlay.cpp

@@ -1,4 +1,4 @@
-#include "graphics/presentation/imeOverlay.h"
+#include "graphics/presentation/systemOverlay.h"
 
 #include "SDL.h"
 #include "common/assert.h"
@@ -7,6 +7,7 @@
 #include "imgui.h"
 #include "imgui_impl_vulkan.h"
 #include "libs/controller.h"
+#include "libs/dialog.h"
 #include "libs/ime.h"
 #include "libs/imeDialog.h"
 
@@ -29,6 +30,7 @@ namespace {
 
 namespace CoreIme   = Libs::Ime;
 namespace DialogIme = Libs::Dialog::ImeDialog;
+namespace ErrorDialog = Libs::Dialog::ErrorDialog;
 
 namespace Ime {
 
@@ -101,6 +103,34 @@ bool HostQueueExternalInput(uint64_t generation, ExternalInput input) {
 
 } // namespace Ime
 
+enum class OverlayKind : uint8_t { None, Ime, Error };
+
+struct OverlaySession {
+	OverlayKind kind       = OverlayKind::None;
+	uint64_t    generation = 0;
+
+	bool operator==(const OverlaySession&) const = default;
+};
+
+struct OverlaySnapshot {
+	OverlaySession            session;
+	Ime::HostSnapshot         ime;
+	ErrorDialog::HostSnapshot error;
+};
+
+bool GetOverlaySnapshot(OverlaySnapshot* snapshot) {
+	if (ErrorDialog::GetHostSnapshot(&snapshot->error)) {
+		snapshot->session = {OverlayKind::Error, snapshot->error.generation};
+		return true;
+	}
+	if (Ime::GetHostSnapshot(&snapshot->ime)) {
+		snapshot->session = {OverlayKind::Ime, snapshot->ime.generation};
+		return true;
+	}
+	snapshot->session = {};
+	return false;
+}
+
 constexpr size_t INPUT_QUEUE_CAPACITY = 128;
 
 enum class InputKind : uint8_t {
@@ -113,19 +143,20 @@ enum class InputKind : uint8_t {
 };
 
 struct InputEvent {
-	InputKind kind;
-	uint64_t  generation;
-	int       id;
-	float     x;
-	float     y;
+	InputKind      kind;
+	OverlaySession session;
+	int            id;
+	float          x;
+	float          y;
 };
 
 struct VisibilityUpdate {
-	uint64_t generation;
-	bool     visible;
-	bool     capture_controller;
-	bool     capture_keyboard;
-	bool     multiline;
+	OverlaySession session;
+	bool           visible;
+	bool           capture_controller;
+	bool           capture_keyboard;
+	bool           text_input;
+	bool           multiline;
 };
 
 std::atomic<Uint32>          g_visibility_event {static_cast<Uint32>(-1)};
@@ -137,15 +168,14 @@ size_t                       g_missing_visibility_wakeups = 0;
 bool                         g_input_reset_requested      = false;
 uint16_t                     g_last_external_keycode      = 0;
 uint32_t                     g_last_external_status       = 0;
-uint64_t                     g_input_generation           = 0;
+OverlaySession               g_input_session;
 bool                         g_input_active               = false;
 bool                         g_input_controller           = false;
 bool                         g_input_keyboard             = false;
 bool                         g_input_multiline            = false;
 bool                         g_input_lifecycle_active     = false;
-bool                         g_active                     = false;
 bool                         g_controller_captured        = false;
-uint64_t                     g_generation                 = 0;
+OverlaySession               g_session;
 
 void ClearInputEvents() {
 	std::scoped_lock lock(g_input_mutex);
@@ -159,7 +189,7 @@ void QueueInput(InputEvent event) {
 	    event.kind == InputKind::Axis || event.kind == InputKind::MousePosition;
 	if (replaceable && !g_input_events.empty()) {
 		auto& last = g_input_events.back();
-		if (last.generation == event.generation && last.kind == event.kind && last.id == event.id) {
+		if (last.session == event.session && last.kind == event.kind && last.id == event.id) {
 			last = event;
 			return;
 		}
@@ -192,34 +222,36 @@ void RefreshVisibility() {
 	if (!g_input_lifecycle_active) {
 		return;
 	}
-	Ime::HostSnapshot snapshot;
-	const bool        visible    = Ime::GetHostSnapshot(&snapshot);
-	const uint64_t    generation = visible ? snapshot.generation : g_generation;
-	if (visible == g_active && (!visible || generation == g_generation)) {
+	OverlaySnapshot snapshot;
+	const bool      visible = GetOverlaySnapshot(&snapshot);
+	if (snapshot.session == g_session) {
 		return;
 	}
-	g_generation            = generation;
+	g_session               = snapshot.session;
 	bool capture_controller = false;
 	bool capture_keyboard   = false;
+	bool text_input         = false;
 	bool multiline          = false;
-	if (visible) {
-		capture_controller = (snapshot.disable_device & Ime::DISABLE_DEVICE_CONTROLLER) == 0;
-		capture_keyboard   = (snapshot.disable_device & Ime::DISABLE_DEVICE_EXT_KEYBOARD) == 0;
-		multiline          = (snapshot.option & Ime::OPTION_MULTILINE) != 0;
+	if (snapshot.session.kind == OverlayKind::Error) {
+		capture_controller = true;
+		capture_keyboard   = true;
+	} else if (snapshot.session.kind == OverlayKind::Ime) {
+		capture_controller = (snapshot.ime.disable_device & Ime::DISABLE_DEVICE_CONTROLLER) == 0;
+		capture_keyboard   = (snapshot.ime.disable_device & Ime::DISABLE_DEVICE_EXT_KEYBOARD) == 0;
+		text_input         = capture_keyboard;
+		multiline          = (snapshot.ime.option & Ime::OPTION_MULTILINE) != 0;
 	}
 	const bool was_controller = std::exchange(g_controller_captured, capture_controller);
 	if (capture_controller || was_controller) {
-		Controller::ControllerResetInputState();
+		Controller::ResetInputState();
 	}
-	g_active = visible;
-
 	const Uint32 type = g_visibility_event.load(std::memory_order_acquire);
 	if (type != static_cast<Uint32>(-1)) {
 		SDL_Event event {};
 		event.type = type;
 		std::scoped_lock input_lock(g_input_mutex);
-		g_visibility_updates.push_back(
-		    {generation, visible, capture_controller, capture_keyboard, multiline});
+		g_visibility_updates.push_back({snapshot.session, visible, capture_controller,
+		                                capture_keyboard, text_input, multiline});
 		if (g_missing_visibility_wakeups != 0 || SDL_PushEvent(&event) <= 0) {
 			g_missing_visibility_wakeups++;
 		}
@@ -383,7 +415,7 @@ void CheckVulkanResult(VkResult result) {
 
 } // namespace
 
-void InitializeImeInput() {
+void InitializeSystemOverlayInput() {
 	const Uint32 type = SDL_RegisterEvents(1);
 	EXIT_IF(type == static_cast<Uint32>(-1));
 	{
@@ -393,18 +425,20 @@ void InitializeImeInput() {
 	}
 	CoreIme::SetVisibilityCallback(OnCoreVisibilityChanged);
 	DialogIme::SetVisibilityCallback(OnDialogVisibilityChanged);
+	ErrorDialog::SetVisibilityCallback(RefreshVisibility);
 	RefreshVisibility();
 }
 
-void ShutdownImeInput() {
+void ShutdownSystemOverlayInput() {
 	CoreIme::SetVisibilityCallback(nullptr);
 	DialogIme::SetVisibilityCallback(nullptr);
+	ErrorDialog::SetVisibilityCallback(nullptr);
 	{
 		std::scoped_lock lock(g_visibility_mutex);
 		g_input_lifecycle_active = false;
-		g_active                 = false;
+		g_session                = {};
 		if (std::exchange(g_controller_captured, false)) {
-			Controller::ControllerResetInputState();
+			Controller::ResetInputState();
 		}
 		g_visibility_event.store(static_cast<Uint32>(-1), std::memory_order_release);
 	}
@@ -415,6 +449,7 @@ void ShutdownImeInput() {
 		g_missing_visibility_wakeups = 0;
 	}
 	g_input_active     = false;
+	g_input_session    = {};
 	g_input_controller = false;
 	g_input_keyboard   = false;
 	g_input_multiline  = false;
@@ -423,13 +458,15 @@ void ShutdownImeInput() {
 	}
 }
 
-ImeVisualState GetImeVisualState() noexcept {
+SystemOverlayVisualState GetSystemOverlayVisualState() noexcept {
 	const auto core   = CoreIme::GetVisualState();
 	const auto dialog = DialogIme::GetVisualState();
-	return {core.active || dialog.active, core.revision + dialog.revision};
+	const auto error  = ErrorDialog::GetVisualState();
+	return {core.active || dialog.active || error.active,
+	        core.revision + dialog.revision + error.revision};
 }
 
-bool ProcessImeInput(const SDL_Event& event) {
+bool ProcessSystemOverlayInput(const SDL_Event& event) {
 	RetryVisibilityWakeup();
 	if (event.type == g_visibility_event.load(std::memory_order_acquire)) {
 		VisibilityUpdate update {};
@@ -444,12 +481,12 @@ bool ProcessImeInput(const SDL_Event& event) {
 		ClearInputEvents();
 		g_last_external_keycode = 0;
 		g_last_external_status  = 0;
-		g_input_generation      = update.generation;
+		g_input_session         = update.session;
 		g_input_active          = update.visible;
 		g_input_controller      = update.capture_controller;
 		g_input_keyboard        = update.capture_keyboard;
 		g_input_multiline       = update.multiline;
-		if (g_input_keyboard) {
+		if (update.text_input) {
 			SDL_StartTextInput();
 		} else if (SDL_IsTextInputActive() == SDL_TRUE) {
 			SDL_StopTextInput();
@@ -458,7 +495,7 @@ bool ProcessImeInput(const SDL_Event& event) {
 	}
 	if (event.type == SDL_CONTROLLERDEVICEREMOVED) {
 		if (g_input_active && g_input_controller) {
-			QueueInput({InputKind::ResetController, g_input_generation, 0, 0.0f, 0.0f});
+			QueueInput({InputKind::ResetController, g_input_session, 0, 0.0f, 0.0f});
 		}
 		return false;
 	}
@@ -466,7 +503,8 @@ bool ProcessImeInput(const SDL_Event& event) {
 		return false;
 	}
 
-	const uint64_t generation       = g_input_generation;
+	const auto     session          = g_input_session;
+	const uint64_t generation       = session.generation;
 	const bool     controller_event = event.type == SDL_CONTROLLERBUTTONDOWN ||
 	                                  event.type == SDL_CONTROLLERBUTTONUP ||
 	                                  event.type == SDL_CONTROLLERAXISMOTION;
@@ -478,6 +516,13 @@ bool ProcessImeInput(const SDL_Event& event) {
 	if (keyboard_event && !g_input_keyboard) {
 		return false;
 	}
+	if (keyboard_event && session.kind == OverlayKind::Error) {
+		if (event.type == SDL_KEYDOWN && event.key.repeat == 0 &&
+		    (event.key.keysym.sym == SDLK_RETURN || event.key.keysym.sym == SDLK_KP_ENTER)) {
+			ErrorDialog::HostAccept(generation);
+		}
+		return true;
+	}
 	switch (event.type) {
 		case SDL_TEXTINPUT: {
 			const auto text = Utf8ToUtf16(event.text.text);
@@ -524,31 +569,33 @@ bool ProcessImeInput(const SDL_Event& event) {
 		case SDL_KEYUP: return true;
 		case SDL_CONTROLLERBUTTONDOWN:
 		case SDL_CONTROLLERBUTTONUP:
-			QueueInput({InputKind::Button, generation, event.cbutton.button,
+			QueueInput({InputKind::Button, session, event.cbutton.button,
 			            event.type == SDL_CONTROLLERBUTTONDOWN ? 1.0f : 0.0f, 0.0f});
 			return true;
 		case SDL_CONTROLLERAXISMOTION:
-			QueueInput({InputKind::Axis, generation, event.caxis.axis,
+			QueueInput({InputKind::Axis, session, event.caxis.axis,
 			            static_cast<float>(event.caxis.value), 0.0f});
 			return true;
 		case SDL_MOUSEMOTION:
-			QueueInput({InputKind::MousePosition, generation, 0, static_cast<float>(event.motion.x),
+			QueueInput({InputKind::MousePosition, session, 0, static_cast<float>(event.motion.x),
 			            static_cast<float>(event.motion.y)});
 			return true;
 		case SDL_MOUSEBUTTONDOWN:
 		case SDL_MOUSEBUTTONUP:
-			QueueInput({InputKind::MouseButton, generation, event.button.button,
+			QueueInput({InputKind::MousePosition, session, 0, static_cast<float>(event.button.x),
+			            static_cast<float>(event.button.y)});
+			QueueInput({InputKind::MouseButton, session, event.button.button,
 			            event.type == SDL_MOUSEBUTTONDOWN ? 1.0f : 0.0f, 0.0f});
 			return true;
 		case SDL_MOUSEWHEEL:
-			QueueInput({InputKind::MouseWheel, generation, 0, static_cast<float>(event.wheel.x),
+			QueueInput({InputKind::MouseWheel, session, 0, static_cast<float>(event.wheel.x),
 			            static_cast<float>(event.wheel.y)});
 			return true;
 		default: return false;
 	}
 }
 
-struct ImeOverlay::Impl {
+struct SystemOverlay::Impl {
 	explicit Impl(GraphicContext& context): graphics(context) {}
 
 	~Impl() {
@@ -570,8 +617,9 @@ struct ImeOverlay::Impl {
 		io.IniFilename = nullptr;
 		io.LogFilename = nullptr;
 		io.ConfigFlags |= ImGuiConfigFlags_NavEnableGamepad;
+		io.ConfigNavCursorVisibleAlways = true;
 		io.BackendFlags |= ImGuiBackendFlags_HasGamepad;
-		io.BackendPlatformName = "Kyty IME input";
+		io.BackendPlatformName = "Kyty system overlay input";
 		ImGui::StyleColorsDark();
 		auto& style          = ImGui::GetStyle();
 		style.WindowRounding = 10.0f;
@@ -609,7 +657,7 @@ struct ImeOverlay::Impl {
 		vulkan_initialized = true;
 	}
 
-	void DrainInput(uint64_t generation) {
+	void DrainInput(OverlaySession session) {
 		std::deque<InputEvent> events;
 		bool                   reset = false;
 		{
@@ -626,16 +674,18 @@ struct ImeOverlay::Impl {
 			right_stick = {};
 		}
 		for (const auto& event: events) {
-			if (event.generation != generation) {
+			if (event.session != session) {
 				continue;
 			}
 			switch (event.kind) {
 				case InputKind::Button: {
 					const bool down = event.x != 0.0f;
-					if (down && event.id == SDL_CONTROLLER_BUTTON_B) {
-						Ime::HostCancel(generation);
-					} else if (down && event.id == SDL_CONTROLLER_BUTTON_Y) {
-						Ime::HostBackspace(generation);
+					if (session.kind == OverlayKind::Ime && down) {
+						if (event.id == SDL_CONTROLLER_BUTTON_B) {
+							Ime::HostCancel(session.generation);
+						} else if (event.id == SDL_CONTROLLER_BUTTON_Y) {
+							Ime::HostBackspace(session.generation);
+						}
 					}
 					const ImGuiKey key = ControllerButtonToKey(event.id);
 					if (key != ImGuiKey_None) {
@@ -732,7 +782,7 @@ struct ImeOverlay::Impl {
 		focus_pending = false;
 	}
 
-	void DrawDialog(const Ime::HostSnapshot& snapshot, vk::Extent2D extent) {
+	void DrawIme(const Ime::HostSnapshot& snapshot, vk::Extent2D extent) {
 		const ImVec2 display(static_cast<float>(extent.width), static_cast<float>(extent.height));
 		const bool   over_2k          = (snapshot.option & Ime::OPTION_USE_OVER_2K) != 0;
 		const float  reference_width  = over_2k ? 3840.0f : 1920.0f;
@@ -851,17 +901,58 @@ struct ImeOverlay::Impl {
 		ImGui::End();
 	}
 
+	void DrawError(const ErrorDialog::HostSnapshot& snapshot, vk::Extent2D extent) {
+		const ImVec2 display(static_cast<float>(extent.width), static_cast<float>(extent.height));
+		const float  scale = std::max(std::min(display.x / 1280.0f, display.y / 720.0f), 0.5f);
+		ImGui::GetBackgroundDrawList()->AddRectFilled({0.0f, 0.0f}, display,
+		                                              IM_COL32(0, 0, 0, 160));
+		ImGui::SetNextWindowPos({display.x * 0.5f, display.y * 0.5f}, ImGuiCond_Always,
+		                        {0.5f, 0.5f});
+		ImGui::SetNextWindowSize(
+		    {std::max(std::min(620.0f * scale, display.x - 32.0f), 1.0f), 0.0f}, ImGuiCond_Always);
+		if (focus_pending) {
+			ImGui::SetNextWindowFocus();
+		}
+		ImGui::PushStyleVar(ImGuiStyleVar_WindowPadding, {24.0f * scale, 24.0f * scale});
+		ImGui::PushStyleVar(ImGuiStyleVar_ItemSpacing, {12.0f * scale, 16.0f * scale});
+		ImGui::PushFont(nullptr, 20.0f * scale);
+		constexpr ImGuiWindowFlags flags = ImGuiWindowFlags_NoDecoration | ImGuiWindowFlags_NoMove |
+		                                   ImGuiWindowFlags_NoSavedSettings |
+		                                   ImGuiWindowFlags_AlwaysAutoResize;
+		ImGui::Begin("##SystemError", nullptr, flags);
+		ImGui::TextUnformatted("Error");
+		ImGui::Separator();
+		const auto error_code = static_cast<uint32_t>(snapshot.error_code);
+		ImGui::TextWrapped("%s", error_code == 0x80550006u
+		                             ? "You are not signed in to PlayStation Network."
+		                             : "An error has occurred.");
+		ImGui::TextDisabled("Error code: 0x%08X", error_code);
+		const float button_width = std::min(140.0f * scale, ImGui::GetContentRegionAvail().x);
+		ImGui::SetCursorPosX((ImGui::GetWindowSize().x - button_width) * 0.5f);
+		const bool accepted = ImGui::Button("OK", {button_width, 44.0f * scale});
+		if (focus_pending) {
+			ImGui::SetItemDefaultFocus();
+			focus_pending = false;
+		}
+		ImGui::End();
+		ImGui::PopFont();
+		ImGui::PopStyleVar(2);
+		if (accepted) {
+			ErrorDialog::HostAccept(snapshot.generation);
+		}
+	}
+
 	bool PrepareFrame(vk::Extent2D frame_extent, vk::Format format, uint32_t image_count) {
-		Ime::HostSnapshot snapshot;
-		if (!Ime::GetHostSnapshot(&snapshot)) {
+		OverlaySnapshot snapshot;
+		if (!GetOverlaySnapshot(&snapshot)) {
 			return false;
 		}
-		const uint64_t prepared_generation = snapshot.generation;
+		const auto prepared_session = snapshot.session;
 		EnsureVulkan(format, image_count);
-		if (generation != snapshot.generation) {
-			generation    = snapshot.generation;
+		if (session != snapshot.session) {
+			session       = snapshot.session;
 			focus_pending = true;
-			shift         = (snapshot.option & Ime::OPTION_NO_AUTO_CAPITALIZE) == 0;
+			shift         = (snapshot.ime.option & Ime::OPTION_NO_AUTO_CAPITALIZE) == 0;
 			symbol_mode   = false;
 			panel_offset  = {};
 			right_stick   = {};
@@ -870,7 +961,7 @@ struct ImeOverlay::Impl {
 			io.ClearInputKeys();
 			io.ClearInputMouse();
 		}
-		DrainInput(snapshot.generation);
+		DrainInput(snapshot.session);
 
 		auto& io       = ImGui::GetIO();
 		io.DisplaySize = {static_cast<float>(frame_extent.width),
@@ -883,11 +974,15 @@ struct ImeOverlay::Impl {
 		last_frame     = now;
 		ImGui_ImplVulkan_NewFrame();
 		ImGui::NewFrame();
-		if (!Ime::GetHostSnapshot(&snapshot) || snapshot.generation != prepared_generation) {
+		if (!GetOverlaySnapshot(&snapshot) || snapshot.session != prepared_session) {
 			ImGui::EndFrame();
 			return false;
 		}
-		DrawDialog(snapshot, frame_extent);
+		if (snapshot.session.kind == OverlayKind::Error) {
+			DrawError(snapshot.error, frame_extent);
+		} else {
+			DrawIme(snapshot.ime, frame_extent);
+		}
 		ImGui::Render();
 		extent = frame_extent;
 		return true;
@@ -934,24 +1029,24 @@ struct ImeOverlay::Impl {
 	float                                 button_height      = 42.0f;
 	ImVec2                                panel_offset {};
 	ImVec2                                right_stick {};
-	uint64_t                              generation = 0;
+	OverlaySession                        session;
 	vk::Extent2D                          extent {};
 	std::chrono::steady_clock::time_point last_frame;
 };
 
-ImeOverlay::ImeOverlay(GraphicContext& graphics): m_impl(std::make_unique<Impl>(graphics)) {}
+SystemOverlay::SystemOverlay(GraphicContext& graphics): m_impl(std::make_unique<Impl>(graphics)) {}
 
-ImeOverlay::~ImeOverlay() = default;
+SystemOverlay::~SystemOverlay() = default;
 
-bool ImeOverlay::PrepareFrame(vk::Extent2D extent, vk::Format format, uint32_t image_count) {
+bool SystemOverlay::PrepareFrame(vk::Extent2D extent, vk::Format format, uint32_t image_count) {
 	return m_impl->PrepareFrame(extent, format, image_count);
 }
 
-void ImeOverlay::Record(vk::CommandBuffer command, vk::ImageView target) {
+void SystemOverlay::Record(vk::CommandBuffer command, vk::ImageView target) {
 	m_impl->Record(command, target);
 }
 
-void ImeOverlay::ReleaseVulkan() {
+void SystemOverlay::ReleaseVulkan() {
 	m_impl->ReleaseVulkan();
 }
 

src/graphics/presentation/systemOverlay.h

@@ -0,0 +1,42 @@
+#ifndef EMULATOR_SRC_GRAPHICS_PRESENTATION_SYSTEMOVERLAY_H_
+#define EMULATOR_SRC_GRAPHICS_PRESENTATION_SYSTEMOVERLAY_H_
+
+#include "common/common.h"
+#include "graphics/host_gpu/vulkanCommon.h"
+
+#include <memory>
+
+union SDL_Event;
+
+namespace Libs::Graphics {
+
+struct GraphicContext;
+
+struct SystemOverlayVisualState {
+	bool     active;
+	uint64_t revision;
+};
+
+void                     InitializeSystemOverlayInput();
+void                     ShutdownSystemOverlayInput();
+bool                     ProcessSystemOverlayInput(const SDL_Event& event);
+SystemOverlayVisualState GetSystemOverlayVisualState() noexcept;
+
+class SystemOverlay final {
+public:
+	explicit SystemOverlay(GraphicContext& graphics);
+	~SystemOverlay();
+	KYTY_CLASS_NO_COPY(SystemOverlay);
+
+	[[nodiscard]] bool PrepareFrame(vk::Extent2D extent, vk::Format format, uint32_t image_count);
+	void               Record(vk::CommandBuffer command, vk::ImageView target);
+	void               ReleaseVulkan();
+
+private:
+	struct Impl;
+	std::unique_ptr<Impl> m_impl;
+};
+
+} // namespace Libs::Graphics
+
+#endif // EMULATOR_SRC_GRAPHICS_PRESENTATION_SYSTEMOVERLAY_H_

src/graphics/presentation/videoOut.cpp

@@ -809,7 +809,7 @@ void VideoOutDriver::Impl::PresentThread(std::stop_token token) {
 
 		VblankBegin();
 		bool presented = m_flip_queue.Flip(0);
-		if (!presented && m_presenter.NeedsImeRefresh()) {
+		if (!presented && m_presenter.NeedsSystemOverlayRefresh()) {
 			if (auto* frame = m_presenter.PrepareLastFrame(); frame != nullptr) {
 				m_presenter.Present(*frame, true);
 				presented = true;
@@ -1123,7 +1123,6 @@ bool FlipQueue::Flip(uint32_t micros) {
 	m_mutex.Unlock();
 
 	m_presenter.Present(*r.frame);
-	Graphics::RenderDocOnGuestFlip();
 
 	m_mutex.Lock();
 	if (m_requests.empty() || m_requests.front().id != r.id ||
@@ -1150,6 +1149,8 @@ bool FlipQueue::Flip(uint32_t micros) {
 	m_mutex.Unlock();
 	r.cfg->mutex.Unlock();
 
+	Graphics::RenderDocOnGuestFlip(m_presenter.Renderer());
+
 	if (Config::GraphicsDebugDumpEnabled() &&
 	    Config::GetPrintfDirection() != Config::OutputDirection::Silent) {
 		LOGF("Flip done: %d\n", r.index);

src/graphics/presentation/window/hostInput.cpp

@@ -194,18 +194,18 @@ const InputMap& GetInputMap() {
 
 void SetButton(uint32_t button, bool down) {
 	if (button == Controller::PAD_BUTTON_L2) {
-		Controller::ControllerAxis(Controller::HOST_INPUT_CONTROLLER_ID,
-		                           Controller::Axis::TriggerLeft, down ? 255 : 0);
+		Controller::SetAxis(Controller::HOST_INPUT_CONTROLLER_ID, Controller::Axis::TriggerLeft,
+		                    down ? 255 : 0);
 	} else if (button == Controller::PAD_BUTTON_R2) {
-		Controller::ControllerAxis(Controller::HOST_INPUT_CONTROLLER_ID,
-		                           Controller::Axis::TriggerRight, down ? 255 : 0);
+		Controller::SetAxis(Controller::HOST_INPUT_CONTROLLER_ID, Controller::Axis::TriggerRight,
+		                    down ? 255 : 0);
 	} else if (button != 0) {
-		Controller::ControllerButton(Controller::HOST_INPUT_CONTROLLER_ID, button, down);
+		Controller::SetButton(Controller::HOST_INPUT_CONTROLLER_ID, button, down);
 	}
 }
 
 void SetTouchPad(float x, bool down) {
-	Controller::ControllerTouchPad(Controller::HOST_INPUT_CONTROLLER_ID, 0, down, x, 0.5f);
+	Controller::SetTouchPad(Controller::HOST_INPUT_CONTROLLER_ID, 0, down, x, 0.5f);
 }
 
 uint32_t DefaultKeyboardButton(int key_code) {
@@ -242,7 +242,7 @@ void SetStickAxis(Controller::Axis axis, bool negative, bool positive) {
 	} else if (positive && !negative) {
 		value = 255;
 	}
-	Controller::ControllerAxis(Controller::HOST_INPUT_CONTROLLER_ID, axis, value);
+	Controller::SetAxis(Controller::HOST_INPUT_CONTROLLER_ID, axis, value);
 }
 
 void SetControl(std::size_t control, bool down) {
@@ -326,15 +326,15 @@ void MouseToJoystick(int delta_x, int delta_y) {
 	const auto map_axis = [scale](int delta) {
 		return std::clamp(128 + static_cast<int>(std::lround(delta * scale)), 0, 255);
 	};
-	Controller::ControllerRightStick(Controller::HOST_INPUT_CONTROLLER_ID, map_axis(delta_x),
-	                                 map_axis(delta_y));
+	Controller::SetRightStick(Controller::HOST_INPUT_CONTROLLER_ID, map_axis(delta_x),
+	                          map_axis(delta_y));
 }
 
 void CenterMouseStick() {
 	if (!g_mouse.output) {
 		return;
 	}
-	Controller::ControllerRightStick(Controller::HOST_INPUT_CONTROLLER_ID, 128, 128);
+	Controller::SetRightStick(Controller::HOST_INPUT_CONTROLLER_ID, 128, 128);
 	g_mouse.output = false;
 }
 

src/graphics/presentation/window/swapchain.cpp

@@ -7,10 +7,9 @@
 #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/imeOverlay.h"
 #include "graphics/presentation/presenter.h"
+#include "graphics/presentation/systemOverlay.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window/windowInternal.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));
 	}
 }
 
@@ -293,8 +283,9 @@ public:
 	void                 Create();
 	void                 Recreate(bool surface_lost = false);
 	[[nodiscard]] Status AcquireNextImage();
-	[[nodiscard]] bool   PrepareImeOverlay();
-	void RecordPresentCommands(CommandBuffer& command, VulkanImage& source, bool draw_ime_overlay);
+	[[nodiscard]] bool   PrepareSystemOverlay();
+	void                 RecordPresentCommands(CommandBuffer& command, VulkanImage& source,
+	                                           bool draw_system_overlay);
 	uint64_t             Submit(CommandScheduler& scheduler);
 	[[nodiscard]] Status Present();
 
@@ -314,7 +305,7 @@ private:
 	std::vector<vk::ImageView>  m_image_views;
 	std::vector<vk::Semaphore>  m_image_acquired;
 	std::vector<vk::Semaphore>  m_render_complete;
-	std::unique_ptr<ImeOverlay> m_ime_overlay;
+	std::unique_ptr<SystemOverlay> m_system_overlay;
 	uint32_t                    m_image_index = static_cast<uint32_t>(-1);
 	uint32_t                    m_frame_index = 0;
 };
@@ -361,7 +352,7 @@ struct Presenter::Impl {
 	Swapchain             swapchain;
 	CommandScheduler      present_scheduler;
 	FramePool             frames;
-	std::atomic<uint64_t> presented_ime_revision {0};
+	std::atomic<uint64_t> presented_overlay_revision {0};
 };
 
 void Swapchain::Create() {
@@ -504,8 +495,8 @@ void Swapchain::Destroy() {
 		Common::LockGuard queue_lock(graphics.queue_mutex);
 		RequireVulkanSuccess(graphics.queue.waitIdle(), "wait for swapchain queue");
 	}
-	if (m_ime_overlay != nullptr) {
-		m_ime_overlay->ReleaseVulkan();
+	if (m_system_overlay != nullptr) {
+		m_system_overlay->ReleaseVulkan();
 	}
 
 	for (const auto semaphore: m_image_acquired) {
@@ -573,21 +564,21 @@ Swapchain::Status Swapchain::AcquireNextImage() {
 		case vk::Result::eErrorSurfaceLostKHR:
 			LOGF("vkAcquireNextImageKHR returned vk::Result::eErrorSurfaceLostKHR\n");
 			return Status::SurfaceLost;
-		default: EXIT("vkAcquireNextImageKHR failed: %s\n", VulkanToString(result).c_str());
+		default: EXIT("vkAcquireNextImageKHR failed: %s\n", vk::to_string(result).c_str());
 	}
 	EXIT_IF(m_image_index >= m_images.size());
 	return Status::Success;
 }
 
-bool Swapchain::PrepareImeOverlay() {
-	if (m_ime_overlay == nullptr) {
-		m_ime_overlay = std::make_unique<ImeOverlay>(m_window.graphic_ctx);
+bool Swapchain::PrepareSystemOverlay() {
+	if (m_system_overlay == nullptr) {
+		m_system_overlay = std::make_unique<SystemOverlay>(m_window.graphic_ctx);
 	}
-	return m_ime_overlay->PrepareFrame(m_extent, m_format, ImageCount());
+	return m_system_overlay->PrepareFrame(m_extent, m_format, ImageCount());
 }
 
 void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& source,
-                                      bool draw_ime_overlay) {
+                                      bool draw_system_overlay) {
 	if (source.state.layout != vk::ImageLayout::eTransferSrcOptimal) {
 		EXIT("invalid prepared presentation image, vk_image=%p layout=%d\n",
 		     static_cast<void*>(source.image), static_cast<int>(source.state.layout));
@@ -608,7 +599,8 @@ void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& sourc
 	to_transfer.subresourceRange.levelCount     = 1;
 	to_transfer.subresourceRange.baseArrayLayer = 0;
 	to_transfer.subresourceRange.layerCount     = 1;
-	vk_command.pipelineBarrier(vk::PipelineStageFlagBits::eTopOfPipe,
+	// Match the acquire wait stage so the layout transition cannot precede acquisition.
+	vk_command.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer,
 	                           vk::PipelineStageFlagBits::eTransfer, vk::DependencyFlags {}, 0,
 	                           nullptr, 0, nullptr, 1, &to_transfer);
 
@@ -634,12 +626,12 @@ void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& sourc
 	vk::ImageMemoryBarrier to_present {};
 	to_present.sType         = vk::StructureType::eImageMemoryBarrier;
 	to_present.srcAccessMask = vk::AccessFlagBits::eTransferWrite;
-	to_present.dstAccessMask = draw_ime_overlay ? vk::AccessFlagBits::eColorAttachmentRead |
-	                                                  vk::AccessFlagBits::eColorAttachmentWrite
-	                                            : vk::AccessFlagBits::eMemoryRead;
+	to_present.dstAccessMask = draw_system_overlay ? vk::AccessFlagBits::eColorAttachmentRead |
+	                                                     vk::AccessFlagBits::eColorAttachmentWrite
+	                                               : vk::AccessFlagBits::eMemoryRead;
 	to_present.oldLayout     = vk::ImageLayout::eTransferDstOptimal;
-	to_present.newLayout     = draw_ime_overlay ? vk::ImageLayout::eColorAttachmentOptimal
-	                                            : vk::ImageLayout::ePresentSrcKHR;
+	to_present.newLayout     = draw_system_overlay ? vk::ImageLayout::eColorAttachmentOptimal
+	                                               : vk::ImageLayout::ePresentSrcKHR;
 	to_present.srcQueueFamilyIndex             = VK_QUEUE_FAMILY_IGNORED;
 	to_present.dstQueueFamilyIndex             = VK_QUEUE_FAMILY_IGNORED;
 	to_present.image                           = m_images[m_image_index];
@@ -648,13 +640,13 @@ void Swapchain::RecordPresentCommands(CommandBuffer& command, VulkanImage& sourc
 	to_present.subresourceRange.levelCount     = 1;
 	to_present.subresourceRange.baseArrayLayer = 0;
 	to_present.subresourceRange.layerCount     = 1;
-	vk_command.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer,
-	                           draw_ime_overlay ? vk::PipelineStageFlagBits::eColorAttachmentOutput
-	                                            : vk::PipelineStageFlagBits::eAllCommands,
-	                           vk::DependencyFlagBits::eByRegion, 0, nullptr, 0, nullptr, 1,
-	                           &to_present);
-	if (draw_ime_overlay) {
-		m_ime_overlay->Record(vk_command, m_image_views[m_image_index]);
+	vk_command.pipelineBarrier(
+	    vk::PipelineStageFlagBits::eTransfer,
+	    draw_system_overlay ? vk::PipelineStageFlagBits::eColorAttachmentOutput
+	                        : vk::PipelineStageFlagBits::eAllCommands,
+	    vk::DependencyFlagBits::eByRegion, 0, nullptr, 0, nullptr, 1, &to_present);
+	if (draw_system_overlay) {
+		m_system_overlay->Record(vk_command, m_image_views[m_image_index]);
 		to_present.srcAccessMask = vk::AccessFlagBits::eColorAttachmentWrite;
 		to_present.dstAccessMask = vk::AccessFlagBits::eMemoryRead;
 		to_present.oldLayout     = vk::ImageLayout::eColorAttachmentOptimal;
@@ -701,7 +693,7 @@ Swapchain::Status Swapchain::Present() {
 		case vk::Result::eErrorSurfaceLostKHR:
 			LOGF("vkQueuePresentKHR returned vk::Result::eErrorSurfaceLostKHR\n");
 			return Status::SurfaceLost;
-		default: EXIT("vkQueuePresentKHR failed: %s\n", VulkanToString(result).c_str());
+		default: EXIT("vkQueuePresentKHR failed: %s\n", vk::to_string(result).c_str());
 	}
 	m_frame_index = (m_frame_index + 1u) % static_cast<uint32_t>(m_images.size());
 	return Status::Success;
@@ -761,10 +753,10 @@ bool Presenter::IsGuestPaused() const noexcept {
 	return m_impl->window.loop.paused.load(std::memory_order_acquire);
 }
 
-bool Presenter::NeedsImeRefresh() const noexcept {
-	const auto visual = GetImeVisualState();
+bool Presenter::NeedsSystemOverlayRefresh() const noexcept {
+	const auto visual = GetSystemOverlayVisualState();
 	return visual.active ||
-	       visual.revision != m_impl->presented_ime_revision.load(std::memory_order_acquire);
+	       visual.revision != m_impl->presented_overlay_revision.load(std::memory_order_acquire);
 }
 
 RenderContext& Presenter::Renderer() const noexcept {
@@ -775,7 +767,7 @@ void Presenter::Present(Frame& frame, bool reuse) {
 	KYTY_PROFILER_FUNCTION();
 	m_impl->frames.ValidateForPresent(&frame, reuse);
 
-	const auto ime_visual = GetImeVisualState();
+	const auto overlay_visual = GetSystemOverlayVisualState();
 	auto&      swapchain  = m_impl->swapchain;
 	for (uint32_t attempt = 0; attempt < 2; attempt++) {
 		auto status = swapchain.AcquireNextImage();
@@ -786,8 +778,9 @@ void Presenter::Present(Frame& frame, bool reuse) {
 		{
 			Common::LockGuard render_lock(m_impl->renderer.GetMutex());
 			auto&             command          = m_impl->present_scheduler.BeginCommand();
-			const bool        draw_ime_overlay = ime_visual.active && swapchain.PrepareImeOverlay();
-			swapchain.RecordPresentCommands(command, frame.image, draw_ime_overlay);
+			const bool        draw_system_overlay =
+			    overlay_visual.active && swapchain.PrepareSystemOverlay();
+			swapchain.RecordPresentCommands(command, frame.image, draw_system_overlay);
 			frame.present_tick = swapchain.Submit(m_impl->present_scheduler);
 		}
 		status = swapchain.Present();
@@ -796,7 +789,8 @@ void Presenter::Present(Frame& frame, bool reuse) {
 			continue;
 		}
 
-		m_impl->presented_ime_revision.store(ime_visual.revision, std::memory_order_release);
+		m_impl->presented_overlay_revision.store(overlay_visual.revision,
+		                                         std::memory_order_release);
 		m_impl->window.UpdateTitle();
 		m_impl->frames.Release(&frame, true);
 		return;

src/graphics/presentation/window/vulkanWindow.cpp

@@ -27,10 +27,9 @@
 #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/imeOverlay.h"
 #include "graphics/presentation/presenter.h"
+#include "graphics/presentation/systemOverlay.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window.h"
 #include "graphics/presentation/window/windowInternal.h"
@@ -63,6 +62,18 @@ struct VulkanExtensions {
 	std::vector<vk::LayerProperties>     available_layers;
 };
 
+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;
+	features.shaderOutputViewportIndex = VK_TRUE;
+	features.bufferDeviceAddress       = VK_TRUE;
+	features.shaderBufferInt64Atomics  = VK_TRUE;
+	return features;
+}
+
 vk::PhysicalDeviceVulkan13Features WindowContext::RequiredVulkan13Features() noexcept {
 	vk::PhysicalDeviceVulkan13Features features {};
 	features.sType            = vk::StructureType::ePhysicalDeviceVulkan13Features;
@@ -132,7 +143,7 @@ static uint32_t VulkanFindQueueFamily(vk::PhysicalDevice device, vk::SurfaceKHR
 		                     "vkGetPhysicalDeviceSurfaceSupportKHR");
 
 		LOGF("\tqueue family: %s [count = %u], [present = %s]\n",
-		     VulkanToString(properties.queueFlags).c_str(), properties.queueCount,
+		     vk::to_string(properties.queueFlags).c_str(), properties.queueCount,
 		     (presentation_supported == VK_TRUE ? "true" : "false"));
 		if (properties.queueCount != 0 && (properties.queueFlags & required) == required &&
 		    presentation_supported == VK_TRUE) {
@@ -158,7 +169,12 @@ 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;
@@ -217,6 +233,7 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 		device_features2.pNext = &features13;
 
 		device.getFeatures2(&device_features2);
+		const auto required_features12 = WindowContext::RequiredVulkan12Features();
 		const auto required_features13 = WindowContext::RequiredVulkan13Features();
 
 		const auto queue_family = VulkanFindQueueFamily(device, surface);
@@ -249,14 +266,36 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 		}
 #endif
 
-		if (features12.samplerMirrorClampToEdge != VK_TRUE) {
+		if (required_features12.samplerMirrorClampToEdge == VK_TRUE &&
+		    features12.samplerMirrorClampToEdge != VK_TRUE) {
 			LOGF("samplerMirrorClampToEdge is not supported\n");
 			skip_device = true;
 		}
-		if (features12.timelineSemaphore != VK_TRUE) {
+		if (required_features12.timelineSemaphore == VK_TRUE &&
+		    features12.timelineSemaphore != VK_TRUE) {
 			LOGF("timelineSemaphore is not supported\n");
 			skip_device = true;
 		}
+		if (required_features12.shaderOutputLayer == VK_TRUE &&
+		    features12.shaderOutputLayer != VK_TRUE) {
+			LOGF("shaderOutputLayer is not supported\n");
+			skip_device = true;
+		}
+		if (required_features12.shaderOutputViewportIndex == VK_TRUE &&
+		    features12.shaderOutputViewportIndex != VK_TRUE) {
+			LOGF("shaderOutputViewportIndex is not supported\n");
+			skip_device = true;
+		}
+		if (required_features12.bufferDeviceAddress == VK_TRUE &&
+		    features12.bufferDeviceAddress != VK_TRUE) {
+			LOGF("bufferDeviceAddress is not supported\n");
+			skip_device = true;
+		}
+		if (required_features12.shaderBufferInt64Atomics == VK_TRUE &&
+		    features12.shaderBufferInt64Atomics != VK_TRUE) {
+			LOGF("shaderBufferInt64Atomics is not supported\n");
+			skip_device = true;
+		}
 		if (features13.robustImageAccess != VK_TRUE) {
 			LOGF("robustImageAccess is not supported\n");
 			skip_device = true;
@@ -291,11 +330,14 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 			LOGF("largePoints is not supported\n");
 			skip_device = true;
 		}
-		if (features12.shaderOutputLayer != VK_TRUE) {
-			LOGF("shaderOutputLayer is not supported\n");
+		if (device_features2.features.multiViewport != VK_TRUE) {
+			LOGF("multiViewport is not supported\n");
+			skip_device = true;
+		}
+		if (device_features2.features.fillModeNonSolid != VK_TRUE) {
+			LOGF("fillModeNonSolid is not supported\n");
 			skip_device = true;
 		}
-
 		if (device_features2.features.fragmentStoresAndAtomics != VK_TRUE) {
 			LOGF("fragmentStoresAndAtomics is not supported\n");
 			skip_device = true;
@@ -501,14 +543,12 @@ static void VulkanInitSubgroupSizeControl(vk::PhysicalDevice physical_device,
 	    (graphics.required_subgroup_size_stages & vk::ShaderStageFlagBits::eCompute) &&
 	    subgroup_size_control.minSubgroupSize <= 64 &&
 	    subgroup_size_control.maxSubgroupSize >= 64;
-	graphics.compute_wave64_supported =
-	    graphics.subgroup_size == 64u || graphics.compute_subgroup_size_control_enabled;
 
 	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.compute_wave64_supported ? "true" : "false");
+	     graphics.SupportsComputeWave64() ? "true" : "false");
 }
 
 static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const VulkanExtensions& r,
@@ -547,10 +587,8 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 #endif
 	depth_clip_control.depthClipControl = VK_TRUE;
 
-	vk::PhysicalDeviceVulkan12Features features12 {};
-	features12.sType                    = vk::StructureType::ePhysicalDeviceVulkan12Features;
-	features12.pNext                    = &depth_clip_control;
-	features12.samplerMirrorClampToEdge = VK_TRUE;
+	auto features12  = WindowContext::RequiredVulkan12Features();
+	features12.pNext = &depth_clip_control;
 
 	vk::PhysicalDeviceVulkan13Features supported_features13 {};
 	supported_features13.sType = vk::StructureType::ePhysicalDeviceVulkan13Features;
@@ -582,12 +620,54 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 #endif
 	}
 
+	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 = &supported_features13;
+	supported_features2.pNext = mesh_extension ? static_cast<void*>(&supported_mesh)
+	                                           : static_cast<void*>(&supported_features13);
+	const bool feedback_extensions =
+	    HasExtension(device_extensions, VK_EXT_ATTACHMENT_FEEDBACK_LOOP_LAYOUT_EXTENSION_NAME) &&
+	    HasExtension(device_extensions, VK_EXT_ATTACHMENT_FEEDBACK_LOOP_DYNAMIC_STATE_EXTENSION_NAME);
+	vk::PhysicalDeviceAttachmentFeedbackLoopLayoutFeaturesEXT feedback_layout {};
+	vk::PhysicalDeviceAttachmentFeedbackLoopDynamicStateFeaturesEXT feedback_dynamic {};
+	if (feedback_extensions) {
+		feedback_dynamic.pNext = supported_features2.pNext;
+		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.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(supported_features12.timelineSemaphore != VK_TRUE);
+	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 &&
@@ -597,17 +677,21 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	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);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.shaderInt64 != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.vertexPipelineStoresAndAtomics != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features12.shaderOutputLayer != VK_TRUE);
-	EXIT_NOT_IMPLEMENTED(supported_features12.bufferDeviceAddress != 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
-	features12.timelineSemaphore = VK_TRUE;
-	features12.shaderOutputLayer = VK_TRUE;
-	features12.bufferDeviceAddress = VK_TRUE;
-
 	vk::PhysicalDeviceFeatures device_features {};
 	device_features.fragmentStoresAndAtomics = VK_TRUE;
 	device_features.samplerAnisotropy        = VK_TRUE;
@@ -624,6 +708,8 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	device_features.shaderClipDistance                   = VK_TRUE;
 	device_features.shaderCullDistance                   = VK_TRUE;
 	device_features.largePoints                          = VK_TRUE;
+	device_features.multiViewport                        = VK_TRUE;
+	device_features.fillModeNonSolid                      = VK_TRUE;
 	device_features.vertexPipelineStoresAndAtomics       = VK_TRUE;
 	graphics.sample_rate_shading_enabled                 = true;
 	device_features.shaderInt64 = VK_TRUE;
@@ -666,8 +752,20 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	     robustness2_ext_enabled && robustness2.robustImageAccess2 == VK_TRUE ? "true" : "false");
 
 	vk::DeviceCreateInfo create_info {};
+	vk::PhysicalDeviceMeshShaderFeaturesEXT mesh_features {};
+	mesh_features.pNext                 = &features13;
+	mesh_features.meshShader            = graphics.mesh_shader_enabled;
 	create_info.sType                   = vk::StructureType::eDeviceCreateInfo;
-	create_info.pNext                   = &features13;
+	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;
+	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.flags                   = {};
 	create_info.pQueueCreateInfos       = &queue_create_info;
 	create_info.queueCreateInfoCount    = 1;
@@ -679,7 +777,7 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 
 	auto result = physical_device.createDevice(&create_info, nullptr, &device);
 	if (result != vk::Result::eSuccess) {
-		LOGF("vkCreateDevice failed: %s\n", VulkanToString(result).c_str());
+		LOGF("vkCreateDevice failed: %s\n", vk::to_string(result).c_str());
 		return nullptr;
 	}
 
@@ -858,7 +956,7 @@ static void VulkanCheckInstanceVersion() {
 	if (VULKAN_HPP_DEFAULT_DISPATCHER.vkEnumerateInstanceVersion != nullptr) {
 		auto result = vk::enumerateInstanceVersion(&version);
 		if (result != vk::Result::eSuccess) {
-			EXIT("Could not query Vulkan loader version: %s\n", VulkanToString(result).c_str());
+			EXIT("Could not query Vulkan loader version: %s\n", vk::to_string(result).c_str());
 		}
 	}
 
@@ -1035,6 +1133,18 @@ 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",
@@ -1047,8 +1157,18 @@ void WindowContext::CreateVulkan() {
 			device_extensions.push_back(VK_EXT_MEMORY_BUDGET_EXTENSION_NAME);
 			graphic_ctx.memory_budget_ext_enabled = true;
 		}
-		if (HasExtension(available_extensions, VK_EXT_ROBUSTNESS_2_EXTENSION_NAME)) {
-			device_extensions.push_back(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME);
+		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)) {
+				device_extensions.push_back(extension);
+			}
+		}
+		if (HasExtension(available_extensions, VK_EXT_ATTACHMENT_FEEDBACK_LOOP_LAYOUT_EXTENSION_NAME) &&
+		    HasExtension(available_extensions, VK_EXT_ATTACHMENT_FEEDBACK_LOOP_DYNAMIC_STATE_EXTENSION_NAME)) {
+			device_extensions.push_back(VK_EXT_ATTACHMENT_FEEDBACK_LOOP_LAYOUT_EXTENSION_NAME);
+			device_extensions.push_back(VK_EXT_ATTACHMENT_FEEDBACK_LOOP_DYNAMIC_STATE_EXTENSION_NAME);
 		}
 	}
 
@@ -1093,7 +1213,7 @@ void WindowContext::RecreateSurface() {
 }
 
 WindowContext::~WindowContext() {
-	ShutdownImeInput();
+	ShutdownSystemOverlayInput();
 	presenter.reset();
 	LibKernel::Memory::InstallGpuResources(nullptr);
 	render_context.reset();

src/graphics/presentation/window/window.cpp

@@ -30,10 +30,9 @@
 #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/imeOverlay.h"
 #include "graphics/presentation/renderDoc.h"
+#include "graphics/presentation/systemOverlay.h"
 #include "graphics/presentation/window/hostInput.h"
 #include "graphics/presentation/window/windowInternal.h"
 #include "kytyGitVersion.h"
@@ -351,26 +350,26 @@ static void GameEventController([[maybe_unused]] const EventController& f) {
 		auto* pad = SDL_GameControllerOpen(f.id);
 		EXIT_NOT_IMPLEMENTED(pad == nullptr);
 		int id = SDL_JoystickInstanceID(SDL_GameControllerGetJoystick(pad));
-		Controller::ControllerConnect(id);
+		Controller::Connect(id);
 	}
 
 	if (f.removed) {
-		Controller::ControllerDisconnect(f.id);
+		Controller::Disconnect(f.id);
 		SDL_GameControllerClose(SDL_GameControllerFromInstanceID(f.id));
 	}
 
 	if (f.down || f.up) {
 		const auto button = ControllerButtonToPadButton(f.button);
 		if (button != 0) {
-			Controller::ControllerButton(f.id, button, f.down);
+			Controller::SetButton(f.id, button, f.down);
 		}
 	}
 
 	if (f.axis) {
 		const auto axis = ControllerAxisFromSdl(f.axis_id);
 		if (axis != Controller::Axis::AxisMax) {
-			Controller::ControllerAxis(f.id, axis,
-			                           ControllerAxisValueFromSdl(f.axis_id, f.axis_value));
+			Controller::SetAxis(f.id, axis,
+			                    ControllerAxisValueFromSdl(f.axis_id, f.axis_value));
 		}
 	}
 }
@@ -513,7 +512,7 @@ void WindowContext::ProcessEvent(double time_s) {
 		}
 		return;
 	}
-	if (ProcessImeInput(*event)) {
+	if (ProcessSystemOverlayInput(*event)) {
 		return;
 	}
 
@@ -708,9 +707,9 @@ void WindowContext::ProcessEvent(double time_s) {
 		case SDL_CONTROLLERTOUCHPADMOTION:
 		case SDL_CONTROLLERTOUCHPADUP:
 			if (event->ctouchpad.touchpad == 0) {
-				Controller::ControllerTouchPad(event->ctouchpad.which, event->ctouchpad.finger,
-				                               event->ctouchpad.type != SDL_CONTROLLERTOUCHPADUP,
-				                               event->ctouchpad.x, event->ctouchpad.y);
+				Controller::SetTouchPad(event->ctouchpad.which, event->ctouchpad.finger,
+				                        event->ctouchpad.type != SDL_CONTROLLERTOUCHPADUP,
+				                        event->ctouchpad.x, event->ctouchpad.y);
 			}
 			break;
 
@@ -814,6 +813,7 @@ static void WindowCreate(WindowContext& context) {
 	int width  = static_cast<int>(context.graphic_ctx.screen_width);
 	int height = static_cast<int>(context.graphic_ctx.screen_height);
 
+	SDL_SetHint(SDL_HINT_JOYSTICK_HIDAPI_PS4_RUMBLE, "1");
 	SDL_SetHint(SDL_HINT_JOYSTICK_HIDAPI_PS5_RUMBLE, "1");
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	SDL_SetHint(SDL_HINT_WINDOWS_DPI_SCALING, "0");
@@ -823,7 +823,7 @@ static void WindowCreate(WindowContext& context) {
 		EXIT("%s\n", SDL_GetError());
 	}
 	HostInputInit();
-	InitializeImeInput();
+	InitializeSystemOverlayInput();
 
 	LOGF("WindowCreate(): width = %d, height = %d\n", width, height);
 
@@ -899,6 +899,7 @@ void WindowRun() {
 
 void WindowShutdown() {
 	if (g_window != nullptr) {
+		Controller::EmergencyShutdown();
 		g_window.reset();
 	}
 }
@@ -1015,7 +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/presentation/window/windowInternal.h

@@ -35,6 +35,7 @@ struct WindowContext {
 	~WindowContext();
 	KYTY_CLASS_NO_COPY(WindowContext);
 
+	[[nodiscard]] static vk::PhysicalDeviceVulkan12Features RequiredVulkan12Features() noexcept;
 	[[nodiscard]] static vk::PhysicalDeviceVulkan13Features RequiredVulkan13Features() noexcept;
 	[[nodiscard]] static uint32_t InitialWindowFlags(bool fullscreen) noexcept;
 	void                                                    CreateVulkan();

src/graphics/shader/recompiler/BufferFormat.h

@@ -26,6 +26,28 @@ struct BufferFormatInfo {
 	bool                   packed_bitfield         = false;
 };
 
+enum class FormattedSourceKind { Memory, Zero, One, Invalid };
+
+struct FormattedSource {
+	FormattedSourceKind kind      = FormattedSourceKind::Zero;
+	uint32_t            component = 0;
+};
+
+constexpr FormattedSource ResolveFormattedSource(const BufferFormatInfo& info, uint32_t selector) {
+	if (selector == 0u) return {};
+	if (selector == 1u) return {FormattedSourceKind::One, 0};
+	if (selector < 4u || selector > 7u || info.component_count == 0u) {
+		return {FormattedSourceKind::Invalid, 0};
+	}
+	// Formatted decoding expands source channels before applying the swizzle.
+	return {FormattedSourceKind::Memory, (selector - 4u) % info.component_count};
+}
+
+constexpr uint32_t FormattedConstantBits(const BufferFormatInfo& info, FormattedSourceKind kind) {
+	if (kind != FormattedSourceKind::One) return 0;
+	return info.type == ComponentType::Uint || info.type == ComponentType::Sint ? 1u : 0x3f800000u;
+}
+
 constexpr Prospero::BufferFormat DecodeTBufferFormat(uint32_t data_format, uint32_t number_format) {
 	return static_cast<Prospero::BufferFormat>(((number_format & 0x7u) << 4u) |
 	                                           (data_format & 0xfu));
@@ -216,28 +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

@@ -14,7 +14,6 @@
 #include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 #include "graphics/shader/recompiler/ir/passes/ResourceTracking.h"
 #include "graphics/shader/recompiler/ir/passes/ShaderInfoCollection.h"
-#include "graphics/shader/recompiler/ir/passes/SharedMemoryBarrier.h"
 #include "graphics/shader/recompiler/ir/passes/SrtWalker.h"
 #include "graphics/shader/recompiler/ir/passes/SsaRewrite.h"
 
@@ -34,9 +33,9 @@ const char* GetDumpLabel(const CompileOptions& options) {
 	return options.dump_label != nullptr ? options.dump_label : "ShaderRecompiler";
 }
 
-std::string MakeIrDump(const CFG::Graph& cfg, const IR::Program& ir) {
+std::string MakeIrDump(std::string_view cfg, const IR::Program& ir) {
 	std::string dump = "CFG:\n";
-	dump += CFG::GraphToString(cfg);
+	dump += cfg;
 	dump += "\nIR:\n";
 	dump += fmt::format("mode={} scratch_dwords={}\n",
 	                    ir.dispatcher_fallback ? "dispatcher" : "structured", ir.scratch_dwords);
@@ -48,6 +47,7 @@ const char* StageName(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Compute: return "CS";
 		case ShaderType::Vertex: return "VS";
+		case ShaderType::Mesh: return "MS";
 		case ShaderType::Pixel: return "PS";
 		default: return "unknown";
 	}
@@ -77,8 +77,7 @@ enum class EmbeddedFetchValueType {
 	AttribTable,
 	Attrib,
 	BufferTable,
-	Buffer,
-	Index
+	Buffer
 };
 
 struct EmbeddedFetchSgprInfo {
@@ -88,10 +87,6 @@ struct EmbeddedFetchSgprInfo {
 	std::vector<uint32_t>  prolog_loads;
 };
 
-struct EmbeddedFetchVgprInfo {
-	EmbeddedFetchValueType type = EmbeddedFetchValueType::Unknown;
-};
-
 using EmbeddedFetchVectorLanes = std::map<uint64_t, EmbeddedFetchSgprInfo>;
 
 uint64_t EmbeddedFetchVectorLaneKey(uint32_t reg, uint32_t lane) {
@@ -108,7 +103,6 @@ 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) {
@@ -234,15 +228,17 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
                                             uint32_t wave_size) {
 	EmbeddedFetchData data;
 	data.loads.reserve(input_info->resources_num);
-	int32_t offset_candidate = -1;
-	bool    offset_conflict  = false;
+	int32_t vertex_offset_candidate   = -1;
+	int32_t instance_offset_candidate = -1;
+	bool    vertex_offset_conflict    = false;
+	bool    instance_offset_conflict  = false;
 
 	const int shift_regs = 8;
 	const int attrib_reg = input_info->fetch_attrib_reg + shift_regs;
 	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(),
@@ -263,34 +259,42 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 
 	for (const auto& inst: decoded.instructions) {
 		// Fetch shaders accumulate the draw's vertex offset in v0. The PS5 NGG ABI
-		// seeds S_NGG_VERTEX_INDEX in v5 and applies the same offset there before fetching.
+		// seeds S_NGG_VERTEX_INDEX in v5 and S_NGG_INSTANCE_INDEX in v8, then applies the
+		// 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));
+		const bool instance_index_accumulator =
+		    IsDecodedVgpr(inst.dst) && (inst.dst.reg == (user_data_base == 8 ? 8u : 3u));
 		uint32_t   sad_zero = 0;
-		const bool vertex_offset_add =
-		    vertex_index_accumulator && IsDecodedSgpr(inst.src0) &&
+		const bool index_offset_add =
+		    (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 &&
+		     (user_data_base == 8 && (inst.dst.reg == 5 || inst.dst.reg == 8) &&
 		      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));
-		if (data.loads.empty() && vertex_offset_add) {
+		if (data.loads.empty() && index_offset_add) {
 			const auto reg = DecodedSgprReg(inst.src0);
 			if (reg >= user_data_base && reg - user_data_base < user_data_count) {
-				if (offset_candidate >= 0 && offset_candidate != static_cast<int32_t>(reg)) {
-					offset_conflict = true;
+				auto& candidate = vertex_index_accumulator ? vertex_offset_candidate
+				                                           : instance_offset_candidate;
+				auto& conflict  = vertex_index_accumulator ? vertex_offset_conflict
+				                                           : instance_offset_conflict;
+				if (candidate >= 0 && candidate != static_cast<int32_t>(reg)) {
+					conflict = true;
 				} else {
-					offset_candidate = static_cast<int32_t>(reg);
+					candidate = static_cast<int32_t>(reg);
 				}
 			}
 		}
 		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() &&
@@ -395,13 +399,13 @@ 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() &&
+					if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgpr_is_index.size() &&
 					    IsDecodedVgpr(inst.src0) && inst.src0.reg == 8 &&
 					    IsDecodedVgpr(inst.src1) && inst.src1.reg == 5) {
-						vgprs[inst.dst.reg].type = EmbeddedFetchValueType::Index;
+						vgpr_is_index[inst.dst.reg] = true;
 					}
 				} else if (IsEmbeddedFetchAttribPropagationAlu(inst)) {
 					if (IsDecodedSgpr(inst.dst) && IsDecodedSgpr(inst.src0) &&
@@ -431,20 +435,24 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 						}
 					}
 				} 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;
+						const auto& buffer = sgprs[DecodedSgprReg(inst.src1)];
+						if (data.loads.empty()) {
+							if (!vertex_offset_conflict) {
+								data.vertex_offset_sgpr = vertex_offset_candidate;
+							}
+							if (!instance_offset_conflict) {
+								data.instance_offset_sgpr = instance_offset_candidate;
+							}
+						}
+						auto& load        = data.loads.emplace_back();
 						load.pc           = inst.pc;
 						load.attrib_id    = buffer.attrib_id;
 						load.components   = DecodedDstSize(inst);
 						load.prolog_loads = buffer.prolog_loads;
-						if (data.loads.empty() && !offset_conflict) {
-							data.vertex_offset_sgpr = offset_candidate;
-						}
-						data.loads.push_back(load);
 					}
 				}
 				break;
@@ -452,7 +460,8 @@ 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);
 			}
@@ -462,14 +471,53 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 	return data;
 }
 
+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);
+	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.
+	joined_code[front_words - 1u] = 0xbf820000u; // s_branch to the following instruction
+	result.instructions.back()    = {};
+	Decoder::DecodeInstruction(joined_code, front_words - 1u, result.instructions.back());
+	Decoder::Program back_program;
+	Decoder::DecodeProgram(back, back_program);
+	const auto back_pc = front_words * sizeof(uint32_t);
+	for (auto& inst: back_program.instructions) {
+		// A back-stage PC-relative data reference requires its guest code address.
+		EXIT_NOT_IMPLEMENTED(inst.opcode == Decoder::Opcode::S_GETPC_B64);
+		inst.pc += back_pc;
+		inst.branch_target += back_pc;
+		result.instructions.push_back(std::move(inst));
+	}
+	result.code = joined_code;
+	return result;
+}
+
 } // namespace
 
-CompileResult Recompile(std::span<const uint32_t> code, const CompileOptions& options) {
+TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOptions& options) {
 	if (code.empty()) {
 		EXIT("shader recompiler input is empty\n");
 	}
 	if (options.stage != ShaderType::Compute && options.stage != ShaderType::Vertex &&
-	    options.stage != ShaderType::Pixel) {
+	    options.stage != ShaderType::Pixel && options.stage != ShaderType::Mesh) {
 		EXIT("shader recompiler received unsupported stage %u\n",
 		     static_cast<unsigned>(options.stage));
 	}
@@ -486,7 +534,12 @@ CompileResult Recompile(std::span<const uint32_t> code, const CompileOptions& op
 	     static_cast<uint64_t>(code.size()));
 
 	Decoder::Program decoded;
-	Decoder::DecodeProgram(code, decoded);
+	std::vector<uint32_t> joined_code;
+	if (!options.back_code.empty()) {
+		decoded = DecodeFusedProgram(code, options.back_code, joined_code);
+	} else {
+		Decoder::DecodeProgram(code, decoded);
+	}
 	LOGF("%s phase end: stage=%s hash=0x%016" PRIx64 " decode instructions=%" PRIu64
 	     " elapsed_ms=%" PRIu64 "\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
@@ -545,8 +598,7 @@ CompileResult Recompile(std::span<const uint32_t> code, const CompileOptions& op
 	const ShaderComputeInputInfo* compute = nullptr;
 	switch (options.stage) {
 		case ShaderType::Vertex:
-			vertex = options.input_info.vertex;
-			break;
+		case ShaderType::Mesh: vertex = options.input_info.vertex; break;
 		case ShaderType::Pixel:
 			pixel = options.input_info.pixel;
 			break;
@@ -602,46 +654,49 @@ CompileResult Recompile(std::span<const uint32_t> code, const CompileOptions& op
 		IR::RemoveIdentities(ir.blocks);
 		IR::EliminateDeadCode(ir.blocks);
 	}
-	if (options.stage == ShaderType::Compute) {
-		const auto lds_barriers =
-		    IR::InsertSharedMemoryBarriers(ir, ir.wave_size, *compute);
-		if (lds_barriers.inserted_barriers != 0) {
-			LOGF("%s wave64 LDS synchronization: barriers=%" PRIu32 "\n", GetDumpLabel(options),
-			     lds_barriers.inserted_barriers);
-		}
-	}
 	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;
+	TranslateResult result;
+	result.program = std::move(ir);
+	if (options.dump_ir) {
+		result.decoded_dump = std::move(decoded_dump);
+		result.cfg_dump     = CFG::GraphToString(cfg);
 	}
+	return result;
+}
 
-	const IR::SrtRuntime runtime {
-	    .user_data                  = options.user_data,
-	    .shader_base                = reinterpret_cast<uint64_t>(code.data()),
-	    .read_memory                = options.read_memory,
-	    .userdata                   = options.read_memory_data,
-	    .read_specialization_memory = options.read_specialization_memory,
-	};
-	IR::ResourceSnapshot resources;
-	if (!IR::MaterializeResources(ir, runtime, resources)) {
-		EXIT("shader resource materialization failed: stage=%s hash=0x%016" PRIx64 "\n",
-		     StageName(options.stage), options.shader_hash);
+CompileResult CompileProgram(TranslateResult translated, const CompileOptions& options,
+                             const IR::ResourceSpecialization& specialization,
+                             uint32_t push_data_start_dword) {
+	const auto emit_begin = std::chrono::steady_clock::now();
+	auto& ir = translated.program;
+	IR::ApplyResourceSpecialization(ir, specialization);
+	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::SpecializeResources(ir, resources);
 
 	IR::ShaderInfoOptions info_options;
 	info_options.vertex  = vertex;
 	info_options.pixel   = pixel;
 	info_options.compute = compute;
 	IR::CollectShaderInfo(ir, info_options);
-	IR::AllocateBindings(ir, options.push_constant_offset);
+	IR::AllocateBindings(ir, push_data_start_dword);
 	Spirv::AnalyzeProgramRequirements(ir);
 	std::string ir_dump;
 	if (options.dump_ir) {
-		ir_dump = MakeIrDump(cfg, ir);
+		ir_dump = MakeIrDump(translated.cfg_dump, ir);
 		if (options.early_dump) {
 			LOGF("%s native IR and bindings (early):\n%s", GetDumpLabel(options), ir_dump.c_str());
 		}
@@ -649,17 +704,19 @@ CompileResult Recompile(std::span<const uint32_t> code, const CompileOptions& op
 
 	LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " SPIR-V EmitProgram\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
-	auto spirv = Spirv::EmitProgram(ir, resources, options.input_info);
+	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,
-	     static_cast<uint64_t>(spirv.size()), phase_ms());
+	     static_cast<uint64_t>(spirv.size()),
+	     static_cast<uint64_t>(std::chrono::duration_cast<std::chrono::milliseconds>(
+	                               std::chrono::steady_clock::now() - emit_begin)
+	                               .count()));
 	CompileResult result;
-	result.spirv     = std::move(spirv);
-	result.program   = std::move(ir);
-	result.resources = std::move(resources);
+	result.spirv   = std::move(spirv);
+	result.program = std::move(ir);
 	if (options.dump_ir) {
-		result.decoded_dump = std::move(decoded_dump);
+		result.decoded_dump = std::move(translated.decoded_dump);
 		result.ir_dump      = std::move(ir_dump);
 	}
 	return result;

src/graphics/shader/recompiler/ShaderRecompiler.h

@@ -17,27 +17,33 @@ struct CompileOptions {
 	uint32_t                    user_data_base  = 0;
 	uint32_t                    scratch_dwords  = 0;
 	uint64_t                    shader_hash     = 0;
-	uint32_t                    push_constant_offset      = 0;
 	bool                        dump_ir                    = true;
 	bool                        early_dump                 = false;
 	const char*                 dump_label                 = nullptr;
 	std::span<const uint32_t>   user_data;
-	IR::SrtMemoryReader         read_memory                = nullptr;
-	IR::SrtMemoryReader         read_specialization_memory = nullptr;
-	void*                       read_memory_data           = nullptr;
+	std::span<const uint32_t>   back_code;
 	ShaderStageInputInfo        input_info;
 };
 
-struct CompileResult {
-	std::vector<uint32_t> spirv;
-	std::string           decoded_dump;
-	std::string           ir_dump;
-	IR::Program           program;
-	IR::ResourceSnapshot  resources;
+struct TranslateResult {
+	IR::Program program;
+	std::string decoded_dump;
+	std::string cfg_dump;
 };
 
-[[nodiscard]] CompileResult Recompile(std::span<const uint32_t> code,
-                                      const CompileOptions& options);
+struct CompileResult {
+	std::vector<uint32_t>  spirv;
+	std::string            decoded_dump;
+	std::string            ir_dump;
+	IR::Program            program;
+};
+
+[[nodiscard]] TranslateResult TranslateProgram(std::span<const uint32_t> code,
+                                               const CompileOptions& options);
+[[nodiscard]] CompileResult CompileProgram(TranslateResult translated,
+                                           const CompileOptions& options,
+                                           const IR::ResourceSpecialization& specialization,
+	                                       uint32_t push_data_start_dword = 0);
 
 } // namespace Libs::Graphics::ShaderRecompiler
 

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

@@ -2,6 +2,7 @@
 
 #include "common/debug.h"
 
+#include <algorithm>
 #include <cstring>
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv {
@@ -31,6 +32,10 @@ uint32_t Builder::AllocateId() {
 	return m_next_id++;
 }
 
+void Builder::RequireVersion(uint32_t version) {
+	m_version = std::max(m_version, version);
+}
+
 void Builder::RequireCapability(uint32_t capability) {
 	if (m_required_capabilities.insert(capability).second) {
 		AddCapability({capability});
@@ -190,6 +195,18 @@ void Builder::AddEntryPoint(uint32_t execution_model, uint32_t entry_point, cons
 	std::vector<uint32_t> 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 id = m_declarations[offset + 2u];
+				if (std::find(interfaces.begin(), interfaces.end(), id) == interfaces.end()) {
+					operands.push_back(id);
+				}
+			}
+			offset += count;
+		}
+	}
 	AppendInstruction(m_entry_points, 15u, operands);
 }
 

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

@@ -28,6 +28,7 @@ public:
 	KYTY_CLASS_DEFAULT_COPY(Builder);
 
 	uint32_t AllocateId();
+	void     RequireVersion(uint32_t version);
 	void     RequireCapability(uint32_t capability);
 	void     RequireExtension(const char* name);
 	uint32_t Import(const char* name);

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

@@ -81,13 +81,13 @@ void ValidateNativeProgram(const IR::Program& program) {
 	const bool uses_flattened_runtime =
 	    !program.srt_reads.empty() ||
 	     std::ranges::any_of(program.info.images, [](const IR::ImageResource& image) {
-		     return image.indirect_mapping_capacity != 0u;
+		     return image.indirect_search_iterations != 0u;
 	     });
 	if (uses_flattened_runtime) {
 		Expect(Kind::FlattenedSrt);
 	}
-	if (program.bindings.ShaderDataDwords() != 0 && program.bindings.push_constant_size == 0) {
-		Expect(Kind::UserData);
+	if (program.bindings.ShaderDataDwords() != 0 && !program.bindings.UsesPushData()) {
+		Expect(Kind::ShaderData);
 	}
 
 	std::array<bool, KindCount> seen {};
@@ -104,21 +104,19 @@ void ValidateNativeProgram(const IR::Program& program) {
 			Fail(program, "native shader plan is missing a required descriptor group");
 		}
 	}
-	const auto has_user_storage = present[static_cast<size_t>(Kind::UserData)];
-	if (program.bindings.push_constant_offset % sizeof(uint32_t) != 0 ||
-	    program.bindings.push_constant_offset + program.bindings.push_constant_size >
-	        IR::NativePushConstantSize ||
+	const auto has_shader_data_storage = present[static_cast<size_t>(Kind::ShaderData)];
+	const auto shader_data_dwords = program.bindings.ShaderDataDwords();
+	if ((program.bindings.UsesPushData() &&
+	     !IR::PushData::CanFit(program.bindings.push_data_start_dword, shader_data_dwords)) ||
 	    program.bindings.memory_offset_dword != program.bindings.user_data_registers.size() ||
 	    program.bindings.memory_offset_count != program.info.buffers.size() ||
-	    (!has_user_storage &&
-	     program.bindings.push_constant_size != program.bindings.ShaderDataDwords() * 4u) ||
-	    (has_user_storage && program.bindings.push_constant_size != 0) ||
+	    has_shader_data_storage != (shader_data_dwords != 0 && !program.bindings.UsesPushData()) ||
 	    !std::is_sorted(program.bindings.user_data_registers.begin(),
 	                    program.bindings.user_data_registers.end()) ||
 	    std::adjacent_find(program.bindings.user_data_registers.begin(),
 	                       program.bindings.user_data_registers.end()) !=
 	        program.bindings.user_data_registers.end()) {
-		Fail(program, "native user-data layout is inconsistent");
+		Fail(program, "native shader-data layout is inconsistent");
 	}
 
 	const auto planning_only_handle = [&](const IR::Inst& handle) {
@@ -191,6 +189,10 @@ void AnalyzeProgramRequirements(IR::Program& program) {
 	const auto MarkBallot = [&] { requirements.subgroup_ballot = true; };
 	for (const auto* block: program.blocks) {
 		for (const auto& inst: *block) {
+			if (IR::BufferAccessOf(inst.GetOpcode()) == IR::BufferAccess::Atomic &&
+			    inst.GetType() == IR::Type::U64) {
+				requirements.buffer_int64_atomics = true;
+			}
 			const auto address_access = IR::AddressOpcodeInfoOf(inst.GetOpcode()).access;
 			if (address_access != IR::AddressAccess::None) {
 				const auto memory_index = inst.Flags<IR::MemoryFlags>().index;
@@ -234,7 +236,8 @@ void AnalyzeProgramRequirements(IR::Program& program) {
 				if (kind != IR::ResourceKind::Lds && kind != IR::ResourceKind::Gds) {
 					Fail(program, "shared operation has invalid resource kind");
 				}
-				if (program.stage != ShaderType::Compute && kind == IR::ResourceKind::Lds) {
+				if (program.stage != ShaderType::Compute && program.stage != ShaderType::Mesh &&
+				    kind == IR::ResourceKind::Lds) {
 					requirements.function_lds = true;
 				}
 				if (shared_access == IR::SharedAccess::Append ||
@@ -257,7 +260,6 @@ void AnalyzeProgramRequirements(IR::Program& program) {
 					break;
 				}
 				case IR::ValueOpcode::DppUpdateU32:
-				case IR::ValueOpcode::WqmMask:
 				case IR::ValueOpcode::WriteLane: {
 					MarkBallot();
 					requirements.subgroup_local_invocation_id = true;
@@ -302,12 +304,11 @@ void AnalyzeProgramRequirements(IR::Program& program) {
 }
 
 std::vector<uint32_t> EmitProgram(const IR::Program& program,
-                                  const IR::ResourceSnapshot& resources,
                                   ShaderStageInputInfo input_info) {
 	using namespace Emitter;
 
 	if (program.stage != ShaderType::Compute && program.stage != ShaderType::Vertex &&
-	    program.stage != ShaderType::Pixel) {
+	    program.stage != ShaderType::Pixel && program.stage != ShaderType::Mesh) {
 		Fail(program, "binary SPIR-V emitter supports compute, vertex, and pixel shaders");
 	}
 	if (!program.srt_plan_complete || !program.resource_tracking_complete ||
@@ -315,17 +316,15 @@ std::vector<uint32_t> EmitProgram(const IR::Program& program,
 	    !program.spirv_requirements.has_value()) {
 		Fail(program, "SPIR-V emitter requires a fully planned native shader program");
 	}
-	if (!IR::ValidateResourceSnapshot(program, resources)) {
-		Fail(program, "resource snapshot is inconsistent with the native shader plan");
-	}
-	if (!IR::ValidateResourceSpecialization(program, resources)) {
-		Fail(program, "resource specialization is inconsistent with the native shader plan");
-	}
 	ValidateNativeProgram(program);
 	IR::ValidateProgram(program, true);
-	EmitterState state(program, resources, input_info);
-	state.stage     = program.stage;
-	state.wave_size = program.wave_size;
+	EmitterState state(program, input_info);
+	state.stage = program.stage;
+	const auto* workgroup = ShaderWorkgroupInput(program.stage, input_info);
+	state.lane_count =
+	    workgroup != nullptr && program.wave_size == 64u && workgroup->host_subgroup_size == 32u
+	        ? 2u
+	        : 1u;
 	state.inputs.reserve(program.info.inputs.size());
 	state.outputs.reserve(program.info.outputs.size());
 	state.interface_variables.reserve(program.info.inputs.size() + program.info.outputs.size());
@@ -333,7 +332,9 @@ std::vector<uint32_t> EmitProgram(const IR::Program& program,
 	AllocateInputVariables(state);
 	AllocateOutputVariables(state);
 	DefineModule(state);
-	EmitProgram(state, program);
+	EmitProgram(state);
+	state.builder.AddEntryPoint(ExecutionModelForStage(state.stage), state.main_func, "main",
+	                            state.interface_variables);
 
 	auto binary = state.builder.Build();
 	if (binary.empty()) {

src/graphics/shader/recompiler/backend/spirv/SpirvEmitter.h

@@ -3,7 +3,7 @@
 
 #include "common/common.h"
 #include "common/stringUtils.h"
-#include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
+#include "graphics/shader/recompiler/ir/ShaderIR.h"
 
 #include <vector>
 
@@ -12,7 +12,6 @@ namespace Libs::Graphics::ShaderRecompiler::Spirv {
 void AnalyzeProgramRequirements(IR::Program& program);
 
 std::vector<uint32_t> EmitProgram(const IR::Program& program,
-                                  const IR::ResourceSnapshot& resources,
                                   ShaderStageInputInfo input_info);
 
 } // namespace Libs::Graphics::ShaderRecompiler::Spirv

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

@@ -10,26 +10,6 @@ struct Pair {
 	uint32_t high = 0;
 };
 
-uint32_t NewUnary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t value) {
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, type, result, value});
-	return result;
-}
-
-uint32_t NewBinary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t lhs,
-                   uint32_t rhs) {
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, type, result, lhs, rhs});
-	return result;
-}
-
-uint32_t NewSelect(EmitterState& state, uint32_t type, uint32_t condition, uint32_t true_value,
-                   uint32_t false_value) {
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({OpSelect, type, result, condition, true_value, false_value});
-	return result;
-}
-
 Pair ExtractPair(EmitterState& state, uint32_t value) {
 	Pair result {state.builder.AllocateId(), state.builder.AllocateId()};
 	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), result.low, value, 0});
@@ -45,20 +25,20 @@ uint32_t MakePair(EmitterState& state, uint32_t low, uint32_t high) {
 
 uint32_t CompareEqual64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value,
                         bool not_equal) {
-	const auto compare = NewBinary(state, not_equal ? OpINotEqual : OpIEqual,
+	const auto compare = Binary(state, not_equal ? OpINotEqual : OpIEqual,
 	                               TypeBoolVector(state, 2), lhs_value, rhs_value);
-	return NewUnary(state, not_equal ? OpAny : OpAll, TypeBool(state), compare);
+	return Unary(state, not_equal ? OpAny : OpAll, TypeBool(state), compare);
 }
 
 uint32_t CompareOrdered64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value,
                           uint32_t high_compare, uint32_t low_compare) {
 	const auto lhs         = ExtractPair(state, lhs_value);
 	const auto rhs         = ExtractPair(state, rhs_value);
-	const auto high_equal  = NewBinary(state, OpIEqual, TypeBool(state), lhs.high, rhs.high);
-	const auto high_result = NewBinary(state, high_compare, TypeBool(state), lhs.high, rhs.high);
-	const auto low_result  = NewBinary(state, low_compare, TypeBool(state), lhs.low, rhs.low);
-	const auto low_path = NewBinary(state, OpLogicalAnd, TypeBool(state), high_equal, low_result);
-	return NewBinary(state, OpLogicalOr, TypeBool(state), high_result, low_path);
+	const auto high_equal  = Binary(state, OpIEqual, TypeBool(state), lhs.high, rhs.high);
+	const auto high_result = Binary(state, high_compare, TypeBool(state), lhs.high, rhs.high);
+	const auto low_result  = Binary(state, low_compare, TypeBool(state), lhs.low, rhs.low);
+	const auto low_path = Binary(state, OpLogicalAnd, TypeBool(state), high_equal, low_result);
+	return Binary(state, OpLogicalOr, TypeBool(state), high_result, low_path);
 }
 
 uint32_t EmitAdd64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value) {
@@ -70,20 +50,20 @@ uint32_t EmitAdd64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value)
 	state.builder.AddFunction({OpIAddCarry, TypeU32Pair(state), low_pair, lhs.low, rhs.low});
 	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), low, low_pair, 0});
 	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), carry, low_pair, 1});
-	const auto high0 = NewBinary(state, OpIAdd, TypeU32(state), lhs.high, rhs.high);
-	const auto high  = NewBinary(state, OpIAdd, TypeU32(state), high0, carry);
+	const auto high0 = Binary(state, OpIAdd, TypeU32(state), lhs.high, rhs.high);
+	const auto high  = Binary(state, OpIAdd, TypeU32(state), high0, carry);
 	return MakePair(state, low, high);
 }
 
 uint32_t EmitSub64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value) {
 	const auto lhs    = ExtractPair(state, lhs_value);
 	const auto rhs    = ExtractPair(state, rhs_value);
-	const auto low    = NewBinary(state, OpISub, TypeU32(state), lhs.low, rhs.low);
-	const auto borrow = NewBinary(state, OpULessThan, TypeBool(state), lhs.low, rhs.low);
+	const auto low    = Binary(state, OpISub, TypeU32(state), lhs.low, rhs.low);
+	const auto borrow = Binary(state, OpULessThan, TypeBool(state), lhs.low, rhs.low);
 	const auto borrow_u32 =
-	    NewSelect(state, TypeU32(state), borrow, ConstantU32(state, 1), ConstantU32(state, 0));
-	const auto high0 = NewBinary(state, OpISub, TypeU32(state), lhs.high, rhs.high);
-	const auto high  = NewBinary(state, OpISub, TypeU32(state), high0, borrow_u32);
+	    Select(state, TypeU32(state), borrow, ConstantU32(state, 1), ConstantU32(state, 0));
+	const auto high0 = Binary(state, OpISub, TypeU32(state), lhs.high, rhs.high);
+	const auto high  = Binary(state, OpISub, TypeU32(state), high0, borrow_u32);
 	return MakePair(state, low, high);
 }
 
@@ -93,27 +73,27 @@ uint32_t EmitMulHigh(EmitterState& state, uint32_t lhs, uint32_t rhs, bool signe
 	uint32_t   lhs_operand  = lhs;
 	uint32_t   rhs_operand  = rhs;
 	if (signed_value) {
-		lhs_operand = NewUnary(state, OpBitcast, TypeI32(state), lhs);
-		rhs_operand = NewUnary(state, OpBitcast, TypeI32(state), rhs);
+		lhs_operand = Unary(state, OpBitcast, TypeI32(state), lhs);
+		rhs_operand = Unary(state, OpBitcast, TypeI32(state), rhs);
 	}
 	const auto extended = state.builder.AllocateId();
 	state.builder.AddFunction({signed_value ? OpSMulExtended : OpUMulExtended, pair_type, extended,
 	                           lhs_operand, rhs_operand});
 	const auto high = state.builder.AllocateId();
 	state.builder.AddFunction({OpCompositeExtract, operand_type, high, extended, 1});
-	return signed_value ? NewUnary(state, OpBitcast, TypeU32(state), high) : high;
+	return signed_value ? Unary(state, OpBitcast, TypeU32(state), high) : high;
 }
 
 uint32_t EmitMul64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value) {
 	const auto lhs   = ExtractPair(state, lhs_value);
 	const auto rhs   = ExtractPair(state, rhs_value);
-	const auto low   = NewBinary(state, OpIMul, TypeU32(state), lhs.low, rhs.low);
+	const auto low   = Binary(state, OpIMul, TypeU32(state), lhs.low, rhs.low);
 	const auto high0 = EmitMulHigh(state, lhs.low, rhs.low, false);
-	const auto high1 = NewBinary(state, OpIMul, TypeU32(state), lhs.low, rhs.high);
-	const auto high2 = NewBinary(state, OpIMul, TypeU32(state), lhs.high, rhs.low);
+	const auto high1 = Binary(state, OpIMul, TypeU32(state), lhs.low, rhs.high);
+	const auto high2 = Binary(state, OpIMul, TypeU32(state), lhs.high, rhs.low);
 	return MakePair(state, low,
-	                NewBinary(state, OpIAdd, TypeU32(state),
-	                          NewBinary(state, OpIAdd, TypeU32(state), high0, high1), high2));
+	                Binary(state, OpIAdd, TypeU32(state),
+	                          Binary(state, OpIAdd, TypeU32(state), high0, high1), high2));
 }
 
 uint32_t EmitShift64(EmitterState& state, uint32_t opcode, uint32_t value, uint32_t shift) {
@@ -126,33 +106,33 @@ uint32_t EmitShift64(EmitterState& state, uint32_t opcode, uint32_t value, uint3
 		EmitShiftRightLogicalU64Values(state, pair.low, pair.high, shift, low, high);
 	} else {
 		const auto amount =
-		    NewBinary(state, OpBitwiseAnd, TypeU32(state), shift, ConstantU32(state, 63));
+		    Binary(state, OpBitwiseAnd, TypeU32(state), shift, ConstantU32(state, 63));
 		const auto word_shift =
-		    NewBinary(state, OpBitwiseAnd, TypeU32(state), amount, ConstantU32(state, 31));
+		    Binary(state, OpBitwiseAnd, TypeU32(state), amount, ConstantU32(state, 31));
 		const auto at_least_32 =
-		    NewBinary(state, OpUGreaterThanEqual, TypeBool(state), amount, ConstantU32(state, 32));
+		    Binary(state, OpUGreaterThanEqual, TypeBool(state), amount, ConstantU32(state, 32));
 		const auto nonzero =
-		    NewBinary(state, OpINotEqual, TypeBool(state), amount, ConstantU32(state, 0));
+		    Binary(state, OpINotEqual, TypeBool(state), amount, ConstantU32(state, 0));
 		const auto high_shifted =
-		    NewBinary(state, OpShiftRightArithmetic, TypeU32(state), pair.high, word_shift);
+		    Binary(state, OpShiftRightArithmetic, TypeU32(state), pair.high, word_shift);
 		const auto low_shifted =
-		    NewBinary(state, OpShiftRightLogical, TypeU32(state), pair.low, word_shift);
+		    Binary(state, OpShiftRightLogical, TypeU32(state), pair.low, word_shift);
 		const auto carry_count =
-		    NewBinary(state, OpBitwiseAnd, TypeU32(state),
-		              NewBinary(state, OpISub, TypeU32(state), ConstantU32(state, 32), word_shift),
+		    Binary(state, OpBitwiseAnd, TypeU32(state),
+		              Binary(state, OpISub, TypeU32(state), ConstantU32(state, 32), word_shift),
 		              ConstantU32(state, 31));
 		const auto carry =
-		    NewSelect(state, TypeU32(state), nonzero,
-		              NewBinary(state, OpShiftLeftLogical, TypeU32(state), pair.high, carry_count),
+		    Select(state, TypeU32(state), nonzero,
+		              Binary(state, OpShiftLeftLogical, TypeU32(state), pair.high, carry_count),
 		              ConstantU32(state, 0));
-		const auto low_below_32 = NewBinary(state, OpBitwiseOr, TypeU32(state), low_shifted, carry);
+		const auto low_below_32 = Binary(state, OpBitwiseOr, TypeU32(state), low_shifted, carry);
 		const auto negative =
-		    NewBinary(state, OpSLessThan, TypeBool(state), pair.high, ConstantU32(state, 0));
-		const auto sign_fill = NewSelect(state, TypeU32(state), negative,
+		    Binary(state, OpSLessThan, TypeBool(state), pair.high, ConstantU32(state, 0));
+		const auto sign_fill = Select(state, TypeU32(state), negative,
 		                                 ConstantU32(state, 0xffffffffu), ConstantU32(state, 0));
 		return MakePair(state,
-		                NewSelect(state, TypeU32(state), at_least_32, high_shifted, low_below_32),
-		                NewSelect(state, TypeU32(state), at_least_32, sign_fill, high_shifted));
+		                Select(state, TypeU32(state), at_least_32, high_shifted, low_below_32),
+		                Select(state, TypeU32(state), at_least_32, sign_fill, high_shifted));
 	}
 	return MakePair(state, low, high);
 }
@@ -165,36 +145,36 @@ uint32_t EmitConstantShift64(EmitterState& state, uint32_t opcode, uint32_t valu
 	const auto pair = ExtractPair(state, value);
 	if (opcode == OpShiftLeftLogical) {
 		if (shift < 32u) {
-			const auto low  = NewBinary(state, OpShiftLeftLogical, TypeU32(state), pair.low,
+			const auto low  = Binary(state, OpShiftLeftLogical, TypeU32(state), pair.low,
 			                            ConstantU32(state, shift));
-			const auto high = NewBinary(state, OpBitwiseOr, TypeU32(state),
-			                            NewBinary(state, OpShiftLeftLogical, TypeU32(state),
+			const auto high = Binary(state, OpBitwiseOr, TypeU32(state),
+			                            Binary(state, OpShiftLeftLogical, TypeU32(state),
 			                                      pair.high, ConstantU32(state, shift)),
-			                            NewBinary(state, OpShiftRightLogical, TypeU32(state),
+			                            Binary(state, OpShiftRightLogical, TypeU32(state),
 			                                      pair.low, ConstantU32(state, 32u - shift)));
 			return MakePair(state, low, high);
 		}
 		return MakePair(state, ConstantU32(state, 0),
 		                shift == 32u ? pair.low
-		                             : NewBinary(state, OpShiftLeftLogical, TypeU32(state),
+		                             : Binary(state, OpShiftLeftLogical, TypeU32(state),
 		                                         pair.low, ConstantU32(state, shift - 32u)));
 	}
 	if (shift < 32u) {
-		const auto low = NewBinary(state, OpBitwiseOr, TypeU32(state),
-		                           NewBinary(state, OpShiftRightLogical, TypeU32(state), pair.low,
+		const auto low = Binary(state, OpBitwiseOr, TypeU32(state),
+		                           Binary(state, OpShiftRightLogical, TypeU32(state), pair.low,
 		                                     ConstantU32(state, shift)),
-		                           NewBinary(state, OpShiftLeftLogical, TypeU32(state), pair.high,
+		                           Binary(state, OpShiftLeftLogical, TypeU32(state), pair.high,
 		                                     ConstantU32(state, 32u - shift)));
 		const auto high =
-		    NewBinary(state, opcode, TypeU32(state), pair.high, ConstantU32(state, shift));
+		    Binary(state, opcode, TypeU32(state), pair.high, ConstantU32(state, shift));
 		return MakePair(state, low, high);
 	}
 	const auto high = opcode == OpShiftRightArithmetic
-	                      ? NewBinary(state, OpShiftRightArithmetic, TypeU32(state), pair.high,
+	                      ? Binary(state, OpShiftRightArithmetic, TypeU32(state), pair.high,
 	                                  ConstantU32(state, 31u))
 	                      : ConstantU32(state, 0);
 	const auto low  = shift == 32u ? pair.high
-	                               : NewBinary(state, opcode, TypeU32(state), pair.high,
+	                               : Binary(state, opcode, TypeU32(state), pair.high,
 	                                           ConstantU32(state, shift - 32u));
 	return MakePair(state, low, high);
 }
@@ -207,12 +187,12 @@ uint32_t EmitFindMsb64(EmitterState& state, uint32_t value) {
 	    {OpExtInst, TypeI32(state), high_i, GlslStd450(state), GlslFindUMsb, pair.high});
 	state.builder.AddFunction(
 	    {OpExtInst, TypeI32(state), low_i, GlslStd450(state), GlslFindUMsb, pair.low});
-	const auto high = NewUnary(state, OpBitcast, TypeU32(state), high_i);
-	const auto low  = NewUnary(state, OpBitcast, TypeU32(state), low_i);
+	const auto high = Unary(state, OpBitcast, TypeU32(state), high_i);
+	const auto low  = Unary(state, OpBitcast, TypeU32(state), low_i);
 	const auto high_nonzero =
-	    NewBinary(state, OpINotEqual, TypeBool(state), pair.high, ConstantU32(state, 0));
-	return NewSelect(state, TypeU32(state), high_nonzero,
-	                 NewBinary(state, OpIAdd, TypeU32(state), high, ConstantU32(state, 32)), low);
+	    Binary(state, OpINotEqual, TypeBool(state), pair.high, ConstantU32(state, 0));
+	return Select(state, TypeU32(state), high_nonzero,
+	                 Binary(state, OpIAdd, TypeU32(state), high, ConstantU32(state, 32)), low);
 }
 
 uint32_t EmitMinMax3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c, bool signed_value,
@@ -226,10 +206,10 @@ uint32_t EmitMinMax3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c, bo
 uint32_t EmitMed3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c, bool signed_value) {
 	const auto minimum = EmitMinMax3(state, a, b, c, signed_value, false);
 	const auto maximum = EmitMinMax3(state, a, b, c, signed_value, true);
-	const auto ab      = NewBinary(state, OpIAdd, TypeU32(state), a, b);
-	const auto abc     = NewBinary(state, OpIAdd, TypeU32(state), ab, c);
-	return NewBinary(state, OpISub, TypeU32(state),
-	                 NewBinary(state, OpISub, TypeU32(state), abc, minimum), maximum);
+	const auto ab      = Binary(state, OpIAdd, TypeU32(state), a, b);
+	const auto abc     = Binary(state, OpIAdd, TypeU32(state), ab, c);
+	return Binary(state, OpISub, TypeU32(state),
+	                 Binary(state, OpISub, TypeU32(state), abc, minimum), maximum);
 }
 
 uint32_t EmitFMinMax3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c, bool max_value) {
@@ -242,11 +222,11 @@ uint32_t EmitFMed3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c) {
 	const auto max_ab   = EmitMinMaxF32Value(state, a, b, true);
 	const auto high_min = EmitMinMaxF32Value(state, max_ab, c, false);
 	const auto median   = EmitMinMaxF32Value(state, min_ab, high_min, true);
-	const auto nan_ab   = NewBinary(state, OpLogicalOr, TypeBool(state),
+	const auto nan_ab   = Binary(state, OpLogicalOr, TypeBool(state),
 	                                EmitClassifyF32(state, a).nan, EmitClassifyF32(state, b).nan);
 	const auto any_nan =
-	    NewBinary(state, OpLogicalOr, TypeBool(state), nan_ab, EmitClassifyF32(state, c).nan);
-	return NewSelect(state, TypeF32(state), any_nan, min3, median);
+	    Binary(state, OpLogicalOr, TypeBool(state), nan_ab, EmitClassifyF32(state, c).nan);
+	return Select(state, TypeF32(state), any_nan, min3, median);
 }
 
 uint32_t EmitExt(EmitterState& state, uint32_t type, uint32_t opcode,
@@ -270,37 +250,30 @@ uint32_t EmitF32ToU32(EmitterState& state, uint32_t src, bool signed_value) {
 	}
 	const auto nan = EmitClassifyF32(state, src).nan;
 	if (signed_value) {
-		const auto below = NewBinary(state, OpFOrdLessThanEqual, TypeBool(state), src,
+		const auto below = Binary(state, OpFOrdLessThanEqual, TypeBool(state), src,
 		                             ConstantF32(state, 0xcf000000u));
-		const auto above = NewBinary(state, OpFOrdGreaterThanEqual, TypeBool(state), src,
+		const auto above = Binary(state, OpFOrdGreaterThanEqual, TypeBool(state), src,
 		                             ConstantF32(state, 0x4f000000u));
 		const auto high =
-		    NewSelect(state, TypeU32(state), above, ConstantU32(state, 0x7fffffffu), converted_raw);
+		    Select(state, TypeU32(state), above, ConstantU32(state, 0x7fffffffu), converted_raw);
 		const auto low =
-		    NewSelect(state, TypeU32(state), below, ConstantU32(state, 0x80000000u), high);
-		return NewSelect(state, TypeU32(state), nan, ConstantU32(state, 0), low);
+		    Select(state, TypeU32(state), below, ConstantU32(state, 0x80000000u), high);
+		return Select(state, TypeU32(state), nan, ConstantU32(state, 0), low);
 	}
 	const auto below =
-	    NewBinary(state, OpFOrdLessThanEqual, TypeBool(state), src, ConstantF32(state, 0));
-	const auto above = NewBinary(state, OpFOrdGreaterThanEqual, TypeBool(state), src,
+	    Binary(state, OpFOrdLessThanEqual, TypeBool(state), src, ConstantF32(state, 0));
+	const auto above = Binary(state, OpFOrdGreaterThanEqual, TypeBool(state), src,
 	                             ConstantF32(state, 0x4f800000u));
-	const auto zero  = NewBinary(state, OpLogicalOr, TypeBool(state), nan, below);
+	const auto zero  = Binary(state, OpLogicalOr, TypeBool(state), nan, below);
 	const auto high =
-	    NewSelect(state, TypeU32(state), above, ConstantU32(state, 0xffffffffu), converted_raw);
-	return NewSelect(state, TypeU32(state), zero, ConstantU32(state, 0), high);
+	    Select(state, TypeU32(state), above, ConstantU32(state, 0xffffffffu), converted_raw);
+	return Select(state, TypeU32(state), zero, ConstantU32(state, 0), high);
 }
 
 uint32_t EmitPackHalf(EmitterState& state, uint32_t src) {
 	return EmitExt(state, TypeU32(state), GlslPackHalf2x16, {src});
 }
 
-uint32_t EmitUnpackHalf(EmitterState& state, uint32_t bits) {
-	const auto pair   = EmitExt(state, TypeF32Vector(state, 2), GlslUnpackHalf2x16, {bits});
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({OpCompositeExtract, TypeF32(state), result, pair, 0});
-	return result;
-}
-
 } // namespace
 
 bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
@@ -341,7 +314,7 @@ bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
 			return true;
 		}
 		case IR::ValueOpcode::ConvertF32F16:
-			ctx.Define(inst, EmitUnpackHalf(state, ctx.Arg(inst, 0)));
+			ctx.Define(inst, EmitF16BitsToF32(state, ctx.Arg(inst, 0)));
 			return true;
 		case IR::ValueOpcode::ConvertS32F32:
 			ctx.Define(inst, EmitF32ToU32(state, ctx.Arg(inst, 0), true));
@@ -350,7 +323,7 @@ bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
 			ctx.Define(inst, EmitF32ToU32(state, ctx.Arg(inst, 0), false));
 			return true;
 		case IR::ValueOpcode::ConvertF32S32: {
-			const auto signed_value = NewUnary(state, OpBitcast, TypeI32(state), ctx.Arg(inst, 0));
+			const auto signed_value = Unary(state, OpBitcast, TypeI32(state), ctx.Arg(inst, 0));
 			ctx.Emit(inst, OpConvertSToF, IR::Type::F32, {signed_value});
 			return true;
 		}
@@ -394,9 +367,9 @@ bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
 		case IR::ValueOpcode::PackFloat2x16Rtz: {
 			const auto low = EmitF32ToF16RtzBits(state, ctx.Arg(inst, 0));
 			const auto high =
-			    NewBinary(state, OpShiftLeftLogical, TypeU32(state),
+			    Binary(state, OpShiftLeftLogical, TypeU32(state),
 			              EmitF32ToF16RtzBits(state, ctx.Arg(inst, 1)), ConstantU32(state, 16));
-			ctx.Define(inst, NewBinary(state, OpBitwiseOr, TypeU32(state), low, high));
+			ctx.Define(inst, Binary(state, OpBitwiseOr, TypeU32(state), low, high));
 			return true;
 		}
 		case IR::ValueOpcode::FPAbs32:
@@ -458,10 +431,10 @@ bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
 			return true;
 		case IR::ValueOpcode::IAbs32: {
 			const auto value = ctx.Arg(inst, 0);
-			const auto neg   = NewUnary(state, OpSNegate, TypeU32(state), value);
+			const auto neg   = Unary(state, OpSNegate, TypeU32(state), value);
 			const auto negative =
-			    NewBinary(state, OpSLessThan, TypeBool(state), value, ConstantU32(state, 0));
-			ctx.Define(inst, NewSelect(state, TypeU32(state), negative, neg, value));
+			    Binary(state, OpSLessThan, TypeBool(state), value, ConstantU32(state, 0));
+			ctx.Define(inst, Select(state, TypeU32(state), negative, neg, value));
 			return true;
 		}
 		case IR::ValueOpcode::ShiftLeftLogical32: return binary(OpShiftLeftLogical, IR::Type::U32);
@@ -496,25 +469,25 @@ bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
 		case IR::ValueOpcode::BitwiseXor32: return binary(OpBitwiseXor, IR::Type::U32);
 		case IR::ValueOpcode::BitwiseNot32: return unary(OpNot, IR::Type::U32);
 		case IR::ValueOpcode::BitwiseAnd64:
-			ctx.Define(inst, NewBinary(state, OpBitwiseAnd, TypeU64(state), ctx.Arg(inst, 0),
+			ctx.Define(inst, Binary(state, OpBitwiseAnd, TypeU64(state), ctx.Arg(inst, 0),
 			                           ctx.Arg(inst, 1)));
 			return true;
 		case IR::ValueOpcode::BitReverse32: return unary(OpBitReverse, IR::Type::U32);
 		case IR::ValueOpcode::BitCount32: return unary(OpBitCount, IR::Type::U32);
 		case IR::ValueOpcode::BitCount64: {
 			const auto pair =
-			    ExtractPair(state, NewUnary(state, OpBitCount, TypeU64(state), ctx.Arg(inst, 0)));
-			ctx.Define(inst, NewBinary(state, OpIAdd, TypeU32(state), pair.low, pair.high));
+			    ExtractPair(state, Unary(state, OpBitCount, TypeU64(state), ctx.Arg(inst, 0)));
+			ctx.Define(inst, Binary(state, OpIAdd, TypeU32(state), pair.low, pair.high));
 			return true;
 		}
 		case IR::ValueOpcode::FindILsb32: {
 			const auto value = EmitExt(state, TypeI32(state), GlslFindILsb, {ctx.Arg(inst, 0)});
-			ctx.Define(inst, NewUnary(state, OpBitcast, TypeU32(state), value));
+			ctx.Define(inst, Unary(state, OpBitcast, TypeU32(state), value));
 			return true;
 		}
 		case IR::ValueOpcode::FindUMsb32: {
 			const auto value = EmitExt(state, TypeI32(state), GlslFindUMsb, {ctx.Arg(inst, 0)});
-			ctx.Define(inst, NewUnary(state, OpBitcast, TypeU32(state), value));
+			ctx.Define(inst, Unary(state, OpBitcast, TypeU32(state), value));
 			return true;
 		}
 		case IR::ValueOpcode::FindUMsb64:
@@ -638,13 +611,13 @@ bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
 			return true;
 		case IR::ValueOpcode::FPRecip32: {
 			const auto source = EmitFlushF32DenormToSignedZero(state, ctx.Arg(inst, 0));
-			ctx.Define(inst, NewBinary(state, OpFDiv, TypeF32(state),
+			ctx.Define(inst, Binary(state, OpFDiv, TypeF32(state),
 			                           ConstantF32(state, 0x3f800000u), source));
 			return true;
 		}
 		case IR::ValueOpcode::FPRecipIFlag32:
 			// Integer-to-float inputs used by IFLAG cannot be denormal.
-			ctx.Define(inst, NewBinary(state, OpFDiv, TypeF32(state),
+			ctx.Define(inst, Binary(state, OpFDiv, TypeF32(state),
 			                           ConstantF32(state, 0x3f800000u), ctx.Arg(inst, 0)));
 			return true;
 		case IR::ValueOpcode::FPRecipSqrt32:
@@ -659,7 +632,7 @@ bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
 		case IR::ValueOpcode::FPCos: {
 			auto source = EmitTrigCycleF32(state, ctx.Arg(inst, 0), op == IR::ValueOpcode::FPSin);
 			source =
-			    NewBinary(state, OpFMul, TypeF32(state), source, ConstantF32(state, 0x40c90fdbu));
+			    Binary(state, OpFMul, TypeF32(state), source, ConstantF32(state, 0x40c90fdbu));
 			ctx.Define(inst, EmitExt(state, TypeF32(state),
 			                         op == IR::ValueOpcode::FPSin ? GlslSin : GlslCos, {source}));
 			return true;
@@ -673,7 +646,7 @@ bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
 			                         {EmitFlushF32DenormToSignedZero(state, ctx.Arg(inst, 0))}));
 			return true;
 		case IR::ValueOpcode::FPLdexp: {
-			const auto exponent = NewUnary(state, OpBitcast, TypeI32(state), ctx.Arg(inst, 1));
+			const auto exponent = Unary(state, OpBitcast, TypeI32(state), ctx.Arg(inst, 1));
 			ctx.Define(inst,
 			           EmitExt(state, TypeF32(state), GlslLdexp, {ctx.Arg(inst, 0), exponent}));
 			return true;

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

@@ -215,9 +215,9 @@ uint32_t EmitLogicalNotBool(EmitterState& state, uint32_t value) {
 	return ret;
 }
 
-F32Class EmitClassifyF32(EmitterState& state, uint32_t value) {
+F32Class EmitClassifyF32Bits(EmitterState& state, uint32_t bits) {
 	F32Class cls;
-	cls.bits                 = EmitBitcastF32ToU32(state, value);
+	cls.bits                 = bits;
 	const auto abs_bits      = EmitAndConstant(state, cls.bits, 0x7fffffffu);
 	const auto exponent_bits = EmitAndConstant(state, abs_bits, 0x7f800000u);
 	const auto mantissa_bits = EmitAndConstant(state, abs_bits, 0x007fffffu);
@@ -228,6 +228,10 @@ F32Class EmitClassifyF32(EmitterState& state, uint32_t value) {
 	return cls;
 }
 
+F32Class EmitClassifyF32(EmitterState& state, uint32_t value) {
+	return EmitClassifyF32Bits(state, EmitBitcastF32ToU32(state, value));
+}
+
 uint32_t EmitClassMaskBitMatch(EmitterState& state, uint32_t mask, uint32_t bit,
                                uint32_t class_match) {
 	const auto selected =

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

@@ -6,13 +6,6 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
-uint32_t PixelParameterMappedLocation(const EmitterState& state, uint32_t attr) {
-	if (state.stage != ShaderType::Pixel) {
-		return attr;
-	}
-	return ShaderPixelParameterMappedLocation(*state.input_info.pixel, attr);
-}
-
 uint32_t PixelParameterLocation(const EmitterState& state, uint32_t attr) {
 	std::array<uint32_t, 32> active_inputs {};
 	uint32_t                 active_count = 0;
@@ -86,8 +79,8 @@ uint32_t          ConstantU32(EmitterState& state, uint32_t value);
 	std::abort();
 }
 
-DescriptorResourceBinding ResourceForDescriptor(const EmitterState&       state,
-                                                IR::DescriptorBindingKind kind, uint32_t resource) {
+uint32_t ResourceForDescriptor(const EmitterState& state, IR::DescriptorBindingKind kind,
+                               uint32_t resource) {
 	const auto* descriptor = IR::FindBinding(state.program.bindings, kind);
 	if (descriptor == nullptr) {
 		ExitDescriptorBindingFailure(state, kind, resource, "descriptor group was not allocated");
@@ -98,7 +91,7 @@ DescriptorResourceBinding ResourceForDescriptor(const EmitterState&       state,
 		ExitDescriptorBindingFailure(state, kind, resource,
 		                             "resource is absent from descriptor group");
 	}
-	return {descriptor, static_cast<uint32_t>(found - descriptor->resources.begin())};
+	return static_cast<uint32_t>(found - descriptor->resources.begin());
 }
 
 uint32_t DescriptorElementPointer(EmitterState& state, uint32_t result_ptr_type,
@@ -114,68 +107,57 @@ uint32_t DescriptorElementPointer(EmitterState& state, uint32_t result_ptr_type,
 	return pointer;
 }
 
-ImageViewKind ImageViewKindFromDimension(Decoder::ImageDimension dimension) {
-	switch (dimension) {
-		case Decoder::ImageDimension::Dim1D: return ImageViewKind::Dim1D;
-		case Decoder::ImageDimension::Dim1DArray: return ImageViewKind::Dim1DArray;
-		case Decoder::ImageDimension::Dim2DArray: return ImageViewKind::Dim2DArray;
-		case Decoder::ImageDimension::Dim3D: return ImageViewKind::Dim3D;
-		case Decoder::ImageDimension::Dim2DMsaa: return ImageViewKind::Dim2DMsaa;
-		case Decoder::ImageDimension::Dim2DMsaaArray: return ImageViewKind::Dim2DMsaaArray;
-		default: return ImageViewKind::Dim2D;
+const ImageDimensionInfo& ImageDimensionInfoFor(ImageDimension dimension) {
+	for (const auto& info: ImageDimensions) {
+		if (info.dimension == dimension) {
+			return info;
+		}
 	}
+	EXIT("SPIR-V image dimension %u is invalid\n", static_cast<uint32_t>(dimension));
+	std::abort();
 }
 
-ImageViewKind SampledImageViewKind(const EmitterState& state, const IR::MemoryInfo& mem,
-                                   uint32_t use_pc) {
-	(void)state;
-	(void)use_pc;
-	return ImageViewKindFromDimension(mem.image_dimension);
-}
-
-ImageViewKind StorageImageViewKind(const IR::MemoryInfo& mem) {
-	return ImageViewKindFromDimension(mem.image_dimension);
-}
-
-uint32_t ImageViewCoordinateComponents(ImageViewKind view) {
-	switch (view) {
-		case ImageViewKind::Dim1D: return 1u;
-		case ImageViewKind::Dim1DArray:
-		case ImageViewKind::Dim2D: return 2u;
-		case ImageViewKind::Dim2DArray:
-		case ImageViewKind::Dim2DMsaaArray:
-		case ImageViewKind::Dim3D: return 3u;
-		case ImageViewKind::Dim2DMsaa: return 2u;
-		default: return 0u;
+uint32_t ImageScalarType(EmitterState& state, Prospero::TextureNumericClass numeric_class) {
+	switch (numeric_class) {
+		case Prospero::TextureNumericClass::Float: return TypeF32(state);
+		case Prospero::TextureNumericClass::Uint: return TypeU32(state);
+		case Prospero::TextureNumericClass::Sint: return TypeI32(state);
+		case Prospero::TextureNumericClass::Unsupported: break;
 	}
+	EXIT("invalid image numeric class");
 }
 
-uint32_t ImageViewSpatialComponents(ImageViewKind view) {
-	switch (view) {
-		case ImageViewKind::Dim1D:
-		case ImageViewKind::Dim1DArray: return 1u;
-		case ImageViewKind::Dim2D:
-		case ImageViewKind::Dim2DArray:
-		case ImageViewKind::Dim2DMsaa:
-		case ImageViewKind::Dim2DMsaaArray: return 2u;
-		case ImageViewKind::Dim3D: return 3u;
-		default: return 0u;
+uint32_t ImageVectorType(EmitterState& state, Prospero::TextureNumericClass numeric_class,
+                         uint32_t components) {
+	return state.builder.Type(OpTypeVector, {ImageScalarType(state, numeric_class), components});
+}
+
+uint32_t ImageType(EmitterState& state, const IR::ImageResource& image) {
+	uint32_t sampled = 0;
+	uint32_t format  = ImageFormatUnknown;
+	if (image.resource_class == IR::ImageResourceClass::Sampled) {
+		EXIT_IF(image.atomic);
+		sampled = 1;
+	} else if (image.resource_class == IR::ImageResourceClass::Storage) {
+		EXIT_IF(image.numeric_class == Prospero::TextureNumericClass::Sint ||
+		        image.numeric_class == Prospero::TextureNumericClass::Unsupported);
+		sampled = 2;
+		if (image.atomic) {
+			EXIT_IF(image.numeric_class != Prospero::TextureNumericClass::Uint);
+			format = ImageFormatR32ui;
+		}
+	} else {
+		EXIT("invalid image resource class");
 	}
-}
-
-uint32_t ImageViewImageType(EmitterState& state, ImageViewKind view, bool integer) {
-	const auto component = integer ? TypeU32(state) : TypeF32(state);
+	const auto& info = ImageDimensionInfoFor(image.dimension);
 	return state.builder.Type(OpTypeImage,
-	                          {component, ImageSpirvDimension(view), 0, ImageSpirvArrayed(view),
-	                           ImageSpirvMultisampled(view), 1, ImageFormatUnknown});
+	                          {ImageScalarType(state, image.numeric_class), info.spirv_dimension,
+	                           image.depth_compare ? 1u : 0u,
+	                           info.arrayed, info.multisampled, sampled, format});
 }
 
-uint32_t ImageViewSampledImageType(EmitterState& state, ImageViewKind view, bool integer) {
-	return state.builder.Type(OpTypeSampledImage, {ImageViewImageType(state, view, integer)});
-}
-
-uint32_t ImageViewSizeType(EmitterState& state, ImageViewKind view) {
-	switch (ImageViewCoordinateComponents(view)) {
+uint32_t ImageViewSizeType(EmitterState& state, ImageDimension dimension) {
+	switch (ImageDimensionInfoFor(dimension).coordinate_components) {
 		case 1u: return TypeU32(state);
 		case 2u: return TypeU32Vector(state, 2);
 		case 3u: return TypeU32Vector(state, 3);
@@ -183,128 +165,89 @@ uint32_t ImageViewSizeType(EmitterState& state, ImageViewKind view) {
 	}
 }
 
-uint32_t StorageImageType(EmitterState& state, StorageImageClass image_class, ImageViewKind view) {
-	uint32_t component = TypeF32(state);
-	uint32_t format    = ImageFormatUnknown;
-	if (image_class == StorageImageClass::FormatlessUint) {
-		component = TypeU32(state);
-	} else if (image_class == StorageImageClass::AtomicUint) {
-		component = TypeU32(state);
-		format    = ImageFormatR32ui;
-	}
-	return state.builder.Type(OpTypeImage, {component, ImageSpirvDimension(view), 0,
-	                                        ImageSpirvArrayed(view), 0, 2, format});
-}
-
-uint32_t StorageImagePointerType(EmitterState& state, StorageImageClass image_class,
-                                 ImageViewKind view) {
-	return state.builder.Type(
-	    OpTypePointer, {StorageClassUniformConstant, StorageImageType(state, image_class, view)});
-}
-
-StorageImageClass StorageImageClassForResource(const EmitterState& state, uint32_t resource) {
-	const auto& image      = state.program.info.images.at(resource);
-	const bool  uint_image = image.kind == IR::ResourceKind::StorageImageUint;
-	EXIT_IF(image.kind != IR::ResourceKind::StorageImage && !uint_image);
-	EXIT_IF(image.atomic && !uint_image);
-	return StorageImageClassFor(uint_image, image.atomic);
-}
-
-uint32_t StorageImageVariable(const EmitterState& state, StorageImageClass image_class,
-                              ImageViewKind view) {
-	return state.storage_image_variables[StorageImageIndex(image_class, view)];
-}
-
-uint32_t LoadSampledImageDescriptor(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
-                                    ImageViewKind view) {
-	(void)use_pc;
-	const bool integer      = mem.kind == IR::ResourceKind::ImageUint;
-	const auto kind         = SampledBindingKind(integer, view);
-	const auto binding      = ResourceForDescriptor(state, kind, mem.resource);
-	const auto variable     = state.sampled_image_variables[SampledImageIndex(integer, view)];
+uint32_t LoadSampledImageDescriptor(EmitterState& state, uint32_t resource) {
+	const auto& image_resource = state.program.info.images.at(resource);
+	EXIT_IF(image_resource.resource_class != IR::ImageResourceClass::Sampled);
+	const auto kind = IR::DescriptorBindingForImage(image_resource);
+	EXIT_IF(!kind.has_value());
+	const auto array_index  = ResourceForDescriptor(state, *kind, resource);
+	const auto variable     = state.image_variables[IR::ImageBindingIndex(*kind)];
 	const auto pointer_type = state.builder.Type(
-	    OpTypePointer, {StorageClassUniformConstant, ImageViewImageType(state, view, integer)});
+	    OpTypePointer, {StorageClassUniformConstant, ImageType(state, image_resource)});
 	const auto pointer =
-	    DescriptorElementPointer(state, pointer_type, variable, binding.array_index, kind,
-	                             mem.resource, "sampled image descriptor array was not emitted");
+	    DescriptorElementPointer(state, pointer_type, variable, array_index, *kind, resource,
+	                             "sampled image descriptor array was not emitted");
 	const auto image = state.builder.AllocateId();
-	state.builder.AddFunction({OpLoad, ImageViewImageType(state, view, integer), image, pointer});
+	state.builder.AddFunction({OpLoad, ImageType(state, image_resource), image, pointer});
 	return image;
 }
 
-uint32_t LoadSamplerDescriptor(EmitterState& state, uint32_t sampler, uint32_t use_pc) {
-	(void)use_pc;
-	const auto binding = ResourceForDescriptor(state, IR::DescriptorBindingKind::Samplers, sampler);
+uint32_t LoadSamplerDescriptor(EmitterState& state, uint32_t sampler) {
+	const auto array_index =
+	    ResourceForDescriptor(state, IR::DescriptorBindingKind::Samplers, sampler);
 	const auto sampler_type = state.builder.Type(OpTypeSampler);
 	const auto pointer_type =
 	    state.builder.Type(OpTypePointer, {StorageClassUniformConstant, sampler_type});
 	const auto pointer = DescriptorElementPointer(
-	    state, pointer_type, state.sampler_variable, binding.array_index,
+	    state, pointer_type, state.sampler_variable, array_index,
 	    IR::DescriptorBindingKind::Samplers, sampler, "sampler descriptor array was not emitted");
 	const auto sampler_id = state.builder.AllocateId();
 	state.builder.AddFunction({OpLoad, sampler_type, sampler_id, pointer});
 	return sampler_id;
 }
 
-uint32_t MakeSampledImage(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
-                          ImageViewKind view) {
-	const auto image   = LoadSampledImageDescriptor(state, mem, use_pc, view);
-	const auto sampler = LoadSamplerDescriptor(state, mem.sampler, use_pc);
-	if (image == 0 || sampler == 0) {
-		ExitDescriptorBindingFailure(
-		    state, SampledBindingKind(mem.kind == IR::ResourceKind::ImageUint, view), mem.resource,
-		    "sampled image or sampler descriptor load failed");
-	}
-	const auto sampled_image = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpSampledImage,
-	     ImageViewSampledImageType(state, view, mem.kind == IR::ResourceKind::ImageUint),
-	     sampled_image, image, sampler});
+uint32_t MakeSampledImage(EmitterState& state, uint32_t resource, uint32_t sampler) {
+	const auto& image_resource = state.program.info.images.at(resource);
+	const auto  image          = LoadSampledImageDescriptor(state, resource);
+	const auto  sampler_id     = LoadSamplerDescriptor(state, sampler);
+	const auto  sampled_image = state.builder.AllocateId();
+	const auto  sampled_type  =
+	    state.builder.Type(OpTypeSampledImage, {ImageType(state, image_resource)});
+	state.builder.AddFunction({OpSampledImage, sampled_type, sampled_image, image, sampler_id});
 	return sampled_image;
 }
 
-uint32_t MakeSampledImage(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
-                          ImageViewKind view, uint32_t image_resource) {
-	auto selected     = mem;
-	selected.resource = image_resource;
-	return MakeSampledImage(state, selected, use_pc, view);
-}
-
-uint32_t StorageImageDescriptorPointer(EmitterState& state, uint32_t resource, ImageViewKind view) {
-	const auto image_class = StorageImageClassForResource(state, resource);
-	const auto kind        = StorageBindingKind(image_class, view);
-	const auto binding  = ResourceForDescriptor(state, kind, resource);
-	const auto ptr_type    = StorageImagePointerType(state, image_class, view);
-	const auto variable    = StorageImageVariable(state, image_class, view);
-	return DescriptorElementPointer(state, ptr_type, variable, binding.array_index, kind, resource,
+uint32_t StorageImageDescriptorPointer(EmitterState& state, uint32_t resource) {
+	const auto& image = state.program.info.images.at(resource);
+	EXIT_IF(image.resource_class != IR::ImageResourceClass::Storage);
+	const auto kind = IR::DescriptorBindingForImage(image);
+	EXIT_IF(!kind.has_value());
+	const auto array_index = ResourceForDescriptor(state, *kind, resource);
+	const auto pointer_type =
+	    state.builder.Type(OpTypePointer, {StorageClassUniformConstant, ImageType(state, image)});
+	const auto variable = state.image_variables[IR::ImageBindingIndex(*kind)];
+	return DescriptorElementPointer(state, pointer_type, variable, array_index, *kind, resource,
 	                                "storage image descriptor array was not emitted");
 }
 
-void EmitStorageImageWrite(EmitterState& state, uint32_t resource, ImageViewKind view,
-                           uint32_t mip_lod, uint32_t coord, uint32_t texel) {
-	const auto image_class = StorageImageClassForResource(state, resource);
-	if (image_class != StorageImageClass::AtomicUint) {
+void EmitStorageImageWrite(EmitterState& state, uint32_t resource, uint32_t mip_lod, uint32_t coord,
+                           uint32_t texel) {
+	const auto& image = state.program.info.images.at(resource);
+	EXIT_IF(image.resource_class != IR::ImageResourceClass::Storage);
+	if (!image.atomic) {
 		state.builder.RequireCapability(CapabilityStorageImageWriteWithoutFormat);
 	}
-	const auto  kind    = StorageBindingKind(image_class, view);
-	const auto  binding = ResourceForDescriptor(state, kind, resource);
-	const auto& image   = state.program.info.images.at(resource);
-	const auto  LoadAt  = [&](uint32_t array_index) {
-		const auto pointer = DescriptorElementPointer(
-		    state, StorageImagePointerType(state, image_class, view),
-		    StorageImageVariable(state, image_class, view), array_index, kind, resource,
-		    "storage image descriptor array was not emitted");
+	const auto kind = IR::DescriptorBindingForImage(image);
+	EXIT_IF(!kind.has_value());
+	const auto array_index = ResourceForDescriptor(state, *kind, resource);
+	const auto image_type  = ImageType(state, image);
+	const auto pointer_type =
+	    state.builder.Type(OpTypePointer, {StorageClassUniformConstant, image_type});
+	const auto variable = state.image_variables[IR::ImageBindingIndex(*kind)];
+	const auto LoadAt   = [&](uint32_t array_index) {
+		const auto pointer =
+		    DescriptorElementPointer(state, pointer_type, variable, array_index, *kind, resource,
+		                             "storage image descriptor array was not emitted");
 		const auto descriptor = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {OpLoad, StorageImageType(state, image_class, view), descriptor, pointer});
+		state.builder.AddFunction({OpLoad, image_type, descriptor, pointer});
 		return descriptor;
 	};
 	if (image.mip_mode != IR::ImageMipMode::DynamicStorage) {
-		state.builder.AddFunction({OpImageWrite, LoadAt(binding.array_index), coord, texel});
+		state.builder.AddFunction({OpImageWrite, LoadAt(array_index), coord, texel});
 		return;
 	}
 	if (image.mip_count == 0u) {
-		ExitDescriptorBindingFailure(state, kind, resource,
+		ExitDescriptorBindingFailure(state, *kind, resource,
 		                             "dynamic storage image has no mip descriptors");
 	}
 
@@ -320,7 +263,7 @@ void EmitStorageImageWrite(EmitterState& state, uint32_t resource, ImageViewKind
 	state.builder.AddFunction(words);
 	for (uint32_t mip = 0; mip < image.mip_count; mip++) {
 		EmitLabel(state, labels[mip]);
-		state.builder.AddFunction({OpImageWrite, LoadAt(binding.array_index + mip), coord, texel});
+		state.builder.AddFunction({OpImageWrite, LoadAt(array_index + mip), coord, texel});
 		state.builder.AddFunction({OpBranch, merge_label});
 	}
 	EmitLabel(state, merge_label);
@@ -329,6 +272,7 @@ void EmitStorageImageWrite(EmitterState& state, uint32_t resource, ImageViewKind
 uint32_t ExecutionModelForStage(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Vertex: return ExecutionModelVertex;
+		case ShaderType::Mesh: return 5365u; // MeshEXT
 		case ShaderType::Pixel: return ExecutionModelFragment;
 		default: return ExecutionModelGLCompute;
 	}

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

@@ -16,26 +16,6 @@ bool UserDataDwordIndex(const EmitterState& state, IR::ScalarReg reg, uint32_t&
 	return true;
 }
 
-uint32_t EmitWqmWordU32(EmitterState& state, uint32_t value) {
-	const auto shifted_one = state.builder.AllocateId();
-	const auto merged_one  = state.builder.AllocateId();
-	const auto shifted_two = state.builder.AllocateId();
-	const auto merged_two  = state.builder.AllocateId();
-	const auto quad_bits   = state.builder.AllocateId();
-	const auto result      = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpShiftRightLogical, TypeU32(state), shifted_one, value, ConstantU32(state, 1)});
-	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), merged_one, value, shifted_one});
-	state.builder.AddFunction(
-	    {OpShiftRightLogical, TypeU32(state), shifted_two, merged_one, ConstantU32(state, 2)});
-	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), merged_two, merged_one, shifted_two});
-	state.builder.AddFunction(
-	    {OpBitwiseAnd, TypeU32(state), quad_bits, merged_two, ConstantU32(state, 0x11111111u)});
-	state.builder.AddFunction(
-	    {OpIMul, TypeU32(state), result, quad_bits, ConstantU32(state, 0x0fu)});
-	return result;
-}
-
 uint32_t EmitWqmU64(EmitterState& state, uint32_t value) {
 	const auto shifted_one = state.builder.AllocateId();
 	const auto merged_one  = state.builder.AllocateId();
@@ -61,6 +41,24 @@ uint32_t EmitBuiltinU32(ValueEmitContext& ctx, IR::StageInputKind kind, uint32_t
 	if (kind == IR::StageInputKind::LocalInvocationIndex) {
 		return EmitLocalInvocationIndex(state);
 	}
+	if (state.lane_count == 2 && (kind == IR::StageInputKind::LocalInvocationId ||
+	                              kind == IR::StageInputKind::GlobalInvocationId)) {
+		const auto* cs      = ShaderWorkgroupInput(state.stage, state.input_info);
+		uint32_t    divisor = 1;
+		for (uint32_t axis = 0; axis < component; axis++) {
+			divisor *= std::max(cs->threads_num[axis], 1u);
+		}
+		const auto size    = std::max(cs->threads_num[component], 1u);
+		const auto divided = EmitBinaryU32(state, OpUDiv, EmitLocalInvocationIndex(state),
+		                                   ConstantU32(state, divisor));
+		const auto local   = EmitBinaryU32(state, OpUMod, divided, ConstantU32(state, size));
+		if (kind == IR::StageInputKind::LocalInvocationId) {
+			return local;
+		}
+		const auto group = EmitInputComponentU32(state, IR::StageInputKind::WorkgroupId, component);
+		return EmitAddU32(state, local,
+		                  EmitBinaryU32(state, OpIMul, group, ConstantU32(state, size)));
+	}
 	const auto variable = InputVariableForKind(state, kind);
 	if (variable == 0) {
 		return ConstantU32(state, 0);
@@ -73,7 +71,8 @@ uint32_t EmitBuiltinU32(ValueEmitContext& ctx, IR::StageInputKind kind, uint32_t
 		    {OpSelect, TypeU32(state), bits, value, ConstantU32(state, 1), ConstantU32(state, 0)});
 		return bits;
 	}
-	if (kind == IR::StageInputKind::VertexIndex || kind == IR::StageInputKind::InstanceIndex) {
+	if (kind == IR::StageInputKind::VertexIndex || kind == IR::StageInputKind::InstanceIndex ||
+	    kind == IR::StageInputKind::Layer || kind == IR::StageInputKind::SampleId) {
 		const auto value = state.builder.AllocateId();
 		const auto bits  = state.builder.AllocateId();
 		state.builder.AddFunction({OpLoad, TypeI32(state), value, variable});
@@ -106,40 +105,6 @@ uint32_t EmitBuiltinU32(ValueEmitContext& ctx, IR::StageInputKind kind, uint32_t
 	return EmitInputComponentU32(state, kind, component);
 }
 
-uint32_t EmitWqm(ValueEmitContext& ctx, uint32_t active) {
-	auto&      state  = ctx.state;
-	const auto ballot = state.builder.AllocateId();
-	state.builder.AddFunction({OpGroupNonUniformBallot, TypeU32Vector(state, 4), ballot,
-	                           ConstantU32(state, ScopeSubgroup), active});
-	const auto low = state.builder.AllocateId();
-	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), low, ballot, 0});
-	const auto wqm_low  = EmitWqmWordU32(state, low);
-	const auto lane     = EmitSubgroupLocalInvocationId(state);
-	uint32_t   mask     = wqm_low;
-	uint32_t   bit_lane = lane;
-	if (state.wave_size == 64u) {
-		const auto high = state.builder.AllocateId();
-		state.builder.AddFunction({OpCompositeExtract, TypeU32(state), high, ballot, 1});
-		const auto wqm_high = EmitWqmWordU32(state, high);
-		const auto upper    = state.builder.AllocateId();
-		mask                = state.builder.AllocateId();
-		bit_lane            = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {OpUGreaterThanEqual, TypeBool(state), upper, lane, ConstantU32(state, 32)});
-		state.builder.AddFunction({OpSelect, TypeU32(state), mask, upper, wqm_high, wqm_low});
-		state.builder.AddFunction(
-		    {OpBitwiseAnd, TypeU32(state), bit_lane, lane, ConstantU32(state, 31)});
-	}
-	const auto bit    = state.builder.AllocateId();
-	const auto hit    = state.builder.AllocateId();
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpShiftLeftLogical, TypeU32(state), bit, ConstantU32(state, 1), bit_lane});
-	state.builder.AddFunction({OpBitwiseAnd, TypeU32(state), hit, mask, bit});
-	state.builder.AddFunction({OpINotEqual, TypeBool(state), result, hit, ConstantU32(state, 0)});
-	return result;
-}
-
 uint32_t EmitDppWriteCondition(ValueEmitContext& ctx, const IR::DppMoveFlags& flags,
                                uint32_t exec) {
 	auto&      state      = ctx.state;
@@ -373,12 +338,24 @@ void EmitAuxPositionExport(ValueEmitContext& ctx, uint32_t data, const IR::Expor
 		}
 		const auto output = IR::DecodePositionExportComponent(
 		    state.input_info.vertex->pa_cl_vs_out_cntl, exp.index, component);
-		if (output.layer) {
-			const auto raw   = ExportRawComponent(ctx, data, component);
-			const auto layer = state.builder.AllocateId();
-			state.builder.AddFunction({OpBitwiseAnd, TypeU32(state), layer, raw,
-			                           ConstantU32(state, 0x7ffu)});
-			state.builder.AddFunction({OpStore, state.layer_variable, layer});
+		if (output.layer || output.viewport) {
+			const auto raw = ExportRawComponent(ctx, data, component);
+			if (output.layer) {
+				const auto layer = state.builder.AllocateId();
+				state.builder.AddFunction({OpBitwiseAnd, TypeU32(state), layer, raw,
+				                           ConstantU32(state, 0x7ffu)});
+				const auto pointer = state.stage == ShaderType::Mesh
+				                         ? MeshOutputPointer(state, IR::StageOutputKind::Layer)
+				                         : state.layer_variable;
+				state.builder.AddFunction({OpStore, pointer, layer});
+			}
+			if (output.viewport) {
+				// GFX10 MISC.z packs the viewport index in bits 16..19 alongside the layer.
+				const auto viewport = state.builder.AllocateId();
+				state.builder.AddFunction({OpBitFieldUExtract, TypeU32(state), viewport, raw,
+				                           ConstantU32(state, 16), ConstantU32(state, 4)});
+				state.builder.AddFunction({OpStore, state.viewport_index_variable, viewport});
+			}
 			continue;
 		}
 		if (!output.point_size && output.clip_distance == UINT32_MAX &&
@@ -454,12 +431,18 @@ void EmitExport(ValueEmitContext& ctx, const IR::Inst& inst) {
 		    {OpSelect, TypeU32(state), value, exec, ConstantU32(state, 1), ConstantU32(state, 0)});
 		state.builder.AddFunction({OpStore, state.pixel_valid_mask_variable, value});
 	}
-	if (exp.kind == IR::ExportTargetKind::Null || exp.kind == IR::ExportTargetKind::Primitive ||
-	    exp.en == 0u) {
+	if (exp.kind == IR::ExportTargetKind::Null || exp.en == 0u) {
 		return;
 	}
 	EmitIfCondition(state, exec, [&]() {
 		const auto data = ctx.Arg(inst, 0);
+		if (exp.kind == IR::ExportTargetKind::Primitive) {
+			if (state.stage == ShaderType::Mesh) {
+				state.builder.AddFunction(
+				    {OpStore, MeshPrimitivePointer(state), ExportRawComponent(ctx, data, 0)});
+			}
+			return;
+		}
 		if (exp.kind == IR::ExportTargetKind::Position && exp.index != 0) {
 			EmitAuxPositionExport(ctx, data, exp);
 			return;
@@ -483,8 +466,9 @@ void EmitExport(ValueEmitContext& ctx, const IR::Inst& inst) {
 			}
 			return;
 		}
-		const auto variable = OutputVariableForExport(state, exp);
-		if (variable == 0) {
+		const auto variable =
+		    state.stage == ShaderType::Mesh ? 0u : OutputVariableForExport(state, exp);
+		if (state.stage != ShaderType::Mesh && variable == 0) {
 			return;
 		}
 		const bool uint_output = MrtOutputMode(state, exp) == 7u;
@@ -501,10 +485,16 @@ void EmitExport(ValueEmitContext& ctx, const IR::Inst& inst) {
 				value = mapped;
 			}
 		}
-		if (exp.kind == IR::ExportTargetKind::Position) {
-			if (state.stage == ShaderType::Vertex && state.input_info.vertex->clip_space.enabled) {
-				value = ConvertPositionToClipSpace(state, value);
-			}
+		if (exp.kind == IR::ExportTargetKind::Position &&
+		    state.input_info.vertex->clip_space.enabled) {
+			value = ConvertPositionToClipSpace(state, value);
+		}
+		if (state.stage == ShaderType::Mesh) {
+			const auto kind = exp.kind == IR::ExportTargetKind::Position
+			                      ? IR::StageOutputKind::Position
+			                      : IR::StageOutputKind::Parameter;
+			state.builder.AddFunction({OpStore, MeshOutputPointer(state, kind, exp.index), value});
+		} else if (exp.kind == IR::ExportTargetKind::Position) {
 			const auto pointer = state.builder.AllocateId();
 			state.builder.AddFunction(
 			    {OpAccessChain, TypePointer(state, StorageClassOutput, TypeF32Vector(state, 4)),
@@ -537,6 +527,19 @@ bool EmitValueFlow(ValueEmitContext& ctx, const IR::Inst& inst) {
 			                           ConstantU32(state, semantics)});
 			return true;
 		}
+		case IR::ValueOpcode::MeshAllocate: EmitMeshAllocate(ctx, inst); return true;
+		case IR::ValueOpcode::MeshDrawParameter: {
+			const auto index = inst.Arg(0).U32();
+			if (state.stage != ShaderType::Mesh || index >= IR::PushData::MeshDrawDwordCount) {
+				ctx.Fail(inst, "invalid mesh draw parameter");
+			}
+			const auto pointer = state.builder.AllocateId();
+			state.builder.AddFunction({OpAccessChain, TypePushConstantElementPointer(state),
+			                           pointer, state.push_constant_variable, ConstantU32(state, 0),
+			                           ConstantU32(state, index)});
+			state.builder.AddFunction({OpLoad, TypeU32(state), ctx.Result(inst), pointer});
+			return true;
+		}
 		case IR::ValueOpcode::GetUserData: {
 			const auto reg   = inst.Arg(0).ScalarRegister();
 			uint32_t   dword = 0;
@@ -561,17 +564,12 @@ bool EmitValueFlow(ValueEmitContext& ctx, const IR::Inst& inst) {
 		case IR::ValueOpcode::DppMoveU32: {
 			const auto flags    = inst.Flags<IR::DppMoveFlags>();
 			const auto target   = EmitDppTargetLane(state, flags.control);
-			const auto shuffled = state.builder.AllocateId();
-			state.builder.AddFunction({OpGroupNonUniformShuffle, TypeU32(state), shuffled,
-			                           ConstantU32(state, ScopeSubgroup), ctx.Arg(inst, 0),
-			                           target.lane});
+			const auto shuffled = ctx.Shuffle(inst, 0, target.lane);
 			if (flags.fetch_inactive) {
 				ctx.Define(inst, shuffled);
 				return true;
 			}
-			const auto ballot = state.builder.AllocateId();
-			state.builder.AddFunction({OpGroupNonUniformBallot, TypeU32Vector(state, 4), ballot,
-			                           ConstantU32(state, ScopeSubgroup), ctx.Arg(inst, 1)});
+			const auto ballot        = ctx.Ballot(inst.Arg(1));
 			const auto source_active = EmitBallotLaneActiveBool(state, ballot, target.lane);
 			const auto can_fetch     = state.builder.AllocateId();
 			state.builder.AddFunction(
@@ -588,30 +586,18 @@ bool EmitValueFlow(ValueEmitContext& ctx, const IR::Inst& inst) {
 		case IR::ValueOpcode::WqmU64:
 			ctx.Define(inst, EmitWqmU64(ctx.state, ctx.Arg(inst, 0)));
 			return true;
-		case IR::ValueOpcode::WqmMask:
-			ctx.Define(inst, EmitWqm(ctx, ctx.Arg(inst, 0)));
-			return true;
 		case IR::ValueOpcode::LaneId:
 			ctx.Define(inst, EmitSubgroupLocalInvocationId(state));
 			return true;
-		case IR::ValueOpcode::Ballot:
-			ctx.Emit(inst, OpGroupNonUniformBallot, IR::Type::U32x4,
-			         {ConstantU32(state, ScopeSubgroup), ctx.Arg(inst, 0)});
-			return true;
+		case IR::ValueOpcode::Ballot: ctx.Define(inst, ctx.Ballot(inst.Arg(0))); return true;
 		case IR::ValueOpcode::ReadFirstLane: {
-			const auto ballot = state.builder.AllocateId();
-			const auto lane   = state.builder.AllocateId();
-			state.builder.AddFunction({OpGroupNonUniformBallot, TypeU32Vector(state, 4), ballot,
-			                           ConstantU32(state, ScopeSubgroup), ctx.Arg(inst, 1)});
-			state.builder.AddFunction({OpGroupNonUniformBallotFindLSB, TypeU32(state), lane,
-			                           ConstantU32(state, ScopeSubgroup), ballot});
-			ctx.Emit(inst, OpGroupNonUniformShuffle, IR::Type::U32,
-			         {ConstantU32(state, ScopeSubgroup), ctx.Arg(inst, 0), lane});
+			const auto ballot = ctx.Ballot(inst.Arg(1));
+			const auto lane   = ctx.FirstLane(ballot);
+			ctx.Define(inst, ctx.Shuffle(inst, 0, lane));
 			return true;
 		}
 		case IR::ValueOpcode::ReadLane:
-			ctx.Emit(inst, OpGroupNonUniformShuffle, IR::Type::U32,
-			         {ConstantU32(state, ScopeSubgroup), ctx.Arg(inst, 0), ctx.Arg(inst, 1)});
+			ctx.Define(inst, ctx.Shuffle(inst, 0, ctx.Arg(inst, 1)));
 			return true;
 		case IR::ValueOpcode::WriteLane: {
 			const auto hit = state.builder.AllocateId();
@@ -656,16 +642,10 @@ bool EmitValueFlow(ValueEmitContext& ctx, const IR::Inst& inst) {
 			state.builder.AddFunction(
 			    {OpBitwiseAnd, TypeU32(state), index, shifted, ConstantU32(state, 15)});
 			state.builder.AddFunction({OpBitwiseOr, TypeU32(state), target, row_value, index});
-			const auto shuffled = state.builder.AllocateId();
-			state.builder.AddFunction({OpGroupNonUniformShuffle, TypeU32(state), shuffled,
-			                           ConstantU32(state, ScopeSubgroup), ctx.Arg(inst, 0),
-			                           target});
+			const auto shuffled = ctx.Shuffle(inst, 0, target);
 			uint32_t result = shuffled;
 			if (!flags.fetch_inactive) {
-				const auto source_exec = state.builder.AllocateId();
-				state.builder.AddFunction({OpGroupNonUniformShuffle, TypeBool(state), source_exec,
-				                           ConstantU32(state, ScopeSubgroup), ctx.Arg(inst, 3),
-				                           target});
+				const auto source_exec = ctx.Shuffle(inst, 3, target);
 				result = state.builder.AllocateId();
 				state.builder.AddFunction({OpSelect, TypeU32(state), result, source_exec, shuffled,
 				                           ConstantU32(state, 0)});

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

@@ -2,10 +2,6 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
-uint32_t EmitTrueBool(EmitterState& state) {
-	return ConstantBool(state, true);
-}
-
 DppTargetLane EmitDppQuadPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control) {
 	const auto quad_base = state.builder.AllocateId();
 	const auto lane      = state.builder.AllocateId();
@@ -23,7 +19,7 @@ DppTargetLane EmitDppQuadPermTargetLane(EmitterState& state, uint32_t subid, uin
 	state.builder.AddFunction(
 	    {OpBitwiseAnd, TypeU32(state), selected, selected0, ConstantU32(state, 3)});
 	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), target, quad_base, selected});
-	return {target, EmitTrueBool(state)};
+	return {target, ConstantBool(state, true)};
 }
 
 DppTargetLane EmitDppRowShiftTargetLane(EmitterState& state, uint32_t subid, uint32_t amount,
@@ -70,7 +66,7 @@ DppTargetLane EmitDppRowRotateRightTargetLane(EmitterState& state, uint32_t subi
 	    {OpIAdd, TypeU32(state), plus, lane, ConstantU32(state, 16u - amount)});
 	state.builder.AddFunction({OpSelect, TypeU32(state), selected, in_high, minus, plus});
 	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), target, row, selected});
-	return {target, EmitTrueBool(state)};
+	return {target, ConstantBool(state, true)};
 }
 
 DppTargetLane EmitDppMirrorTargetLane(EmitterState& state, uint32_t subid, bool half_row) {
@@ -87,7 +83,7 @@ DppTargetLane EmitDppMirrorTargetLane(EmitterState& state, uint32_t subid, bool
 	state.builder.AddFunction(
 	    {OpISub, TypeU32(state), mirrored, ConstantU32(state, lane_mask), lane});
 	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), target, base, mirrored});
-	return {target, EmitTrueBool(state)};
+	return {target, ConstantBool(state, true)};
 }
 
 DppTargetLane EmitDppTargetLane(EmitterState& state, uint32_t control) {
@@ -114,9 +110,9 @@ DppTargetLane EmitDppTargetLane(EmitterState& state, uint32_t control) {
 		const auto target = state.builder.AllocateId();
 		state.builder.AddFunction(
 		    {OpBitwiseXor, TypeU32(state), target, subid, ConstantU32(state, control & 0xfu)});
-		return {target, EmitTrueBool(state)};
+		return {target, ConstantBool(state, true)};
 	}
-	return {subid, EmitTrueBool(state)};
+	return {subid, ConstantBool(state, true)};
 }
 
 uint32_t EmitSubgroupLocalInvocationId(EmitterState& state) {
@@ -126,7 +122,7 @@ uint32_t EmitSubgroupLocalInvocationId(EmitterState& state) {
 	const auto value = state.builder.AllocateId();
 	state.builder.AddFunction(
 	    {OpLoad, TypeU32(state), value, state.subgroup_local_invocation_id_variable});
-	return value;
+	return state.lane_half == 0 ? value : EmitAddU32(state, value, ConstantU32(state, 32));
 }
 
 uint32_t InputVariableForKind(const EmitterState& state, IR::StageInputKind kind) {
@@ -167,31 +163,24 @@ uint32_t EmitLocalInvocationIndex(EmitterState& state) {
 	}
 	const auto value = state.builder.AllocateId();
 	state.builder.AddFunction({OpLoad, TypeU32(state), value, variable});
-	return value;
-}
-
-uint32_t VertexInputDefaultComponentU32(EmitterState& state, VertexInputScalarKind kind,
-                                        uint32_t component) {
-	if ((component & 3u) != 3u) {
-		return ConstantU32(state, 0);
+	if (state.lane_count == 2) {
+		const auto wave_base = EmitBinaryU32(state, OpBitwiseAnd, value, ConstantU32(state, ~31u));
+		return EmitAddU32(state, EmitAddU32(state, value, wave_base),
+		                  ConstantU32(state, state.lane_half * 32));
 	}
-	return ConstantU32(state, kind == VertexInputScalarKind::Float ? 0x3f800000u : 1u);
+	return value;
 }
 
 uint32_t EmitVertexParameterComponentU32(EmitterState& state, const InputBinding& input,
                                          uint32_t component) {
-	const auto count = VertexParameterComponentCount(state, input);
+	const auto count = VertexParameterComponentCount(input);
 	const auto kind  = VertexParameterScalarKind(state, input.location);
-	if (component >= count) {
-		return VertexInputDefaultComponentU32(state, kind, component);
-	}
-
 	const auto scalar_type = VertexParameterScalarType(state, kind);
 	uint32_t   raw         = state.builder.AllocateId();
 	if (count == 1u) {
 		state.builder.AddFunction({OpLoad, scalar_type, raw, input.variable_id});
 	} else {
-		const auto pointer_type = VertexParameterScalarPointerType(state, kind);
+		const auto pointer_type = TypePointer(state, StorageClassInput, scalar_type);
 		const auto pointer      = state.builder.AllocateId();
 		state.builder.AddFunction({OpAccessChain, pointer_type, pointer, input.variable_id,
 		                           ConstantU32(state, component)});
@@ -210,14 +199,14 @@ uint32_t EmitVertexParameterComponentU32(EmitterState& state, const InputBinding
 uint32_t EmitSubgroupLaneActiveBool(EmitterState& state, uint32_t lane) {
 	const auto active_ballot = state.builder.AllocateId();
 	state.builder.AddFunction({OpGroupNonUniformBallot, TypeU32Vector(state, 4), active_ballot,
-	                           ConstantU32(state, ScopeSubgroup), EmitTrueBool(state)});
+	                           ConstantU32(state, ScopeSubgroup), ConstantBool(state, true)});
 	return EmitBallotLaneActiveBool(state, active_ballot, lane);
 }
 uint32_t EmitBallotLaneActiveBool(EmitterState& state, uint32_t active_ballot, uint32_t lane) {
 	const auto low = state.builder.AllocateId();
 	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), low, active_ballot, 0});
 	uint32_t mask = low;
-	if (state.wave_size == 64u) {
+	if (state.program.wave_size == 64u) {
 		const auto high     = state.builder.AllocateId();
 		const auto in_high  = state.builder.AllocateId();
 		const auto selected = state.builder.AllocateId();
@@ -242,7 +231,7 @@ uint32_t EmitBallotLaneActiveBool(EmitterState& state, uint32_t active_ballot, u
 	state.builder.AddFunction({OpBitwiseAnd, TypeU32(state), hit, mask, bit});
 	state.builder.AddFunction({OpINotEqual, TypeBool(state), active, hit, ConstantU32(state, 0)});
 	state.builder.AddFunction(
-	    {OpULessThan, TypeBool(state), in_range, lane, ConstantU32(state, state.wave_size)});
+	    {OpULessThan, TypeBool(state), in_range, lane, ConstantU32(state, state.program.wave_size)});
 	state.builder.AddFunction({OpLogicalAnd, TypeBool(state), ret, active, in_range});
 	return ret;
 }

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

@@ -1,5 +1,6 @@
 #include "graphics/guest_gpu/gpu_format.h"
 #include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
+#include "graphics/shader/recompiler/frontend/decode/ImageOps.h"
 
 #include <algorithm>
 #include <bit>
@@ -31,61 +32,18 @@ uint32_t ImageGatherComponent(uint32_t dmask) {
 	}
 }
 
-uint32_t Unary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t value) {
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, type, result, value});
-	return result;
-}
-
-uint32_t Binary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t lhs, uint32_t rhs) {
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, type, result, lhs, rhs});
-	return result;
-}
-
 bool HasFlag(const IR::MemoryInfo& mem, uint32_t flag) {
 	return (mem.image_sample_flags & flag) != 0u;
 }
 
-bool UsesA16(const IR::MemoryInfo& mem) {
-	return HasFlag(mem, Decoder::ImageSampleFlagA16);
-}
-
-bool ComponentUsesA16(const IR::MemoryInfo& mem, uint32_t component) {
-	if (!UsesA16(mem)) return false;
-	uint32_t cursor = 0;
-	if (HasFlag(mem, Decoder::ImageSampleFlagOffset)) {
-		if (component == cursor) return false;
-		cursor++;
-	}
-	if (HasFlag(mem, Decoder::ImageSampleFlagBias)) {
-		if (component == cursor) return true;
-		cursor++;
-	}
-	if (HasFlag(mem, Decoder::ImageSampleFlagCompare) && component == cursor) return false;
-	return true;
-}
-
-uint32_t HalfComponent(const IR::MemoryInfo& mem, uint32_t component) {
-	if (!UsesA16(mem)) return component * 2u;
-	uint32_t half = 0;
-	for (uint32_t index = 0; index < component; index++) {
-		const auto width = ComponentUsesA16(mem, index) ? 1u : 2u;
-		if (width == 2u && (half & 1u) != 0u) half++;
-		half += width;
-	}
-	if (!ComponentUsesA16(mem, component) && (half & 1u) != 0u) half++;
-	return half;
-}
-
 uint32_t AddressU32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::Inst& address,
                     uint32_t component) {
-	const auto half   = HalfComponent(mem, component);
-	const auto packed = half / 2u;
+	const auto layout = Decoder::ImageAddressComponentLayout(mem.image_sample_flags, component);
+	const auto packed = layout.bit_offset / 32u;
 	if (packed >= address.NumArgs()) return ConstantU32(ctx.state, 0);
 	auto value = ctx.Def(address.Arg(packed));
-	if (ComponentUsesA16(mem, component)) {
-		if ((half & 1u) != 0u) {
+	if (layout.bit_width == 16u) {
+		if ((layout.bit_offset & 31u) != 0u) {
 			value = Binary(ctx.state, OpShiftRightLogical, TypeU32(ctx.state), value,
 			               ConstantU32(ctx.state, 16));
 		}
@@ -98,25 +56,26 @@ uint32_t AddressU32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::
 uint32_t AddressF32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::Inst& address,
                     uint32_t component) {
 	const auto value = AddressU32(ctx, mem, address, component);
-	return ComponentUsesA16(mem, component)
+	return Decoder::ImageAddressComponentLayout(mem.image_sample_flags, component).bit_width == 16u
 	           ? EmitF16BitsToF32(ctx.state, value)
 	           : Unary(ctx.state, OpBitcast, TypeF32(ctx.state), value);
 }
 
-ImageSampleLayout Layout(const IR::MemoryInfo& mem, ImageViewKind view) {
+ImageSampleLayout Layout(const IR::MemoryInfo& mem, ImageDimension dimension) {
 	ImageSampleLayout layout;
 	uint32_t          cursor = 0;
+	const auto&       info   = ImageDimensionInfoFor(dimension);
 	if (HasFlag(mem, Decoder::ImageSampleFlagOffset)) layout.offset = cursor++;
 	if (HasFlag(mem, Decoder::ImageSampleFlagBias)) layout.bias = cursor++;
 	if (HasFlag(mem, Decoder::ImageSampleFlagCompare)) layout.dref = cursor++;
 	if (HasFlag(mem, Decoder::ImageSampleFlagDerivative)) {
 		layout.grad_x = cursor;
-		cursor += ImageViewSpatialComponents(view);
+		cursor += info.spatial_components;
 		layout.grad_y = cursor;
-		cursor += ImageViewSpatialComponents(view);
+		cursor += info.spatial_components;
 	}
 	layout.coord = cursor;
-	cursor += ImageViewCoordinateComponents(view);
+	cursor += info.coordinate_components;
 	if (HasFlag(mem, Decoder::ImageSampleFlagLod)) layout.lod = cursor++;
 	return layout;
 }
@@ -144,11 +103,12 @@ uint32_t CubeLayer(EmitterState& state, uint32_t value) {
 
 uint32_t CoordF32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::Inst& address,
                   uint32_t first, uint32_t components) {
+	const bool cube = ctx.state.program.info.images.at(mem.resource).cube;
 	auto x = AddressF32(ctx, mem, address, first);
 	if (components == 1u) return x;
 	auto y = mem.image_address_components > first + 1u ? AddressF32(ctx, mem, address, first + 1u)
 	                                                   : ZeroF32(ctx.state);
-	if (mem.image_cube) {
+	if (cube) {
 		x = CubeAxis(ctx.state, x);
 		y = CubeAxis(ctx.state, y);
 	}
@@ -157,7 +117,7 @@ uint32_t CoordF32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::In
 		auto z = mem.image_address_components > first + 2u
 		             ? AddressF32(ctx, mem, address, first + 2u)
 		             : ZeroF32(ctx.state);
-		if (mem.image_cube) z = CubeLayer(ctx.state, z);
+		if (cube) z = CubeLayer(ctx.state, z);
 		ctx.state.builder.AddFunction(
 		    {OpCompositeConstruct, TypeF32Vector(ctx.state, 3), result, x, y, z});
 	} else {
@@ -168,8 +128,8 @@ uint32_t CoordF32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::In
 }
 
 uint32_t CoordU32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::Inst& address,
-                  ImageViewKind view) {
-	const auto components = ImageViewCoordinateComponents(view);
+                  ImageDimension dimension) {
+	const auto components = ImageDimensionInfoFor(dimension).coordinate_components;
 	const auto x          = AddressU32(ctx, mem, address, 0);
 	if (components == 1u) return x;
 	const auto y      = mem.image_address_components > 1u ? AddressU32(ctx, mem, address, 1)
@@ -188,8 +148,8 @@ uint32_t CoordU32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::In
 }
 
 uint32_t LodU32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::Inst& address,
-                ImageViewKind view) {
-	const auto component = ImageViewCoordinateComponents(view);
+                ImageDimension dimension) {
+	const auto component = ImageDimensionInfoFor(dimension).coordinate_components;
 	return mem.image_has_mip && mem.image_address_components > component
 	           ? AddressU32(ctx, mem, address, component)
 	           : ConstantU32(ctx.state, 0);
@@ -199,8 +159,33 @@ uint32_t FloatBits(ValueEmitContext& ctx, uint32_t value) {
 	return Unary(ctx.state, OpBitcast, TypeU32(ctx.state), value);
 }
 
-uint32_t ResultVector(ValueEmitContext& ctx, uint32_t value, bool integer, bool dref,
+uint32_t SampledComponentBits(ValueEmitContext& ctx, uint32_t value,
+                              Prospero::TextureNumericClass numeric_class) {
+	if (numeric_class == Prospero::TextureNumericClass::Uint) {
+		return value;
+	}
+	return Unary(ctx.state, OpBitcast, TypeU32(ctx.state), value);
+}
+
+uint32_t SampledComponentZero(EmitterState& state, Prospero::TextureNumericClass numeric_class) {
+	switch (numeric_class) {
+		case Prospero::TextureNumericClass::Float: return ZeroF32(state);
+		case Prospero::TextureNumericClass::Uint: return ConstantU32(state, 0);
+		case Prospero::TextureNumericClass::Sint: return ConstantI32(state, 0);
+		case Prospero::TextureNumericClass::Unsupported: break;
+	}
+	EXIT("invalid sampled image numeric class");
+}
+
+uint32_t ResultVector(ValueEmitContext& ctx, uint32_t value,
+                      Prospero::TextureNumericClass numeric_class, bool dref,
                       const IR::MemoryInfo& mem, bool gather = false) {
+	auto value_class = numeric_class;
+	if (dref) {
+		value_class = Prospero::TextureNumericClass::Float;
+	}
+	const bool integer = value_class == Prospero::TextureNumericClass::Uint ||
+	                     value_class == Prospero::TextureNumericClass::Sint;
 	if (mem.data_bits == 16u) {
 		uint32_t   packed[4] = {ConstantU32(ctx.state, 0), ConstantU32(ctx.state, 0),
 		                        ConstantU32(ctx.state, 0), ConstantU32(ctx.state, 0)};
@@ -209,25 +194,27 @@ uint32_t ResultVector(ValueEmitContext& ctx, uint32_t value, bool integer, bool
 			const auto component = gather ? index : DmaskComponent(mem.dmask, index);
 			const auto result    = ctx.state.builder.AllocateId();
 			ctx.state.builder.AddFunction({OpCompositeExtract,
-			                               integer ? TypeU32(ctx.state) : TypeF32(ctx.state),
-			                               result, value, component});
+			                               ImageScalarType(ctx.state, value_class), result, value,
+			                               component});
 			return result;
 		};
 		for (uint32_t word = 0; word < mem.data_dwords; word++) {
 			const auto low_index  = word * 2u;
 			const auto high_index = low_index + 1u;
 			const auto low        = scalar(low_index);
-			const auto high = high_index < mem.component_count
-			                      ? scalar(high_index)
-			                      : (integer ? ConstantU32(ctx.state, 0) : ZeroF32(ctx.state));
+			const auto high       = high_index < mem.component_count
+			                            ? scalar(high_index)
+			                            : SampledComponentZero(ctx.state, value_class);
 			if (integer) {
-				const auto mask = ConstantU32(ctx.state, 0xffffu);
-				packed[word] =
-				    Binary(ctx.state, OpBitwiseOr, TypeU32(ctx.state),
-				           Binary(ctx.state, OpBitwiseAnd, TypeU32(ctx.state), low, mask),
-				           Binary(ctx.state, OpShiftLeftLogical, TypeU32(ctx.state),
-				                  Binary(ctx.state, OpBitwiseAnd, TypeU32(ctx.state), high, mask),
-				                  ConstantU32(ctx.state, 16u)));
+				const auto low_bits  = SampledComponentBits(ctx, low, value_class);
+				const auto high_bits = SampledComponentBits(ctx, high, value_class);
+				const auto mask      = ConstantU32(ctx.state, 0xffffu);
+				packed[word]         = Binary(
+				    ctx.state, OpBitwiseOr, TypeU32(ctx.state),
+				    Binary(ctx.state, OpBitwiseAnd, TypeU32(ctx.state), low_bits, mask),
+				    Binary(ctx.state, OpShiftLeftLogical, TypeU32(ctx.state),
+				           Binary(ctx.state, OpBitwiseAnd, TypeU32(ctx.state), high_bits, mask),
+				           ConstantU32(ctx.state, 16u)));
 			} else {
 				const auto pair = ctx.state.builder.AllocateId();
 				ctx.state.builder.AddFunction(
@@ -249,10 +236,9 @@ uint32_t ResultVector(ValueEmitContext& ctx, uint32_t value, bool integer, bool
 			continue;
 		}
 		const auto scalar = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction({OpCompositeExtract,
-		                               integer ? TypeU32(ctx.state) : TypeF32(ctx.state), scalar,
-		                               value, index});
-		component[index] = integer ? scalar : FloatBits(ctx, scalar);
+		ctx.state.builder.AddFunction(
+		    {OpCompositeExtract, ImageScalarType(ctx.state, value_class), scalar, value, index});
+		component[index] = SampledComponentBits(ctx, scalar, value_class);
 	}
 	const auto result = ctx.state.builder.AllocateId();
 	ctx.state.builder.AddFunction({OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result,
@@ -260,20 +246,20 @@ uint32_t ResultVector(ValueEmitContext& ctx, uint32_t value, bool integer, bool
 	return result;
 }
 
-uint32_t QueryDimensions(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
+uint32_t QueryDimensions(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
                          const IR::Inst& address) {
-	const auto pc    = inst.Flags<IR::MemoryFlags>().pc;
-	const auto view  = SampledImageViewKind(ctx.state, mem, pc);
-	const auto image = LoadSampledImageDescriptor(ctx.state, mem, pc, view);
-	const auto size  = ctx.state.builder.AllocateId();
-	if (ImageSpirvMultisampled(view) != 0u) {
+	const auto  dimension = ctx.state.program.info.images.at(mem.resource).dimension;
+	const auto& info      = ImageDimensionInfoFor(dimension);
+	const auto  image     = LoadSampledImageDescriptor(ctx.state, mem.resource);
+	const auto  size      = ctx.state.builder.AllocateId();
+	if (info.multisampled != 0u) {
 		ctx.state.builder.AddFunction(
-		    {OpImageQuerySize, ImageViewSizeType(ctx.state, view), size, image});
+		    {OpImageQuerySize, ImageViewSizeType(ctx.state, dimension), size, image});
 	} else {
-		ctx.state.builder.AddFunction({OpImageQuerySizeLod, ImageViewSizeType(ctx.state, view),
+		ctx.state.builder.AddFunction({OpImageQuerySizeLod, ImageViewSizeType(ctx.state, dimension),
 		                               size, image, AddressU32(ctx, mem, address, 0)});
 	}
-	const auto components = ImageViewCoordinateComponents(view);
+	const auto components = info.coordinate_components;
 	uint32_t   result[4]  = {ConstantU32(ctx.state, 0), ConstantU32(ctx.state, 0),
 	                         ConstantU32(ctx.state, 0), ConstantU32(ctx.state, 0)};
 	if (components == 1u) {
@@ -285,7 +271,7 @@ uint32_t QueryDimensions(ValueEmitContext& ctx, const IR::Inst& inst, const IR::
 			    {OpCompositeExtract, TypeU32(ctx.state), result[index], size, index});
 		}
 	}
-	if (ImageSpirvMultisampled(view) == 0u) {
+	if (info.multisampled == 0u) {
 		result[3] = ctx.state.builder.AllocateId();
 		ctx.state.builder.AddFunction({OpImageQueryLevels, TypeU32(ctx.state), result[3], image});
 	}
@@ -296,8 +282,8 @@ uint32_t QueryDimensions(ValueEmitContext& ctx, const IR::Inst& inst, const IR::
 }
 
 uint32_t PackedOffset(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::Inst& address,
-                      const ImageSampleLayout& layout, ImageViewKind view) {
-	const auto components = ImageViewSpatialComponents(view);
+                      const ImageSampleLayout& layout, ImageDimension dimension) {
+	const auto components = ImageDimensionInfoFor(dimension).spatial_components;
 	const auto zero       = ConstantI32(ctx.state, 0);
 	if (layout.offset == NoImageComponent || mem.image_address_components <= layout.offset) {
 		if (components == 1u) return zero;
@@ -332,8 +318,8 @@ uint32_t PackedOffset(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR
 	return result;
 }
 
-uint32_t HorizontalOffsets(EmitterState& state, ImageViewKind view) {
-	const auto components   = ImageViewSpatialComponents(view);
+uint32_t HorizontalOffsets(EmitterState& state, ImageDimension dimension) {
+	const auto components   = ImageDimensionInfoFor(dimension).spatial_components;
 	const auto count        = ConstantU32(state, 4);
 	const auto element_type = components == 1u ? TypeI32(state) : TypeI32Vector(state, 2);
 	const auto array_type   = state.builder.Type(OpTypeArray, {element_type, count});
@@ -363,8 +349,7 @@ Format::BufferFormatInfo ImageConversionFormat(const EmitterState&   state,
 	const auto format = state.program.info.images[mem.resource].conversion_format;
 	if (format == Prospero::BufferFormat::kInvalid) return {};
 	const auto info = Format::GetFormatInfo(format);
-	EXIT_IF(!Prospero::IsSampledTextureFormat(format) ||
-	        !Prospero::IsUintTextureFormat(format) ||
+	EXIT_IF(Prospero::SampledTextureNumericClass(format) != Prospero::TextureNumericClass::Uint ||
 	        Prospero::RemapTextureFormat(format) == format ||
 	        info.type != Format::ComponentType::Uint || !info.packed_bitfield ||
 	        info.byte_size != sizeof(uint32_t) || info.component_count == 0u ||
@@ -439,46 +424,46 @@ uint32_t UnpackImageGather(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uin
 	return result;
 }
 
-uint32_t EmitOneDimensionalGatherLz(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint32_t pc,
-	                                 uint32_t coord, bool integer) {
+uint32_t EmitOneDimensionalGatherLz(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
+                                    uint32_t coord, Prospero::TextureNumericClass numeric_class) {
 	auto& state = ctx.state;
 	state.builder.RequireCapability(CapabilityImageQuery);
-	const auto image = LoadSampledImageDescriptor(state, mem, pc, ImageViewKind::Dim1D);
+	const auto image = LoadSampledImageDescriptor(state, mem.resource);
 	const auto width = state.builder.AllocateId();
 	state.builder.AddFunction(
 	    {OpImageQuerySizeLod, TypeU32(state), width, image, ConstantU32(state, 0)});
 	const auto width_f32 = state.builder.AllocateId();
 	state.builder.AddFunction({OpConvertUToF, TypeF32(state), width_f32, width});
 	const auto left = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpExtInst, TypeF32(state), left, GlslStd450(state), GlslFloor,
-	     Binary(state, OpFSub, TypeF32(state),
-	            Binary(state, OpFMul, TypeF32(state), coord, width_f32),
-	            ConstantF32(state, 0x3f000000u))});
+	state.builder.AddFunction({OpExtInst, TypeF32(state), left, GlslStd450(state), GlslFloor,
+	                           Binary(state, OpFSub, TypeF32(state),
+	                                  Binary(state, OpFMul, TypeF32(state), coord, width_f32),
+	                                  ConstantF32(state, 0x3f000000u))});
 
-	const auto sampled = MakeSampledImage(state, mem, pc, ImageViewKind::Dim1D);
-	const auto vector_type = integer ? TypeU32Vector(state, 4) : TypeF32Vector(state, 4);
-	const auto scalar_type = integer ? TypeU32(state) : TypeF32(state);
-	const auto component = ImageConversionFormat(state, mem).format == Prospero::BufferFormat::kInvalid
-	                           ? ImageGatherComponent(mem.dmask)
-	                           : 0u;
+	const auto sampled     = MakeSampledImage(state, mem.resource, mem.sampler);
+	const auto vector_type = ImageVectorType(state, numeric_class, 4);
+	const auto scalar_type = ImageScalarType(state, numeric_class);
+	const auto component =
+	    ImageConversionFormat(state, mem).format == Prospero::BufferFormat::kInvalid
+	        ? ImageGatherComponent(mem.dmask)
+	        : 0u;
 	uint32_t values[2] {};
 	for (uint32_t index = 0; index < 2u; index++) {
-		const auto sample_coord = Binary(
-		    state, OpFDiv, TypeF32(state),
-		    Binary(state, OpFAdd, TypeF32(state), left,
-		           ConstantF32(state, index == 0u ? 0x3f000000u : 0x3fc00000u)),
-		    width_f32);
+		const auto sample_coord =
+		    Binary(state, OpFDiv, TypeF32(state),
+		           Binary(state, OpFAdd, TypeF32(state), left,
+		                  ConstantF32(state, index == 0u ? 0x3f000000u : 0x3fc00000u)),
+		           width_f32);
 		const auto texel = state.builder.AllocateId();
-		state.builder.AddFunction({OpImageSampleExplicitLod, vector_type, texel, sampled, sample_coord,
-		                           ImageOperandsLodMask, ZeroF32(state)});
+		state.builder.AddFunction({OpImageSampleExplicitLod, vector_type, texel, sampled,
+		                           sample_coord, ImageOperandsLodMask, ZeroF32(state)});
 		values[index] = state.builder.AllocateId();
 		state.builder.AddFunction(
 		    {OpCompositeExtract, scalar_type, values[index], texel, component});
 	}
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({OpCompositeConstruct, vector_type, result, values[0], values[1],
-	                           values[1], values[0]});
+	state.builder.AddFunction(
+	    {OpCompositeConstruct, vector_type, result, values[0], values[1], values[1], values[0]});
 	return result;
 }
 
@@ -549,6 +534,7 @@ uint32_t StoreTexel(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint32_t d
 
 uint32_t ImageAtomicOpcode(IR::ValueOpcode opcode) {
 	switch (opcode) {
+		case IR::ValueOpcode::ImageAtomicSwap32: return OpAtomicExchange;
 		case IR::ValueOpcode::ImageAtomicIAdd32: return OpAtomicIAdd;
 		case IR::ValueOpcode::ImageAtomicUMin32: return OpAtomicUMin;
 		case IR::ValueOpcode::ImageAtomicUMax32: return OpAtomicUMax;
@@ -569,24 +555,24 @@ bool EmitValueImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 	}
 	auto&       state     = ctx.state;
 	const auto& mem       = ctx.Memory(inst);
-	const auto  pc        = inst.Flags<IR::MemoryFlags>().pc;
 	const auto  image_arg = inst.Arg(0);
 	ctx.ResourceIndex(image_arg, IR::ValueOpcode::GetImageResource);
+	const auto& image   = state.program.info.images.at(mem.resource);
 	const auto* address = ctx.ImageAddress(inst.Arg(image_info.needs_sampler ? 2 : 1));
 	if (address == nullptr) return true;
 	if (op == IR::ValueOpcode::ImageQueryDimensions) {
 		state.builder.RequireCapability(CapabilityImageQuery);
-		ctx.Define(inst, QueryDimensions(ctx, inst, mem, *address));
+		ctx.Define(inst, QueryDimensions(ctx, mem, *address));
 		return true;
 	}
 	if (op == IR::ValueOpcode::ImageQueryLod) {
 		state.builder.RequireCapability(CapabilityImageQuery);
-		const auto view    = SampledImageViewKind(state, mem, pc);
-		const auto sampled = MakeSampledImage(state, mem, pc, view);
-		const auto lod     = state.builder.AllocateId();
+		const auto dimension = image.dimension;
+		const auto sampled   = MakeSampledImage(state, mem.resource, mem.sampler);
+		const auto lod       = state.builder.AllocateId();
 		state.builder.AddFunction(
 		    {OpImageQueryLod, TypeF32Vector(state, 2), lod, sampled,
-		     CoordF32(ctx, mem, *address, 0, ImageViewSpatialComponents(view))});
+		     CoordF32(ctx, mem, *address, 0, ImageDimensionInfoFor(dimension).spatial_components)});
 		uint32_t values[4] = {ConstantU32(state, 0), ConstantU32(state, 0), ConstantU32(state, 0),
 		                      ConstantU32(state, 0)};
 		for (uint32_t index = 0; index < 2u; index++) {
@@ -601,143 +587,153 @@ bool EmitValueImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		return true;
 	}
 	if (op == IR::ValueOpcode::ImageRead) {
-		const auto view      = SampledImageViewKind(state, mem, pc);
-		const bool integer   = mem.kind == IR::ResourceKind::ImageUint;
-		const auto condition = ctx.Arg(inst, 2);
+		const auto  dimension      = image.dimension;
+		const auto& dimension_info = ImageDimensionInfoFor(dimension);
+		const auto  numeric_class  = image.numeric_class;
+		const auto  condition      = ctx.Arg(inst, 2);
 		ctx.Define(
 		    inst,
 		    EmitValueOrDefaultIfCondition(
 		        state, condition, TypeU32Vector(state, 4), ConstantU32CompositeZero(state, 4),
 		        [&]() {
-			        const auto image = LoadSampledImageDescriptor(state, mem, pc, view);
-			        const auto color = state.builder.AllocateId();
-			        const auto coord = CoordU32(ctx, mem, *address, view);
-			        if (ImageSpirvMultisampled(view) != 0u) {
+			        const auto descriptor = LoadSampledImageDescriptor(state, mem.resource);
+			        const auto color      = state.builder.AllocateId();
+			        const auto coord      = CoordU32(ctx, mem, *address, dimension);
+			        if (dimension_info.multisampled != 0u) {
 				        state.builder.AddFunction(
-				            {OpImageFetch,
-				             integer ? TypeU32Vector(state, 4) : TypeF32Vector(state, 4), color,
-				             image, coord, ImageOperandsSampleMask,
-				             AddressU32(ctx, mem, *address, ImageViewCoordinateComponents(view))});
+				            {OpImageFetch, ImageVectorType(state, numeric_class, 4), color,
+				             descriptor, coord, ImageOperandsSampleMask,
+				             AddressU32(ctx, mem, *address, dimension_info.coordinate_components)});
 			        } else {
-				        state.builder.AddFunction(
-				            {OpImageFetch,
-				             integer ? TypeU32Vector(state, 4) : TypeF32Vector(state, 4), color,
-				             image, coord, ImageOperandsLodMask, LodU32(ctx, mem, *address, view)});
+				        state.builder.AddFunction({OpImageFetch,
+				                                   ImageVectorType(state, numeric_class, 4), color,
+				                                   descriptor, coord, ImageOperandsLodMask,
+				                                   LodU32(ctx, mem, *address, dimension)});
 			        }
-			        return ResultVector(ctx, UnpackImageTexel(ctx, mem, color), integer, false,
-			                            mem);
+			        return ResultVector(ctx, UnpackImageTexel(ctx, mem, color), numeric_class,
+			                            false, mem);
 		        }));
 		return true;
 	}
 	if (op == IR::ValueOpcode::ImageWrite) {
-		const auto uint_image = mem.kind == IR::ResourceKind::StorageImageUint;
-		const auto view       = StorageImageViewKind(mem);
+		const bool uint_image = image.numeric_class == Prospero::TextureNumericClass::Uint;
+		const auto dimension  = image.dimension;
 		EmitIfCondition(state, ctx.Arg(inst, 3), [&]() {
 			const auto mip_lod =
 			    state.program.info.images[mem.resource].mip_mode == IR::ImageMipMode::DynamicStorage
-			        ? LodU32(ctx, mem, *address, view)
+			        ? LodU32(ctx, mem, *address, dimension)
 			        : 0u;
-			const auto coord = CoordU32(ctx, mem, *address, view);
+			const auto coord = CoordU32(ctx, mem, *address, dimension);
 			const auto texel = StoreTexel(ctx, mem, ctx.Arg(inst, 2), uint_image);
-			EmitStorageImageWrite(state, mem.resource, view, mip_lod, coord, texel);
+			EmitStorageImageWrite(state, mem.resource, mip_lod, coord, texel);
 		});
 		return true;
 	}
 	if (op == IR::ValueOpcode::ImageSampleRaw || op == IR::ValueOpcode::ImageGatherRaw) {
-		const auto view    = SampledImageViewKind(state, mem, pc);
-		const auto layout  = Layout(mem, view);
-		const bool integer = mem.kind == IR::ResourceKind::ImageUint;
-		const bool dref    = HasFlag(mem, Decoder::ImageSampleFlagCompare);
+		const auto  dimension      = image.dimension;
+		const auto& dimension_info = ImageDimensionInfoFor(dimension);
+		const auto  layout         = Layout(mem, dimension);
+		const auto  numeric_class  = image.numeric_class;
+		const bool  dref           = HasFlag(mem, Decoder::ImageSampleFlagCompare);
 		if (dref && state.program.info.images[mem.resource].conversion_format !=
 		                Prospero::BufferFormat::kInvalid) {
 			ctx.Fail(inst, "uses depth comparison with a packed integer image");
 			return true;
 		}
 		const auto coord =
-		    CoordF32(ctx, mem, *address, layout.coord, ImageViewCoordinateComponents(view));
+		    CoordF32(ctx, mem, *address, layout.coord, dimension_info.coordinate_components);
 		if (op == IR::ValueOpcode::ImageGatherRaw) {
-			if (view == ImageViewKind::Dim1D) {
+			if (dimension == ImageDimension::Dim1D) {
 				if (dref || !HasFlag(mem, Decoder::ImageSampleFlagLevelZero) ||
 				    HasFlag(mem, Decoder::ImageSampleFlagOffset) ||
 				    HasFlag(mem, Decoder::ImageSampleFlagGatherHorizontal)) {
 					ctx.Fail(inst, "has an unsupported 1D gather variant");
 					return true;
 				}
-				const auto sample = EmitOneDimensionalGatherLz(ctx, mem, pc, coord, integer);
-				ctx.Define(inst, ResultVector(ctx, UnpackImageGather(ctx, mem, sample), integer,
-				                              false, mem, true));
+				const auto sample = EmitOneDimensionalGatherLz(ctx, mem, coord, numeric_class);
+				ctx.Define(inst, ResultVector(ctx, UnpackImageGather(ctx, mem, sample),
+				                              numeric_class, false, mem, true));
 				return true;
 			}
-			if (view == ImageViewKind::Dim1DArray) {
+			if (dimension == ImageDimension::Dim1DArray) {
 				ctx.Fail(inst, "has an unsupported 1D-array gather");
 				return true;
 			}
-			const auto            sampled = MakeSampledImage(state, mem, pc, view);
+			const auto            sampled = MakeSampledImage(state, mem.resource, mem.sampler);
 			const auto            sample  = state.builder.AllocateId();
-			std::vector<uint32_t> words =
-			    dref ? std::vector<uint32_t> {OpImageDrefGather,
-			                                  TypeF32Vector(state, 4),
-			                                  sample,
-			                                  sampled,
-			                                  coord,
-			                                  layout.dref != NoImageComponent
-			                                      ? AddressF32(ctx, mem, *address, layout.dref)
-			                                      : ZeroF32(state)}
-			         : std::vector<uint32_t> {
-			               OpImageGather,
-			               integer ? TypeU32Vector(state, 4) : TypeF32Vector(state, 4),
-			               sample,
-			               sampled,
-			               coord,
-			               ConstantU32(state, ImageConversionFormat(state, mem).format ==
-			                                          Prospero::BufferFormat::kInvalid
-			                                      ? ImageGatherComponent(mem.dmask)
-			                                      : 0u)};
+			std::vector<uint32_t> words;
+			if (dref) {
+				auto dref_value = ZeroF32(state);
+				if (layout.dref != NoImageComponent) {
+					dref_value = AddressF32(ctx, mem, *address, layout.dref);
+				}
+				words = {OpImageDrefGather, TypeF32Vector(state, 4), sample, sampled, coord,
+				         dref_value};
+			} else {
+				uint32_t component = 0;
+				if (ImageConversionFormat(state, mem).format == Prospero::BufferFormat::kInvalid) {
+					component = ImageGatherComponent(mem.dmask);
+				}
+				words = {OpImageGather, ImageVectorType(state, numeric_class, 4),
+				         sample,        sampled,
+				         coord,         ConstantU32(state, component)};
+			}
 			if (HasFlag(mem, Decoder::ImageSampleFlagGatherHorizontal)) {
 				words.push_back(ImageOperandsConstOffsetsMask);
-				words.push_back(HorizontalOffsets(state, view));
+				words.push_back(HorizontalOffsets(state, dimension));
 			} else if (layout.offset != NoImageComponent) {
 				words.push_back(ImageOperandsOffsetMask);
-				words.push_back(PackedOffset(ctx, mem, *address, layout, view));
+				words.push_back(PackedOffset(ctx, mem, *address, layout, dimension));
 			}
 			state.builder.AddFunction(words);
+			auto result_numeric_class = numeric_class;
+			if (dref) {
+				result_numeric_class = Prospero::TextureNumericClass::Float;
+			}
 			ctx.Define(inst, ResultVector(ctx, UnpackImageGather(ctx, mem, sample),
-			                              integer && !dref, false, mem, true));
+			                              result_numeric_class, false, mem, true));
 			return true;
 		}
 		const bool explicit_lod = HasFlag(mem, Decoder::ImageSampleFlagDerivative) ||
 		                          HasFlag(mem, Decoder::ImageSampleFlagLod) ||
 		                          HasFlag(mem, Decoder::ImageSampleFlagLevelZero) ||
 		                          state.stage != ShaderType::Pixel;
-		const auto opcode = explicit_lod
-		                        ? (dref ? OpImageSampleDrefExplicitLod : OpImageSampleExplicitLod)
-		                        : (dref ? OpImageSampleDrefImplicitLod : OpImageSampleImplicitLod);
-		const auto result_type =
-		    dref ? TypeF32(state) : (integer ? TypeU32Vector(state, 4) : TypeF32Vector(state, 4));
-		const auto dref_value =
-		    dref ? (layout.dref != NoImageComponent ? AddressF32(ctx, mem, *address, layout.dref)
-		                                            : ZeroF32(state))
-		         : 0u;
+		uint32_t opcode = OpImageSampleImplicitLod;
+		if (explicit_lod) {
+			opcode = dref ? OpImageSampleDrefExplicitLod : OpImageSampleExplicitLod;
+		} else if (dref) {
+			opcode = OpImageSampleDrefImplicitLod;
+		}
+		uint32_t result_type = ImageVectorType(state, numeric_class, 4);
+		uint32_t dref_value  = 0;
+		if (dref) {
+			result_type = TypeF32(state);
+			dref_value  = ZeroF32(state);
+			if (layout.dref != NoImageComponent) {
+				dref_value = AddressF32(ctx, mem, *address, layout.dref);
+			}
+		}
 		uint32_t              operand_mask = 0;
 		std::vector<uint32_t> operands;
 		if (HasFlag(mem, Decoder::ImageSampleFlagDerivative)) {
 			operand_mask |= ImageOperandsGradMask;
 			operands.push_back(
-			    CoordF32(ctx, mem, *address, layout.grad_x, ImageViewSpatialComponents(view)));
+			    CoordF32(ctx, mem, *address, layout.grad_x, dimension_info.spatial_components));
 			operands.push_back(
-			    CoordF32(ctx, mem, *address, layout.grad_y, ImageViewSpatialComponents(view)));
+			    CoordF32(ctx, mem, *address, layout.grad_y, dimension_info.spatial_components));
 		} else if (explicit_lod) {
 			operand_mask |= ImageOperandsLodMask;
-			operands.push_back(HasFlag(mem, Decoder::ImageSampleFlagLod) &&
-			                           layout.lod != NoImageComponent
-			                       ? AddressF32(ctx, mem, *address, layout.lod)
-			                       : ZeroF32(state));
+			auto lod = ZeroF32(state);
+			if (HasFlag(mem, Decoder::ImageSampleFlagLod) && layout.lod != NoImageComponent) {
+				lod = AddressF32(ctx, mem, *address, layout.lod);
+			}
+			operands.push_back(lod);
 		} else if (layout.bias != NoImageComponent) {
 			operand_mask |= ImageOperandsBiasMask;
 			operands.push_back(AddressF32(ctx, mem, *address, layout.bias));
 		}
 		const auto EmitSample = [&](uint32_t resource) {
-			const auto            sampled = MakeSampledImage(state, mem, pc, view, resource);
+			const auto            sampled = MakeSampledImage(state, resource, mem.sampler);
 			const auto            sample  = state.builder.AllocateId();
 			std::vector<uint32_t> words {opcode, result_type, sample, sampled, coord};
 			if (dref) {
@@ -750,16 +746,18 @@ bool EmitValueImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			state.builder.AddFunction(words);
 			return sample;
 		};
-		const auto& image = state.program.info.images[mem.resource];
 		if (image.indirect_root != mem.resource) {
 			const auto sample = EmitSample(mem.resource);
-			ctx.Define(inst, ResultVector(ctx, dref ? sample : UnpackImageTexel(ctx, mem, sample),
-			                              integer, dref, mem));
+			auto       result = sample;
+			if (!dref) {
+				result = UnpackImageTexel(ctx, mem, sample);
+			}
+			ctx.Define(inst, ResultVector(ctx, result, numeric_class, dref, mem));
 			return true;
 		}
 		const auto* handle = image_arg.ResolveInstruction();
-		const auto* source = image.source < ctx.program.descriptor_sources.size()
-		                         ? &ctx.program.descriptor_sources[image.source]
+		const auto* source = image.source < state.program.descriptor_sources.size()
+		                         ? &state.program.descriptor_sources[image.source]
 		                         : nullptr;
 		if (handle == nullptr || source == nullptr || !source->indirect_image.has_value() ||
 		    source->indirect_image->key_arg >= handle->NumArgs()) {
@@ -767,7 +765,7 @@ bool EmitValueImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			return true;
 		}
 		const auto key = ctx.Def(handle->Arg(source->indirect_image->key_arg));
-		if (state.flattened_srt_variable == 0 || image.indirect_mapping_capacity == 0u ||
+		if (state.flattened_srt_variable == 0 || image.indirect_search_iterations == 0u ||
 		    image.indirect_resources.size() < 2u) {
 			ctx.Fail(inst, "has no indirect image runtime mapping");
 			return true;
@@ -785,7 +783,7 @@ bool EmitValueImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		auto       low      = ConstantU32(state, 0u);
 		auto       high     = LoadMapping(mapping);
 		auto       selected = ConstantU32(state, 0u);
-		for (uint32_t iteration = 0; iteration < std::bit_width(image.indirect_mapping_capacity);
+		for (uint32_t iteration = 0; iteration < image.indirect_search_iterations;
 		     iteration++) {
 			const auto active = Binary(state, OpULessThan, TypeBool(state), low, high);
 			const auto mid =
@@ -845,22 +843,24 @@ bool EmitValueImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		}
 		EmitLabel(state, merge_label);
 		state.builder.AddFunction(phi_words);
-		ctx.Define(inst,
-		           ResultVector(ctx, dref ? phi_words[2] : UnpackImageTexel(ctx, mem, phi_words[2]),
-		                        integer, dref, mem));
+		auto result = phi_words[2];
+		if (!dref) {
+			result = UnpackImageTexel(ctx, mem, result);
+		}
+		ctx.Define(inst, ResultVector(ctx, result, numeric_class, dref, mem));
 		return true;
 	}
 	const auto atomic_opcode = ImageAtomicOpcode(op);
 	if (atomic_opcode != 0u) {
-		const auto view = StorageImageViewKind(mem);
+		const auto dimension = image.dimension;
 		ctx.Define(inst, EmitValueOrZeroIfCondition(state, ctx.Arg(inst, 3), [&]() {
 			           const auto pointer = state.builder.AllocateId();
 			           const auto pointer_type =
 			               state.builder.Type(OpTypePointer, {StorageClassImage, TypeU32(state)});
 			           state.builder.AddFunction(
 			               {OpImageTexelPointer, pointer_type, pointer,
-			                StorageImageDescriptorPointer(state, mem.resource, view),
-			                CoordU32(ctx, mem, *address, view), ConstantU32(state, 0)});
+			                StorageImageDescriptorPointer(state, mem.resource),
+			                CoordU32(ctx, mem, *address, dimension), ConstantU32(state, 0)});
 			           const auto old = state.builder.AllocateId();
 			           state.builder.AddFunction({atomic_opcode, TypeU32(state), old, pointer,
 			                                      ConstantU32(state, ScopeDevice),

src/graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h

@@ -6,7 +6,6 @@
 #include "graphics/shader/recompiler/BufferFormat.h"
 #include "graphics/shader/recompiler/backend/spirv/SpirvBuilder.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
-#include "graphics/shader/recompiler/ir/ShaderIR.h"
 #include "graphics/shader/recompiler/ir/passes/BindingLayout.h"
 #include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 
@@ -40,9 +39,11 @@ enum : uint32_t {
 	MemoryModelGLSL450                       = 1,
 	CapabilityShader                         = 1,
 	CapabilityInt64                          = 11,
+	CapabilityInt64Atomics                   = 12,
 	CapabilityImageGatherExtended            = 25,
 	CapabilityClipDistance                   = 32,
 	CapabilityCullDistance                   = 33,
+	CapabilitySampleRateShading              = 35,
 	CapabilitySampled1D                      = 43,
 	CapabilityImage1D                        = 44,
 	CapabilityImageQuery                     = 50,
@@ -50,8 +51,9 @@ enum : uint32_t {
 	CapabilityGroupNonUniform                = 61,
 	CapabilityGroupNonUniformBallot          = 64,
 	CapabilityGroupNonUniformShuffle         = 65,
+	CapabilityShaderLayer                    = 69,
+	CapabilityShaderViewportIndex            = 70,
 	CapabilitySignedZeroInfNanPreserve       = 4466,
-	CapabilityShaderViewportIndexLayerEXT    = 5254,
 	CapabilityFragmentBarycentricKHR         = 5284,
 	CapabilityComputeDerivativeGroupQuadsKHR = 5288,
 	CapabilityPhysicalStorageBufferAddresses = 5347,
@@ -59,6 +61,7 @@ enum : uint32_t {
 	StorageClassInput                        = 1,
 	StorageClassOutput                       = 3,
 	StorageClassWorkgroup                    = 4,
+	StorageClassPrivate                      = 6,
 	StorageClassFunction                     = 7,
 	StorageClassPushConstant                 = 9,
 	StorageClassImage                        = 11,
@@ -74,6 +77,7 @@ enum : uint32_t {
 	DecorationBuiltIn       = 11,
 	DecorationNoPerspective = 13,
 	DecorationFlat          = 14,
+	DecorationAliased       = 20,
 	DecorationLocation      = 30,
 	DecorationArrayStride   = 6,
 	DecorationBinding       = 33,
@@ -88,8 +92,10 @@ enum : uint32_t {
 	BuiltInClipDistance              = 3,
 	BuiltInCullDistance              = 4,
 	BuiltInLayer                     = 9,
+	BuiltInViewportIndex              = 10,
 	BuiltInFragCoord                 = 15,
 	BuiltInFrontFacing               = 17,
+	BuiltInSampleId                  = 18,
 	BuiltInSampleMask                = 20,
 	BuiltInFragDepth                 = 22,
 	BuiltInWorkgroupId               = 26,
@@ -129,6 +135,7 @@ enum : uint32_t {
 	MemorySemanticsAcquireRelease  = 0x00000008u,
 	MemorySemanticsUniformMemory   = 0x00000040u,
 	MemorySemanticsWorkgroupMemory = 0x00000100u,
+	MemorySemanticsImageMemory     = 0x00000800u,
 };
 
 enum : uint32_t {
@@ -196,6 +203,7 @@ enum : uint32_t {
 	OpIMul                         = 132,
 	OpFMul                         = 133,
 	OpUDiv                         = 134,
+	OpUMod                         = 137,
 	OpFDiv                         = 136,
 	OpIAddCarry                    = 149,
 	OpUMulExtended                 = 151,
@@ -316,67 +324,68 @@ struct OutputBinding {
 	uint32_t            location    = 0;
 	uint32_t            variable_id = 0;
 	std::string         debug_name;
+	uint32_t            mesh_data_variable = 0;
 };
 
-struct DescriptorResourceBinding {
-	const IR::DescriptorBinding* descriptor  = nullptr;
-	uint32_t                     array_index = 0;
+using ImageDimension = Decoder::ImageDimension;
+
+struct ImageDimensionInfo {
+	ImageDimension dimension;
+	uint32_t       spirv_dimension;
+	uint32_t       coordinate_components;
+	uint32_t       spatial_components;
+	uint32_t       arrayed;
+	uint32_t       multisampled;
 };
 
-enum class ImageViewKind {
-	Dim1D,
-	Dim1DArray,
-	Dim2D,
-	Dim2DArray,
-	Dim3D,
-	Dim2DMsaa,
-	Dim2DMsaaArray,
-	Count,
-};
+constexpr std::array<ImageDimensionInfo, 7> ImageDimensions {{
+    {ImageDimension::Dim1D, Dim1D, 1, 1, 0, 0},
+    {ImageDimension::Dim1DArray, Dim1D, 2, 1, 1, 0},
+    {ImageDimension::Dim2D, Dim2D, 2, 2, 0, 0},
+    {ImageDimension::Dim2DArray, Dim2D, 3, 2, 1, 0},
+    {ImageDimension::Dim3D, Dim3D, 3, 3, 0, 0},
+    {ImageDimension::Dim2DMsaa, Dim2D, 2, 2, 0, 1},
+    {ImageDimension::Dim2DMsaaArray, Dim2D, 3, 2, 1, 1},
+}};
 
-constexpr uint32_t SampledImageViewKindCount = static_cast<uint32_t>(ImageViewKind::Count);
-constexpr uint32_t StorageImageViewKindCount = static_cast<uint32_t>(ImageViewKind::Dim2DMsaa);
-
-enum class StorageImageClass : uint32_t {
-	FormatlessFloat,
-	FormatlessUint,
-	AtomicUint,
-	Count,
-};
+const ImageDimensionInfo& ImageDimensionInfoFor(ImageDimension dimension);
 
 struct EmitterState {
-	EmitterState(const IR::Program& program_, const IR::ResourceSnapshot& resources_,
-	             ShaderStageInputInfo input_info_)
-	    : program(program_), resources(resources_), input_info(input_info_),
-	      requirements(*program_.spirv_requirements) {}
+	EmitterState(const IR::Program& program_, ShaderStageInputInfo input_info_)
+	    : builder(program_.stage == ShaderType::Mesh ? 0x00010400u : 0x00010300u),
+	      program(program_), input_info(input_info_), requirements(*program_.spirv_requirements) {}
 
-	Builder                      builder;
-	const IR::Program&           program;
-	const IR::ResourceSnapshot&  resources;
-	ShaderStageInputInfo         input_info;
-	const IR::SpirvRequirements& requirements;
-	ShaderType                   stage                   = ShaderType::Unknown;
-	uint32_t                     wave_size               = 64;
-	uint32_t                     storage_buffer_variable = 0;
-	std::array<uint32_t, IR::ShaderInfo::MaxBuffers>      memory_byte_offsets {};
-	uint32_t                                             bda_pagetable_variable = 0;
-	uint32_t                                             fault_buffer_variable  = 0;
-	uint32_t                                             bda_pointer_function   = 0;
-	uint32_t                                             gds_variable            = 0;
-	uint32_t                                             gds_length              = 0;
-	uint32_t                                             push_constant_variable  = 0;
-	uint32_t                                             vsharp_storage_variable = 0;
-	uint32_t                                             flattened_srt_variable  = 0;
-	uint32_t                                             lds_variable            = 0;
-	uint32_t                                             scratch_variable        = 0;
-	std::array<uint32_t, 2u * SampledImageViewKindCount> sampled_image_variables {};
-	std::array<uint32_t,
-	           static_cast<uint32_t>(StorageImageClass::Count) * StorageImageViewKindCount>
-	                           storage_image_variables {};
+	Builder                                          builder;
+	const IR::Program&                               program;
+	ShaderStageInputInfo                             input_info;
+	const IR::SpirvRequirements&                     requirements;
+	ShaderType                                       stage                   = ShaderType::Unknown;
+	uint32_t                                         lane_count              = 1;
+	uint32_t                                         lane_half               = 0;
+	uint32_t                                         storage_buffer_variable = 0;
+	uint32_t                                         storage_buffer_u64_variable = 0;
+	std::array<uint32_t, IR::ShaderInfo::MaxBuffers> memory_byte_offsets {};
+	uint32_t                                         bda_pagetable_variable  = 0;
+	uint32_t                                         fault_buffer_variable   = 0;
+	uint32_t                                         bda_pointer_function    = 0;
+	uint32_t                                         gds_variable            = 0;
+	uint32_t                                         gds_length              = 0;
+	uint32_t                                         push_constant_variable  = 0;
+	uint32_t                                         shader_data_storage_variable = 0;
+	uint32_t                                         flattened_srt_variable  = 0;
+	uint32_t                                         lds_variable            = 0;
+	std::array<uint32_t, 2>                          scratch_variable {};
+	std::array<uint32_t, IR::ImageBindingCount>      image_variables {};
 	uint32_t                   sampler_variable                      = 0;
 	uint32_t                   main_func                             = 0;
+	uint32_t                   mesh_guest_func                       = 0;
+	uint32_t                   mesh_allocation                       = 0;
+	uint32_t                   mesh_primitive_data                   = 0;
+	uint32_t                   mesh_primitives                       = 0;
+	uint32_t                   mesh_cull                             = 0;
 	uint32_t                   entry_label                           = 0;
 	uint32_t                   current_label                         = 0;
+	const IR::Block*           current_block                         = nullptr;
 	uint32_t                   pixel_valid_mask_variable             = 0;
 	uint32_t                   subgroup_local_invocation_id_variable = 0;
 	uint32_t                   per_vertex_variable                   = 0;
@@ -384,6 +393,7 @@ struct EmitterState {
 	uint32_t                   clip_distance_variable                = 0;
 	uint32_t                   cull_distance_variable                = 0;
 	uint32_t                   layer_variable                        = 0;
+	uint32_t                   viewport_index_variable               = 0;
 	uint32_t                   clip_distance_count                   = 0;
 	uint32_t                   cull_distance_count                   = 0;
 	uint32_t                   depth_variable                        = 0;
@@ -391,6 +401,7 @@ struct EmitterState {
 	std::vector<InputBinding>  inputs;
 	std::vector<OutputBinding> outputs;
 	std::vector<uint32_t>      interface_variables;
+	std::unordered_map<const IR::Block*, uint32_t> labels;
 };
 
 uint32_t TypeVoid(EmitterState& state);
@@ -399,6 +410,7 @@ uint32_t TypeBoolVector(EmitterState& state, uint32_t components);
 uint32_t TypeU32(EmitterState& state);
 uint32_t TypeU64(EmitterState& state);
 uint32_t TypeDeviceAddress(EmitterState& state);
+uint32_t TypeScalarU64(EmitterState& state);
 uint32_t TypeU32Pair(EmitterState& state);
 uint32_t TypeI32(EmitterState& state);
 uint32_t TypeI32Pair(EmitterState& state);
@@ -412,6 +424,8 @@ uint32_t TypePointer(EmitterState& state, uint32_t storage_class, uint32_t point
 uint32_t TypeFunction(EmitterState& state);
 uint32_t TypeStorageBufferPointer(EmitterState& state);
 uint32_t TypeStorageBufferElementPointer(EmitterState& state);
+uint32_t TypeStorageBufferU64Pointer(EmitterState& state);
+uint32_t TypeStorageBufferU64ElementPointer(EmitterState& state);
 uint32_t TypeDeviceAddressStoragePointer(EmitterState& state);
 uint32_t TypePhysicalU32Pointer(EmitterState& state);
 uint32_t TypePushConstantElementPointer(EmitterState& state);
@@ -423,12 +437,35 @@ inline void EmitLabel(EmitterState& state, uint32_t label) {
 	state.builder.AddFunction({OpLabel, label});
 }
 
+inline uint32_t Unary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t value) {
+	const auto result = state.builder.AllocateId();
+	state.builder.AddFunction({opcode, type, result, value});
+	return result;
+}
+
+inline uint32_t Binary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t lhs,
+                            uint32_t rhs) {
+	const auto result = state.builder.AllocateId();
+	state.builder.AddFunction({opcode, type, result, lhs, rhs});
+	return result;
+}
+
+inline uint32_t Select(EmitterState& state, uint32_t type, uint32_t condition,
+                            uint32_t true_value, uint32_t false_value) {
+	const auto result = state.builder.AllocateId();
+	state.builder.AddFunction({OpSelect, type, result, condition, true_value, false_value});
+	return result;
+}
+
 struct ValueEmitContext {
-	ValueEmitContext(EmitterState& state_, const IR::Program& program_)
-	    : state(state_), program(program_) {}
+	explicit ValueEmitContext(EmitterState& state_): state(state_) {}
 
 	uint32_t              Def(IR::Value value);
 	uint32_t              Arg(const IR::Inst& inst, size_t index);
+	uint32_t              HalfArg(const IR::Inst& inst, size_t index, uint32_t half);
+	uint32_t              Ballot(IR::Value predicate);
+	uint32_t              FirstLane(uint32_t ballot);
+	uint32_t              Shuffle(const IR::Inst& inst, size_t index, uint32_t lane);
 	uint32_t              Result(const IR::Inst& inst);
 	uint32_t              TypeId(IR::Type type) const;
 	uint32_t              Emit(const IR::Inst& inst, uint32_t opcode, IR::Type type,
@@ -443,13 +480,12 @@ struct ValueEmitContext {
 	[[noreturn]] void     Fail(const IR::Inst& inst, const char* reason) const;
 
 	EmitterState&                                                      state;
-	const IR::Program&                                            program;
 	std::unordered_map<const IR::Inst*, uint32_t>                      definitions;
-	std::unordered_map<const IR::Block*, uint32_t>                     labels;
 	const std::unordered_map<const IR::Inst*, uint32_t>*               dispatcher_spills = nullptr;
 	std::unordered_map<const IR::Inst*, std::pair<uint32_t, uint32_t>> dispatcher_block_loads;
-	const IR::Block*                                                   current_block = nullptr;
 	uint32_t                                                           scratch_u32_variable = 0;
+	ValueEmitContext*                                                  other_half = nullptr;
+	uint32_t                                                           half       = 0;
 };
 
 enum class VertexInputScalarKind { Float, Sint, Uint };
@@ -471,127 +507,12 @@ struct ImageSampleLayout {
 	uint32_t grad_y = NoImageComponent;
 };
 
-constexpr uint32_t SampledImageIndex(bool integer, ImageViewKind view) {
-	return static_cast<uint32_t>(view) + (integer ? SampledImageViewKindCount : 0u);
-}
-
-constexpr StorageImageClass StorageImageClassFor(bool uint_image, bool atomic) {
-	if (atomic) {
-		return StorageImageClass::AtomicUint;
-	}
-	if (uint_image) {
-		return StorageImageClass::FormatlessUint;
-	}
-	return StorageImageClass::FormatlessFloat;
-}
-
-constexpr uint32_t StorageImageIndex(StorageImageClass image_class, ImageViewKind view) {
-	return static_cast<uint32_t>(view) +
-	       static_cast<uint32_t>(image_class) * StorageImageViewKindCount;
-}
-
-constexpr IR::DescriptorBindingKind SampledBindingKind(bool integer, ImageViewKind view) {
-	if (integer) {
-		switch (view) {
-			case ImageViewKind::Dim1D: return IR::DescriptorBindingKind::SampledUint1D;
-			case ImageViewKind::Dim1DArray: return IR::DescriptorBindingKind::SampledUint1DArray;
-			case ImageViewKind::Dim2D: return IR::DescriptorBindingKind::SampledUint2D;
-			case ImageViewKind::Dim2DArray: return IR::DescriptorBindingKind::SampledUint2DArray;
-			case ImageViewKind::Dim3D: return IR::DescriptorBindingKind::SampledUint3D;
-			case ImageViewKind::Dim2DMsaa: return IR::DescriptorBindingKind::SampledUint2DMsaa;
-			case ImageViewKind::Dim2DMsaaArray:
-				return IR::DescriptorBindingKind::SampledUint2DMsaaArray;
-			default: break;
-		}
-	}
-	switch (view) {
-		case ImageViewKind::Dim1D: return IR::DescriptorBindingKind::Sampled1D;
-		case ImageViewKind::Dim1DArray: return IR::DescriptorBindingKind::Sampled1DArray;
-		case ImageViewKind::Dim2D: return IR::DescriptorBindingKind::Sampled2D;
-		case ImageViewKind::Dim2DArray: return IR::DescriptorBindingKind::Sampled2DArray;
-		case ImageViewKind::Dim3D: return IR::DescriptorBindingKind::Sampled3D;
-		case ImageViewKind::Dim2DMsaa: return IR::DescriptorBindingKind::Sampled2DMsaa;
-		case ImageViewKind::Dim2DMsaaArray: return IR::DescriptorBindingKind::Sampled2DMsaaArray;
-		default: break;
-	}
-	return IR::DescriptorBindingKind::Count;
-}
-
-constexpr IR::DescriptorBindingKind StorageBindingKind(StorageImageClass image_class,
-                                                       ImageViewKind     view) {
-	switch (image_class) {
-		case StorageImageClass::FormatlessFloat:
-			switch (view) {
-				case ImageViewKind::Dim1D: return IR::DescriptorBindingKind::Storage1D;
-				case ImageViewKind::Dim1DArray: return IR::DescriptorBindingKind::Storage1DArray;
-				case ImageViewKind::Dim2D: return IR::DescriptorBindingKind::Storage2D;
-				case ImageViewKind::Dim2DArray: return IR::DescriptorBindingKind::Storage2DArray;
-				case ImageViewKind::Dim3D: return IR::DescriptorBindingKind::Storage3D;
-				default: break;
-			}
-			break;
-		case StorageImageClass::FormatlessUint:
-			switch (view) {
-				case ImageViewKind::Dim1D: return IR::DescriptorBindingKind::StorageUint1D;
-				case ImageViewKind::Dim1DArray:
-					return IR::DescriptorBindingKind::StorageUint1DArray;
-				case ImageViewKind::Dim2D: return IR::DescriptorBindingKind::StorageUint2D;
-				case ImageViewKind::Dim2DArray:
-					return IR::DescriptorBindingKind::StorageUint2DArray;
-				case ImageViewKind::Dim3D: return IR::DescriptorBindingKind::StorageUint3D;
-				default: break;
-			}
-			break;
-		case StorageImageClass::AtomicUint:
-			switch (view) {
-				case ImageViewKind::Dim1D: return IR::DescriptorBindingKind::StorageAtomic1D;
-				case ImageViewKind::Dim1DArray:
-					return IR::DescriptorBindingKind::StorageAtomic1DArray;
-				case ImageViewKind::Dim2D: return IR::DescriptorBindingKind::StorageAtomic2D;
-				case ImageViewKind::Dim2DArray:
-					return IR::DescriptorBindingKind::StorageAtomic2DArray;
-				case ImageViewKind::Dim3D: return IR::DescriptorBindingKind::StorageAtomic3D;
-				default: break;
-			}
-			break;
-		case StorageImageClass::Count: break;
-	}
-	return IR::DescriptorBindingKind::Count;
-}
-
-constexpr uint32_t ImageSpirvDimension(ImageViewKind view) {
-	switch (view) {
-		case ImageViewKind::Dim1D:
-		case ImageViewKind::Dim1DArray: return Dim1D;
-		case ImageViewKind::Dim2D:
-		case ImageViewKind::Dim2DArray:
-		case ImageViewKind::Dim2DMsaa:
-		case ImageViewKind::Dim2DMsaaArray:
-		case ImageViewKind::Count: return Dim2D;
-		case ImageViewKind::Dim3D: return Dim3D;
-	}
-	return Dim2D;
-}
-
-constexpr uint32_t ImageSpirvArrayed(ImageViewKind view) {
-	return view == ImageViewKind::Dim1DArray || view == ImageViewKind::Dim2DArray ||
-	               view == ImageViewKind::Dim2DMsaaArray
-	           ? 1u
-	           : 0u;
-}
-
-constexpr uint32_t ImageSpirvMultisampled(ImageViewKind view) {
-	return view == ImageViewKind::Dim2DMsaa || view == ImageViewKind::Dim2DMsaaArray ? 1u : 0u;
-}
-
 struct F32Class {
 	uint32_t bits = 0;
 	uint32_t nan  = 0;
 	uint32_t zero = 0;
 };
 
-uint32_t PixelParameterMappedLocation(const EmitterState& state, uint32_t attr);
-
 uint32_t PixelParameterLocation(const EmitterState& state, uint32_t attr);
 
 bool PixelParameterIsFlat(const EmitterState& state, uint32_t attr);
@@ -600,15 +521,10 @@ bool PixelParameterIsCustom(const EmitterState& state, uint32_t attr);
 
 VertexInputScalarKind VertexParameterScalarKind(const EmitterState& state, uint32_t location);
 
-uint32_t VertexParameterComponentCount(const EmitterState& state, const InputBinding& input);
+uint32_t VertexParameterComponentCount(const InputBinding& input);
 
 uint32_t VertexParameterScalarType(EmitterState& state, VertexInputScalarKind kind);
 
-uint32_t VertexParameterScalarPointerType(EmitterState& state, VertexInputScalarKind kind);
-
-uint32_t VertexParameterVectorOrScalarType(EmitterState& state, VertexInputScalarKind kind,
-                                           uint32_t components);
-
 uint32_t VertexParameterInputPointerType(EmitterState& state, VertexInputScalarKind kind,
                                          uint32_t components);
 
@@ -626,48 +542,33 @@ uint32_t EmitSubgroupLocalInvocationId(EmitterState& state);
                                                IR::DescriptorBindingKind kind, uint32_t resource,
                                                const char* reason);
 
-DescriptorResourceBinding ResourceForDescriptor(const EmitterState&       state,
-                                                IR::DescriptorBindingKind kind, uint32_t resource);
+uint32_t ResourceForDescriptor(const EmitterState& state, IR::DescriptorBindingKind kind,
+                               uint32_t resource);
 
 uint32_t DescriptorElementPointer(EmitterState& state, uint32_t result_ptr_type,
                                   uint32_t variable_id, uint32_t array_index,
                                   IR::DescriptorBindingKind kind, uint32_t resource,
                                   const char* variable_name);
 
-ImageViewKind SampledImageViewKind(const EmitterState& state, const IR::MemoryInfo& mem,
-                                   uint32_t use_pc);
+uint32_t ImageScalarType(EmitterState& state, Prospero::TextureNumericClass numeric_class);
 
-uint32_t ImageViewCoordinateComponents(ImageViewKind view);
+uint32_t ImageVectorType(EmitterState& state, Prospero::TextureNumericClass numeric_class,
+                         uint32_t components);
 
-uint32_t ImageViewSpatialComponents(ImageViewKind view);
+uint32_t ImageType(EmitterState& state, const IR::ImageResource& image);
 
-uint32_t ImageViewImageType(EmitterState& state, ImageViewKind view, bool integer);
+uint32_t ImageViewSizeType(EmitterState& state, ImageDimension dimension);
 
-uint32_t ImageViewSampledImageType(EmitterState& state, ImageViewKind view, bool integer);
+uint32_t LoadSampledImageDescriptor(EmitterState& state, uint32_t resource);
 
-uint32_t ImageViewSizeType(EmitterState& state, ImageViewKind view);
+uint32_t LoadSamplerDescriptor(EmitterState& state, uint32_t sampler);
 
-uint32_t StorageImageType(EmitterState& state, StorageImageClass image_class, ImageViewKind view);
+uint32_t MakeSampledImage(EmitterState& state, uint32_t resource, uint32_t sampler);
 
-uint32_t StorageImagePointerType(EmitterState& state, StorageImageClass image_class,
-                                 ImageViewKind view);
+uint32_t StorageImageDescriptorPointer(EmitterState& state, uint32_t resource);
 
-uint32_t LoadSampledImageDescriptor(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
-                                    ImageViewKind view);
-
-uint32_t LoadSamplerDescriptor(EmitterState& state, uint32_t sampler, uint32_t use_pc);
-
-uint32_t MakeSampledImage(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
-                          ImageViewKind view);
-uint32_t MakeSampledImage(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
-                          ImageViewKind view, uint32_t image_resource);
-
-ImageViewKind StorageImageViewKind(const IR::MemoryInfo& mem);
-
-uint32_t StorageImageDescriptorPointer(EmitterState& state, uint32_t resource, ImageViewKind view);
-
-void EmitStorageImageWrite(EmitterState& state, uint32_t resource, ImageViewKind view,
-                           uint32_t mip_lod, uint32_t coord, uint32_t texel);
+void EmitStorageImageWrite(EmitterState& state, uint32_t resource, uint32_t mip_lod, uint32_t coord,
+                           uint32_t texel);
 
 uint32_t ExecutionModelForStage(ShaderType stage);
 
@@ -705,8 +606,11 @@ void DecorateDescriptor(EmitterState& state, uint32_t variable, const char* name
 void AddDescriptorAnnotationsAndNames(EmitterState& state);
 
 void DefineModule(EmitterState& state);
-
-uint32_t EmitTrueBool(EmitterState& state);
+void     DefineMeshOutputs(EmitterState& state);
+void     EmitMeshEntryPoint(EmitterState& state);
+void     EmitMeshAllocate(ValueEmitContext& ctx, const IR::Inst& inst);
+uint32_t MeshOutputPointer(EmitterState& state, IR::StageOutputKind kind, uint32_t index = 0);
+uint32_t MeshPrimitivePointer(EmitterState& state);
 
 DppTargetLane EmitDppQuadPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control);
 
@@ -755,11 +659,17 @@ struct MemoryResourceAccess {
 	uint32_t         object_pointer   = 0;
 	uint32_t         length           = 0;
 	uint32_t         index_offset     = 0;
+	uint32_t         byte_offset      = 0;
 	bool             add_index_offset = false;
 };
 
 MemoryResourceAccess PrepareMemoryResourceAccess(EmitterState& state, const IR::MemoryInfo& mem);
 
+MemoryResourceAccess PrepareStorageBufferResourceAccess(EmitterState& state,
+                                                         const IR::MemoryInfo& mem,
+                                                         uint32_t variable,
+                                                         uint32_t pointer_type);
+
 uint32_t EmitMemoryElementIndex(EmitterState& state, const MemoryResourceAccess& access,
                                 uint32_t raw_index);
 
@@ -769,22 +679,17 @@ uint32_t EmitMemoryElementInBounds(EmitterState& state, const MemoryResourceAcce
 uint32_t EmitMemoryElementPointer(EmitterState& state, const MemoryResourceAccess& access,
                                   uint32_t index);
 
-uint32_t EmitTBufferBitcastF32ToU32(EmitterState& state, uint32_t value);
-
-uint32_t EmitTBufferBitcastU32ToF32(EmitterState& state, uint32_t value);
+uint32_t EmitStorageBufferElementPointer(EmitterState& state,
+                                         const MemoryResourceAccess& access, uint32_t index,
+                                         uint32_t pointer_type);
 
 uint32_t EmitTBufferBitcastU32ToI32(EmitterState& state, uint32_t value);
 
-uint32_t EmitTBufferCompareU32Constant(EmitterState& state, uint32_t opcode, uint32_t value,
-                                       uint32_t constant);
-
 uint32_t EmitTBufferSelectF32(EmitterState& state, uint32_t condition, uint32_t true_value,
                               uint32_t false_value);
 
 bool IsSignedFormatComponent(Format::ComponentType type);
 
-uint32_t EmitHalfToF32Bits(EmitterState& state, uint32_t raw);
-
 uint32_t EmitUFloatToF32Bits(EmitterState& state, uint32_t raw, uint32_t bits);
 
 uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferFormatInfo& info,
@@ -792,6 +697,9 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 
 void EmitDeviceAtomicMemoryBarrier(EmitterState& state);
 
+uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t source,
+                                    bool max_value);
+
 uint32_t EmitDsSwizzleTargetLane(EmitterState& state, uint32_t subid, uint32_t control);
 
 uint32_t EmitSelectValueU32(EmitterState& state, uint32_t cond, uint32_t true_value,
@@ -832,6 +740,8 @@ uint32_t EmitLogicalOrBool(EmitterState& state, uint32_t lhs, uint32_t rhs);
 
 uint32_t EmitLogicalNotBool(EmitterState& state, uint32_t value);
 
+F32Class EmitClassifyF32Bits(EmitterState& state, uint32_t bits);
+
 F32Class EmitClassifyF32(EmitterState& state, uint32_t value);
 
 uint32_t EmitClassMaskBitMatch(EmitterState& state, uint32_t mask, uint32_t bit,
@@ -861,7 +771,7 @@ bool EmitValueMemory(ValueEmitContext& ctx, const IR::Inst& inst);
 
 bool EmitValueImage(ValueEmitContext& ctx, const IR::Inst& inst);
 
-void EmitProgram(EmitterState& state, const IR::Program& program);
+void EmitProgram(EmitterState& state);
 
 void DefineGetBdaPointer(EmitterState& state);
 
@@ -907,6 +817,47 @@ uint32_t EmitValueOrZeroIfCondition(EmitterState& state, uint32_t condition, Fn&
 	                                     std::forward<Fn>(fn));
 }
 
+template <typename Fn>
+uint32_t AtomicUpdate(EmitterState& state, uint32_t pointer, IR::ResourceKind kind, Fn&& desired) {
+	const auto scope = kind == IR::ResourceKind::Lds ? ScopeWorkgroup : ScopeDevice;
+	const auto memory = [&] {
+		switch (kind) {
+			case IR::ResourceKind::Lds: return MemorySemanticsWorkgroupMemory;
+			case IR::ResourceKind::Image: return MemorySemanticsImageMemory;
+			default: return MemorySemanticsUniformMemory;
+		}
+	}();
+	const auto preheader = state.builder.AllocateId();
+	const auto header    = state.builder.AllocateId();
+	const auto cont      = state.builder.AllocateId();
+	const auto merge     = state.builder.AllocateId();
+	const auto initial   = state.builder.AllocateId();
+	const auto observed  = state.builder.AllocateId();
+	const auto exchanged = state.builder.AllocateId();
+	state.builder.AddFunction({OpBranch, preheader});
+	EmitLabel(state, preheader);
+	state.builder.AddFunction({OpAtomicLoad, TypeU32(state), initial, pointer,
+	                           ConstantU32(state, scope), ConstantU32(state, MemorySemanticsNone)});
+	state.builder.AddFunction({OpBranch, header});
+	EmitLabel(state, header);
+	state.builder.AddFunction(
+	    {OpPhi, TypeU32(state), observed, initial, preheader, exchanged, cont});
+	const auto next = desired(observed);
+	state.builder.AddFunction({OpAtomicCompareExchange, TypeU32(state), exchanged, pointer,
+	                           ConstantU32(state, scope), ConstantU32(state, MemorySemanticsNone),
+	                           ConstantU32(state, MemorySemanticsNone), next, observed});
+	const auto success = state.builder.AllocateId();
+	state.builder.AddFunction({OpIEqual, TypeBool(state), success, exchanged, observed});
+	state.builder.AddFunction({OpLoopMerge, merge, cont, LoopControlNone});
+	state.builder.AddFunction({OpBranchConditional, success, merge, cont});
+	EmitLabel(state, cont);
+	state.builder.AddFunction({OpBranch, header});
+	EmitLabel(state, merge);
+	state.builder.AddFunction({OpMemoryBarrier, ConstantU32(state, scope),
+	                           ConstantU32(state, MemorySemanticsAcquireRelease | memory)});
+	return observed;
+}
+
 } // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter
 
 #endif /* EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_SPIRVEMITTER_INTERNAL_H_ */

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

@@ -7,31 +7,15 @@
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 namespace {
 
-uint32_t Unary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t value) {
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, type, result, value});
-	return result;
-}
-
-uint32_t Binary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t lhs, uint32_t rhs) {
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, type, result, lhs, rhs});
-	return result;
-}
-
-uint32_t Select(EmitterState& state, uint32_t type, uint32_t condition, uint32_t true_value,
-                uint32_t false_value) {
-	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({OpSelect, type, result, condition, true_value, false_value});
-	return result;
-}
-
 uint32_t AndCondition(EmitterState& state, uint32_t lhs, uint32_t rhs) {
 	return Binary(state, OpLogicalAnd, TypeBool(state), lhs, rhs);
 }
 
 uint32_t EmitDsMaskedLaneRead(EmitterState& state, uint32_t source, uint32_t target,
                               uint32_t exec) {
+	if (state.lane_count == 2) {
+		target = Binary(state, OpBitwiseAnd, TypeU32(state), target, ConstantU32(state, 31));
+	}
 	const auto shuffled = state.builder.AllocateId();
 	state.builder.AddFunction({OpGroupNonUniformShuffle, TypeU32(state), shuffled,
 	                           ConstantU32(state, ScopeSubgroup), source, target});
@@ -287,12 +271,6 @@ PreparedMemoryElement PrepareMemoryElement(ValueEmitContext& ctx, const IR::Memo
 	return {.resource = resource, .index = index};
 }
 
-PreparedMemoryElement PrepareMemoryElement(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
-                                           const MemoryResourceAccess& resource,
-                                           uint32_t                    raw_index) {
-	return {.resource = resource, .index = EmitMemoryElementIndex(ctx.state, resource, raw_index)};
-}
-
 uint32_t LoadWordInBounds(ValueEmitContext& ctx, const MemoryResourceAccess& resource,
                           uint32_t index);
 
@@ -301,10 +279,10 @@ uint32_t LoadSubwordInBounds(ValueEmitContext& ctx, const MemoryResourceAccess&
 
 uint32_t LoadWordPrepared(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
                           const MemoryResourceAccess& resource) {
-	const auto access = PrepareMemoryElement(ctx, mem, resource, DwordIndex(ctx, inst, mem));
+	const auto index = EmitMemoryElementIndex(ctx.state, resource, DwordIndex(ctx, inst, mem));
 	return EmitValueOrZeroIfCondition(
-	    ctx.state, EmitMemoryElementInBounds(ctx.state, access.resource, access.index),
-	    [&]() { return LoadWordInBounds(ctx, access.resource, access.index); });
+	    ctx.state, EmitMemoryElementInBounds(ctx.state, resource, index),
+	    [&]() { return LoadWordInBounds(ctx, resource, index); });
 }
 
 uint32_t LoadWord(ValueEmitContext& ctx, const IR::Inst& inst, IR::MemoryInfo mem) {
@@ -320,11 +298,10 @@ uint32_t LoadSubwordPrepared(ValueEmitContext& ctx, const IR::Inst& inst, const
 	const auto address   = ByteAddress(ctx, inst, mem);
 	const auto raw_index = Binary(ctx.state, OpShiftRightLogical, TypeU32(ctx.state), address,
 	                              ConstantU32(ctx.state, 2));
-	const auto access    = PrepareMemoryElement(ctx, mem, resource, raw_index);
+	const auto index     = EmitMemoryElementIndex(ctx.state, resource, raw_index);
 	return EmitValueOrZeroIfCondition(
-	    ctx.state, EmitMemoryElementInBounds(ctx.state, access.resource, access.index), [&]() {
-		    return LoadSubwordInBounds(ctx, access.resource, address, access.index, bits,
-		                               sign_extend);
+	    ctx.state, EmitMemoryElementInBounds(ctx.state, resource, index), [&]() {
+		    return LoadSubwordInBounds(ctx, resource, address, index, bits, sign_extend);
 	    });
 }
 
@@ -362,15 +339,14 @@ uint32_t LoadSubword(ValueEmitContext& ctx, const IR::Inst& inst, IR::MemoryInfo
 	});
 }
 
-Prospero::BufferFormat BufferFormat(const ValueEmitContext& ctx, const IR::Inst& inst,
-                                    const IR::MemoryInfo& mem) {
+Prospero::BufferFormat BufferFormat(const ValueEmitContext& ctx, const IR::MemoryInfo& mem) {
 	return mem.typed ? Format::DecodeTBufferFormat(mem.data_format, mem.number_format)
 	                 : StorageBufferFormat(ctx.state, mem);
 }
 
-IR::MemoryInfo RebaseFormattedComponent(IR::MemoryInfo mem, Prospero::BufferFormat format,
+IR::MemoryInfo RebaseFormattedComponent(IR::MemoryInfo mem, const Format::BufferFormatInfo& info,
                                         uint32_t component) {
-	mem.offset += Format::GetFormatComponentByteOffset(format, component);
+	mem.offset += Format::GetFormatComponentByteOffset(info, component);
 	mem.data_dwords     = 1u;
 	mem.component_index = component;
 	return mem;
@@ -383,12 +359,8 @@ IR::MemoryInfo RebaseRawComponent(IR::MemoryInfo mem, uint32_t component) {
 	return mem;
 }
 
-enum class FormattedSourceKind { Memory, Zero, One };
-
-struct FormattedSource {
-	FormattedSourceKind kind      = FormattedSourceKind::Zero;
-	uint32_t            component = 0;
-};
+using Format::FormattedSource;
+using Format::FormattedSourceKind;
 
 FormattedSource ResolveFormattedSource(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
                                        const Format::BufferFormatInfo& info,
@@ -400,29 +372,22 @@ FormattedSource ResolveFormattedSource(ValueEmitContext& ctx, const IR::MemoryIn
 	}
 	const auto selector = GetDstSel(ctx.state.program.info.buffers[mem.resource].descriptor_swizzle,
 	                                output_component);
-	if (selector == 0u) return {};
-	if (selector == 1u) return {FormattedSourceKind::One, 0};
-	if (selector < 4u) {
+	const auto source = Format::ResolveFormattedSource(info, selector);
+	if (source.kind == FormattedSourceKind::Invalid) {
 		ExitDescriptorBindingFailure(ctx.state, IR::DescriptorBindingKind::Buffers, mem.resource,
 		                             "buffer descriptor has reserved dst_sel");
 	}
-	const auto component = selector - 4u;
-	return component < info.component_count
-	           ? FormattedSource {FormattedSourceKind::Memory, component}
-	           : FormattedSource {};
+	return source;
 }
 
 uint32_t FormattedConstant(ValueEmitContext& ctx, const Format::BufferFormatInfo& info,
                            FormattedSourceKind kind) {
-	if (kind != FormattedSourceKind::One) return ConstantU32(ctx.state, 0);
-	const auto integer =
-	    info.type == Format::ComponentType::Uint || info.type == Format::ComponentType::Sint;
-	return ConstantU32(ctx.state, integer ? 1u : 0x3f800000u);
+	return ConstantU32(ctx.state, Format::FormattedConstantBits(info, kind));
 }
 
 template <typename LoadWordFn, typename LoadSubwordFn>
 uint32_t LoadFormattedComponent(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
-                                Prospero::BufferFormat format, const Format::BufferFormatInfo& info,
+                                const Format::BufferFormatInfo& info,
                                 uint32_t output_component, LoadWordFn&& load_word,
                                 LoadSubwordFn&& load_subword) {
 	const auto source = ResolveFormattedSource(ctx, mem, info, output_component);
@@ -457,19 +422,18 @@ uint32_t LoadFormattedComponent(ValueEmitContext& ctx, const IR::MemoryInfo& mem
 uint32_t FormattedLoadPrepared(ValueEmitContext& ctx, const IR::Inst& inst,
                                const IR::MemoryInfo& mem, uint32_t output_component,
                                const MemoryResourceAccess& resource) {
-	const auto format = BufferFormat(ctx, inst, mem);
-	if (!Format::IsKnownFormat(format)) {
+	const auto info = Format::GetFormatInfo(BufferFormat(ctx, mem));
+	if (info.type == Format::ComponentType::Unknown) {
 		return LoadWordPrepared(ctx, inst, RebaseRawComponent(mem, output_component), resource);
 	}
-	const auto info = Format::GetFormatInfo(format);
 	return LoadFormattedComponent(
-	    ctx, mem, format, info, output_component,
+	    ctx, mem, info, output_component,
 	    [&](uint32_t component) {
-		    return LoadWordPrepared(ctx, inst, RebaseFormattedComponent(mem, format, component),
+		    return LoadWordPrepared(ctx, inst, RebaseFormattedComponent(mem, info, component),
 		                            resource);
 	    },
 	    [&](uint32_t component, uint32_t bits, bool sign_extend) {
-		    return LoadSubwordPrepared(ctx, inst, RebaseFormattedComponent(mem, format, component),
+		    return LoadSubwordPrepared(ctx, inst, RebaseFormattedComponent(mem, info, component),
 		                               resource, bits, sign_extend);
 	    });
 }
@@ -481,41 +445,6 @@ uint32_t FormattedLoad(ValueEmitContext& ctx, const IR::Inst& inst, const IR::Me
 	});
 }
 
-template <typename Fn>
-uint32_t AtomicUpdate(EmitterState& state, uint32_t pointer, IR::ResourceKind kind, Fn&& desired) {
-	const auto scope     = kind == IR::ResourceKind::Lds ? ScopeWorkgroup : ScopeDevice;
-	const auto memory    = kind == IR::ResourceKind::Lds ? MemorySemanticsWorkgroupMemory
-	                                                     : MemorySemanticsUniformMemory;
-	const auto preheader = state.builder.AllocateId();
-	const auto header    = state.builder.AllocateId();
-	const auto cont      = state.builder.AllocateId();
-	const auto merge     = state.builder.AllocateId();
-	const auto initial   = state.builder.AllocateId();
-	const auto observed  = state.builder.AllocateId();
-	const auto exchanged = state.builder.AllocateId();
-	state.builder.AddFunction({OpBranch, preheader});
-	EmitLabel(state, preheader);
-	state.builder.AddFunction({OpAtomicLoad, TypeU32(state), initial, pointer,
-	                           ConstantU32(state, scope), ConstantU32(state, MemorySemanticsNone)});
-	state.builder.AddFunction({OpBranch, header});
-	EmitLabel(state, header);
-	state.builder.AddFunction(
-	    {OpPhi, TypeU32(state), observed, initial, preheader, exchanged, cont});
-	const auto next = desired(observed);
-	state.builder.AddFunction({OpAtomicCompareExchange, TypeU32(state), exchanged, pointer,
-	                           ConstantU32(state, scope), ConstantU32(state, MemorySemanticsNone),
-	                           ConstantU32(state, MemorySemanticsNone), next, observed});
-	const auto success = Binary(state, OpIEqual, TypeBool(state), exchanged, observed);
-	state.builder.AddFunction({OpLoopMerge, merge, cont, LoopControlNone});
-	state.builder.AddFunction({OpBranchConditional, success, merge, cont});
-	EmitLabel(state, cont);
-	state.builder.AddFunction({OpBranch, header});
-	EmitLabel(state, merge);
-	state.builder.AddFunction({OpMemoryBarrier, ConstantU32(state, scope),
-	                           ConstantU32(state, MemorySemanticsAcquireRelease | memory)});
-	return observed;
-}
-
 void StoreSubwordInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
                           const MemoryResourceAccess& resource, uint32_t address, uint32_t index,
                           uint32_t bits, uint32_t data) {
@@ -550,10 +479,10 @@ void StoreSubwordPrepared(ValueEmitContext& ctx, const IR::Inst& inst, const IR:
 	const auto address   = ByteAddress(ctx, inst, mem);
 	const auto raw_index = Binary(ctx.state, OpShiftRightLogical, TypeU32(ctx.state), address,
 	                              ConstantU32(ctx.state, 2));
-	const auto access    = PrepareMemoryElement(ctx, mem, resource, raw_index);
+	const auto index     = EmitMemoryElementIndex(ctx.state, resource, raw_index);
 	EmitIfCondition(
-	    ctx.state, EmitMemoryElementInBounds(ctx.state, access.resource, access.index), [&]() {
-		    StoreSubwordInBounds(ctx, mem, access.resource, address, access.index, bits, data);
+	    ctx.state, EmitMemoryElementInBounds(ctx.state, resource, index), [&]() {
+		    StoreSubwordInBounds(ctx, mem, resource, address, index, bits, data);
 	    });
 }
 
@@ -566,12 +495,11 @@ void StoreSubword(ValueEmitContext& ctx, const IR::Inst& inst, IR::MemoryInfo me
 
 void StoreWordPrepared(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
                        const MemoryResourceAccess& resource, uint32_t data) {
-	const auto access = PrepareMemoryElement(ctx, mem, resource, DwordIndex(ctx, inst, mem));
+	const auto index = EmitMemoryElementIndex(ctx.state, resource, DwordIndex(ctx, inst, mem));
 	EmitIfCondition(
-	    ctx.state, EmitMemoryElementInBounds(ctx.state, access.resource, access.index), [&]() {
+	    ctx.state, EmitMemoryElementInBounds(ctx.state, resource, index), [&]() {
 		    ctx.state.builder.AddFunction(
-		        {OpStore, EmitMemoryElementPointer(ctx.state, access.resource, access.index),
-		         data});
+		        {OpStore, EmitMemoryElementPointer(ctx.state, resource, index), data});
 	    });
 }
 
@@ -591,15 +519,14 @@ void StoreWord(ValueEmitContext& ctx, const IR::Inst& inst, IR::MemoryInfo mem)
 void FormattedStorePrepared(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
                             uint32_t component, const MemoryResourceAccess& resource,
                             uint32_t data) {
-	const auto format = BufferFormat(ctx, inst, mem);
-	if (!Format::IsKnownFormat(format)) {
+	const auto info = Format::GetFormatInfo(BufferFormat(ctx, mem));
+	if (info.type == Format::ComponentType::Unknown) {
 		StoreWordPrepared(ctx, inst, RebaseRawComponent(mem, component), resource, data);
 		return;
 	}
-	const auto info = Format::GetFormatInfo(format);
 	if (component >= info.component_count) return;
 	const auto bits          = info.component_bits[component];
-	const auto component_mem = RebaseFormattedComponent(mem, format, component);
+	const auto component_mem = RebaseFormattedComponent(mem, info, component);
 	if (bits == 8u || bits == 16u) {
 		StoreSubwordPrepared(ctx, inst, component_mem, resource, bits, data);
 	} else {
@@ -614,9 +541,11 @@ void FormattedStore(ValueEmitContext& ctx, const IR::Inst& inst, const IR::Memor
 	});
 }
 
-uint32_t AtomicOpcode(IR::ValueOpcode opcode) {
+uint32_t SpirvAtomicOpcode(IR::ValueOpcode opcode) {
 	switch (opcode) {
+		case IR::ValueOpcode::BufferAtomicCmpSwap32: return OpAtomicCompareExchange;
 		case IR::ValueOpcode::BufferAtomicSwap32:
+		case IR::ValueOpcode::BufferAtomicSwap64:
 		case IR::ValueOpcode::SharedAtomicSwap32: return OpAtomicExchange;
 		case IR::ValueOpcode::BufferAtomicIAdd32:
 		case IR::ValueOpcode::SharedAtomicIAdd32: return OpAtomicIAdd;
@@ -633,6 +562,7 @@ uint32_t AtomicOpcode(IR::ValueOpcode opcode) {
 		case IR::ValueOpcode::BufferAtomicAnd32:
 		case IR::ValueOpcode::SharedAtomicAnd32: return OpAtomicAnd;
 		case IR::ValueOpcode::BufferAtomicOr32:
+		case IR::ValueOpcode::BufferAtomicOr64:
 		case IR::ValueOpcode::SharedAtomicOr32: return OpAtomicOr;
 		case IR::ValueOpcode::BufferAtomicXor32:
 		case IR::ValueOpcode::SharedAtomicXor32: return OpAtomicXor;
@@ -640,87 +570,116 @@ uint32_t AtomicOpcode(IR::ValueOpcode opcode) {
 	}
 }
 
-uint32_t EmitAtomic(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
-                    bool return_value) {
-	const auto result =
-	    EmitValueOrZeroIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
-		    const auto access = PrepareMemoryElement(ctx, mem, DwordIndex(ctx, inst, mem));
-		    return EmitValueOrZeroIfCondition(
-		        ctx.state, EmitMemoryElementInBounds(ctx.state, access.resource, access.index),
-		        [&]() {
-			        const auto value = ctx.Arg(inst, inst.NumArgs() - 2);
-			        const auto old   = ctx.state.builder.AllocateId();
-			        const auto scope =
-			            mem.kind == IR::ResourceKind::Lds ? ScopeWorkgroup : ScopeDevice;
-			        ctx.state.builder.AddFunction(
-			            {AtomicOpcode(inst.GetOpcode()), TypeU32(ctx.state), old,
-			             EmitMemoryElementPointer(ctx.state, access.resource, access.index),
-			             ConstantU32(ctx.state, scope), ConstantU32(ctx.state, MemorySemanticsNone),
-			             value});
-			        if (mem.kind == IR::ResourceKind::Lds) {
-				        const auto semantics =
-				            MemorySemanticsAcquireRelease | MemorySemanticsWorkgroupMemory;
-				        ctx.state.builder.AddFunction({OpMemoryBarrier,
-				                                       ConstantU32(ctx.state, ScopeWorkgroup),
-				                                       ConstantU32(ctx.state, semantics)});
-			        } else {
-				        EmitDeviceAtomicMemoryBarrier(ctx.state);
-			        }
-			        return old;
-		        });
-	    });
-	return return_value ? result : ConstantU32(ctx.state, 0);
+uint32_t EmitAtomicOperation(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t pointer,
+                             uint32_t scope) {
+	const auto old = ctx.state.builder.AllocateId();
+	if (inst.GetOpcode() == IR::ValueOpcode::BufferAtomicCmpSwap32) {
+		const auto desired    = ctx.Arg(inst, inst.NumArgs() - 3);
+		const auto comparator = ctx.Arg(inst, inst.NumArgs() - 2);
+		ctx.state.builder.AddFunction(
+		    {OpAtomicCompareExchange, TypeU32(ctx.state), old, pointer, ConstantU32(ctx.state, scope),
+		     ConstantU32(ctx.state, MemorySemanticsNone),
+		     ConstantU32(ctx.state, MemorySemanticsNone), desired, comparator});
+	} else {
+		const auto value = ctx.Arg(inst, inst.NumArgs() - 2);
+		ctx.state.builder.AddFunction(
+		    {SpirvAtomicOpcode(inst.GetOpcode()), TypeU32(ctx.state), old, pointer,
+		     ConstantU32(ctx.state, scope), ConstantU32(ctx.state, MemorySemanticsNone), value});
+	}
+	return old;
 }
 
-uint32_t FloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t source,
-                                bool max_value) {
-	struct OrderedBits {
-		uint32_t nan;
-		uint32_t zero;
-		uint32_t key;
-	};
-	const auto classify = [&](uint32_t bits) {
-		const auto value    = Unary(state, OpBitcast, TypeF32(state), bits);
-		const auto cls      = EmitClassifyF32(state, value);
-		const auto negative = Binary(
-		    state, OpINotEqual, TypeBool(state),
-		    Binary(state, OpBitwiseAnd, TypeU32(state), bits, ConstantU32(state, 0x80000000u)),
-		    ConstantU32(state, 0));
-		const auto negative_key = Unary(state, OpNot, TypeU32(state), bits);
-		const auto positive_key =
-		    Binary(state, OpBitwiseXor, TypeU32(state), bits, ConstantU32(state, 0x80000000u));
-		return OrderedBits {cls.nan, cls.zero,
-		                    Select(state, TypeU32(state), negative, negative_key, positive_key)};
-	};
-	const auto source_class = classify(source);
-	const auto old_class    = classify(old);
-	const auto unordered =
-	    Binary(state, OpLogicalOr, TypeBool(state),
-	           Binary(state, OpLogicalOr, TypeBool(state), source_class.nan, old_class.nan),
-	           Binary(state, OpLogicalAnd, TypeBool(state), source_class.zero, old_class.zero));
-	const auto ordered = Unary(state, OpLogicalNot, TypeBool(state), unordered);
-	const auto compare = Binary(state, max_value ? OpUGreaterThan : OpULessThan, TypeBool(state),
-	                            source_class.key, old_class.key);
-	return Select(state, TypeU32(state),
-	              Binary(state, OpLogicalAnd, TypeBool(state), ordered, compare), source, old);
-}
-
-uint32_t FloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
-                     bool max_value) {
+template <typename Fn>
+uint32_t EmitAtomicAccess(ValueEmitContext& ctx, const IR::Inst& inst,
+                          const IR::MemoryInfo& mem, Fn&& operation) {
 	return EmitValueOrZeroIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
 		const auto access = PrepareMemoryElement(ctx, mem, DwordIndex(ctx, inst, mem));
 		return EmitValueOrZeroIfCondition(
 		    ctx.state, EmitMemoryElementInBounds(ctx.state, access.resource, access.index), [&]() {
-			    const auto pointer =
-			        EmitMemoryElementPointer(ctx.state, access.resource, access.index);
-			    const auto source = ctx.Arg(inst, inst.NumArgs() - 2);
-			    return AtomicUpdate(ctx.state, pointer, mem.kind, [&](uint32_t old) {
-				    return FloatAtomicReplacement(ctx.state, old, source, max_value);
-			    });
+			    return operation(EmitMemoryElementPointer(ctx.state, access.resource, access.index));
 		    });
 	});
 }
 
+uint32_t EmitAtomic(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
+	return EmitAtomicAccess(ctx, inst, mem, [&](uint32_t pointer) {
+		const auto scope = mem.kind == IR::ResourceKind::Lds ? ScopeWorkgroup : ScopeDevice;
+		const auto old   = EmitAtomicOperation(ctx, inst, pointer, scope);
+		if (mem.kind == IR::ResourceKind::Lds) {
+			const auto semantics = MemorySemanticsAcquireRelease | MemorySemanticsWorkgroupMemory;
+			ctx.state.builder.AddFunction({OpMemoryBarrier, ConstantU32(ctx.state, scope),
+			                               ConstantU32(ctx.state, semantics)});
+		} else {
+			EmitDeviceAtomicMemoryBarrier(ctx.state);
+		}
+		return old;
+	});
+}
+
+template <typename Fn>
+uint32_t EmitAtomicUpdate(ValueEmitContext& ctx, const IR::Inst& inst,
+                          const IR::MemoryInfo& mem, Fn&& replacement) {
+	const auto value = ctx.Arg(inst, inst.NumArgs() - 2);
+	return EmitAtomicAccess(ctx, inst, mem, [&](uint32_t pointer) {
+		return AtomicUpdate(ctx.state, pointer, mem.kind, [&](uint32_t old) {
+			return replacement(ctx.state, old, value);
+		});
+	});
+}
+
+uint32_t AtomicIncrement(EmitterState& state, uint32_t old, uint32_t limit) {
+	// old >= limit ? 0 : old + 1 (unsigned).
+	const auto wrap = Binary(state, OpUGreaterThanEqual, TypeBool(state), old, limit);
+	const auto next = Binary(state, OpIAdd, TypeU32(state), old, ConstantU32(state, 1));
+	return Select(state, TypeU32(state), wrap, ConstantU32(state, 0), next);
+}
+
+uint32_t AtomicDecrement(EmitterState& state, uint32_t old, uint32_t limit) {
+	// old == 0 || old > limit ? limit : old - 1 (unsigned).
+	const auto zero  = Binary(state, OpIEqual, TypeBool(state), old, ConstantU32(state, 0));
+	const auto above = Binary(state, OpUGreaterThan, TypeBool(state), old, limit);
+	const auto wrap  = Binary(state, OpLogicalOr, TypeBool(state), zero, above);
+	const auto next  = Binary(state, OpISub, TypeU32(state), old, ConstantU32(state, 1));
+	return Select(state, TypeU32(state), wrap, limit, next);
+}
+
+uint32_t EmitBufferAtomic64(ValueEmitContext& ctx, const IR::Inst& inst,
+                            const IR::MemoryInfo& mem) {
+	auto& state = ctx.state;
+	return EmitValueOrDefaultIfCondition(
+	    state, ctx.Arg(inst, inst.NumArgs() - 1), TypeU64(state), ConstantU64(state, 0), [&]() {
+		    const auto resource = PrepareStorageBufferResourceAccess(
+		        state, mem, state.storage_buffer_u64_variable, TypeStorageBufferU64Pointer(state));
+		    const auto byte_address = Binary(state, OpIAdd, TypeU32(state),
+		                                     ByteAddress(ctx, inst, mem), resource.byte_offset);
+		    const auto index = Binary(state, OpShiftRightLogical, TypeU32(state), byte_address,
+		                              ConstantU32(state, 3u));
+		    return EmitValueOrDefaultIfCondition(
+		        state, EmitMemoryElementInBounds(state, resource, index), TypeU64(state),
+		        ConstantU64(state, 0), [&]() {
+			        const auto value = Unary(state, OpBitcast, TypeScalarU64(state),
+			                                 ctx.Arg(inst, inst.NumArgs() - 2));
+			        const auto old   = state.builder.AllocateId();
+			        state.builder.AddFunction(
+			            {SpirvAtomicOpcode(inst.GetOpcode()), TypeScalarU64(state), old,
+			             EmitStorageBufferElementPointer(
+			                 state, resource, index, TypeStorageBufferU64ElementPointer(state)),
+			             ConstantU32(state, ScopeDevice),
+			             ConstantU32(state, MemorySemanticsNone), value});
+			        EmitDeviceAtomicMemoryBarrier(state);
+			        return Unary(state, OpBitcast, TypeU64(state), old);
+		        });
+	    });
+}
+
+uint32_t FloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
+                     bool max_value) {
+	return EmitAtomicUpdate(ctx, inst, mem, [max_value](EmitterState& state, uint32_t old,
+	                                                  uint32_t value) {
+		return EmitFloatAtomicReplacement(state, old, value, max_value);
+	});
+}
+
 uint32_t SharedFloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
                            bool max_value) {
 	EmitIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
@@ -753,6 +712,9 @@ uint32_t SharedFloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst, const IR
 
 uint32_t AppendConsume(ValueEmitContext& ctx, const IR::Inst& inst, bool append) {
 	auto&      state = ctx.state;
+	if (ctx.half == 1) {
+		return ctx.other_half->Def(IR::Value(const_cast<IR::Inst*>(&inst)));
+	}
 	const auto m0    = ctx.Arg(inst, 0);
 	const auto base =
 	    Binary(state, OpShiftRightLogical, TypeU32(state), m0, ConstantU32(state, 16));
@@ -765,9 +727,7 @@ uint32_t AppendConsume(ValueEmitContext& ctx, const IR::Inst& inst, bool append)
 	const auto access = PrepareMemoryResourceAccess(state, mem);
 	const auto index  = EmitMemoryElementIndex(state, access, raw_index);
 	const auto exec   = ctx.Arg(inst, 1);
-	const auto ballot = state.builder.AllocateId();
-	state.builder.AddFunction({OpGroupNonUniformBallot, TypeU32Vector(state, 4), ballot,
-	                           ConstantU32(state, ScopeSubgroup), exec});
+	const auto ballot = ctx.Ballot(inst.Arg(1));
 	const auto low  = state.builder.AllocateId();
 	const auto high = state.builder.AllocateId();
 	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), low, ballot, 0});
@@ -775,11 +735,12 @@ uint32_t AppendConsume(ValueEmitContext& ctx, const IR::Inst& inst, bool append)
 	const auto count =
 	    Binary(state, OpIAdd, TypeU32(state), Unary(state, OpBitCount, TypeU32(state), low),
 	           Unary(state, OpBitCount, TypeU32(state), high));
-	const auto first = state.builder.AllocateId();
-	state.builder.AddFunction({OpGroupNonUniformBallotFindLSB, TypeU32(state), first,
-	                           ConstantU32(state, ScopeSubgroup), ballot});
+	const auto first       = ctx.FirstLane(ballot);
+	const auto source_lane = state.lane_count == 2 ? Binary(state, OpBitwiseAnd, TypeU32(state),
+	                                                        first, ConstantU32(state, 31))
+	                                               : first;
 	const auto is_first =
-	    Binary(state, OpIEqual, TypeBool(state), EmitSubgroupLocalInvocationId(state), first);
+	    Binary(state, OpIEqual, TypeBool(state), EmitSubgroupLocalInvocationId(state), source_lane);
 	const auto storage_bounds = EmitMemoryElementInBounds(state, access, index);
 	const auto m0_bounds =
 	    mem.kind == IR::ResourceKind::Gds
@@ -787,7 +748,12 @@ uint32_t AppendConsume(ValueEmitContext& ctx, const IR::Inst& inst, bool append)
 	        : Binary(state, OpULessThan, TypeBool(state),
 	                 ConstantU32(state, ctx.Memory(inst).offset + 3u), size);
 	const auto condition = AndCondition(
-	    state, is_first, AndCondition(state, exec, AndCondition(state, storage_bounds, m0_bounds)));
+	    state, is_first,
+	    AndCondition(state,
+	                 state.lane_count == 2
+	                     ? Binary(state, OpINotEqual, TypeBool(state), count, ConstantU32(state, 0))
+	                     : exec,
+	                 AndCondition(state, storage_bounds, m0_bounds)));
 	const auto atomic = EmitValueOrZeroIfCondition(state, condition, [&]() {
 		const auto value = state.builder.AllocateId();
 		state.builder.AddFunction(
@@ -799,12 +765,11 @@ uint32_t AppendConsume(ValueEmitContext& ctx, const IR::Inst& inst, bool append)
 	});
 	const auto result = state.builder.AllocateId();
 	state.builder.AddFunction({OpGroupNonUniformShuffle, TypeU32(state), result,
-	                           ConstantU32(state, ScopeSubgroup), atomic, first});
+	                           ConstantU32(state, ScopeSubgroup), atomic, source_lane});
 	return result;
 }
 
 struct PreparedFormattedMemory {
-	Prospero::BufferFormat   format = Prospero::BufferFormat::kInvalid;
 	Format::BufferFormatInfo info;
 	MemoryResourceAccess     resource;
 	std::array<uint32_t, 4>  addresses {};
@@ -817,10 +782,10 @@ enum class FormattedAccess { Load, Store };
 PreparedFormattedMemory PrepareFormattedMemory(ValueEmitContext& ctx, const IR::Inst& inst,
                                                const IR::MemoryInfo&       mem,
                                                const MemoryResourceAccess& resource,
+                                               const Format::BufferFormatInfo& info,
                                                uint32_t components, FormattedAccess access) {
 	PreparedFormattedMemory plan;
-	plan.format   = BufferFormat(ctx, inst, mem);
-	plan.info     = Format::GetFormatInfo(plan.format);
+	plan.info     = info;
 	plan.resource = resource;
 	std::array<bool, 4> required_components {};
 	if (access == FormattedAccess::Load) {
@@ -839,11 +804,11 @@ PreparedFormattedMemory PrepareFormattedMemory(ValueEmitContext& ctx, const IR::
 	bool first_bound = true;
 	for (uint32_t component = 0; component < plan.info.component_count; component++) {
 		if (!required_components[component]) continue;
-		const auto byte_offset = Format::GetFormatComponentByteOffset(plan.format, component);
+		const auto byte_offset = Format::GetFormatComponentByteOffset(plan.info, component);
 		bool       reused      = false;
 		for (uint32_t previous = 0; previous < component; previous++) {
 			if (required_components[previous] &&
-			    Format::GetFormatComponentByteOffset(plan.format, previous) == byte_offset) {
+			    Format::GetFormatComponentByteOffset(plan.info, previous) == byte_offset) {
 				plan.addresses[component] = plan.addresses[previous];
 				plan.indices[component]   = plan.indices[previous];
 				reused                    = true;
@@ -851,7 +816,7 @@ PreparedFormattedMemory PrepareFormattedMemory(ValueEmitContext& ctx, const IR::
 			}
 		}
 		if (reused) continue;
-		const auto component_mem  = RebaseFormattedComponent(mem, plan.format, component);
+		const auto component_mem  = RebaseFormattedComponent(mem, plan.info, component);
 		plan.addresses[component] = ByteAddress(ctx, inst, component_mem);
 		const auto raw_index      = Binary(ctx.state, OpShiftRightLogical, TypeU32(ctx.state),
 		                                   plan.addresses[component], ConstantU32(ctx.state, 2));
@@ -872,7 +837,7 @@ PreparedFormattedMemory PrepareFormattedMemory(ValueEmitContext& ctx, const IR::
 uint32_t LoadFormattedInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
                                const PreparedFormattedMemory& plan, uint32_t output_component) {
 	return LoadFormattedComponent(
-	    ctx, mem, plan.format, plan.info, output_component,
+	    ctx, mem, plan.info, output_component,
 	    [&](uint32_t component) {
 		    return LoadWordInBounds(ctx, plan.resource, plan.indices[component]);
 	    },
@@ -921,10 +886,10 @@ uint32_t LoadWideBuffer(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t co
 	    ConstantU32CompositeZero(state, components), [&]() {
 		    const auto mem      = ctx.Memory(inst);
 		    const auto resource = PrepareMemoryResourceAccess(state, mem);
-		    const auto format =
-		        mem.formatted ? BufferFormat(ctx, inst, mem) : Prospero::BufferFormat::kInvalid;
-		    if (Format::IsKnownFormat(format)) {
-			    const auto plan = PrepareFormattedMemory(ctx, inst, mem, resource, components,
+		    const auto info = Format::GetFormatInfo(
+		        mem.formatted ? BufferFormat(ctx, mem) : Prospero::BufferFormat::kInvalid);
+		    if (info.type != Format::ComponentType::Unknown) {
+			    const auto plan = PrepareFormattedMemory(ctx, inst, mem, resource, info, components,
 			                                             FormattedAccess::Load);
 			    return EmitValueOrDefaultIfCondition(
 			        state, plan.in_bounds, TypeU32Composite(state, components),
@@ -951,10 +916,10 @@ void StoreWideBuffer(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t compo
 		const auto mem       = ctx.Memory(inst);
 		const auto resource  = PrepareMemoryResourceAccess(state, mem);
 		const auto composite = ctx.Arg(inst, inst.NumArgs() - 2);
-		const auto format =
-		    mem.formatted ? BufferFormat(ctx, inst, mem) : Prospero::BufferFormat::kInvalid;
-		if (Format::IsKnownFormat(format)) {
-			const auto plan = PrepareFormattedMemory(ctx, inst, mem, resource, components,
+		const auto info = Format::GetFormatInfo(
+		    mem.formatted ? BufferFormat(ctx, mem) : Prospero::BufferFormat::kInvalid);
+		if (info.type != Format::ComponentType::Unknown) {
+			const auto plan = PrepareFormattedMemory(ctx, inst, mem, resource, info, components,
 			                                         FormattedAccess::Store);
 			EmitIfCondition(state, plan.in_bounds, [&]() {
 				for (uint32_t component = 0; component < components; component++) {
@@ -990,10 +955,10 @@ uint32_t LoadWideShared(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t co
 			                                                    ConstantU32(state, component * 4u));
 			    const auto raw_index = Binary(state, OpShiftRightLogical, TypeU32(state), address,
 			                                  ConstantU32(state, 2));
-			    const auto access    = PrepareMemoryElement(ctx, mem, resource, raw_index);
+			    const auto index     = EmitMemoryElementIndex(state, resource, raw_index);
 			    values[component]    = EmitValueOrZeroIfCondition(
-			        state, EmitMemoryElementInBounds(state, access.resource, access.index),
-			        [&]() { return LoadWordInBounds(ctx, access.resource, access.index); });
+			        state, EmitMemoryElementInBounds(state, resource, index),
+			        [&]() { return LoadWordInBounds(ctx, resource, index); });
 		    }
 		    return ConstructU32Composite(state, components, values);
 	    });
@@ -1011,11 +976,10 @@ void StoreWideShared(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t compo
 			                                              ConstantU32(state, component * 4u));
 			const auto raw_index =
 			    Binary(state, OpShiftRightLogical, TypeU32(state), address, ConstantU32(state, 2));
-			const auto access = PrepareMemoryElement(ctx, mem, resource, raw_index);
-			EmitIfCondition(state, EmitMemoryElementInBounds(state, access.resource, access.index),
+			const auto index = EmitMemoryElementIndex(state, resource, raw_index);
+			EmitIfCondition(state, EmitMemoryElementInBounds(state, resource, index),
 			                [&]() {
-				                StoreWordInBounds(ctx, access.resource, access.index,
-				                                  ctx.Arg(inst, component + 1u));
+				                StoreWordInBounds(ctx, resource, index, ctx.Arg(inst, component + 1u));
 			                });
 		}
 	});
@@ -1080,12 +1044,15 @@ bool EmitValueMemory(ValueEmitContext& ctx, const IR::Inst& inst) {
 	const auto op                = inst.GetOpcode();
 	const auto buffer_components = IR::BufferComponentCount(op);
 	if (buffer_components > 1u) {
-		if (IR::BufferAccessOf(op) == IR::BufferAccess::Read) {
+		const auto access = IR::BufferAccessOf(op);
+		if (access == IR::BufferAccess::Read) {
 			ctx.Define(inst, LoadWideBuffer(ctx, inst, buffer_components));
-		} else {
+			return true;
+		}
+		if (access == IR::BufferAccess::Write) {
 			StoreWideBuffer(ctx, inst, buffer_components);
+			return true;
 		}
-		return true;
 	}
 	const auto shared_components = IR::SharedComponentCount(op);
 	if (shared_components > 1u) {
@@ -1176,9 +1143,13 @@ bool EmitValueMemory(ValueEmitContext& ctx, const IR::Inst& inst) {
 			StoreWord(ctx, inst, mem);
 		return true;
 	}
-	const auto atomic_opcode = AtomicOpcode(op);
+	const auto atomic_opcode = SpirvAtomicOpcode(op);
+	if (atomic_opcode != 0 && inst.GetType() == IR::Type::U64) {
+		ctx.Define(inst, EmitBufferAtomic64(ctx, inst, ctx.Memory(inst)));
+		return true;
+	}
 	if (atomic_opcode != 0) {
-		ctx.Define(inst, EmitAtomic(ctx, inst, ctx.Memory(inst), true));
+		ctx.Define(inst, EmitAtomic(ctx, inst, ctx.Memory(inst)));
 		return true;
 	}
 	if (op == IR::ValueOpcode::BufferAtomicFMin32 || op == IR::ValueOpcode::BufferAtomicFMax32) {
@@ -1186,6 +1157,12 @@ bool EmitValueMemory(ValueEmitContext& ctx, const IR::Inst& inst) {
 		                             op == IR::ValueOpcode::BufferAtomicFMax32));
 		return true;
 	}
+	if (op == IR::ValueOpcode::SharedAtomicInc32 || op == IR::ValueOpcode::SharedAtomicDec32) {
+		const auto replacement = op == IR::ValueOpcode::SharedAtomicInc32
+		                             ? AtomicIncrement : AtomicDecrement;
+		ctx.Define(inst, EmitAtomicUpdate(ctx, inst, ctx.Memory(inst), replacement));
+		return true;
+	}
 	if (op == IR::ValueOpcode::SharedAtomicFMin32 || op == IR::ValueOpcode::SharedAtomicFMax32) {
 		SharedFloatAtomic(ctx, inst, ctx.Memory(inst), op == IR::ValueOpcode::SharedAtomicFMax32);
 		return true;

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

@@ -3,7 +3,8 @@
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
 uint32_t EmitShaderDataDwordLoad(EmitterState& state, uint32_t dword_index) {
-	if (state.push_constant_variable != 0) {
+	if (state.program.bindings.UsesPushData()) {
+		dword_index += state.program.bindings.push_data_start_dword;
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
 		state.builder.AddFunction({OpAccessChain, TypePushConstantElementPointer(state), pointer,
@@ -12,11 +13,11 @@ uint32_t EmitShaderDataDwordLoad(EmitterState& state, uint32_t dword_index) {
 		state.builder.AddFunction({OpLoad, TypeU32(state), value, pointer});
 		return value;
 	}
-	if (state.vsharp_storage_variable != 0) {
+	if (state.shader_data_storage_variable != 0) {
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
 		state.builder.AddFunction({OpAccessChain, TypeStorageBufferElementPointer(state), pointer,
-		                           state.vsharp_storage_variable, ConstantU32(state, 0),
+		                           state.shader_data_storage_variable, ConstantU32(state, 0),
 		                           ConstantU32(state, dword_index)});
 		state.builder.AddFunction({OpLoad, TypeU32(state), value, pointer});
 		return value;
@@ -64,14 +65,15 @@ void EmitMemoryOffsets(EmitterState& state) {
 }
 
 uint32_t LdsDwordCount(const EmitterState& state) {
-	return state.stage == ShaderType::Compute ? state.input_info.compute->lds_size_dwords : 8192u;
+	const auto* workgroup = ShaderWorkgroupInput(state.stage, state.input_info);
+	return workgroup != nullptr ? workgroup->lds_size_dwords : 8192u;
 }
 
 static void EnsureLdsStorage(EmitterState& state) {
 	if (state.lds_variable != 0) {
 		return;
 	}
-	if (state.stage != ShaderType::Compute) {
+	if (ShaderWorkgroupInput(state.stage, state.input_info) == nullptr) {
 		EXIT("function LDS was not prepared before SPIR-V function emission\n");
 	}
 	state.lds_variable = state.builder.DefineGlobalVariable(
@@ -80,6 +82,27 @@ static void EnsureLdsStorage(EmitterState& state) {
 	state.builder.AddName(state.lds_variable, "lds_dwords");
 }
 
+MemoryResourceAccess PrepareStorageBufferResourceAccess(EmitterState& state,
+                                                         const IR::MemoryInfo& mem,
+                                                         uint32_t variable,
+                                                         uint32_t pointer_type) {
+	if (variable == 0) {
+		ExitDescriptorBindingFailure(state, IR::DescriptorBindingKind::Buffers, mem.resource,
+		                             "storage buffer descriptor array was not emitted");
+	}
+	const auto array_index =
+	    ResourceForDescriptor(state, IR::DescriptorBindingKind::Buffers, mem.resource);
+	MemoryResourceAccess access {.kind = mem.kind};
+	access.object_pointer = state.builder.AllocateId();
+	state.builder.AddFunction({OpAccessChain, pointer_type, access.object_pointer, variable,
+	                           ConstantU32(state, array_index)});
+	access.byte_offset = state.memory_byte_offsets[array_index];
+	access.length      = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpArrayLength, TypeU32(state), access.length, access.object_pointer, 0});
+	return access;
+}
+
 MemoryResourceAccess PrepareMemoryResourceAccess(EmitterState& state, const IR::MemoryInfo& mem) {
 	MemoryResourceAccess access {.kind = mem.kind};
 	switch (mem.kind) {
@@ -100,10 +123,10 @@ MemoryResourceAccess PrepareMemoryResourceAccess(EmitterState& state, const IR::
 			access.length = state.gds_length;
 			return access;
 		case IR::ResourceKind::Scratch:
-			if (state.scratch_variable == 0) {
+			if (state.scratch_variable[state.lane_half] == 0) {
 				EXIT("scratch storage was not prepared before SPIR-V function emission\n");
 			}
-			access.object_pointer = state.scratch_variable;
+			access.object_pointer = state.scratch_variable[state.lane_half];
 			access.length         = ConstantU32(state, state.program.scratch_dwords);
 			return access;
 		case IR::ResourceKind::ScalarAddress:
@@ -112,22 +135,13 @@ MemoryResourceAccess PrepareMemoryResourceAccess(EmitterState& state, const IR::
 			EXIT("physical address memory must use the BDA emitter\n");
 		case IR::ResourceKind::ScalarBuffer:
 		case IR::ResourceKind::Buffer: {
-			if (state.storage_buffer_variable == 0) {
-				ExitDescriptorBindingFailure(state, IR::DescriptorBindingKind::Buffers,
-				                             mem.resource,
-				                             "storage buffer descriptor array was not emitted");
-			}
-			const auto binding =
-			    ResourceForDescriptor(state, IR::DescriptorBindingKind::Buffers, mem.resource);
-			access.object_pointer = state.builder.AllocateId();
-			state.builder.AddFunction({OpAccessChain, TypeStorageBufferPointer(state),
-			                           access.object_pointer, state.storage_buffer_variable,
-			                           ConstantU32(state, binding.array_index)});
-			access.index_offset     = EmitBinaryU32(state, OpShiftRightLogical,
-			                                        state.memory_byte_offsets[binding.array_index],
-			                                        ConstantU32(state, 2u));
+			access = PrepareStorageBufferResourceAccess(
+			    state, mem, state.storage_buffer_variable, TypeStorageBufferPointer(state));
+			access.index_offset =
+			    EmitBinaryU32(state, OpShiftRightLogical, access.byte_offset,
+			                  ConstantU32(state, 2u));
 			access.add_index_offset = true;
-			break;
+			return access;
 		}
 		default: EXIT("unsupported memory resource kind: %u\n", static_cast<unsigned>(mem.kind));
 	}
@@ -151,45 +165,35 @@ uint32_t EmitMemoryElementInBounds(EmitterState& state, const MemoryResourceAcce
 
 uint32_t EmitMemoryElementPointer(EmitterState& state, const MemoryResourceAccess& access,
                                   uint32_t index) {
-	const auto pointer = state.builder.AllocateId();
 	if (access.kind == IR::ResourceKind::Lds || access.kind == IR::ResourceKind::Scratch) {
+		const auto pointer = state.builder.AllocateId();
 		const auto storage_class = access.kind == IR::ResourceKind::Scratch ? StorageClassFunction
-		                           : state.stage == ShaderType::Compute     ? StorageClassWorkgroup
-		                                                                    : StorageClassFunction;
+		                           : ShaderWorkgroupInput(state.stage, state.input_info) != nullptr
+		                               ? StorageClassWorkgroup
+		                               : StorageClassFunction;
 		state.builder.AddFunction({OpAccessChain, TypeU32ElementPointer(state, storage_class),
 		                           pointer, access.object_pointer, index});
-	} else {
-		state.builder.AddFunction({OpAccessChain, TypeStorageBufferElementPointer(state), pointer,
-		                           access.object_pointer, ConstantU32(state, 0), index});
+		return pointer;
 	}
+	return EmitStorageBufferElementPointer(state, access, index,
+	                                       TypeStorageBufferElementPointer(state));
+}
+
+uint32_t EmitStorageBufferElementPointer(EmitterState& state,
+                                         const MemoryResourceAccess& access, uint32_t index,
+                                         uint32_t pointer_type) {
+	const auto pointer = state.builder.AllocateId();
+	state.builder.AddFunction({OpAccessChain, pointer_type, pointer, access.object_pointer,
+	                           ConstantU32(state, 0), index});
 	return pointer;
 }
 
-uint32_t EmitTBufferBitcastF32ToU32(EmitterState& state, uint32_t value) {
-	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction({OpBitcast, TypeU32(state), ret, value});
-	return ret;
-}
-
-uint32_t EmitTBufferBitcastU32ToF32(EmitterState& state, uint32_t value) {
-	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction({OpBitcast, TypeF32(state), ret, value});
-	return ret;
-}
-
 uint32_t EmitTBufferBitcastU32ToI32(EmitterState& state, uint32_t value) {
 	const auto ret = state.builder.AllocateId();
 	state.builder.AddFunction({OpBitcast, TypeI32(state), ret, value});
 	return ret;
 }
 
-uint32_t EmitTBufferCompareU32Constant(EmitterState& state, uint32_t opcode, uint32_t value,
-                                       uint32_t constant) {
-	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, TypeBool(state), ret, value, ConstantU32(state, constant)});
-	return ret;
-}
-
 uint32_t EmitTBufferSelectF32(EmitterState& state, uint32_t condition, uint32_t true_value,
                               uint32_t false_value) {
 	const auto ret = state.builder.AllocateId();
@@ -202,15 +206,6 @@ bool IsSignedFormatComponent(Format::ComponentType type) {
 	       type == Format::ComponentType::Sscaled;
 }
 
-uint32_t EmitHalfToF32Bits(EmitterState& state, uint32_t raw) {
-	const auto unpacked = state.builder.AllocateId();
-	const auto value    = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpExtInst, TypeF32Vector(state, 2), unpacked, GlslStd450(state), GlslUnpackHalf2x16, raw});
-	state.builder.AddFunction({OpCompositeExtract, TypeF32(state), value, unpacked, 0});
-	return EmitTBufferBitcastF32ToU32(state, value);
-}
-
 uint32_t EmitUFloatToF32Bits(EmitterState& state, uint32_t raw, uint32_t bits) {
 	const auto mantissa_bits = bits == 11u ? 6u : 5u;
 	const auto mantissa_mask = (1u << mantissa_bits) - 1u;
@@ -227,11 +222,11 @@ uint32_t EmitUFloatToF32Bits(EmitterState& state, uint32_t raw, uint32_t bits) {
 	const auto mantissa_bits_32 =
 	    EmitBinaryU32(state, OpShiftLeftLogical, mantissa, ConstantU32(state, 23u - mantissa_bits));
 	const auto normal_bits = EmitBinaryU32(state, OpBitwiseOr, exponent_bits, mantissa_bits_32);
-	const auto normal      = EmitTBufferBitcastU32ToF32(state, normal_bits);
+	const auto normal      = EmitBitcastU32ToF32(state, normal_bits);
 
 	const auto special_bits =
 	    EmitBinaryU32(state, OpBitwiseOr, ConstantU32(state, 0x7f800000u), mantissa_bits_32);
-	const auto special = EmitTBufferBitcastU32ToF32(state, special_bits);
+	const auto special = EmitBitcastU32ToF32(state, special_bits);
 
 	const auto mantissa_f32 = state.builder.AllocateId();
 	const auto subnormal    = state.builder.AllocateId();
@@ -240,11 +235,11 @@ uint32_t EmitUFloatToF32Bits(EmitterState& state, uint32_t raw, uint32_t bits) {
 	    {OpFMul, TypeF32(state), subnormal, mantissa_f32,
 	     ConstantF32Value(state, std::ldexp(1.0f, 1 - 15 - static_cast<int>(mantissa_bits)))});
 
-	const auto zero_exp    = EmitTBufferCompareU32Constant(state, OpIEqual, exponent, 0);
-	const auto special_exp = EmitTBufferCompareU32Constant(state, OpIEqual, exponent, 31);
+	const auto zero_exp    = EmitCompareU32Constant(state, OpIEqual, exponent, 0);
+	const auto special_exp = EmitCompareU32Constant(state, OpIEqual, exponent, 31);
 	const auto finite      = EmitTBufferSelectF32(state, zero_exp, subnormal, normal);
 	const auto result      = EmitTBufferSelectF32(state, special_exp, special, finite);
-	return EmitTBufferBitcastF32ToU32(state, result);
+	return EmitBitcastF32ToU32(state, result);
 }
 
 uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferFormatInfo& info,
@@ -256,13 +251,13 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 		case Format::ComponentType::Uscaled: {
 			const auto value = state.builder.AllocateId();
 			state.builder.AddFunction({OpConvertUToF, TypeF32(state), value, raw});
-			return EmitTBufferBitcastF32ToU32(state, value);
+			return EmitBitcastF32ToU32(state, value);
 		}
 		case Format::ComponentType::Sscaled: {
 			const auto signed_raw = EmitTBufferBitcastU32ToI32(state, raw);
 			const auto value      = state.builder.AllocateId();
 			state.builder.AddFunction({OpConvertSToF, TypeF32(state), value, signed_raw});
-			return EmitTBufferBitcastF32ToU32(state, value);
+			return EmitBitcastF32ToU32(state, value);
 		}
 		case Format::ComponentType::Unorm: {
 			const auto value      = state.builder.AllocateId();
@@ -271,7 +266,7 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 			state.builder.AddFunction({OpConvertUToF, TypeF32(state), value, raw});
 			state.builder.AddFunction(
 			    {OpFDiv, TypeF32(state), normalized, value, ConstantF32Value(state, max_value)});
-			return EmitTBufferBitcastF32ToU32(state, normalized);
+			return EmitBitcastF32ToU32(state, normalized);
 		}
 		case Format::ComponentType::Snorm: {
 			const auto signed_raw = EmitTBufferBitcastU32ToI32(state, raw);
@@ -284,14 +279,14 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 			    {OpFDiv, TypeF32(state), normalized, value, ConstantF32Value(state, max_value)});
 			state.builder.AddFunction({OpExtInst, TypeF32(state), clamped, GlslStd450(state),
 			                           GlslFMax, normalized, ConstantF32Value(state, -1.0f)});
-			return EmitTBufferBitcastF32ToU32(state, clamped);
+			return EmitBitcastF32ToU32(state, clamped);
 		}
 		case Format::ComponentType::Float:
 			if (bits == 32u) {
 				return raw;
 			}
 			if (bits == 16u) {
-				return EmitHalfToF32Bits(state, raw);
+				return EmitBitcastF32ToU32(state, EmitF16BitsToF32(state, raw));
 			}
 			return EmitUFloatToF32Bits(state, raw, bits);
 		default: return raw;
@@ -304,6 +299,38 @@ void EmitDeviceAtomicMemoryBarrier(EmitterState& state) {
 	    {OpMemoryBarrier, ConstantU32(state, ScopeDevice), ConstantU32(state, semantics)});
 }
 
+uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t source,
+                                    bool max_value) {
+	struct OrderedBits {
+		uint32_t nan;
+		uint32_t zero;
+		uint32_t key;
+	};
+	const auto classify = [&](uint32_t bits) {
+		const auto cls = EmitClassifyF32Bits(state, bits);
+		const auto negative = EmitCompareU32Constant(
+		    state, OpINotEqual, EmitAndConstant(state, bits, 0x80000000u), 0u);
+		const auto negative_key = state.builder.AllocateId();
+		state.builder.AddFunction({OpNot, TypeU32(state), negative_key, bits});
+		const auto positive_key = EmitBinaryU32(
+		    state, OpBitwiseXor, bits, ConstantU32(state, 0x80000000u));
+		return OrderedBits {
+		    cls.nan, cls.zero,
+		    EmitSelectValueU32(state, negative, negative_key, positive_key)};
+	};
+	const auto source_class = classify(source);
+	const auto old_class    = classify(old);
+	const auto unordered = EmitLogicalOrBool(
+	    state, EmitLogicalOrBool(state, source_class.nan, old_class.nan),
+	    EmitLogicalAndBool(state, source_class.zero, old_class.zero));
+	const auto compare = state.builder.AllocateId();
+	state.builder.AddFunction({max_value ? OpUGreaterThan : OpULessThan, TypeBool(state), compare,
+	                           source_class.key, old_class.key});
+	return EmitSelectValueU32(
+	    state, EmitLogicalAndBool(state, EmitLogicalNotBool(state, unordered), compare), source,
+	    old);
+}
+
 uint32_t EmitDsSwizzleTargetLane(EmitterState& state, uint32_t subid, uint32_t control) {
 	if ((control & 0xc000u) == 0xc000u) {
 		const uint32_t mask         = control & 0x1fu;

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

@@ -0,0 +1,197 @@
+#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
+namespace {
+
+uint32_t MeshArray(EmitterState& state, uint32_t storage, uint32_t type, uint32_t count) {
+	const auto array = state.builder.Type(OpTypeArray, {type, ConstantU32(state, count)});
+	return state.builder.DefineGlobalVariable(TypePointer(state, storage, array), storage);
+}
+
+uint32_t MeshElement(EmitterState& state, uint32_t variable, uint32_t storage, uint32_t type,
+                     uint32_t index) {
+	const auto pointer = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpAccessChain, TypePointer(state, storage, type), pointer, variable, index});
+	return pointer;
+}
+
+uint32_t MeshLoad(EmitterState& state, uint32_t variable, uint32_t storage, uint32_t type,
+                  uint32_t index) {
+	const auto pointer = MeshElement(state, variable, storage, type, index);
+	const auto value   = state.builder.AllocateId();
+	state.builder.AddFunction({OpLoad, type, value, pointer});
+	return value;
+}
+
+uint32_t MeshOutputType(EmitterState& state, IR::StageOutputKind kind) {
+	return kind == IR::StageOutputKind::Layer ? TypeU32(state) : TypeF32Vector(state, 4);
+}
+
+} // namespace
+
+void DefineMeshOutputs(EmitterState& state) {
+	const auto& mesh = state.input_info.vertex->mesh;
+	for (auto& output: state.outputs) {
+		if (output.kind != IR::StageOutputKind::Position &&
+		    output.kind != IR::StageOutputKind::Parameter &&
+		    output.kind != IR::StageOutputKind::Layer) {
+			EXIT("unsupported mesh output kind=%u\n", static_cast<uint32_t>(output.kind));
+		}
+		const auto type    = MeshOutputType(state, output.kind);
+		output.variable_id = MeshArray(
+		    state, StorageClassOutput, type,
+		    output.kind == IR::StageOutputKind::Layer ? mesh.max_primitives : mesh.max_vertices);
+		// Only Layer is read by another invocation, through the primitive's provoking vertex.
+		const bool shared = output.kind == IR::StageOutputKind::Layer;
+		output.mesh_data_variable = MeshArray(
+		    state, shared ? StorageClassWorkgroup : StorageClassPrivate, type,
+		    shared ? mesh.max_vertices : state.lane_count);
+		state.interface_variables.push_back(output.variable_id);
+		state.builder.AddName(output.variable_id, output.debug_name.c_str());
+		if (output.kind == IR::StageOutputKind::Parameter) {
+			state.builder.AddAnnotation(
+			    {OpDecorate, output.variable_id, DecorationLocation, output.location});
+		} else {
+			state.builder.AddAnnotation(
+			    {OpDecorate, output.variable_id, DecorationBuiltIn,
+			     output.kind == IR::StageOutputKind::Layer ? BuiltInLayer : BuiltInPosition});
+		}
+		if (output.kind == IR::StageOutputKind::Layer) {
+			state.builder.AddAnnotation({OpDecorate, output.variable_id, 5271u}); // PerPrimitiveEXT
+		}
+	}
+	state.mesh_allocation = MeshArray(state, StorageClassWorkgroup, TypeU32(state), 2);
+	state.mesh_primitive_data =
+	    MeshArray(state, StorageClassPrivate, TypeU32(state), state.lane_count);
+	state.mesh_primitives =
+	    MeshArray(state, StorageClassOutput, TypeU32Vector(state, 3), mesh.max_primitives);
+	state.mesh_cull = MeshArray(state, StorageClassOutput, TypeBool(state), mesh.max_primitives);
+	state.interface_variables.push_back(state.mesh_primitives);
+	state.interface_variables.push_back(state.mesh_cull);
+	state.builder.AddAnnotation({OpDecorate, state.mesh_primitives, DecorationBuiltIn,
+	                             5296u}); // PrimitiveTriangleIndicesEXT
+	state.builder.AddAnnotation(
+	    {OpDecorate, state.mesh_cull, DecorationBuiltIn, 5299u}); // CullPrimitiveEXT
+	state.builder.AddAnnotation({OpDecorate, state.mesh_cull, 5271u});
+}
+
+uint32_t MeshOutputPointer(EmitterState& state, IR::StageOutputKind kind, uint32_t index) {
+	const auto output = std::ranges::find_if(state.outputs, [=](const OutputBinding& binding) {
+		return binding.kind == kind && binding.index == index;
+	});
+	if (output == state.outputs.end()) {
+		EXIT("mesh export has no output binding: kind=%u index=%u\n", static_cast<uint32_t>(kind),
+		     index);
+	}
+	const bool shared = kind == IR::StageOutputKind::Layer;
+	return MeshElement(state, output->mesh_data_variable,
+	                   shared ? StorageClassWorkgroup : StorageClassPrivate,
+	                   MeshOutputType(state, kind),
+	                   shared ? EmitLocalInvocationIndex(state)
+	                          : ConstantU32(state, state.lane_half));
+}
+
+uint32_t MeshPrimitivePointer(EmitterState& state) {
+	return MeshElement(state, state.mesh_primitive_data, StorageClassPrivate, TypeU32(state),
+	                   ConstantU32(state, state.lane_half));
+}
+
+void EmitMeshAllocate(ValueEmitContext& ctx, const IR::Inst& inst) {
+	auto&      state = ctx.state;
+	const auto first = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpIEqual, TypeBool(state), first, EmitLocalInvocationIndex(state), ConstantU32(state, 0)});
+	EmitIfCondition(state, first, [&] {
+		const auto allocation = ctx.Arg(inst, 0);
+		for (uint32_t field = 0; field < 2; field++) {
+			const auto value = state.builder.AllocateId();
+			state.builder.AddFunction({OpBitFieldUExtract, TypeU32(state), value, allocation,
+			                           ConstantU32(state, field * 12u),
+			                           ConstantU32(state, field == 0 ? 10u : 11u)});
+			const auto pointer = MeshElement(state, state.mesh_allocation, StorageClassWorkgroup,
+			                                 TypeU32(state), ConstantU32(state, field));
+			state.builder.AddFunction({OpStore, pointer, value});
+		}
+	});
+}
+
+void EmitMeshEntryPoint(EmitterState& state) {
+	state.builder.AddFunction(
+	    {OpFunction, TypeVoid(state), state.main_func, FunctionControlNone, TypeFunction(state)});
+	EmitLabel(state, state.builder.AllocateId());
+	state.builder.AddFunction(
+	    {OpFunctionCall, TypeVoid(state), state.builder.AllocateId(), state.mesh_guest_func});
+	// All guest waves finish before the uniform Vulkan allocation and output stores.
+	state.builder.AddFunction(
+	    {OpControlBarrier, ConstantU32(state, ScopeWorkgroup), ConstantU32(state, ScopeWorkgroup),
+	     ConstantU32(state, MemorySemanticsAcquireRelease | MemorySemanticsWorkgroupMemory)});
+	const auto vertices = MeshLoad(state, state.mesh_allocation, StorageClassWorkgroup,
+	                               TypeU32(state), ConstantU32(state, 0));
+	const auto primitives = MeshLoad(state, state.mesh_allocation, StorageClassWorkgroup,
+	                                 TypeU32(state), ConstantU32(state, 1));
+	state.builder.AddFunction({5295u, vertices, primitives}); // OpSetMeshOutputsEXT
+	for (uint32_t half = 0; half < state.lane_count; half++) {
+		state.lane_half      = half;
+		const auto index     = EmitLocalInvocationIndex(state);
+		const auto is_vertex = state.builder.AllocateId();
+		state.builder.AddFunction({OpULessThan, TypeBool(state), is_vertex, index, vertices});
+		EmitIfCondition(state, is_vertex, [&] {
+			for (const auto& output: state.outputs) {
+				if (output.kind == IR::StageOutputKind::Layer) {
+					continue;
+				}
+				const auto type  = MeshOutputType(state, output.kind);
+				const auto value = MeshLoad(state, output.mesh_data_variable, StorageClassPrivate,
+				                            type, ConstantU32(state, half));
+				const auto pointer =
+				    MeshElement(state, output.variable_id, StorageClassOutput, type, index);
+				state.builder.AddFunction({OpStore, pointer, value});
+			}
+		});
+		const auto is_primitive = state.builder.AllocateId();
+		state.builder.AddFunction({OpULessThan, TypeBool(state), is_primitive, index, primitives});
+		EmitIfCondition(state, is_primitive, [&] {
+			const auto packed = MeshLoad(state, state.mesh_primitive_data, StorageClassPrivate,
+			                             TypeU32(state), ConstantU32(state, half));
+			uint32_t   vertex[3] {};
+			for (uint32_t component = 0; component < 3; component++) {
+				vertex[component] = state.builder.AllocateId();
+				state.builder.AddFunction({OpBitFieldUExtract, TypeU32(state), vertex[component],
+				                           packed, ConstantU32(state, component * 10u),
+				                           ConstantU32(state, 10)});
+			}
+			const auto triangle = state.builder.AllocateId();
+			state.builder.AddFunction({OpCompositeConstruct, TypeU32Vector(state, 3), triangle,
+			                           vertex[0], vertex[1], vertex[2]});
+			const auto triangle_pointer = MeshElement(
+			    state, state.mesh_primitives, StorageClassOutput, TypeU32Vector(state, 3), index);
+			state.builder.AddFunction({OpStore, triangle_pointer, triangle});
+			const auto null_bit =
+			    EmitBinaryU32(state, OpBitwiseAnd, packed, ConstantU32(state, 0x80000000u));
+			const auto culled = state.builder.AllocateId();
+			state.builder.AddFunction(
+			    {OpINotEqual, TypeBool(state), culled, null_bit, ConstantU32(state, 0)});
+			state.builder.AddFunction(
+			    {OpStore,
+			     MeshElement(state, state.mesh_cull, StorageClassOutput, TypeBool(state), index),
+			     culled});
+			for (const auto& output: state.outputs) {
+				if (output.kind != IR::StageOutputKind::Layer) {
+					continue;
+				}
+				const auto layer = MeshLoad(
+				    state, output.mesh_data_variable, StorageClassWorkgroup, TypeU32(state),
+				    vertex[state.input_info.vertex->mesh.provoking_vertex]);
+				const auto pointer = MeshElement(state, output.variable_id, StorageClassOutput,
+				                                 TypeU32(state), index);
+				state.builder.AddFunction({OpStore, pointer, layer});
+			}
+		});
+	}
+	state.lane_half = 0;
+	state.builder.AddFunction({OpReturn});
+	state.builder.AddFunction({OpFunctionEnd});
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter

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

@@ -29,6 +29,10 @@ uint32_t TypeU64(EmitterState& state) {
 }
 
 uint32_t TypeDeviceAddress(EmitterState& state) {
+	return TypeScalarU64(state);
+}
+
+uint32_t TypeScalarU64(EmitterState& state) {
 	return state.builder.Type(OpTypeInt, {64, 0});
 }
 
@@ -94,6 +98,25 @@ uint32_t TypeStorageBufferElementPointer(EmitterState& state) {
 	return TypePointer(state, StorageClassStorageBuffer, TypeU32(state));
 }
 
+uint32_t StorageU64RuntimeArrayType(EmitterState& state) {
+	return state.builder.DecoratedType(OpTypeRuntimeArray, {TypeScalarU64(state)},
+	                                   {{OpDecorate, {DecorationArrayStride, sizeof(uint64_t)}}});
+}
+
+uint32_t StorageBufferU64Type(EmitterState& state) {
+	return state.builder.DecoratedType(
+	    OpTypeStruct, {StorageU64RuntimeArrayType(state)},
+	    {{OpMemberDecorate, {0, DecorationOffset, 0}}, {OpDecorate, {DecorationBlock}}});
+}
+
+uint32_t TypeStorageBufferU64Pointer(EmitterState& state) {
+	return TypePointer(state, StorageClassStorageBuffer, StorageBufferU64Type(state));
+}
+
+uint32_t TypeStorageBufferU64ElementPointer(EmitterState& state) {
+	return TypePointer(state, StorageClassStorageBuffer, TypeScalarU64(state));
+}
+
 uint32_t TypeDeviceAddressStoragePointer(EmitterState& state) {
 	return TypePointer(state, StorageClassStorageBuffer, TypeDeviceAddress(state));
 }
@@ -119,8 +142,7 @@ uint32_t TypeU32ElementPointer(EmitterState& state, uint32_t storage_class) {
 namespace {
 
 uint32_t PushConstantArrayType(EmitterState& state) {
-	const auto count =
-	    ConstantU32(state, state.program.bindings.push_constant_size / sizeof(uint32_t));
+	const auto count = ConstantU32(state, IR::PushData::DwordCount);
 	return state.builder.DecoratedType(OpTypeArray, {TypeU32(state), count},
 	                                   {{OpDecorate, {DecorationArrayStride, sizeof(uint32_t)}}});
 }
@@ -128,7 +150,7 @@ uint32_t PushConstantArrayType(EmitterState& state) {
 uint32_t PushConstantBlockType(EmitterState& state) {
 	return state.builder.DecoratedType(
 	    OpTypeStruct, {PushConstantArrayType(state)},
-	    {{OpMemberDecorate, {0, DecorationOffset, state.program.bindings.push_constant_offset}},
+	    {{OpMemberDecorate, {0, DecorationOffset, 0}},
 	     {OpDecorate, {DecorationBlock}}});
 }
 
@@ -163,6 +185,13 @@ void DefineDescriptorVariables(EmitterState& state) {
 		const auto pointer_type = TypePointer(state, StorageClassStorageBuffer, array_type);
 		state.storage_buffer_variable =
 		    state.builder.DefineGlobalVariable(pointer_type, StorageClassStorageBuffer);
+		if (state.requirements.buffer_int64_atomics) {
+			const auto u64_array_type =
+			    state.builder.Type(OpTypeArray, {StorageBufferU64Type(state), count});
+			state.storage_buffer_u64_variable = state.builder.DefineGlobalVariable(
+			    TypePointer(state, StorageClassStorageBuffer, u64_array_type),
+			    StorageClassStorageBuffer);
+		}
 	}
 	if (DescriptorBinding(state, IR::DescriptorBindingKind::BdaPagetable) != nullptr) {
 		state.bda_pagetable_variable = state.builder.DefineGlobalVariable(
@@ -173,52 +202,37 @@ void DefineDescriptorVariables(EmitterState& state) {
 		state.fault_buffer_variable = state.builder.DefineGlobalVariable(
 		    TypeStorageBufferPointer(state), StorageClassStorageBuffer);
 	}
-	if (state.program.bindings.push_constant_size != 0) {
+	if (state.program.bindings.UsesPushData() || state.stage == ShaderType::Mesh) {
 		const auto pointer_type =
 		    TypePointer(state, StorageClassPushConstant, PushConstantBlockType(state));
 		state.push_constant_variable =
 		    state.builder.DefineGlobalVariable(pointer_type, StorageClassPushConstant);
 	}
-	if (DescriptorBinding(state, IR::DescriptorBindingKind::UserData) != nullptr) {
-		state.vsharp_storage_variable = state.builder.DefineGlobalVariable(
+	if (DescriptorBinding(state, IR::DescriptorBindingKind::ShaderData) != nullptr) {
+		state.shader_data_storage_variable = state.builder.DefineGlobalVariable(
 		    TypeStorageBufferPointer(state), StorageClassStorageBuffer);
 	}
 	if (DescriptorBinding(state, IR::DescriptorBindingKind::FlattenedSrt) != nullptr) {
 		state.flattened_srt_variable = state.builder.DefineGlobalVariable(
 		    TypeStorageBufferPointer(state), StorageClassStorageBuffer);
 	}
-	for (uint32_t i = 0; i < state.sampled_image_variables.size(); i++) {
-		const auto view    = static_cast<ImageViewKind>(i % SampledImageViewKindCount);
-		const bool integer = i >= SampledImageViewKindCount;
-		const auto kind    = SampledBindingKind(integer, view);
-		if (DescriptorBinding(state, kind) == nullptr) {
+	for (const auto& binding: state.program.bindings.descriptors) {
+		if (IR::ImageBindingResourceClass(binding.kind) == IR::ImageResourceClass::None) {
 			continue;
 		}
-		const auto count        = ConstantU32(state, DescriptorCount(state, kind));
-		const auto image_type   = ImageViewImageType(state, view, integer);
-		const auto array_type   = state.builder.Type(OpTypeArray, {image_type, count});
-		const auto pointer_type = TypePointer(state, StorageClassUniformConstant, array_type);
-		state.sampled_image_variables[i] =
+		const auto& image        = state.program.info.images.at(binding.resources.front());
+		const auto  count        = ConstantU32(state, DescriptorCount(state, binding.kind));
+		const auto  image_type   = ImageType(state, image);
+		const auto  array_type   = state.builder.Type(OpTypeArray, {image_type, count});
+		const auto  pointer_type = TypePointer(state, StorageClassUniformConstant, array_type);
+		state.image_variables[IR::ImageBindingIndex(binding.kind)] =
 		    state.builder.DefineGlobalVariable(pointer_type, StorageClassUniformConstant);
-		if (view == ImageViewKind::Dim1D || view == ImageViewKind::Dim1DArray) {
-			state.builder.RequireCapability(CapabilitySampled1D);
-		}
-	}
-	for (uint32_t i = 0; i < state.storage_image_variables.size(); i++) {
-		const auto view        = static_cast<ImageViewKind>(i % StorageImageViewKindCount);
-		const auto image_class = static_cast<StorageImageClass>(i / StorageImageViewKindCount);
-		const auto kind        = StorageBindingKind(image_class, view);
-		if (DescriptorBinding(state, kind) == nullptr) {
-			continue;
-		}
-		const auto count        = ConstantU32(state, DescriptorCount(state, kind));
-		const auto image_type   = StorageImageType(state, image_class, view);
-		const auto array_type   = state.builder.Type(OpTypeArray, {image_type, count});
-		const auto pointer_type = TypePointer(state, StorageClassUniformConstant, array_type);
-		state.storage_image_variables[i] =
-		    state.builder.DefineGlobalVariable(pointer_type, StorageClassUniformConstant);
-		if (view == ImageViewKind::Dim1D || view == ImageViewKind::Dim1DArray) {
-			state.builder.RequireCapability(CapabilityImage1D);
+		if (image.dimension == ImageDimension::Dim1D ||
+		    image.dimension == ImageDimension::Dim1DArray) {
+			const auto capability = image.resource_class == IR::ImageResourceClass::Sampled
+			                            ? CapabilitySampled1D
+			                            : CapabilityImage1D;
+			state.builder.RequireCapability(capability);
 		}
 	}
 	if (DescriptorBinding(state, IR::DescriptorBindingKind::Samplers) != nullptr) {
@@ -322,15 +336,8 @@ VertexInputScalarKind VertexParameterScalarKind(const EmitterState& state, uint3
 	}
 }
 
-uint32_t VertexParameterComponentCount(const EmitterState& state, const InputBinding& input) {
-	uint32_t count = input.component_count;
-	if (state.stage == ShaderType::Vertex && input.location < ShaderVertexInputInfo::RES_MAX &&
-	    input.location < static_cast<uint32_t>(state.input_info.vertex->resources_num) &&
-	    state.input_info.vertex->resources_dst[input.location].registers_num > 0) {
-		count = static_cast<uint32_t>(
-		    state.input_info.vertex->resources_dst[input.location].registers_num);
-	}
-	return std::clamp(count, 1u, 4u);
+uint32_t VertexParameterComponentCount(const InputBinding& input) {
+	return std::clamp(input.component_count, 1u, 4u);
 }
 
 uint32_t VertexParameterScalarType(EmitterState& state, VertexInputScalarKind kind) {
@@ -342,45 +349,6 @@ uint32_t VertexParameterScalarType(EmitterState& state, VertexInputScalarKind ki
 	}
 }
 
-uint32_t VertexParameterScalarPointerType(EmitterState& state, VertexInputScalarKind kind) {
-	switch (kind) {
-		case VertexInputScalarKind::Sint:
-			return TypePointer(state, StorageClassInput, TypeI32(state));
-		case VertexInputScalarKind::Uint:
-			return TypePointer(state, StorageClassInput, TypeU32(state));
-		case VertexInputScalarKind::Float:
-		default: return TypePointer(state, StorageClassInput, TypeF32(state));
-	}
-}
-
-uint32_t VertexParameterVectorOrScalarType(EmitterState& state, VertexInputScalarKind kind,
-                                           uint32_t components) {
-	switch (kind) {
-		case VertexInputScalarKind::Sint:
-			switch (components) {
-				case 1: return TypeI32(state);
-				case 2: return TypeI32Vector(state, 2);
-				case 3: return TypeI32Vector(state, 3);
-				default: return TypeI32Vector(state, 4);
-			}
-		case VertexInputScalarKind::Uint:
-			switch (components) {
-				case 1: return TypeU32(state);
-				case 2: return TypeU32Vector(state, 2);
-				case 3: return TypeU32Vector(state, 3);
-				default: return TypeU32Vector(state, 4);
-			}
-		case VertexInputScalarKind::Float:
-		default:
-			switch (components) {
-				case 1: return TypeF32(state);
-				case 2: return TypeF32Vector(state, 2);
-				case 3: return TypeF32Vector(state, 3);
-				default: return TypeF32Vector(state, 4);
-			}
-	}
-}
-
 uint32_t VertexParameterInputPointerType(EmitterState& state, VertexInputScalarKind kind,
                                          uint32_t components) {
 	switch (kind) {
@@ -416,6 +384,22 @@ static bool MrtUsesUintOutput(const EmitterState& state, uint32_t index) {
 }
 
 void AllocateInputVariables(EmitterState& state) {
+	if (state.lane_count == 2) {
+		const auto add_builtin = [&](IR::StageInputKind kind, uint32_t components,
+		                             const char* name) {
+			if (std::ranges::none_of(state.inputs, [kind](const InputBinding& input) {
+				    return input.kind == kind;
+			    })) {
+				state.inputs.push_back({kind, 0, components, 0, name});
+			}
+		};
+		add_builtin(IR::StageInputKind::LocalInvocationIndex, 1, "gl_LocalInvocationIndex");
+		if (std::ranges::any_of(state.inputs, [](const InputBinding& input) {
+			    return input.kind == IR::StageInputKind::GlobalInvocationId;
+		    })) {
+			add_builtin(IR::StageInputKind::WorkgroupId, 3, "gl_WorkGroupID");
+		}
+	}
 	for (auto& binding: state.inputs) {
 		binding.variable_id = state.builder.AllocateId();
 		state.interface_variables.push_back(binding.variable_id);
@@ -440,6 +424,10 @@ static uint32_t AllocateSharedOutputVariable(EmitterState& state, uint32_t& vari
 }
 
 void AllocateOutputVariables(EmitterState& state) {
+	if (state.stage == ShaderType::Mesh) {
+		DefineMeshOutputs(state);
+		return;
+	}
 	for (auto& binding: state.outputs) {
 		switch (binding.kind) {
 			case IR::StageOutputKind::Position:
@@ -463,6 +451,10 @@ void AllocateOutputVariables(EmitterState& state) {
 			case IR::StageOutputKind::Layer:
 				binding.variable_id = AllocateSharedOutputVariable(state, state.layer_variable);
 				break;
+			case IR::StageOutputKind::ViewportIndex:
+				binding.variable_id =
+				    AllocateSharedOutputVariable(state, state.viewport_index_variable);
+				break;
 			case IR::StageOutputKind::Depth:
 				binding.variable_id = AllocateSharedOutputVariable(state, state.depth_variable);
 				break;
@@ -484,6 +476,8 @@ uint32_t BuiltInForInput(IR::StageInputKind kind) {
 		case IR::StageInputKind::InstanceIndex: return BuiltInInstanceIndex;
 		case IR::StageInputKind::FragCoord: return BuiltInFragCoord;
 		case IR::StageInputKind::FrontFacing: return BuiltInFrontFacing;
+		case IR::StageInputKind::Layer: return BuiltInLayer;
+		case IR::StageInputKind::SampleId: return BuiltInSampleId;
 		case IR::StageInputKind::BaryCoordSmooth: return BuiltInBaryCoordKHR;
 		case IR::StageInputKind::BaryCoordNoPerspective: return BuiltInBaryCoordNoPerspKHR;
 		case IR::StageInputKind::WorkgroupId: return BuiltInWorkgroupId;
@@ -507,6 +501,9 @@ void AddInputAnnotationsAndNames(EmitterState& state) {
 	}
 	for (const auto& input: state.inputs) {
 		state.builder.AddName(input.variable_id, input.debug_name.c_str());
+		if (input.kind == IR::StageInputKind::Layer || input.kind == IR::StageInputKind::SampleId) {
+			state.builder.AddAnnotation({OpDecorate, input.variable_id, DecorationFlat});
+		}
 		if (input.kind == IR::StageInputKind::Parameter) {
 			const auto flat = PixelParameterIsFlat(state, input.location);
 			if (input.per_vertex) {
@@ -534,6 +531,9 @@ void AddInputAnnotationsAndNames(EmitterState& state) {
 }
 
 void AddOutputAnnotationsAndNames(EmitterState& state) {
+	if (state.stage == ShaderType::Mesh) {
+		return;
+	}
 	if (state.per_vertex_variable != 0) {
 		state.builder.AddName(PerVertexType(state), "gl_PerVertex");
 		state.builder.AddName(state.per_vertex_variable, "outPerVertex");
@@ -548,6 +548,7 @@ void AddOutputAnnotationsAndNames(EmitterState& state) {
 	AddBuiltIn(state.clip_distance_variable, "gl_ClipDistance", BuiltInClipDistance);
 	AddBuiltIn(state.cull_distance_variable, "gl_CullDistance", BuiltInCullDistance);
 	AddBuiltIn(state.layer_variable, "gl_Layer", BuiltInLayer);
+	AddBuiltIn(state.viewport_index_variable, "gl_ViewportIndex", BuiltInViewportIndex);
 	if (state.depth_variable != 0) {
 		state.builder.AddName(state.depth_variable, "gl_FragDepth");
 		state.builder.AddAnnotation(
@@ -586,6 +587,14 @@ void AddDescriptorAnnotationsAndNames(EmitterState& state) {
 	if (state.storage_buffer_variable != 0) {
 		Decorate(state.storage_buffer_variable, "buffers", IR::DescriptorBindingKind::Buffers);
 	}
+	if (state.storage_buffer_u64_variable != 0) {
+		Decorate(state.storage_buffer_u64_variable, "buffers_u64",
+		         IR::DescriptorBindingKind::Buffers);
+		state.builder.AddAnnotation(
+		    {OpDecorate, state.storage_buffer_variable, DecorationAliased});
+		state.builder.AddAnnotation(
+		    {OpDecorate, state.storage_buffer_u64_variable, DecorationAliased});
+	}
 	if (state.bda_pagetable_variable != 0) {
 		Decorate(state.bda_pagetable_variable, "bda_pagetable",
 		         IR::DescriptorBindingKind::BdaPagetable);
@@ -594,45 +603,13 @@ void AddDescriptorAnnotationsAndNames(EmitterState& state) {
 		Decorate(state.fault_buffer_variable, "fault_buffer",
 		         IR::DescriptorBindingKind::FaultBuffer);
 	}
-	constexpr const char* SampledNames[] = {"sampled_1d",
-	                                        "sampled_1d_array",
-	                                        "sampled_2d",
-	                                        "sampled_2d_array",
-	                                        "sampled_3d",
-	                                        "sampled_2d_msaa",
-	                                        "sampled_2d_msaa_array",
-	                                        "sampled_uint_1d",
-	                                        "sampled_uint_1d_array",
-	                                        "sampled_uint_2d",
-	                                        "sampled_uint_2d_array",
-	                                        "sampled_uint_3d",
-	                                        "sampled_uint_2d_msaa",
-	                                        "sampled_uint_2d_msaa_array"};
-	for (uint32_t i = 0; i < state.sampled_image_variables.size(); i++) {
-		const auto view = static_cast<ImageViewKind>(i % SampledImageViewKindCount);
-		Decorate(state.sampled_image_variables[i], SampledNames[i],
-		         SampledBindingKind(i >= SampledImageViewKindCount, view));
-	}
-	constexpr const char* StorageNames[] = {"storage_1d",
-	                                        "storage_1d_array",
-	                                        "storage_2d",
-	                                        "storage_2d_array",
-	                                        "storage_3d",
-	                                        "storage_uint_1d",
-	                                        "storage_uint_1d_array",
-	                                        "storage_uint_2d",
-	                                        "storage_uint_2d_array",
-	                                        "storage_uint_3d",
-	                                        "storage_atomic_1d",
-	                                        "storage_atomic_1d_array",
-	                                        "storage_atomic_2d",
-	                                        "storage_atomic_2d_array",
-	                                        "storage_atomic_3d"};
-	for (uint32_t i = 0; i < state.storage_image_variables.size(); i++) {
-		const auto view        = static_cast<ImageViewKind>(i % StorageImageViewKindCount);
-		const auto image_class = static_cast<StorageImageClass>(i / StorageImageViewKindCount);
-		Decorate(state.storage_image_variables[i], StorageNames[i],
-		         StorageBindingKind(image_class, view));
+	for (const auto& binding: state.program.bindings.descriptors) {
+		if (IR::ImageBindingResourceClass(binding.kind) == IR::ImageResourceClass::None) {
+			continue;
+		}
+		const auto name = "image_" + std::to_string(static_cast<uint32_t>(binding.kind));
+		Decorate(state.image_variables[IR::ImageBindingIndex(binding.kind)], name.c_str(),
+		         binding.kind);
 	}
 	if (state.sampler_variable != 0) {
 		Decorate(state.sampler_variable, "samplers", IR::DescriptorBindingKind::Samplers);
@@ -651,9 +628,9 @@ void AddVsharpAnnotationsAndNames(EmitterState& state) {
 		state.builder.AddName(PushConstantBlockType(state), "BufferResource");
 		state.builder.AddName(state.push_constant_variable, "vsharp");
 	}
-	if (state.vsharp_storage_variable != 0) {
-		DecorateDescriptor(state, state.vsharp_storage_variable, "user_data",
-		                   IR::DescriptorBindingKind::UserData);
+	if (state.shader_data_storage_variable != 0) {
+		DecorateDescriptor(state, state.shader_data_storage_variable, "shader_data",
+		                   IR::DescriptorBindingKind::ShaderData);
 	}
 }
 
@@ -663,9 +640,21 @@ void DefineModule(EmitterState& state) {
 		state.lds_variable = state.builder.AllocateId();
 	}
 	if (state.requirements.function_scratch) {
-		state.scratch_variable = state.builder.AllocateId();
+		for (uint32_t half = 0; half < state.lane_count; half++) {
+			state.scratch_variable[half] = state.builder.AllocateId();
+		}
 	}
 	state.main_func   = state.builder.AllocateId();
+	if (state.stage == ShaderType::Mesh) {
+		state.mesh_guest_func = state.builder.AllocateId();
+		state.builder.RequireCapability(5283u); // MeshShadingEXT
+		state.builder.RequireExtension("SPV_EXT_mesh_shader");
+		state.builder.AddExecutionMode({state.main_func, 5298u}); // OutputTrianglesEXT
+		state.builder.AddExecutionMode(
+		    {state.main_func, 26u, state.input_info.vertex->mesh.max_vertices});
+		state.builder.AddExecutionMode(
+		    {state.main_func, 5270u, state.input_info.vertex->mesh.max_primitives});
+	}
 	state.entry_label = state.builder.AllocateId();
 
 	state.builder.RequireCapability(CapabilityShader);
@@ -675,24 +664,35 @@ void DefineModule(EmitterState& state) {
 		state.builder.RequireCapability(CapabilityPhysicalStorageBufferAddresses);
 		state.builder.RequireExtension("SPV_KHR_physical_storage_buffer");
 	}
+	if (state.requirements.buffer_int64_atomics) {
+		state.builder.RequireCapability(CapabilityInt64);
+		state.builder.RequireCapability(CapabilityInt64Atomics);
+	}
 	if (state.clip_distance_variable != 0) {
 		state.builder.RequireCapability(CapabilityClipDistance);
 	}
 	if (state.cull_distance_variable != 0) {
 		state.builder.RequireCapability(CapabilityCullDistance);
 	}
-	if (state.layer_variable != 0) {
-		state.builder.RequireCapability(CapabilityShaderViewportIndexLayerEXT);
-		state.builder.RequireExtension("SPV_EXT_shader_viewport_index_layer");
+	if (state.layer_variable != 0 || InputVariableForKind(state, IR::StageInputKind::Layer) != 0) {
+		state.builder.RequireVersion(0x00010500u);
+		state.builder.RequireCapability(CapabilityShaderLayer);
+	}
+	if (state.viewport_index_variable != 0) {
+		state.builder.RequireVersion(0x00010500u);
+		state.builder.RequireCapability(CapabilityShaderViewportIndex);
+	}
+	if (InputVariableForKind(state, IR::StageInputKind::SampleId) != 0) {
+		state.builder.RequireCapability(CapabilitySampleRateShading);
 	}
 	if (state.requirements.image_gather_extended) {
 		state.builder.RequireCapability(CapabilityImageGatherExtended);
 	}
-	if (state.requirements.subgroup_ballot || state.requirements.subgroup_shuffle ||
-	    state.requirements.subgroup_local_invocation_id) {
+	if (state.lane_count == 2 || state.requirements.subgroup_ballot ||
+	    state.requirements.subgroup_shuffle || state.requirements.subgroup_local_invocation_id) {
 		state.builder.RequireCapability(CapabilityGroupNonUniform);
 	}
-	if (state.requirements.subgroup_ballot) {
+	if (state.lane_count == 2 || state.requirements.subgroup_ballot) {
 		state.builder.RequireCapability(CapabilityGroupNonUniformBallot);
 	}
 	if (state.requirements.subgroup_shuffle) {
@@ -717,19 +717,24 @@ void DefineModule(EmitterState& state) {
 	    {state.program.info.uses_dma ? AddressingModelPhysicalStorageBuffer64
 	                                 : AddressingModelLogical,
 	     MemoryModelGLSL450});
-	state.builder.AddEntryPoint(ExecutionModelForStage(state.stage), state.main_func, "main",
-	                            state.interface_variables);
 	// GCN/RDNA arithmetic preserves 32-bit signed zero, infinity, and NaN. Declaring that
 	// contract prevents host compilers from treating synthesized IEEE values as finite.
 	state.builder.AddExecutionMode({state.main_func, ExecutionModeSignedZeroInfNanPreserve, 32u});
-	if (state.stage == ShaderType::Compute) {
+	if (const auto* cs = ShaderWorkgroupInput(state.stage, state.input_info)) {
 		uint32_t    local_x = state.requirements.compute_derivatives ? 2u : 1u;
 		uint32_t    local_y = state.requirements.compute_derivatives ? 2u : 1u;
 		uint32_t    local_z = 1u;
-		const auto* cs      = state.input_info.compute;
 		local_x             = cs->threads_num[0] != 0u ? cs->threads_num[0] : local_x;
 		local_y             = cs->threads_num[1] != 0u ? cs->threads_num[1] : local_y;
 		local_z             = cs->threads_num[2] != 0u ? cs->threads_num[2] : local_z;
+		if (state.lane_count == 2) {
+			local_x = ((local_x * local_y * local_z + 63u) / 64u) * 32u;
+			local_y = local_z = 1u;
+			if (state.requirements.compute_derivatives) {
+				local_y = local_x / 2u;
+				local_x = 2u;
+			}
+		}
 		state.builder.AddExecutionMode(
 		    {state.main_func, ExecutionModeLocalSize, local_x, local_y, local_z});
 	}
@@ -752,7 +757,9 @@ void DefineModule(EmitterState& state) {
 		state.builder.AddName(state.lds_variable, "lds_dwords");
 	}
 	if (state.requirements.function_scratch) {
-		state.builder.AddName(state.scratch_variable, "scratch_dwords");
+		for (uint32_t half = 0; half < state.lane_count; half++) {
+			state.builder.AddName(state.scratch_variable[half], "scratch_dwords");
+		}
 	}
 	AddInputAnnotationsAndNames(state);
 	AddOutputAnnotationsAndNames(state);
@@ -769,6 +776,8 @@ void DefineModule(EmitterState& state) {
 		switch (input.kind) {
 			case IR::StageInputKind::VertexIndex:
 			case IR::StageInputKind::InstanceIndex:
+			case IR::StageInputKind::Layer:
+			case IR::StageInputKind::SampleId:
 				ptr_type = TypePointer(state, StorageClassInput, TypeI32(state));
 				break;
 			case IR::StageInputKind::WorkgroupId:
@@ -789,7 +798,7 @@ void DefineModule(EmitterState& state) {
 			case IR::StageInputKind::Parameter:
 				if (state.stage == ShaderType::Vertex) {
 					const auto kind       = VertexParameterScalarKind(state, input.location);
-					const auto components = VertexParameterComponentCount(state, input);
+					const auto components = VertexParameterComponentCount(input);
 					ptr_type = VertexParameterInputPointerType(state, kind, components);
 				} else if (input.per_vertex) {
 					const auto array_type = state.builder.Type(
@@ -803,6 +812,9 @@ void DefineModule(EmitterState& state) {
 		}
 		state.builder.DefineGlobalVariable(input.variable_id, ptr_type, StorageClassInput);
 	}
+	if (state.stage == ShaderType::Mesh) {
+		return;
+	}
 	if (state.per_vertex_variable != 0) {
 		state.builder.DefineGlobalVariable(
 		    state.per_vertex_variable, TypePointer(state, StorageClassOutput, PerVertexType(state)),
@@ -832,6 +844,11 @@ void DefineModule(EmitterState& state) {
 		    state.layer_variable, TypePointer(state, StorageClassOutput, TypeU32(state)),
 		    StorageClassOutput);
 	}
+	if (state.viewport_index_variable != 0) {
+		state.builder.DefineGlobalVariable(
+		    state.viewport_index_variable, TypePointer(state, StorageClassOutput, TypeU32(state)),
+		    StorageClassOutput);
+	}
 	for (const auto& binding: state.outputs) {
 		if (binding.kind == IR::StageOutputKind::Parameter ||
 		    binding.kind == IR::StageOutputKind::Mrt) {