Fix PS5 front-face VGPR encoding to restore scene lighting

Clear DCC images using the requested view format.
UNORM alpha 0xffff reads as NaN in a FLOAT view.

src/common/emulatorConfig.cpp

@@ -55,6 +55,10 @@ bool FullscreenEnabled() {
 	return g_config->fullscreen_enabled;
 }
 
+bool VrEnabled() {
+	return g_config->vr_enabled;
+}
+
 uint32_t GetVblankFrequency() {
 	return std::clamp(g_config->vblank_frequency, 30u, 360u);
 }

src/common/emulatorConfig.h

@@ -46,6 +46,7 @@ struct ConfigOptions {
 	PresentMode            present_mode                = PresentMode::Mailbox;
 	int32_t                gpu_index                   = -1;
 	bool                   fullscreen_enabled          = false;
+	bool                   vr_enabled                  = false;
 	uint32_t               vblank_frequency            = 60;
 	uint32_t               console_language            = DEFAULT_CONSOLE_LANGUAGE;
 	bool                   vulkan_validation_enabled   = false;
@@ -79,6 +80,7 @@ int32_t  GetUserId();
 PresentMode GetPresentMode();
 int32_t GetGpuIndex();
 bool     FullscreenEnabled();
+bool     VrEnabled();
 uint32_t GetVblankFrequency();
 uint32_t GetConsoleLanguage();
 bool     VulkanValidationEnabled();

src/common/threads.cpp

@@ -8,7 +8,6 @@
 #include <chrono>             // IWYU pragma: keep
 #include <condition_variable> // IWYU pragma: keep
 #include <mutex>
-#include <vector>
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS && KYTY_COMPILER == KYTY_COMPILER_CLANG
 #define KYTY_WIN_CS
@@ -197,37 +196,7 @@ struct CondVarPrivate {
 #endif
 };
 
-static std::recursive_mutex                         g_cond_waiters_mutex;
-static std::vector<std::pair<int, CondVarPrivate*>> g_cond_waiters;
-static wait_poll_func_t                             g_cond_wait_poll_callback = nullptr;
-
-static void WakeCondVar(CondVarPrivate* cond_var) {
-#ifdef KYTY_WIN_CS
-	static auto func = ResolveWakeAllConditionVariable();
-	EXIT_NOT_IMPLEMENTED(func == nullptr);
-	func(&cond_var->m_cv);
-#else
-	cond_var->m_cv.notify_all();
-#endif
-}
-
-static void RegisterCondWaiter(CondVarPrivate* cond_var) {
-	std::lock_guard lock(g_cond_waiters_mutex);
-	g_cond_waiters.emplace_back(Thread::GetThreadIdUnique(), cond_var);
-}
-
-static void UnregisterCondWaiter(CondVarPrivate* cond_var) {
-	const auto      thread_id = Thread::GetThreadIdUnique();
-	std::lock_guard lock(g_cond_waiters_mutex);
-
-	const auto it = std::find_if(g_cond_waiters.begin(), g_cond_waiters.end(),
-	                             [thread_id, cond_var](const auto& waiter) {
-		                             return waiter.first == thread_id && waiter.second == cond_var;
-	                             });
-	if (it != g_cond_waiters.end()) {
-		g_cond_waiters.erase(it);
-	}
-}
+static wait_poll_func_t g_cond_wait_poll_callback = nullptr;
 
 struct ThreadPrivate {
 	ThreadPrivate(thread_func_t f, void* a): func(f), arg(a), m_thread(&Run, this) {}
@@ -365,7 +334,6 @@ CondVar::~CondVar() {
 }
 
 void CondVar::Wait(Mutex* mutex) {
-	RegisterCondWaiter(m_cond_var.get());
 #ifndef KYTY_WIN_CS
 	std::unique_lock<std::recursive_mutex> cpp_lock(mutex->m_mutex->m_mutex, std::adopt_lock_t());
 #endif
@@ -406,7 +374,6 @@ void CondVar::Wait(Mutex* mutex) {
 	}
 	cpp_lock.release();
 #endif
-	UnregisterCondWaiter(m_cond_var.get());
 }
 
 void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
@@ -415,7 +382,6 @@ void CondVar::SetWaitPollCallback(wait_poll_func_t callback) {
 
 bool CondVar::WaitFor(Mutex* mutex, uint32_t micros) {
 	bool ok = false;
-	RegisterCondWaiter(m_cond_var.get());
 #ifndef KYTY_WIN_CS
 	std::unique_lock<std::recursive_mutex> cpp_lock(mutex->m_mutex->m_mutex, std::adopt_lock_t());
 #endif
@@ -430,7 +396,6 @@ bool CondVar::WaitFor(Mutex* mutex, uint32_t micros) {
 	      std::cv_status::no_timeout);
 	cpp_lock.release();
 #endif
-	UnregisterCondWaiter(m_cond_var.get());
 	return ok;
 }
 
@@ -445,16 +410,13 @@ void CondVar::Signal() {
 }
 
 void CondVar::SignalAll() {
-	WakeCondVar(m_cond_var.get());
-}
-
-void CondVar::SignalThread(int thread_id) {
-	std::lock_guard lock(g_cond_waiters_mutex);
-	for (const auto& waiter: g_cond_waiters) {
-		if (waiter.first == thread_id) {
-			WakeCondVar(waiter.second);
-		}
-	}
+#ifdef KYTY_WIN_CS
+	static auto func = ResolveWakeAllConditionVariable();
+	EXIT_NOT_IMPLEMENTED(func == nullptr);
+	func(&m_cond_var->m_cv);
+#else
+	m_cond_var->m_cv.notify_all();
+#endif
 }
 
 int Thread::GetThreadIdUnique() {

src/common/threads.h

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

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -3635,6 +3635,11 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_RSRC4_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
+	// Toolkit state restoration emits the compiler's GS-front allocation
+	// metadata here. Native GS resource registers already carry that allocation.
+	g_hw_sh_indirect_func[0x0ca] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+		HwShIgnoreShaderRegister(cmd_offset, value);
+	};
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_GS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
@@ -3654,12 +3659,13 @@ void GraphicsInitJmpTablesShIndirect() {
 	g_hw_sh_indirect_func[Pm4::SPI_GRAPHICS_SHADER_CONTROL_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		HwShIgnoreShaderRegister(cmd_offset, value);
 	};
-	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HwShIgnoreShaderRegister(cmd_offset, value);
-	};
-	g_hw_sh_indirect_func[Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HwShIgnoreShaderRegister(cmd_offset, value);
-	};
+	for (uint32_t offset = Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS;
+	     offset <= Pm4::SPI_SHADER_USER_DATA_ADDR_HI_HS; offset++) {
+		g_hw_sh_indirect_func[offset] = [](KYTY_HW_SH_INDIRECT_ARGS) {
+			cp.GetShCtx().SetHsUserDataAddress(cmd_offset - Pm4::SPI_SHADER_USER_DATA_ADDR_LO_HS,
+			                                   value);
+		};
+	}
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_LO_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
 		auto base = cp.GetShCtx().GetVs().hs_regs.data_addr;
 		base &= 0xFFFFFF00000000FFull;
@@ -3818,6 +3824,25 @@ void GraphicsInitJmpTablesUcIndirect() {
 	for (auto& func: g_hw_uc_indirect_func) {
 		func = nullptr;
 	}
+	for (uint32_t i = 0; i < 4; ++i) {
+		g_hw_uc_indirect_func[Pm4::FSR_CONTROL_POINTS_LEFT_X + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrControlPoint(0, cmd_offset - Pm4::FSR_CONTROL_POINTS_LEFT_X, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_CONTROL_POINTS_LEFT_Y + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrControlPoint(1, cmd_offset - Pm4::FSR_CONTROL_POINTS_LEFT_Y, value);
+		};
+	}
+	for (uint32_t i = 0; i < 2; ++i) {
+		g_hw_uc_indirect_func[Pm4::FSR_ALPHA_LEFT_X + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrAlpha(0, cmd_offset - Pm4::FSR_ALPHA_LEFT_X, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_ALPHA_LEFT_Y + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrAlpha(1, cmd_offset - Pm4::FSR_ALPHA_LEFT_Y, value);
+		};
+		g_hw_uc_indirect_func[Pm4::FSR_WINDOW_LEFT + i] = [](KYTY_HW_UC_INDIRECT_ARGS) {
+			cp.GetUcfg().SetFsrWindow(cmd_offset - Pm4::FSR_WINDOW_LEFT, value);
+		};
+	}
 
 	g_hw_uc_indirect_func[Pm4::GE_CNTL] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		HW::GeControl r;

src/graphics/guest_gpu/graphicsRun.cpp

@@ -1292,6 +1292,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 								break;
 							case 0x14:
 							case 0x28:
+							case 0x2f:
 								if (event_index == 0x00) {
 									write64(false);
 									return;
@@ -1299,7 +1300,6 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 								break;
 							case 0x2b:
 							case 0x2d:
-							case 0x2f:
 							case 0x30:
 								if (event_index == 0x00 && !with_interrupt) {
 									write64(false);

src/graphics/guest_gpu/hardwareContext.h

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

src/graphics/guest_gpu/pm4.h

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

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

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

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

@@ -33,22 +33,24 @@ namespace {
 
 constexpr uint64_t NumFramesBeforeRemoval = 32;
 
-[[nodiscard]] bool DecodeDccClear(const TextureCache::ImageDesc& desc, vk::Format format,
-                                  uint32_t fill, vk::ClearColorValue& clear) {
-	const auto code = static_cast<uint8_t>(fill);
-	if (fill != static_cast<uint32_t>(code) * 0x01010101u) {
-		return false;
+[[nodiscard]] bool DecodeDccClear(const TextureCache::ImageDesc& desc, uint8_t code,
+                                  vk::ClearColorValue& clear) {
+	switch (code) {
+		case 0x00:
+		case 0x20:
+		case 0x40:
+		case 0x80:
+		case 0xc0: break;
+		default: return false;
 	}
 	const auto& metadata = desc.info.metadata;
+	const auto  format   = desc.view_info.format;
 	if (code == 0x20) {
 		// Clear-to-register is a color-buffer operation; the texture pipe cannot decode it.
 		return desc.type == TextureCache::BindingType::RenderTarget &&
 		       metadata.dcc_clear_register_valid &&
 		       DecodePackedColorClear(format, metadata.dcc_clear_word, clear);
 	}
-	if (code != 0x00 && code != 0x40 && code != 0x80 && code != 0xc0) {
-		return false;
-	}
 	clear = {};
 	if (code == 0x00) {
 		return true;
@@ -323,10 +325,8 @@ void TextureCache::DeleteImage(ImageId id) {
 	if (image->info.HasMetadata()) {
 		const auto metadata = m_surface_metas.find(image->info.metadata.range.address);
 		if (metadata != m_surface_metas.end() &&
-		    ((image->info.metadata.kind == ImageMetadataKind::Dcc &&
-		      metadata->second.type == MetaDataInfo::Type::Dcc) ||
-		     (image->info.metadata.kind == ImageMetadataKind::Htile &&
-		      metadata->second.type == MetaDataInfo::Type::HTile))) {
+		    image->info.metadata.kind == ImageMetadataKind::Htile &&
+		    metadata->second.type == MetaDataInfo::Type::HTile) {
 			// A later binding may have reused this address for another metadata type.
 			m_surface_metas.erase(metadata);
 		}
@@ -810,17 +810,10 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 	const bool safe_to_delete =
 	    current_tick - std::min(current_tick, cached.tick_accessed_last) > NumFramesBeforeRemoval;
 
-	if (requested.data.address == cached.info.data.address) {
-		const uint32_t requested_block = requested.bytes_per_block * requested.samples;
-		const uint32_t cached_block    = cached.info.bytes_per_block * cached.info.samples;
-		if (requested.BlockExtent() != cached.info.BlockExtent() ||
-		    requested_block != cached_block) {
-			if (safe_to_delete) {
-				FreeImage(cached_id);
-			}
-			return {merged_id};
-		}
-
+	const uint32_t requested_block = requested.bytes_per_block * requested.samples;
+	const uint32_t cached_block    = cached.info.bytes_per_block * cached.info.samples;
+	if (requested.data.address == cached.info.data.address &&
+	    requested.BlockExtent() == cached.info.BlockExtent() && requested_block == cached_block) {
 		if (const auto depth_id = ResolveDepthOverlap(requested, binding, cached_id)) {
 			return {depth_id};
 		}
@@ -873,38 +866,27 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		     static_cast<uint32_t>(cached.info.tile_mode));
 	}
 
-	if (requested.data.address > cached.info.data.address) {
-		const int32_t mip = requested.MipOf(cached.info);
-		if (mip >= 0) {
-			const int32_t layer = requested.SliceOf(cached.info, mip);
-			if (layer >= 0) {
-				return {cached_id, mip, layer};
-			}
-		}
-		if (safe_to_delete) {
-			FreeImage(cached_id);
-		}
-		return {};
+	const int32_t requested_mip = requested.MipOf(cached.info);
+	if (requested_mip >= 0) {
+		const int32_t layer = requested.SliceOf(cached.info, requested_mip);
+		return {cached_id, requested_mip, layer};
 	}
 
 	const int32_t mip = cached.info.MipOf(requested);
 	if (mip >= 0) {
 		const int32_t layer = cached.info.SliceOf(requested, mip);
-		if (layer >= 0) {
-			if (cached.binding.is_target) {
-				cached.binding.needs_rebind = true;
-				if (merged_id) {
-					m_slot_images[merged_id].binding.is_target = true;
-				}
-				FreeImage(cached_id);
-				return {merged_id};
-			}
-			if (merged_id) {
-				CopyImageMip(merged_id, cached_id, static_cast<uint32_t>(mip),
-				             static_cast<uint32_t>(layer));
-				FreeImage(cached_id);
-			}
+		if (!merged_id) {
+			return {ExpandImage(requested, cached_id)};
 		}
+		cached.binding.needs_rebind |= cached.binding.is_bound || cached.binding.is_target;
+		m_slot_images[merged_id].binding.is_target |= cached.binding.is_target;
+		CopyImageMip(merged_id, cached_id, static_cast<uint32_t>(mip),
+		             static_cast<uint32_t>(layer));
+		FreeImage(cached_id);
+		return {merged_id};
+	}
+	if (requested.data.address >= cached.info.data.address && safe_to_delete) {
+		FreeImage(cached_id);
 	}
 	return {merged_id};
 }
@@ -919,7 +901,14 @@ ImageId TextureCache::ExpandImage(const ImageInfo& info, ImageId source_id) {
 		source.binding.needs_rebind = true;
 	}
 	InitializeImage(expanded_id);
-	CopyImage(expanded_id, source_id);
+	const int32_t mip = source.info.MipOf(info);
+	const int32_t layer = source.info.SliceOf(info, mip);
+	if (layer >= 0) {
+		CopyImageMip(expanded_id, source_id, static_cast<uint32_t>(mip),
+		             static_cast<uint32_t>(layer));
+	} else {
+		CopyImage(expanded_id, source_id);
+	}
 	FreeImage(source_id);
 	return expanded_id;
 }
@@ -1135,53 +1124,69 @@ void TextureCache::InitializeImage(ImageId id) {
 	}
 }
 
-void TextureCache::PrepareDccClear(ImageId id, const ImageDesc& desc) {
+void TextureCache::MaterializeDccClear(ImageId id, const ImageDesc& desc,
+                                       uint32_t metadata_base_layer) {
 	if (desc.info.metadata.kind != ImageMetadataKind::Dcc) {
 		return;
 	}
-	auto& image            = m_slot_images[id];
-	image.info.metadata    = desc.info.metadata;
-	auto [entry, inserted] = m_surface_metas.try_emplace(
-	    desc.info.metadata.range.address, MetaDataInfo {.type = MetaDataInfo::Type::Dcc});
-	auto& metadata = entry->second;
-	if (!inserted && metadata.type == MetaDataInfo::Type::PendingDcc) {
-		metadata.type = MetaDataInfo::Type::Dcc;
-	} else if (metadata.type != MetaDataInfo::Type::Dcc) {
-		EXIT("TextureCache: image reuses non-DCC metadata\n");
+	const auto range = desc.info.metadata.range;
+	{
+		std::scoped_lock lock {m_lock};
+		auto& image         = m_slot_images[id];
+		image.info.metadata = desc.info.metadata;
+		// A native DCC allocation must not retain a reused HTile/CMask/FMask interpretation.
+		m_surface_metas.erase(range.address);
+		if (range.size == 0 || desc.info.resources.levels != 1 || image.info.resources.levels != 1) {
+			return;
+		}
 	}
-	if (metadata.clear_mask == 0 || image.info.resources.levels != 1 ||
-	    desc.info.metadata.range.size == 0 || metadata.fill_size < desc.info.metadata.range.size) {
-		return;
-	}
-	vk::ClearValue clear {};
-	if (!DecodeDccClear(desc, image.backing.format, metadata.fill_value, clear.color)) {
-		return;
+	const auto layers = desc.info.TransferLayers();
+	// These one-mip surfaces use complete 4 KiB DCC metadata blocks.
+	constexpr uint64_t MetadataBlockSize = 0x1000;
+	if (!range.Valid() || range.address % MetadataBlockSize != 0 || layers == 0 ||
+	    range.size % layers != 0 || (range.size / layers) % MetadataBlockSize != 0) {
+		EXIT("TextureCache: DCC slices must contain aligned 4 KiB blocks\n");
 	}
 	const auto& view           = desc.view_info;
-	const bool  volume_texture = image.info.IsVolume() && view.type == vk::ImageViewType::e3D;
-	const auto  first          = volume_texture ? 0u : view.base_layer;
-	const auto  count = volume_texture ? std::max(image.info.extent.depth >> view.base_level, 1u)
-	                                   : view.layer_count;
-	if (first >= 32 || count > 32 - first) {
-		return;
+	const bool  volume_texture = desc.info.IsVolume() && view.type == vk::ImageViewType::e3D;
+	const auto  first          = volume_texture ? 0u : metadata_base_layer;
+	const auto  image_first    = volume_texture ? 0u : view.base_layer;
+	const auto  count          = volume_texture ? desc.info.extent.depth : view.layer_count;
+	if (first >= layers || count > layers - first) {
+		EXIT("TextureCache: DCC view exceeds its native metadata slices\n");
 	}
-	// The metadata fill covers the complete allocation. Consume each layer only after its
-	// native image contents exist; already materialized layers may have been rendered since.
-	for (uint32_t layer = first; layer < first + count;) {
-		if ((metadata.clear_mask & (1u << layer)) == 0) {
-			layer++;
+	// Finish native metadata writes before reading backing bytes. This can submit the scheduler,
+	// so discovery runs before final draw uploads and never holds the texture lock across it.
+	if (m_buffer_cache.IsRegionGpuModified(range.address, range.size)) {
+		m_buffer_cache.ReadMemory(range.address, range.size, false);
+	}
+	const auto slice_size = range.size / layers;
+	for (uint32_t slice = 0; slice < count; slice++) {
+		const auto address = range.address + slice_size * (first + slice);
+		uint8_t code = 0;
+		if (!LibKernel::Memory::TryReadBacking(address, &code, sizeof(code))) {
+			EXIT("TextureCache: failed to read DCC metadata backing\n");
+		}
+		vk::ClearValue clear {};
+		if (!DecodeDccClear(desc, code, clear.color)) {
 			continue;
 		}
-		const auto start = layer;
-		uint32_t   mask  = 0;
-		do {
-			mask |= 1u << layer++;
-		} while (layer < first + count && (metadata.clear_mask & (1u << layer)) != 0);
-		ClearImage(m_scheduler.Current(), id,
-		           {vk::ImageAspectFlagBits::eColor, view.base_level, view.level_count, start,
-		            layer - start},
-		           clear);
-		metadata.clear_mask &= ~mask;
+		std::vector<uint8_t> bytes(slice_size);
+		if (!LibKernel::Memory::TryReadBacking(address, bytes.data(), bytes.size())) {
+			EXIT("TextureCache: failed to read DCC metadata slice\n");
+		}
+		if (!std::all_of(bytes.begin(), bytes.end(), [code](uint8_t byte) { return byte == code; })) {
+			continue;
+		}
+		{
+			std::scoped_lock lock {m_lock};
+			ClearImage(m_scheduler.Current(), id, view.format,
+			           {vk::ImageAspectFlagBits::eColor, view.base_level, view.level_count,
+			            image_first + slice, 1}, clear);
+		}
+		// Native expanded keys own consumption. Existing buffer tracking publishes this CPU
+		// write to future GPU readers; FillBuffer can fault and must run outside the texture lock.
+		m_buffer_cache.FillBuffer(address, slice_size, UINT32_MAX, false);
 	}
 }
 
@@ -1246,6 +1251,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		std::scoped_lock lock {m_lock};
 		return GetNullImage(desc);
 	}
+	const auto metadata_base_layer = desc.view_info.base_layer;
 
 	ImageId result {};
 	{
@@ -1313,6 +1319,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		image.tick_accessed_last = m_scheduler.CurrentTick();
 		TouchImage(image);
 	}
+	MaterializeDccClear(result, desc, metadata_base_layer);
 	return result;
 }
 
@@ -1376,7 +1383,6 @@ vk::ImageView TextureCache::FindTexture(ImageId id, const ImageDesc& desc) {
 		image.MarkGpuModified();
 	}
 	if (!image.info.data.Empty()) {
-		PrepareDccClear(id, desc);
 		RefreshImage(id);
 	}
 	switch (desc.type) {
@@ -1406,7 +1412,6 @@ vk::ImageView TextureCache::FindRenderTarget(ImageId id, const ImageDesc& desc)
 	TouchImage(image);
 	image.MarkGpuModified();
 	image.usage.render_target = true;
-	PrepareDccClear(id, desc);
 	RefreshImage(id);
 	CommitGpuWrite(image);
 	TrackImageDownload(id, image);
@@ -1430,16 +1435,9 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 	RefreshImage(id);
 	if (desc.info.HasMetadata()) {
 		image.info.metadata = desc.info.metadata;
-		auto [metadata, inserted] =
-		    m_surface_metas.try_emplace(desc.info.metadata.range.address,
-		                                MetaDataInfo {.type       = MetaDataInfo::Type::HTile,
-		                                              .clear_mask = image.info.htile_clear_mask});
-		if (!inserted && metadata->second.type != MetaDataInfo::Type::HTile) {
-			// PS5 allocations can reuse DCC storage as HTile while the old color image is cached.
-			// The depth binding defines the new type; incompatible fill state cannot carry over.
-			metadata->second = {.type       = MetaDataInfo::Type::HTile,
-			                    .clear_mask = image.info.htile_clear_mask};
-		}
+		m_surface_metas.emplace(desc.info.metadata.range.address,
+		                        MetaDataInfo {.type       = MetaDataInfo::Type::HTile,
+		                                      .clear_mask = image.info.htile_clear_mask});
 	}
 	CommitGpuWrite(image);
 	if (desc.info.HasStencil()) {
@@ -1527,12 +1525,12 @@ bool TextureCache::ClearImageFromBuffer(CommandBuffer& command, uint64_t address
 		}
 		clear.depthStencil.stencil = stencil_clear;
 	}
-	ClearImage(command, selected,
+	ClearImage(command, selected, image.backing.format,
 	           {aspect, 0, image.info.resources.levels, 0, image.info.TransferLayers()}, clear);
 	return true;
 }
 
-void TextureCache::ClearImage(CommandBuffer& command, ImageId id,
+void TextureCache::ClearImage(CommandBuffer& command, ImageId id, vk::Format format,
                               const vk::ImageSubresourceRange& range, const vk::ClearValue& clear) {
 	auto& image = m_slot_images[id];
 	const auto aspects = image.info.IsDepth() ? ImageViewOps::DepthAspectMask(image.backing.format)
@@ -1557,11 +1555,12 @@ void TextureCache::ClearImage(CommandBuffer& command, ImageId id,
 		}
 	}
 	command.EndRendering();
-	if (image.info.IsVolume() && !full_image) {
+	// Transfer clears use the backing format; aliased clears must encode through their view.
+	if (format != image.backing.format || (image.info.IsVolume() && !full_image)) {
 		EXIT_NOT_IMPLEMENTED(range.aspectMask != vk::ImageAspectFlagBits::eColor ||
 		                     range.levelCount != 1);
 		ImageViewInfo view {};
-		view.format = image.backing.format;
+		view.format = format;
 		view.type   = range.layerCount == 1 ? vk::ImageViewType::e2D : vk::ImageViewType::e2DArray;
 		view.base_level  = range.baseMipLevel;
 		view.base_layer  = range.baseArrayLayer;
@@ -1872,72 +1871,32 @@ void TextureCache::InvalidateCpuAliases(uint64_t address, uint64_t size) {
 bool TextureCache::IsMeta(uint64_t address) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	return found != m_surface_metas.end() && found->second.type != MetaDataInfo::Type::PendingDcc;
+	return found != m_surface_metas.end();
 }
 
-bool TextureCache::IsMetaCleared(uint64_t address, uint32_t slice, uint32_t* fill_value) {
+bool TextureCache::IsMetaCleared(uint64_t address, uint32_t slice) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    slice >= 32) {
+	if (found == m_surface_metas.end() || slice >= 32) {
 		return false;
 	}
-	if (fill_value != nullptr) {
-		*fill_value = found->second.fill_value;
-	}
 	return (found->second.clear_mask & (1u << slice)) != 0;
 }
 
 bool TextureCache::ClearMeta(uint64_t address) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    found->second.type == MetaDataInfo::Type::Dcc) {
-		// Preserve the broad metadata-clear operation for CMask/FMask/HTile. DCC requires a
-		// validated fill value, so an arbitrary compute write must not clear it.
+	if (found == m_surface_metas.end()) {
 		return false;
 	}
 	found->second.clear_mask = UINT32_MAX;
 	return true;
 }
 
-void TextureCache::TrackDccFill(uint64_t address, uint64_t size, uint32_t fill_value) {
-	if (!GuestRange {address, size}.Valid()) {
-		EXIT("TextureCache: invalid DCC fill range\n");
-	}
-	// DCC fills use a repeated byte code. Require all four bytes of the detected dword to agree,
-	// and mark only recognized deferred-clear encodings as logically clear.
-	const auto dcc_clear_mask = [fill_value] {
-		const auto code = static_cast<uint8_t>(fill_value);
-		if (fill_value != static_cast<uint32_t>(code) * 0x01010101u) {
-			return 0u;
-		}
-		switch (code) {
-			case 0x00:
-			case 0x20:
-			case 0x40:
-			case 0x80:
-			case 0xc0: return UINT32_MAX;
-			default: return 0u;
-		}
-	}();
-	std::scoped_lock lock {m_lock};
-	// The guest dispatch still writes metadata. An unknown address remains PendingDcc until an
-	// image descriptor confirms its role; never reinterpret CMask/FMask/HTile as DCC.
-	const auto found = m_surface_metas.try_emplace(address).first;
-	if (found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    found->second.type == MetaDataInfo::Type::Dcc) {
-		found->second.clear_mask = dcc_clear_mask;
-		found->second.fill_value = fill_value;
-		found->second.fill_size  = size;
-	}
-}
-
 bool TextureCache::TouchMeta(uint64_t address, uint32_t slice, bool is_clear) {
 	std::scoped_lock lock {m_lock};
 	const auto       found = m_surface_metas.find(address);
-	if (found == m_surface_metas.end() || found->second.type == MetaDataInfo::Type::PendingDcc ||
-	    slice >= 32) {
+	if (found == m_surface_metas.end() || slice >= 32) {
 		return false;
 	}
 	if (is_clear) {
@@ -1953,10 +1912,13 @@ void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 		EXIT("TextureCache: invalid unmap range\n");
 	}
 	std::scoped_lock lock {m_lock};
-	const auto       end = address + size;
-	for (auto metadata = m_surface_metas.lower_bound(address);
-	     metadata != m_surface_metas.end() && metadata->first < end;) {
-		metadata = m_surface_metas.erase(metadata);
+	for (auto metadata = m_surface_metas.begin(); metadata != m_surface_metas.end();) {
+		const auto base = metadata->first;
+		if (base >= address && base < address + size) {
+			metadata = m_surface_metas.erase(metadata);
+		} else {
+			++metadata;
+		}
 	}
 	auto images = FindImagesInRegion(address, size, false);
 	for (const auto id: images) {

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

@@ -64,11 +64,8 @@ public:
 	[[nodiscard]] bool IsRegionGpuModified(uint64_t address, uint64_t size);
 
 	[[nodiscard]] bool IsMeta(uint64_t address);
-	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice,
-	                                 uint32_t* fill_value = nullptr);
+	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice);
 	[[nodiscard]] bool ClearMeta(uint64_t address);
-	// Record deferred DCC state while the original guest dispatch writes the metadata.
-	void               TrackDccFill(uint64_t address, uint64_t size, uint32_t fill_value);
 	[[nodiscard]] bool TouchMeta(uint64_t address, uint32_t slice, bool is_clear);
 
 	void UnmapMemory(uint64_t address, uint64_t size);
@@ -81,17 +78,10 @@ private:
 	struct ImageDownload;
 
 	struct MetaDataInfo {
-		// A guest metadata-fill dispatch may initialize DCC before its render target is bound.
-		// PendingDcc retains that exact fill until an image binding classifies the address,
-		// without exposing an unconfirmed buffer address to the normal metadata heuristics.
-		// Keep all surface metadata in one entry so CMask/FMask can be
-		// registered beside HTile and DCC without introducing parallel tracking paths.
-		enum class Type : uint8_t { PendingDcc, CMask, FMask, HTile, Dcc };
+		enum class Type : uint8_t { CMask, FMask, HTile };
 
-		Type     type       = Type::PendingDcc;
-		uint32_t clear_mask = 0;
-		uint32_t fill_value = 0xffffffffu;
-		uint64_t fill_size  = 0;
+		Type     type;
+		uint32_t clear_mask = UINT32_MAX;
 	};
 
 	struct OverlapResult {
@@ -149,7 +139,8 @@ private:
 	                                                ImageId cached);
 	[[nodiscard]] ImageId       ExpandImage(const ImageInfo& info, ImageId source);
 	void                        RefreshImage(ImageId id);
-	void                        PrepareDccClear(ImageId id, const ImageDesc& desc);
+	void                        MaterializeDccClear(ImageId id, const ImageDesc& desc,
+	                                                uint32_t metadata_base_layer);
 	void                        InitializeImage(ImageId id);
 	[[nodiscard]] TextureTransfer
 	BuildTextureTransfer(const Image& image, BindingType binding, TransferDirection direction) const;
@@ -160,8 +151,8 @@ private:
 	void DownloadDepth(Image& image, Buffer& destination, uint64_t destination_offset);
 	void CommitGpuWrite(Image& image);
 	// Caller holds m_lock. Volume layer ranges select depth slices.
-	void ClearImage(CommandBuffer& command, ImageId id, const vk::ImageSubresourceRange& range,
-	                const vk::ClearValue& clear);
+	void ClearImage(CommandBuffer& command, ImageId id, vk::Format format,
+	                const vk::ImageSubresourceRange& range, const vk::ClearValue& clear);
 	void PrepareImageCopy(Image& image);
 	void RefreshCopySource(ImageId id);
 	[[nodiscard]] bool CopyD16(Image& destination, Image& source);

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

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

src/graphics/host_gpu/renderer/depthRenderTarget.h

@@ -34,10 +34,8 @@ struct RenderDepthInfo {
 	bool                        stencil_clear_enable     = false;
 	uint8_t                     stencil_clear_value      = 0;
 	bool                        stencil_test_enable      = false;
-	PipelineStencilStaticState  stencil_static_front;
-	PipelineStencilStaticState  stencil_static_back;
-	PipelineStencilDynamicState stencil_dynamic_front;
-	PipelineStencilDynamicState stencil_dynamic_back;
+	vk::StencilOpState          stencil_front;
+	vk::StencilOpState          stencil_back;
 	ImageId                     image_id;
 
 	[[nodiscard]] vk::ImageAspectFlags AttachmentWriteAspects() const;

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

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

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

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

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

@@ -94,9 +94,24 @@ struct ImageInfo {
 		       bytes_per_block == other.bytes_per_block;
 	}
 	[[nodiscard]] int32_t MipOf(const ImageInfo& container) const noexcept {
-		if (!IsCompatible(container) || tile_mode != container.tile_mode || resources.levels != 1 ||
-		    container.resources.layers == 0 ||
-		    container.resources.levels > container.mip_layout.size()) {
+		if (container.resources.levels > container.mip_layout.size()) {
+			return -1;
+		}
+		for (uint32_t level = 0; level < container.resources.levels; level++) {
+			if (SliceOf(container, static_cast<int32_t>(level)) >= 0) {
+				return static_cast<int32_t>(level);
+			}
+		}
+		return -1;
+	}
+	[[nodiscard]] int32_t SliceOf(const ImageInfo& container, int32_t mip) const noexcept {
+		if (!IsCompatible(container) || tile_mode != container.tile_mode || type != container.type ||
+		    IsVolume() || resources.levels != 1 || resources.layers == 0 ||
+		    container.resources.layers == 0 || mip < 0 ||
+		    static_cast<uint32_t>(mip) >= container.resources.levels ||
+		    container.resources.levels > container.mip_layout.size() ||
+		    !data.Valid() || !container.data.Valid() || data.address < container.data.address ||
+		    data.End() > container.data.End()) {
 			return -1;
 		}
 		if (HasStencil() != container.HasStencil() ||
@@ -105,75 +120,36 @@ struct ImageInfo {
 			return -1;
 		}
 
-		int32_t mip = -1;
-		for (uint32_t level = 0; level < container.resources.levels; level++) {
-			const auto& layout = container.mip_layout[level];
-			if (layout.size == 0 || layout.size % container.resources.layers != 0 ||
-			    container.data.address > UINT64_MAX - layout.offset) {
-				continue;
-			}
-			const auto mip_base   = container.data.address + layout.offset;
-			const auto slice_size = layout.size / container.resources.layers;
-			if (slice_size == 0 || mip_base > UINT64_MAX - layout.size) {
-				continue;
-			}
-			const auto mip_end = mip_base + layout.size;
-			if (data.address >= mip_base && data.address < mip_end &&
-			    (data.address - mip_base) % slice_size == 0) {
-				mip = static_cast<int32_t>(level);
-				break;
-			}
-		}
-		if (mip < 0) {
+		const auto  level  = static_cast<uint32_t>(mip);
+		const auto& layout = container.mip_layout[level];
+		if (extent.width != std::max(container.extent.width >> level, 1u) ||
+		    extent.height != std::max(container.extent.height >> level, 1u) ||
+		    extent.depth != container.extent.depth || mip_layout[0].pitch != layout.pitch ||
+		    mip_layout[0].height != layout.height || layout.size == 0 ||
+		    layout.size % container.resources.layers != 0 ||
+		    container.data.size % container.resources.layers != 0 ||
+		    data.size % resources.layers != 0) {
 			return -1;
 		}
 
-		const auto level = static_cast<uint32_t>(mip);
-		if (extent.width != std::max(container.extent.width >> level, 1u) ||
-		    extent.height != std::max(container.extent.height >> level, 1u)) {
+		// PS5 block slices contain the entire reversed mip chain.
+		const auto slice_stride = container.data.size / container.resources.layers;
+		const auto child_stride = data.size / resources.layers;
+		if (child_stride != layout.size / container.resources.layers ||
+		    layout.offset >= slice_stride || child_stride > slice_stride - layout.offset ||
+		    (resources.layers > 1 && child_stride != slice_stride)) {
 			return -1;
 		}
-		const auto mip_depth = std::max(container.extent.depth >> level, 1u);
-		if (container.type == Prospero::ImageType::kColor3D &&
-		    type == Prospero::ImageType::kColor2D) {
-			if (resources.layers != mip_depth) {
-				return -1;
-			}
-		} else if (type != container.type) {
+		const auto address_delta = data.address - container.data.address;
+		if (address_delta < layout.offset || (address_delta - layout.offset) % slice_stride != 0) {
 			return -1;
 		}
-		return mip;
-	}
-	[[nodiscard]] int32_t SliceOf(const ImageInfo& container, int32_t mip) const noexcept {
-		if (!IsCompatible(container) || type != container.type || mip < 0 ||
-		    static_cast<uint32_t>(mip) >= container.resources.levels ||
-		    container.resources.levels > container.mip_layout.size() ||
-		    container.resources.layers == 0 || data.size == 0) {
+		const auto layer = (address_delta - layout.offset) / slice_stride;
+		if (layer > INT32_MAX || layer >= container.resources.layers ||
+		    resources.layers > container.resources.layers - layer) {
 			return -1;
 		}
-		const auto level = static_cast<uint32_t>(mip);
-		if (extent.width != std::max(container.extent.width >> level, 1u) ||
-		    extent.height != std::max(container.extent.height >> level, 1u)) {
-			return -1;
-		}
-		const auto& layout = container.mip_layout[level];
-		if (layout.size == 0 || layout.size % container.resources.layers != 0 ||
-		    container.data.address > UINT64_MAX - layout.offset) {
-			return -1;
-		}
-		const auto slice_size = layout.size / container.resources.layers;
-		if (slice_size == 0 || data.size % slice_size != 0) {
-			return -1;
-		}
-		const auto mip_base = container.data.address + layout.offset;
-		if (data.address < mip_base) {
-			return -1;
-		}
-		const auto address_delta = data.address - mip_base;
-		if (address_delta % data.size != 0 || address_delta / data.size > INT32_MAX) {
-			return -1;
-		}
-		return static_cast<int32_t>(address_delta / data.size);
+		return static_cast<int32_t>(layer);
 	}
 };
 

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

@@ -260,12 +260,8 @@ vk::Pipeline TileManager::GetPipeline(uint32_t slot) {
 	const uint32_t                   values[] {1u << element_index, direction_index};
 	const vk::SpecializationMapEntry entries[] {{0, 0, 4}, {1, 4, 4}};
 	const vk::SpecializationInfo     specialization {2, entries, sizeof(values), values};
-	vk::ShaderModuleCreateInfo       module_info {};
-	module_info.codeSize    = shaders[family_index].words * sizeof(uint32_t);
-	module_info.pCode       = shaders[family_index].code;
-	vk::ShaderModule module = nullptr;
-	RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
-	                     "create TileManager shader module");
+	const auto module =
+	    CompileSPV({shaders[family_index].code, shaders[family_index].words}, m_graphics.device);
 	vk::PipelineShaderStageCreateInfo stage {};
 	stage.stage               = vk::ShaderStageFlagBits::eCompute;
 	stage.module              = module;
@@ -472,12 +468,7 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 			code  = GPU_TILER_DEMOTE_D16_SPV;
 			words = std::size(GPU_TILER_DEMOTE_D16_SPV);
 		}
-		vk::ShaderModuleCreateInfo module_info {};
-		module_info.codeSize    = words * sizeof(uint32_t);
-		module_info.pCode       = code;
-		vk::ShaderModule module = nullptr;
-		RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
-		                     "create D16 conversion shader module");
+		const auto module = CompileSPV({code, words}, m_graphics.device);
 		vk::PipelineShaderStageCreateInfo stage {};
 		stage.stage               = vk::ShaderStageFlagBits::eCompute;
 		stage.module              = module;
@@ -611,12 +602,7 @@ void TileManager::ConvertD16(Result source, Result target, D16Direction directio
 
 void TileManager::SwapBgra16(Result input, Result output, uint32_t pixels) {
 	if (m_swap_bgra16 == nullptr) {
-		vk::ShaderModuleCreateInfo module_info {};
-		module_info.codeSize    = std::size(GPU_TILER_SWAP_BGRA16_SPV) * sizeof(uint32_t);
-		module_info.pCode       = GPU_TILER_SWAP_BGRA16_SPV;
-		vk::ShaderModule module = nullptr;
-		RequireVulkanSuccess(m_graphics.device.createShaderModule(&module_info, nullptr, &module),
-		                     "create BGRA16 swap shader module");
+		const auto module = CompileSPV(GPU_TILER_SWAP_BGRA16_SPV, m_graphics.device);
 		vk::PipelineShaderStageCreateInfo stage {};
 		stage.stage  = vk::ShaderStageFlagBits::eCompute;
 		stage.module = module;

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

@@ -15,6 +15,7 @@
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/hostMemory.h"
+#include "graphics/host_gpu/renderer/colorRenderTarget.h"
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
@@ -93,22 +94,15 @@ static const char* ShaderStageResourceName(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Vertex: return "Vertex";
 		case ShaderType::Mesh: return "Mesh";
+		case ShaderType::Local: return "Local";
+		case ShaderType::TessellationControl: return "Hull";
+		case ShaderType::TessellationEvaluation: return "Domain";
 		case ShaderType::Pixel: return "Pixel";
 		case ShaderType::Compute: return "Compute";
 		default: return "Unknown";
 	}
 }
 
-static vk::ShaderStageFlags NativeShaderStage(ShaderType stage) {
-	switch (stage) {
-		case ShaderType::Vertex: return vk::ShaderStageFlagBits::eVertex;
-		case ShaderType::Mesh: return vk::ShaderStageFlagBits::eMeshEXT;
-		case ShaderType::Pixel: return vk::ShaderStageFlagBits::eFragment;
-		case ShaderType::Compute: return vk::ShaderStageFlagBits::eCompute;
-		default: EXIT("unknown native shader stage\n");
-	}
-}
-
 static Prospero::ImageType TextureType(const ShaderTextureResource& descriptor) {
 	const auto type = descriptor.Type();
 	return type == Prospero::ImageType::kCube ? Prospero::ImageType::kColor2DArray : type;
@@ -154,7 +148,7 @@ NativeStorageBuffer(RenderContext& context, const PreparedBindings::BufferSource
 	}
 	buffer_offset = static_cast<uint32_t>(adjustment);
 	const vk::DescriptorBufferInfo result {buffer->Handle(), aligned_offset, size + adjustment};
-	if (resource.formatted && resource.written) {
+	if (resource.written) {
 		context.GetTextureCache().InvalidateMemoryFromGPU(address, size);
 	}
 	const char* access = "Read";
@@ -857,32 +851,39 @@ void RenderExecutor::RebindImages(PreparedBindings& prepared) {
 	}
 }
 
-RenderExecutor::GraphicsBindings
-RenderExecutor::PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
-                                        const ShaderStageRuntime& pixel, bool pixel_active) {
-	GraphicsBindings bindings {
-	    .vertex = PrepareBindings(vertex),
-	};
-	if (pixel_active) {
-		bindings.pixel.emplace(PrepareBindings(pixel));
+void RenderExecutor::PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
+                                             std::span<RenderColorInfo> colors) {
+	bool uses_dma = false;
+	for (auto* stage: stages) {
+		FindBuffers(*stage);
+		uses_dma |= stage->runtime->program->info.uses_dma;
 	}
-	FindBuffers(bindings.vertex);
-	if (bindings.pixel) {
-		FindBuffers(*bindings.pixel);
-	}
-	if (bindings.vertex.runtime->program->info.uses_dma ||
-	    (bindings.pixel && bindings.pixel->runtime->program->info.uses_dma)) {
+	if (uses_dma) {
 		m_context.PrepareBda();
 	}
-	RebindBuffers(bindings.vertex);
-	if (bindings.pixel) {
-		RebindBuffers(*bindings.pixel);
+	for (auto* stage: stages) {
+		RebindImages(*stage);
 	}
-	RebindImages(bindings.vertex);
-	if (bindings.pixel) {
-		RebindImages(*bindings.pixel);
+	auto& cache = m_context.GetTextureCache();
+	for (auto& target: colors) {
+		EXIT_IF(!target.image_id);
+		const auto old_image = cache.m_slot_images.try_get(target.image_id);
+		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
+		    old_image->binding.needs_rebind) {
+			if (old_image != nullptr) {
+				old_image->binding = {};
+			}
+			target.desc.view_info.base_level = target.guest_mip_level;
+			target.desc.view_info.base_layer = target.guest_array_layer;
+			target.image_id = cache.FindImage(target.desc);
+			BindRenderTarget(target.image_id);
+		}
+	}
+	// Discovery can read back PS5 metadata and submit the scheduler. Reserve draw buffers only
+	// after image identities are final; attachment layout transitions follow buffer alias copies.
+	for (auto* stage: stages) {
+		RebindBuffers(*stage);
 	}
-	return bindings;
 }
 
 void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
@@ -895,9 +896,10 @@ void RenderExecutor::CommitBindings(CommandBuffer&                     buffer,
 	size_t write_count      = 0;
 	ShaderRecompiler::IR::PushData push_data;
 	bool                           has_push_data = false;
-	constexpr auto                 GraphicsStages = vk::ShaderStageFlagBits::eVertex |
-	                                                vk::ShaderStageFlagBits::eMeshEXT |
-	                                                vk::ShaderStageFlagBits::eFragment;
+	constexpr auto                 GraphicsStages =
+	    vk::ShaderStageFlagBits::eVertex | vk::ShaderStageFlagBits::eMeshEXT |
+	    vk::ShaderStageFlagBits::eTessellationControl |
+	    vk::ShaderStageFlagBits::eTessellationEvaluation | vk::ShaderStageFlagBits::eFragment;
 	vk::ShaderStageFlags push_stages = pipeline_bind_point == vk::PipelineBindPoint::eGraphics
 	                                       ? vk::ShaderStageFlagBits::eFragment
 	                                       : vk::ShaderStageFlags {};

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

@@ -236,6 +236,9 @@ struct PipelineCache::ProgramCache {
 		switch (options.stage) {
 			case ShaderType::Vertex: stage_name = "vs"; break;
 			case ShaderType::Mesh: stage_name = "ms"; break;
+			case ShaderType::Local: stage_name = "ls"; break;
+			case ShaderType::TessellationControl: stage_name = "hs"; break;
+			case ShaderType::TessellationEvaluation: stage_name = "ds"; break;
 			case ShaderType::Pixel: stage_name = "ps"; break;
 			case ShaderType::Compute: stage_name = "cs"; break;
 			default: EXIT("invalid pipeline shader stage\n");
@@ -250,12 +253,7 @@ struct PipelineCache::ProgramCache {
 		}
 		DumpShaderSpirv(stage_name, options.shader_hash, result.spirv);
 
-		vk::ShaderModuleCreateInfo create_info {};
-		create_info.codeSize    = result.spirv.size() * sizeof(uint32_t);
-		create_info.pCode       = result.spirv.data();
-		vk::ShaderModule module = nullptr;
-		RequireVulkanSuccess(device.createShaderModule(&create_info, nullptr, &module),
-		                     "create recompiled shader module");
+		const auto module = CompileSPV(result.spirv, device);
 		EXIT_IF(module == nullptr);
 		if (options.dump_ir) {
 			LOGF("%s SPIR-V words=%" PRIu64 " wave_size=%u\n", options.dump_label,
@@ -273,7 +271,7 @@ struct PipelineCache::ProgramCache {
 	                  uint32_t& push_data_cursor) {
 		ShaderType stage;
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
-			stage = input_info.mesh.threads_num[0] != 0 ? ShaderType::Mesh : ShaderType::Vertex;
+			stage = input_info.logical_stage;
 		} else if constexpr (std::is_same_v<InputInfo, ShaderPixelInputInfo>) {
 			stage = ShaderType::Pixel;
 		} else {
@@ -325,6 +323,9 @@ struct PipelineCache::ProgramCache {
 		switch (stage) {
 			case ShaderType::Vertex: label = "ShaderRecompiler VS"; break;
 			case ShaderType::Mesh: label = "ShaderRecompiler MS"; break;
+			case ShaderType::Local: label = "ShaderRecompiler LS"; break;
+			case ShaderType::TessellationControl: label = "ShaderRecompiler HS"; break;
+			case ShaderType::TessellationEvaluation: label = "ShaderRecompiler DS"; break;
 			case ShaderType::Pixel: label = "ShaderRecompiler PS"; break;
 			case ShaderType::Compute: label = "ShaderRecompiler CS"; break;
 			default: EXIT("invalid pipeline shader stage\n");
@@ -341,11 +342,11 @@ struct PipelineCache::ProgramCache {
 
 		if constexpr (std::is_same_v<InputInfo, ShaderVertexInputInfo>) {
 			options.user_data_base = 8;
-			if (stage == ShaderType::Mesh) {
+			if (stage == ShaderType::Mesh || stage == ShaderType::TessellationControl) {
 				options.user_data_base = 0;
-				options.wave_size      = input_info.mesh.wave_size;
+				options.wave_size = stage == ShaderType::Mesh ? input_info.mesh.wave_size : 64u;
 			}
-		} else if constexpr (std::is_same_v<InputInfo, ShaderComputeInputInfo>) {
+		} else {
 			options.wave_size = input_info.wave_size;
 		}
 		auto translated = ShaderRecompiler::TranslateProgram(params.code, options);
@@ -361,16 +362,19 @@ struct PipelineCache::ProgramCache {
 		input_info.stage = {.program = &permutation.program, .resources = std::move(resources)};
 		permutation.program.bindings.AdvancePushData(push_data_cursor);
 
-		std::array<size_t, static_cast<size_t>(ShaderType::Mesh) + 1> counts {};
+		std::array<size_t, static_cast<size_t>(ShaderType::TessellationEvaluation) + 1> counts {};
 		for (const auto& [key, source]: programs) {
 			counts[static_cast<size_t>(key.stage)] += source.permutations.size();
 		}
 		// Guest geometry shaders are compiled through the host mesh stage.
-		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu\n",
+		std::printf("Shaders: VS %zu | PS %zu | CS %zu | GS %zu | LS %zu | HS %zu | TES %zu\n",
 		            counts[static_cast<size_t>(ShaderType::Vertex)],
 		            counts[static_cast<size_t>(ShaderType::Pixel)],
 		            counts[static_cast<size_t>(ShaderType::Compute)],
-		            counts[static_cast<size_t>(ShaderType::Mesh)]);
+		            counts[static_cast<size_t>(ShaderType::Mesh)],
+		            counts[static_cast<size_t>(ShaderType::Local)],
+		            counts[static_cast<size_t>(ShaderType::TessellationControl)],
+		            counts[static_cast<size_t>(ShaderType::TessellationEvaluation)]);
 		return permutation.handle;
 	}
 
@@ -564,12 +568,18 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
     const HW::VertexShaderInfo& vertex_regs, const HW::PixelShaderInfo& pixel_regs,
     const HW::ShaderRegisters& sh, const HW::Context& context, const HW::UserConfig& user_config,
     std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping, bool pixel_active,
-    ShaderVertexInputInfo& vertex_info, ShaderPixelInputInfo& pixel_info) {
-	const auto vertex_params = PrepareProgram(vertex_regs, context, user_config, vertex_info);
-	const bool mesh_active   = vertex_info.mesh.threads_num[0] != 0;
+    std::array<ShaderVertexInputInfo, 3>& vertex_info, ShaderPixelInputInfo& pixel_info) {
+	const bool tess_active = user_config.GetPrimType() == Prospero::PrimitiveType::kPatch;
+	std::array<ShaderParams, 3> vertex_params;
+	if (tess_active) {
+		vertex_params = PrepareTessellationPrograms(vertex_regs, context, vertex_info);
+	} else {
+		vertex_params[0] = PrepareProgram(vertex_regs, context, user_config, vertex_info[0]);
+	}
+	const bool mesh_active = vertex_info[0].logical_stage == ShaderType::Mesh;
 	if (mesh_active) {
 		EXIT_NOT_IMPLEMENTED(!m_graphics.mesh_shader_enabled);
-		auto& mesh              = vertex_info.mesh;
+		auto& mesh              = vertex_info[0].mesh;
 		mesh.host_subgroup_size = m_graphics.subgroup_size;
 		const auto& limits      = m_graphics.mesh_shader_properties;
 		const auto  logical_threads =
@@ -592,7 +602,7 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	if (context.GetClipControl().clip_disable) {
 		const auto& viewport = context.GetScreenViewport().viewports[0];
 		const auto& limits   = m_graphics.GetPhysicalDeviceProperties().limits;
-		auto&       clip     = vertex_info.clip_space;
+		auto&       clip     = vertex_info[tess_active ? 2u : 0u].clip_space;
 		clip.scale[0]        = viewport.xscale;
 		clip.scale[1]        = viewport.yscale;
 		clip.offset[0]       = viewport.xoffset;
@@ -610,7 +620,9 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	if (pixel_active) {
 		result.pixel = m_program_cache->Get(pixel_params, pixel_info, push_data_cursor);
 	}
-	result.vertex = m_program_cache->Get(vertex_params, vertex_info, push_data_cursor);
+	for (uint32_t i = 0; i < (tess_active ? 3u : 1u); i++) {
+		result.vertex[i] = m_program_cache->Get(vertex_params[i], vertex_info[i], push_data_cursor);
+	}
 	return result;
 }
 
@@ -630,10 +642,12 @@ bool PipelineStaticParameters::operator==(const PipelineStaticParameters& other)
 
 PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
     std::span<const RenderColorInfo> colors, const RenderDepthInfo& depth,
-    const ShaderVertexInputInfo& vs_input_info, CommandBuffer& command,
+    std::span<const ShaderVertexInputInfo> vertex_info, CommandBuffer& command,
     const ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology,
-    bool primitive_restart_enable, const ShaderProgram& vertex_program,
-    const ShaderProgram& pixel_program) {
+    bool primitive_restart_enable, const GraphicsPrograms& programs) {
+	const auto& vs_input_info  = vertex_info.front();
+	const auto& vertex_program = programs.vertex[0];
+	const auto& pixel_program  = programs.pixel;
 	KYTY_PROFILER_BLOCK("PipelineCache::CreatePipeline(Gfx)", profiler::colors::DeepOrangeA200);
 
 	EXIT_IF(colors.size() > RENDER_COLOR_ATTACHMENTS_MAX);
@@ -651,7 +665,9 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	const auto ps_id = ps_active ? pixel_program.id : 0;
 
 	GraphicsPipelineKey key {};
-	key.vs_shader_id            = vs_id;
+	for (uint32_t i = 0; i < programs.vertex.size(); i++) {
+		key.vertex_shader_ids[i] = programs.vertex[i].id;
+	}
 	key.ps_shader_id            = ps_id;
 	auto& static_params         = key.static_params;
 	auto& rendering             = key.rendering;
@@ -726,10 +742,8 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	static_params.depth_bounds_test_enable = depth.depth_bounds_test_enable;
 	static_params.depth_min_bounds         = depth.depth_min_bounds;
 	static_params.depth_max_bounds         = depth.depth_max_bounds;
-	static_params.stencil_test_enable      = depth.stencil_test_enable;
-	static_params.stencil_front            = depth.stencil_static_front;
-	static_params.stencil_back             = depth.stencil_static_back;
-	const bool rect_list     = topology == vk::PrimitiveTopology::ePatchList;
+	const bool rect_list =
+	    command.GetUserConfig().GetPrimType() == Prospero::PrimitiveType::kRectList;
 	static_params.cull_back  = !rect_list && mc.cull_back;
 	static_params.cull_front = !rect_list && mc.cull_front;
 	static_params.face       = mc.face;
@@ -780,9 +794,8 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 
 	auto cached = std::make_unique<Pipeline>();
 	LogPipelineTrace("CreatePipelineInternal begin", vs_id, ps_id);
-	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vs_input_info,
-	                       vertex_program, ps_input_info, pixel_program, static_params,
-	                       m_driver_cache);
+	CreatePipelineInternal(m_graphics, *cached, rendering, key.vertex_input, vertex_info,
+	                       ps_input_info, programs, static_params, m_driver_cache);
 	LogPipelineTrace("CreatePipelineInternal done", vs_id, ps_id);
 
 	EXIT_NOT_IMPLEMENTED(cached->pipeline == nullptr);

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

@@ -42,9 +42,6 @@ struct PipelineStaticParameters {
 	bool                       depth_bounds_test_enable = false;
 	float                      depth_min_bounds         = 0.0f;
 	float                      depth_max_bounds         = 0.0f;
-	bool                       stencil_test_enable      = false;
-	PipelineStencilStaticState stencil_front;
-	PipelineStencilStaticState stencil_back;
 	uint32_t                   color_mask[RENDER_COLOR_ATTACHMENTS_MAX]           = {};
 	bool                       cull_front                                         = false;
 	bool                       cull_back                                          = false;
@@ -68,7 +65,7 @@ struct PipelineStaticParameters {
 static_assert(std::is_trivially_copyable_v<PipelineStaticParameters>);
 static_assert(std::is_standard_layout_v<PipelineStaticParameters>);
 static_assert(alignof(PipelineStaticParameters) == 1);
-static_assert(sizeof(PipelineStaticParameters) == 158);
+static_assert(sizeof(PipelineStaticParameters) == 125);
 
 struct PipelineRenderingState {
 	std::array<vk::Format, RENDER_COLOR_ATTACHMENTS_MAX> color_formats {};
@@ -121,8 +118,10 @@ public:
 	};
 
 	struct GraphicsPrograms {
-		ShaderProgram vertex;
+		std::array<ShaderProgram, 3> vertex;
 		ShaderProgram pixel;
+
+		[[nodiscard]] uint32_t VertexStageCount() const { return vertex[1] ? 3u : 1u; }
 	};
 
 	GraphicsPrograms
@@ -130,18 +129,18 @@ public:
 	                    const HW::PixelShaderInfo& pixel_regs, const HW::ShaderRegisters& sh,
 	                    const HW::Context& context, const HW::UserConfig& user_config,
 	                    std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
-	                    bool pixel_active, ShaderVertexInputInfo& vertex_info,
+	                    bool pixel_active, std::array<ShaderVertexInputInfo, 3>& vertex_info,
 	                    ShaderPixelInputInfo& pixel_info);
 	ShaderProgram GetComputeProgram(const HW::ComputeShaderInfo& regs,
 	                                const HW::ShaderRegisters&   sh,
 	                                ShaderComputeInputInfo&      input_info);
 
-	Pipeline&
-	GetGraphicsPipeline(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& GetGraphicsPipeline(std::span<const RenderColorInfo>       colors,
+	                              const RenderDepthInfo&                 depth,
+	                              std::span<const ShaderVertexInputInfo> vertex_info,
+	                              CommandBuffer& command, const ShaderPixelInputInfo* ps_input_info,
+	                              vk::PrimitiveTopology topology, bool primitive_restart_enable,
+	                              const GraphicsPrograms& programs);
 	Pipeline& GetComputePipeline(const ShaderComputeInputInfo& input_info,
 	                             const ShaderProgram&          compute_program);
 
@@ -150,13 +149,13 @@ private:
 
 	struct GraphicsPipelineKey {
 		PipelineRenderingState   rendering;
-		uint64_t                 vs_shader_id = 0;
+		std::array<uint64_t, 3>  vertex_shader_ids {};
 		uint64_t                 ps_shader_id = 0;
 		PipelineVertexInputState vertex_input;
 		PipelineStaticParameters static_params;
 
 		bool operator==(const GraphicsPipelineKey& other) const {
-			return rendering == other.rendering && vs_shader_id == other.vs_shader_id &&
+			return rendering == other.rendering && vertex_shader_ids == other.vertex_shader_ids &&
 			       ps_shader_id == other.ps_shader_id && vertex_input == other.vertex_input &&
 			       static_params == other.static_params;
 		}
@@ -189,7 +188,9 @@ private:
 		std::size_t operator()(const GraphicsPipelineKey& key) const {
 			std::size_t hash = 0;
 			PipelineKeyHash::MixRendering(hash, key.rendering);
-			PipelineKeyHash::Mix(hash, key.vs_shader_id);
+			for (const auto id: key.vertex_shader_ids) {
+				PipelineKeyHash::Mix(hash, id);
+			}
 			PipelineKeyHash::Mix(hash, key.ps_shader_id);
 			PipelineKeyHash::Mix(hash, key.vertex_input.binding_count);
 			for (uint32_t i = 0; i < key.vertex_input.binding_count; i++) {
@@ -219,12 +220,14 @@ private:
 };
 
 void LogPipelineTrace(const char* phase, uint64_t vertex_program_id, uint64_t pixel_program_id);
-void CreatePipelineInternal(
-    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
-    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
-    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
-    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
-    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache);
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+                            const PipelineRenderingState&          rendering,
+                            const PipelineVertexInputState&        vertex_input,
+                            std::span<const ShaderVertexInputInfo> vertex_info,
+                            const ShaderPixelInputInfo*            ps_input_info,
+                            const PipelineCache::GraphicsPrograms& programs,
+                            const PipelineStaticParameters&        static_params,
+                            vk::PipelineCache                      driver_cache);
 void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
                             const ShaderComputeInputInfo& input_info,
                             vk::ShaderModule compute_module, vk::PipelineCache driver_cache);

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

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

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

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

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

@@ -8,6 +8,7 @@
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
+#include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/renderTarget.h"
@@ -205,12 +206,18 @@ static void CreateDescriptorLayout(GraphicContext& graphics, PipelineCache::Pipe
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void CreatePipelineInternal(
-    GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
-    const PipelineRenderingState& rendering, const PipelineVertexInputState& vertex_input,
-    const ShaderVertexInputInfo& vs_input_info, const ShaderProgram& vertex_program,
-    const ShaderPixelInputInfo* ps_input_info, const ShaderProgram& pixel_program,
-    const PipelineStaticParameters& static_params, vk::PipelineCache driver_cache) {
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+                            const PipelineRenderingState&          rendering,
+                            const PipelineVertexInputState&        vertex_input,
+                            std::span<const ShaderVertexInputInfo> vertex_info,
+                            const ShaderPixelInputInfo*            ps_input_info,
+                            const PipelineCache::GraphicsPrograms& programs,
+                            const PipelineStaticParameters&        static_params,
+                            vk::PipelineCache                      driver_cache) {
+	const auto& vs_input_info  = vertex_info.front();
+	const auto& vertex_program = programs.vertex[0];
+	const auto& pixel_program  = programs.pixel;
+	const bool  tessellation   = vertex_info.size() == 3;
 	const bool ps_active = ps_input_info != nullptr;
 	EXIT_IF(!vertex_program || (ps_active && !pixel_program));
 	const bool with_depth = rendering.depth_format != vk::Format::eUndefined ||
@@ -218,46 +225,39 @@ void CreatePipelineInternal(
 	EXIT_IF(!vs_input_info.stage);
 	const bool mesh = vs_input_info.stage.program->stage == ShaderType::Mesh;
 	EXIT_NOT_IMPLEMENTED(mesh && !graphics.mesh_shader_enabled);
-	const auto vertex_stage =
-	    mesh ? vk::ShaderStageFlagBits::eMeshEXT : vk::ShaderStageFlagBits::eVertex;
-
-	const bool rect_list = !mesh && static_params.topology == vk::PrimitiveTopology::ePatchList;
+	const bool rect_list =
+	    !mesh && !tessellation && static_params.topology == vk::PrimitiveTopology::ePatchList;
 
 	vk::ShaderModule tess_control_shader_module = nullptr;
 	vk::ShaderModule tess_eval_shader_module    = nullptr;
 
-	vk::ShaderModuleCreateInfo create_info {};
-	vk::Result result {};
 	if (rect_list) {
 		const auto shaders =
 		    BuildRectListShaders(vs_input_info, ps_active ? ps_input_info : nullptr);
-		create_info.codeSize = shaders.control.size() * 4;
-		create_info.pCode    = shaders.control.data();
-		result =
-		    graphics.device.createShaderModule(&create_info, nullptr, &tess_control_shader_module);
+		tess_control_shader_module = CompileSPV(shaders.control, graphics.device);
 		if (graphics_debug_dump_enabled()) {
-			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done result=%s module=%p\n",
-			     vk::to_string(result).c_str(), static_cast<void*>(tess_control_shader_module));
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TCS done module=%p\n",
+			     static_cast<void*>(tess_control_shader_module));
 		}
-		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
-		create_info.codeSize = shaders.evaluation.size() * 4;
-		create_info.pCode    = shaders.evaluation.data();
-		result =
-		    graphics.device.createShaderModule(&create_info, nullptr, &tess_eval_shader_module);
+		tess_eval_shader_module = CompileSPV(shaders.evaluation, graphics.device);
 		if (graphics_debug_dump_enabled()) {
-			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done result=%s module=%p\n",
-			     vk::to_string(result).c_str(), static_cast<void*>(tess_eval_shader_module));
+			LOGF("PipelineTrace: vkCreateShaderModule RectList TES done module=%p\n",
+			     static_cast<void*>(tess_eval_shader_module));
 		}
-		EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 	}
 
 	EXIT_NOT_IMPLEMENTED(
 	    rect_list && (tess_control_shader_module == nullptr || tess_eval_shader_module == nullptr));
 
-	vk::PipelineShaderStageCreateInfo shader_stages[4] = {
-	    {.stage = vertex_stage, .module = vertex_program.module, .pName = "main"}};
-	uint32_t shader_stage_count = 1;
+	vk::PipelineShaderStageCreateInfo shader_stages[4] {};
+	uint32_t                          shader_stage_count = 0;
+	for (uint32_t i = 0; i < vertex_info.size(); i++) {
+		shader_stages[shader_stage_count++] = {.stage =
+		                                           NativeShaderStage(vertex_info[i].logical_stage),
+		                                       .module = programs.vertex[i].module,
+		                                       .pName  = "main"};
+	}
 	if (rect_list) {
 		shader_stages[shader_stage_count++] = {.stage =
 		                                           vk::ShaderStageFlagBits::eTessellationControl,
@@ -432,15 +432,19 @@ void CreatePipelineInternal(
 	color_blending.pAttachments    = color_blend_attachment;
 
 	std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
-	AddLayoutBindings(descriptor_bindings, *vs_input_info.stage.program, vertex_stage);
+	vk::ShaderStageFlags graphics_stages = vk::ShaderStageFlagBits::eFragment;
+	for (const auto& stage: vertex_info) {
+		const auto native_stage = NativeShaderStage(stage.logical_stage);
+		AddLayoutBindings(descriptor_bindings, *stage.stage.program, native_stage);
+		graphics_stages |= native_stage;
+	}
 	if (ps_active) {
 		EXIT_IF(!ps_input_info->stage);
 		AddLayoutBindings(descriptor_bindings, *ps_input_info->stage.program,
 		                  vk::ShaderStageFlagBits::eFragment);
 	}
 	CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
-	const auto                  GraphicsStages = vertex_stage | vk::ShaderStageFlagBits::eFragment;
-	const vk::PushConstantRange push_constants {GraphicsStages, 0,
+	const vk::PushConstantRange push_constants {graphics_stages, 0,
 	                                            ShaderRecompiler::IR::NativePushConstantSize};
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
@@ -456,8 +460,8 @@ void CreatePipelineInternal(
 		     " set_layouts=1 push_constants=%" PRIu32 "\n",
 		     vertex_program.id, ps_active ? pixel_program.id : 0, 1u);
 	}
-	result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
-	                                              &pipeline.pipeline_layout);
+	auto result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
+	                                                   &pipeline.pipeline_layout);
 	if (graphics_debug_dump_enabled()) {
 		LOGF("PipelineTrace: vkCreatePipelineLayout done result=%s layout=%p\n",
 		     vk::to_string(result).c_str(), static_cast<void*>(pipeline.pipeline_layout));
@@ -473,15 +477,6 @@ void CreatePipelineInternal(
 #else
 	    (static_params.depth_bounds_test_enable ? VK_TRUE : VK_FALSE);
 #endif
-	depth_stencil_info.stencilTestEnable = (static_params.stencil_test_enable ? VK_TRUE : VK_FALSE);
-	depth_stencil_info.front.failOp      = static_params.stencil_front.failOp;
-	depth_stencil_info.front.passOp      = static_params.stencil_front.passOp;
-	depth_stencil_info.front.depthFailOp = static_params.stencil_front.depthFailOp;
-	depth_stencil_info.front.compareOp   = static_params.stencil_front.compareOp;
-	depth_stencil_info.back.failOp       = static_params.stencil_back.failOp;
-	depth_stencil_info.back.passOp       = static_params.stencil_back.passOp;
-	depth_stencil_info.back.depthFailOp  = static_params.stencil_back.depthFailOp;
-	depth_stencil_info.back.compareOp    = static_params.stencil_back.compareOp;
 	depth_stencil_info.minDepthBounds    = static_params.depth_min_bounds;
 	depth_stencil_info.maxDepthBounds    = static_params.depth_max_bounds;
 
@@ -494,6 +489,8 @@ void CreatePipelineInternal(
 	    vk::DynamicState::eDepthCompareOp,
 	    vk::DynamicState::eDepthBiasEnable,
 	    vk::DynamicState::eDepthBias,
+	    vk::DynamicState::eStencilTestEnable,
+	    vk::DynamicState::eStencilOp,
 	    vk::DynamicState::eStencilCompareMask,
 	    vk::DynamicState::eStencilReference,
 	    vk::DynamicState::eStencilWriteMask,
@@ -524,8 +521,9 @@ void CreatePipelineInternal(
 	pipeline_info.pVertexInputState        = mesh ? nullptr : &vertex_input_info;
 	pipeline_info.pInputAssemblyState      = mesh ? nullptr : &input_assembly;
 	vk::PipelineTessellationStateCreateInfo tessellation_state {};
-	tessellation_state.patchControlPoints = 3;
-	pipeline_info.pTessellationState      = (rect_list ? &tessellation_state : nullptr);
+	tessellation_state.patchControlPoints =
+	    tessellation ? vs_input_info.tess.input_control_points : 3u;
+	pipeline_info.pTessellationState = (rect_list || tessellation) ? &tessellation_state : nullptr;
 	pipeline_info.pViewportState          = &viewport_state;
 	pipeline_info.pRasterizationState     = &rasterizer;
 	pipeline_info.pMultisampleState       = &multisampling;

src/graphics/host_gpu/renderer/render.h

@@ -170,15 +170,14 @@ private:
 	void DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const DrawAutoArgs& args);
 
 	struct GraphicsBindings {
-		PreparedBindings                vertex;
+		std::array<PreparedBindings, 3> vertex;
 		std::optional<PreparedBindings> pixel;
 	};
 
 	[[nodiscard]] TextureBinding ResolveTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	                                            const ShaderRecompiler::IR::DescriptorValue& value);
-	[[nodiscard]] GraphicsBindings PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
-	                                                       const ShaderStageRuntime& pixel,
-	                                                       bool                      pixel_active);
+	void PrepareGraphicsBindings(std::span<PreparedBindings* const> stages,
+	                             std::span<RenderColorInfo> colors);
 	void ResolveRenderColorTarget(CommandBuffer& buffer, RenderColorInfo& target,
 	                              uint32_t render_target_slice_offset, uint32_t render_target_slot,
 	                              bool ignore_target_mask = false, bool exact_format = false);

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -172,7 +172,7 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
 		                                       1, view.base_layer, view.layer_count};
 		vk::ClearValue clear {};
 		clear.depthStencil = vk::ClearDepthStencilValue {0.0f, fill.value};
-		cache.ClearImage(command, binding.image_id, range, clear);
+		cache.ClearImage(command, binding.image_id, image.backing.format, range, clear);
 		return true;
 	}
 	ShaderBufferResource descriptor;
@@ -183,14 +183,6 @@ bool RenderExecutor::TryConsumeComputeImageClear(const ShaderComputeInputInfo& i
 		return false;
 	}
 	if (!cache.ClearImageFromBuffer(command, descriptor.Base48(), size, packed_clear)) {
-		// Track deferred DCC state while the original dispatch writes the metadata allocation.
-		cache.TrackDccFill(descriptor.Base48(), size, packed_clear);
-		static std::atomic<uint32_t> logged_metadata_clears {0};
-		if (logged_metadata_clears.fetch_add(1, std::memory_order_relaxed) < 32) {
-			LOGF("GraphicsRenderDispatchDirect: metadata fill shader=0x%016" PRIx64
-			     " addr=0x%016" PRIx64 " size=0x%016" PRIx64 " value=0x%08" PRIx32 "\n",
-			     input.stage.program->shader_hash, descriptor.Base48(), size, packed_clear);
-		}
 		return false;
 	}
 	static std::atomic<uint32_t> logged_clears {0};
@@ -374,8 +366,8 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 	if (program.info.uses_dma) {
 		m_context.PrepareBda();
 	}
-	RebindBuffers(bindings);
 	RebindImages(bindings);
+	RebindBuffers(bindings);
 
 	auto              vk_buffer        = buffer.Handle();
 	PreparedBindings* descriptor_stage = &bindings;

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -403,19 +403,16 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 		vk_buffer.setDepthBias(constant_factor, poly_offset.clamp, slope_factor);
 	}
 
+	vk_buffer.setStencilTestEnable(depth.stencil_test_enable ? VK_TRUE : VK_FALSE);
 	if (depth.stencil_test_enable) {
-		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eFront,
-		                                depth.stencil_dynamic_front.compareMask);
-		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eBack,
-		                                depth.stencil_dynamic_back.compareMask);
-		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eFront,
-		                              depth.stencil_dynamic_front.writeMask);
-		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eBack,
-		                              depth.stencil_dynamic_back.writeMask);
-		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eFront,
-		                              depth.stencil_dynamic_front.reference);
-		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eBack,
-		                              depth.stencil_dynamic_back.reference);
+		const auto set_stencil = [&](vk::StencilFaceFlagBits face, const vk::StencilOpState& state) {
+			vk_buffer.setStencilOp(face, state.failOp, state.passOp, state.depthFailOp, state.compareOp);
+			vk_buffer.setStencilCompareMask(face, state.compareMask);
+			vk_buffer.setStencilWriteMask(face, state.writeMask);
+			vk_buffer.setStencilReference(face, state.reference);
+		};
+		set_stencil(vk::StencilFaceFlagBits::eFront, depth.stencil_front);
+		set_stencil(vk::StencilFaceFlagBits::eBack, depth.stencil_back);
 	}
 
 #if defined(__APPLE__)
@@ -453,7 +450,7 @@ struct DrawRenderState {
 	RenderColorInfo       color_info[RENDER_COLOR_ATTACHMENTS_MAX] = {};
 	uint32_t              color_count                              = 0;
 	bool                  ps_active                                = true;
-	ShaderVertexInputInfo vs_input_info;
+	std::array<ShaderVertexInputInfo, 3> vertex_info;
 	ShaderPixelInputInfo  ps_input_info;
 	PipelineCache::GraphicsPrograms programs;
 };
@@ -482,14 +479,10 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 	for (uint32_t i = 0; i < color_count; i++) {
 		auto& target = colors[i];
 		EXIT_IF(!target.image_id);
-		const auto old_image = cache.m_slot_images.try_get(target.image_id);
-		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
-		    old_image->binding.needs_rebind) {
-			if (old_image != nullptr) {
-				old_image->binding = {};
-			}
-			target.image_id = cache.FindImage(target.desc);
-			BindRenderTarget(target.image_id);
+		const auto owner = cache.m_slot_images.try_get(target.image_id);
+		if (owner == nullptr || (!owner->registered && !owner->info.data.Empty()) ||
+		    owner->binding.needs_rebind) {
+			EXIT("color target changed after render-state discovery\n");
 		}
 		const auto image_view = cache.FindRenderTarget(target.image_id, target.desc);
 		auto&      image      = cache.GetImage(target.image_id);
@@ -709,6 +702,7 @@ struct PreparedVertexBuffers {
 
 	std::array<vk::Buffer, MaxBuffers>     buffers {};
 	std::array<vk::DeviceSize, MaxBuffers> offsets {};
+	std::array<vk::DeviceSize, MaxBuffers> sizes {};
 	uint32_t                               count = 0;
 };
 
@@ -793,6 +787,7 @@ static PreparedVertexBuffers AcquireVertexBuffers(CommandBuffer&               b
 
 		prepared.buffers[i] = range->binding.first->Handle();
 		prepared.offsets[i] = range->binding.second + vertex.addr - range->base_address;
+		prepared.sizes[i]   = std::min(size, range->acquired_end - vertex.addr);
 		SetVulkanObjectNameF(
 		    buffer.GetContext().GetGraphics().device, prepared.buffers[i],
 		    "Kyty.VertexBuffer[slot={} guest=0x{:016x} size=0x{:x} stride={} records={}]", i,
@@ -831,6 +826,7 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 		case Prospero::PrimitiveType::kTriStrip:
 			topology = vk::PrimitiveTopology::eTriangleStrip;
 			break;
+		case Prospero::PrimitiveType::kPatch:
 		case Prospero::PrimitiveType::kRectList:
 			topology = vk::PrimitiveTopology::ePatchList;
 			break;
@@ -953,7 +949,7 @@ static void RefreshShaders(CommandBuffer& buffer, const DrawCallInfo& draw,
 	}
 	state.programs = pipeline_cache.GetGraphicsPrograms(
 	    vertex_shader_info, pixel_shader_info, shader_regs, ctx, buffer.GetUserConfig(),
-	    target_export_mapping, state.ps_active, state.vs_input_info, state.ps_input_info);
+	    target_export_mapping, state.ps_active, state.vertex_info, state.ps_input_info);
 }
 
 static PreparedIndexBuffer PrepareIndexBuffer(CommandBuffer&               buffer,
@@ -988,8 +984,9 @@ static void CommitVertexBuffers(vk::CommandBuffer            vk_buffer,
 		EXIT_IF(prepared.buffers[i] == nullptr);
 	}
 	if (prepared.count != 0) {
-		vk_buffer.bindVertexBuffers(0, prepared.count, prepared.buffers.data(),
-		                            prepared.offsets.data());
+		// Guest descriptor bounds must survive allocation merging in the cache.
+		vk_buffer.bindVertexBuffers2(0, prepared.count, prepared.buffers.data(),
+		                             prepared.offsets.data(), prepared.sizes.data(), nullptr);
 	}
 }
 
@@ -1016,7 +1013,7 @@ static void LogDrawStateIfNeeded(const CommandBuffer& buffer, const DrawCallInfo
 		LogDrawTargetState(draw.Name(), state.color_info[0], state.depth_info, buffer,
 		                   state.ps_input_info, draw.index_count, 0);
 	}
-	LogDrawInputState(buffer, state.color_info[0], state.vs_input_info, index_type_and_size,
+	LogDrawInputState(buffer, state.color_info[0], state.vertex_info[0], index_type_and_size,
 	                  draw.index_count, index_addr);
 }
 
@@ -1031,6 +1028,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 		case Prospero::PrimitiveType::kTriFan:
 		case Prospero::PrimitiveType::kTriStrip:
 		case Prospero::PrimitiveType::kRectList:
+		case Prospero::PrimitiveType::kPatch:
 			if (draw.IsIndexed()) {
 				vk_buffer.drawIndexed(draw.index_count, draw.instance_count, 0, emit.vertex_offset,
 				                      emit.first_instance);
@@ -1069,10 +1067,12 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
                                          const DrawIndexBufferSource& index_source,
 	                                     bool primitive_restart_enable) {
 	auto& ucfg = buffer.GetUserConfig();
-	const bool mesh_active = state.vs_input_info.stage.program->stage == ShaderType::Mesh;
+	const auto vertex_stages =
+	    std::span {state.vertex_info.data(), state.programs.VertexStageCount()};
+	const bool mesh_active = state.vertex_info[0].stage.program->stage == ShaderType::Mesh;
 	uint32_t   mesh_groups = 0;
 	if (mesh_active) {
-		const auto& mesh = state.vs_input_info.mesh;
+		const auto& mesh = state.vertex_info[0].mesh;
 		if (primitive_restart_enable || mesh.primitives_per_group == 0) {
 			EXIT("unsupported mesh draw: primitive=%u indexed=%u restart=%u\n",
 			     static_cast<uint32_t>(ucfg.GetPrimType()), draw.IsIndexed(), primitive_restart_enable);
@@ -1100,13 +1100,24 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		                              index_source.guest_element_size);
 	}
 	LogDrawPhase(draw.Name(), "PrepareBindings");
-	auto bindings = PrepareGraphicsBindings(state.vs_input_info.stage, state.ps_input_info.stage,
-	                                        state.ps_active);
+	GraphicsBindings                 bindings;
+	std::array<PreparedBindings*, 4> descriptor_stages {};
+	uint32_t                         stage_count = 0;
+	for (uint32_t i = 0; i < vertex_stages.size(); i++) {
+		bindings.vertex[i]               = PrepareBindings(state.vertex_info[i].stage);
+		descriptor_stages[stage_count++] = &bindings.vertex[i];
+	}
+	if (state.ps_active) {
+		bindings.pixel.emplace(PrepareBindings(state.ps_input_info.stage));
+		descriptor_stages[stage_count++] = &*bindings.pixel;
+	}
+	const auto stages = std::span {descriptor_stages.data(), stage_count};
+	PrepareGraphicsBindings(stages, std::span {state.color_info, state.color_count});
 	PreparedVertexBuffers vertex_bindings;
 	PreparedIndexBuffer   index_binding;
 	if (!mesh_active) {
 		LogDrawPhase(draw.Name(), "PrepareVertexBuffers");
-		vertex_bindings = AcquireVertexBuffers(buffer, state.vs_input_info);
+		vertex_bindings = AcquireVertexBuffers(buffer, state.vertex_info[0]);
 		index_binding   = PrepareIndexBuffer(buffer, index_source);
 	}
 	const auto rendering =
@@ -1117,9 +1128,9 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		LogDrawPhase(draw.Name(), "CreatePipeline");
 	}
 	auto& pipeline = m_context.GetPipelineCache().GetGraphicsPipeline(
-	    std::span {state.color_info, state.color_count}, state.depth_info, state.vs_input_info, buffer,
+	    std::span {state.color_info, state.color_count}, state.depth_info, vertex_stages, buffer,
 	    state.ps_active ? &state.ps_input_info : nullptr, topology, primitive_restart_enable,
-	    state.programs.vertex, state.programs.pixel);
+	    state.programs);
 
 	// Resource preparation above may synchronously finish and restart the scheduler. From this
 	// point onward, every operation targets the current command buffer and cannot touch guest
@@ -1132,13 +1143,7 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 	if (bindings.pixel && !draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x300u);
 	}
-	std::array<PreparedBindings*, 2> descriptor_stages {&bindings.vertex, nullptr};
-	const size_t                     descriptor_stage_count = bindings.pixel.has_value() ? 2u : 1u;
-	if (bindings.pixel) {
-		descriptor_stages[1] = &*bindings.pixel;
-	}
-	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline,
-	               std::span {descriptor_stages.data(), descriptor_stage_count});
+	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline, stages);
 	if (mesh_active) {
 		const uint32_t draw_data[] {
 		    draw.index_count,
@@ -1155,7 +1160,7 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 		CommitIndexBuffer(vk_buffer, index_binding);
 	}
 
-	SetGraphicsDynamicParams(buffer, vk_buffer, state.vs_input_info, state.color_info,
+	SetGraphicsDynamicParams(buffer, vk_buffer, vertex_stages.back(), state.color_info,
 	                         state.color_count, state.depth_info);
 	if (m_context.GetGraphics().attachment_feedback_loop_enabled) {
 		vk_buffer.setAttachmentFeedbackLoopEnableEXT(
@@ -1177,16 +1182,17 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 	if (mesh_active) {
 		vk_buffer.drawMeshTasksEXT(mesh_groups, draw.instance_count, 1);
 	} else {
-		EmitDrawPrimitives(ucfg, vk_buffer, state.vs_input_info, draw, emit);
+		EmitDrawPrimitives(ucfg, vk_buffer, state.vertex_info[0], draw, emit);
 	}
 
 	if (!draw.IsIndexed()) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x600u);
 	}
 	vk::PipelineStageFlags shader_write_stages = {};
-	if (HasShaderBufferWrites(state.vs_input_info.stage)) {
-		shader_write_stages |= mesh_active ? vk::PipelineStageFlagBits::eMeshShaderEXT
-		                                   : vk::PipelineStageFlagBits::eVertexShader;
+	for (const auto& stage: vertex_stages) {
+		if (HasShaderBufferWrites(stage.stage)) {
+			shader_write_stages |= ShaderPipelineStages(NativeShaderStage(stage.logical_stage));
+		}
 	}
 	if (state.ps_active && HasShaderBufferWrites(state.ps_input_info.stage)) {
 		shader_write_stages |= vk::PipelineStageFlagBits::eFragmentShader;
@@ -1302,7 +1308,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 	const bool indirect = args.offset_source == DrawOffsetSource::IndirectArgs;
 	const auto [vertex_offset, instance_offset] =
 	    indirect ? std::pair<int32_t, uint32_t> {0, args.first_instance}
-	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vs_input_info);
+	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vertex_info[0]);
 
 	DrawEmitInfo emit {};
 	emit.vertex_offset  = vertex_offset + args.base_vertex;
@@ -1374,9 +1380,9 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 	}
 	RefreshShaders(buffer, draw, state);
 
-	const bool rect_list = topology == vk::PrimitiveTopology::ePatchList;
-	if (rect_list && state.vs_input_info.buffers_num == 0 &&
-	    state.vs_input_info.stage.program->param_export_mask == 0 &&
+	const bool rect_list = ucfg.GetPrimType() == Prospero::PrimitiveType::kRectList;
+	if (rect_list && state.vertex_info[0].buffers_num == 0 &&
+	    state.vertex_info[0].stage.program->param_export_mask == 0 &&
 	    state.ps_input_info.input_num != 0) {
 		if (graphics_debug_dump_enabled()) {
 			LOGF("DrawIndexAuto: skipping rect-list draw with no VS param exports and PS inputs: "
@@ -1393,7 +1399,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 	const bool indirect = args.offset_source == DrawOffsetSource::IndirectArgs;
 	const auto [vertex_offset, instance_offset] =
 	    indirect ? std::pair<int32_t, uint32_t> {0, args.first_instance}
-	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vs_input_info);
+	             : ResolveDrawOffsets(ucfg.GetIndexOffset(), state.vertex_info[0]);
 	DrawEmitInfo emit {};
 	emit.first_vertex = static_cast<uint32_t>(vertex_offset + static_cast<int32_t>(args.first_vertex));
 	emit.first_instance = instance_offset;

src/graphics/host_gpu/renderer/renderTarget.h

@@ -5,7 +5,6 @@
 
 #include <array>
 #include <cstdint>
-#include <type_traits>
 
 namespace Libs::Graphics {
 
@@ -67,33 +66,6 @@ inline constexpr TargetViewInfo ResolveTargetViewInfo(uint32_t base_layer, uint3
 	        base_layer, last_layer - base_layer + 1u, last_layer + 1u};
 }
 
-#pragma pack(push, 1)
-
-struct PipelineStencilStaticState {
-	vk::StencilOp failOp      = vk::StencilOp::eKeep;
-	vk::StencilOp passOp      = vk::StencilOp::eKeep;
-	vk::StencilOp depthFailOp = vk::StencilOp::eKeep;
-	vk::CompareOp compareOp   = vk::CompareOp::eNever;
-};
-
-struct PipelineStencilDynamicState {
-	uint32_t compareMask = 0;
-	uint32_t writeMask   = 0;
-	uint32_t reference   = 0;
-};
-
-#pragma pack(pop)
-
-static_assert(std::is_trivially_copyable_v<PipelineStencilStaticState>);
-static_assert(std::is_standard_layout_v<PipelineStencilStaticState>);
-static_assert(alignof(PipelineStencilStaticState) == 1);
-static_assert(sizeof(PipelineStencilStaticState) ==
-              sizeof(vk::StencilOp) * 3 + sizeof(vk::CompareOp));
-static_assert(std::is_trivially_copyable_v<PipelineStencilDynamicState>);
-static_assert(std::is_standard_layout_v<PipelineStencilDynamicState>);
-static_assert(alignof(PipelineStencilDynamicState) == 1);
-static_assert(sizeof(PipelineStencilDynamicState) == sizeof(uint32_t) * 3);
-
 } // namespace Libs::Graphics
 
 #endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RENDERTARGET_H_

src/graphics/host_gpu/vulkanCommon.cpp

@@ -123,4 +123,14 @@ void RequireVulkanSuccess(vk::Result result, const char* operation) {
 	}
 }
 
+vk::ShaderModule CompileSPV(std::span<const uint32_t> code, vk::Device device) {
+	vk::ShaderModuleCreateInfo create_info {};
+	create_info.codeSize    = code.size_bytes();
+	create_info.pCode       = code.data();
+	vk::ShaderModule module = nullptr;
+	RequireVulkanSuccess(device.createShaderModule(&create_info, nullptr, &module),
+	                     "create SPIR-V shader module");
+	return module;
+}
+
 } // namespace Libs::Graphics

src/graphics/host_gpu/vulkanCommon.h

@@ -15,6 +15,7 @@
 
 #include <cstdint>
 #include <fmt/format.h>
+#include <span>
 #include <string>
 #include <utility>
 #include <vector>
@@ -27,6 +28,7 @@ using VulkanMemoryBarrier = vk::MemoryBarrier;
 
 vk::Format  VulkanFormat(Prospero::BufferFormat guest_format);
 void        RequireVulkanSuccess(vk::Result result, const char* operation);
+vk::ShaderModule CompileSPV(std::span<const uint32_t> code, vk::Device device);
 
 template <typename Handle, typename... Args>
 void SetVulkanObjectNameF(vk::Device device, Handle handle, fmt::format_string<Args...> format,

src/graphics/presentation/videoOut.cpp

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

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -1,4 +1,5 @@
 #include "graphics/shader/recompiler/ShaderRecompiler.h"
+#include "graphics/shader/recompiler/Tessellation.h"
 
 #include "common/assert.h"
 #include "common/logging/log.h"
@@ -47,6 +48,9 @@ const char* StageName(ShaderType stage) {
 	switch (stage) {
 		case ShaderType::Compute: return "CS";
 		case ShaderType::Vertex: return "VS";
+		case ShaderType::Local: return "LS";
+		case ShaderType::TessellationControl: return "HS";
+		case ShaderType::TessellationEvaluation: return "TES";
 		case ShaderType::Mesh: return "MS";
 		case ShaderType::Pixel: return "PS";
 		default: return "unknown";
@@ -211,6 +215,8 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
                                             const ShaderVertexInputInfo* input_info,
                                             uint32_t user_data_base, uint32_t user_data_count,
                                             uint32_t wave_size) {
+	const uint32_t    vertex_index_reg   = input_info->logical_stage == ShaderType::Local ? 2u : 5u;
+	const uint32_t    instance_index_reg = input_info->logical_stage == ShaderType::Local ? 5u : 8u;
 	EmbeddedFetchData data;
 	data.loads.reserve(input_info->resources_num);
 	int32_t vertex_offset_candidate   = -1;
@@ -251,16 +257,17 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 		// corresponding direct-draw offsets before fetching.
 		const bool vertex_index_accumulator =
 		    IsDecodedVgpr(inst.dst) &&
-		    (inst.dst.reg == 0 || (user_data_base == 8 && inst.dst.reg == 5));
+		    (inst.dst.reg == 0 || (user_data_base == 8 && inst.dst.reg == vertex_index_reg));
 		const bool instance_index_accumulator =
-		    IsDecodedVgpr(inst.dst) && (inst.dst.reg == (user_data_base == 8 ? 8u : 3u));
+		    IsDecodedVgpr(inst.dst) &&
+		    (inst.dst.reg == (user_data_base == 8 ? instance_index_reg : 3u));
 		uint32_t   sad_zero = 0;
 		const bool index_offset_add =
-		    (vertex_index_accumulator || instance_index_accumulator) &&
-		    IsDecodedSgpr(inst.src0) &&
+		    (vertex_index_accumulator || instance_index_accumulator) && IsDecodedSgpr(inst.src0) &&
 		    ((inst.opcode == Decoder::Opcode::V_ADD_I32 && IsDecodedVgpr(inst.src1) &&
 		      inst.src1.reg == inst.dst.reg) ||
-		     (user_data_base == 8 && (inst.dst.reg == 5 || inst.dst.reg == 8) &&
+		     (user_data_base == 8 &&
+		      (inst.dst.reg == vertex_index_reg || inst.dst.reg == instance_index_reg) &&
 		      inst.opcode == Decoder::Opcode::V_SAD_U32 && IsDecodedVgpr(inst.src2) &&
 		      inst.src2.reg == inst.dst.reg &&
 		      TryDecodedOperandConstant(sgprs, inst.src1, sad_zero) && sad_zero == 0));
@@ -391,8 +398,8 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 						ClearEmbeddedFetchVectorLanes(&vector_lanes, inst.dst.reg);
 					}
 					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) {
+					    IsDecodedVgpr(inst.src0) && inst.src0.reg == instance_index_reg &&
+					    IsDecodedVgpr(inst.src1) && inst.src1.reg == vertex_index_reg) {
 						vgpr_is_index[inst.dst.reg] = true;
 					}
 				} else if (IsEmbeddedFetchAttribPropagationAlu(inst)) {
@@ -462,22 +469,9 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 Decoder::Program DecodeFusedProgram(std::span<const uint32_t> front, std::span<const uint32_t> back,
                                     std::vector<uint32_t>& joined_code) {
 	EXIT_IF(back.empty());
-	Decoder::Program result;
-	uint32_t         front_words = 0;
-	while (front_words < front.size()) {
-		auto& inst = result.instructions.emplace_back();
-		Decoder::DecodeInstruction(front, front_words, inst);
-		front_words += inst.word_count;
-		if (inst.opcode == Decoder::Opcode::S_SETPC_B64) {
-			EXIT_NOT_IMPLEMENTED(inst.src0.kind != Decoder::OperandKind::Sgpr ||
-			                     inst.src0.reg != 6u);
-			break;
-		}
-		EXIT_NOT_IMPLEMENTED(inst.opcode == Decoder::Opcode::S_ENDPGM);
-	}
-	EXIT_IF(result.instructions.empty() ||
-	        result.instructions.back().opcode != Decoder::Opcode::S_SETPC_B64);
-	joined_code.assign(front.begin(), front.begin() + front_words);
+	auto       result      = Decoder::DecodeFrontProgram(front);
+	const auto front_words = static_cast<uint32_t>(result.code.size());
+	joined_code.assign(result.code.begin(), result.code.end());
 	joined_code.insert(joined_code.end(), back.begin(), back.end());
 	// The merged-stage ABI passes the back shader in s[6:7]. Give that handoff an
 	// ordinary CFG edge, retaining both bodies in one register and LDS lifetime.
@@ -505,7 +499,9 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		EXIT("shader recompiler input is empty\n");
 	}
 	if (options.stage != ShaderType::Compute && options.stage != ShaderType::Vertex &&
-	    options.stage != ShaderType::Pixel && options.stage != ShaderType::Mesh) {
+	    options.stage != ShaderType::Pixel && options.stage != ShaderType::Mesh &&
+	    options.stage != ShaderType::Local && options.stage != ShaderType::TessellationControl &&
+	    options.stage != ShaderType::TessellationEvaluation) {
 		EXIT("shader recompiler received unsupported stage %u\n",
 		     static_cast<unsigned>(options.stage));
 	}
@@ -525,6 +521,12 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 	std::vector<uint32_t> joined_code;
 	if (!options.back_code.empty()) {
 		decoded = DecodeFusedProgram(code, options.back_code, joined_code);
+	} else if (options.stage == ShaderType::Local) {
+		decoded = Decoder::DecodeFrontProgram(code);
+		// The separately compiled hull half runs in the next Vulkan stage.
+		auto& handoff     = decoded.instructions.back();
+		handoff.opcode    = Decoder::Opcode::S_ENDPGM;
+		handoff.src_count = 0;
 	} else {
 		Decoder::DecodeProgram(code, decoded);
 	}
@@ -568,8 +570,8 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 	}
 
 	EmbeddedFetchData embedded_fetch;
-	if (options.stage == ShaderType::Vertex && options.input_info.vertex != nullptr &&
-	    options.input_info.vertex->fetch_embedded) {
+	if ((options.stage == ShaderType::Vertex || options.stage == ShaderType::Local) &&
+	    options.input_info.vertex != nullptr && options.input_info.vertex->fetch_embedded) {
 		embedded_fetch = DetectEmbeddedVertexFetch(
 		    decoded, options.input_info.vertex, options.user_data_base,
 		    static_cast<uint32_t>(options.user_data.size()), options.wave_size);
@@ -608,6 +610,7 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		IR::RemoveIdentities(ir.blocks);
 		IR::EliminateDeadCode(ir.blocks);
 	}
+	LowerTessellationMemory(ir, options);
 	IR::BuildSrtPlan(ir);
 	IR::EliminateDeadCode(ir.blocks);
 	IR::TrackResources(ir);

src/graphics/shader/recompiler/Tessellation.cpp

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

src/graphics/shader/recompiler/Tessellation.h

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

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

@@ -15,8 +15,8 @@ static void AppendInstructionWords(std::vector<uint32_t>& section, const uint32_
 	if (words_num == 0) {
 		return;
 	}
-	const auto opcode     = words[0];
-	const auto word_count = static_cast<uint32_t>(words_num);
+	const uint32_t opcode     = words[0];
+	const auto     word_count = static_cast<uint32_t>(words_num);
 	section.push_back((word_count << spv::WordCountShift) | opcode);
 	section.insert(section.end(), words + 1, words + words_num);
 }
@@ -38,7 +38,7 @@ void Builder::RequireVersion(uint32_t version) {
 
 void Builder::RequireCapability(spv::Capability capability) {
 	if (m_required_capabilities.insert(capability).second) {
-		AppendInstruction(m_capabilities, spv::OpCapability, {capability});
+		AppendInstruction(m_capabilities, spv::OpCapability, capability);
 	}
 }
 
@@ -62,37 +62,25 @@ uint32_t Builder::Import(const char* name) {
 	return id;
 }
 
-uint32_t Builder::Type(spv::Op opcode, std::initializer_list<uint32_t> operands) {
-	return Type(opcode, std::vector<uint32_t>(operands));
-}
-
-uint32_t Builder::Type(spv::Op opcode, const std::vector<uint32_t>& operands) {
-	std::vector<uint32_t> key;
-	key.reserve(operands.size() + 2u);
-	key.push_back(opcode);
-	key.push_back(static_cast<uint32_t>(operands.size()));
-	key.insert(key.end(), operands.begin(), operands.end());
+uint32_t Builder::DeclareType(spv::Op opcode, std::vector<uint32_t> key) {
 	if (const auto it = m_declaration_ids.find(key); it != m_declaration_ids.end()) {
 		return it->second;
 	}
 	const auto id = AllocateId();
+	AppendInstruction(m_declarations, opcode, id, std::span<const uint32_t>(key).subspan(2));
 	m_declaration_ids.emplace(std::move(key), id);
-	std::vector<uint32_t> words {opcode, id};
-	words.insert(words.end(), operands.begin(), operands.end());
-	AppendInstructionWords(m_declarations, words.data(), words.size());
 	return id;
 }
 
-uint32_t Builder::DecoratedType(spv::Op opcode, std::initializer_list<uint32_t> operands,
-                                std::initializer_list<TypeAnnotation> annotations) {
+uint32_t Builder::DeclareDecoratedType(spv::Op opcode, std::vector<uint32_t> key,
+                                       std::initializer_list<TypeAnnotation> annotations) {
 	if (annotations.size() == 0) {
-		return Type(opcode, operands);
+		return DeclareType(opcode, std::move(key));
 	}
-	std::vector<uint32_t> key {opcode, static_cast<uint32_t>(operands.size())};
-	key.insert(key.end(), operands.begin(), operands.end());
+	const auto operand_count = key[1];
 	key.push_back(static_cast<uint32_t>(annotations.size()));
 	for (const auto& annotation: annotations) {
-		key.push_back(annotation.opcode);
+		AppendOperand(key, annotation.opcode);
 		key.push_back(static_cast<uint32_t>(annotation.operands.size()));
 		key.insert(key.end(), annotation.operands.begin(), annotation.operands.end());
 	}
@@ -100,37 +88,23 @@ uint32_t Builder::DecoratedType(spv::Op opcode, std::initializer_list<uint32_t>
 		return it->second;
 	}
 	const auto id = AllocateId();
+	AppendInstruction(m_declarations, opcode, id,
+	                  std::span<const uint32_t>(key).subspan(2, operand_count));
 	m_declaration_ids.emplace(std::move(key), id);
-	std::vector<uint32_t> words {opcode, id};
-	words.insert(words.end(), operands.begin(), operands.end());
-	AppendInstructionWords(m_declarations, words.data(), words.size());
 	for (const auto& annotation: annotations) {
-		words = {annotation.opcode, id};
-		words.insert(words.end(), annotation.operands.begin(), annotation.operands.end());
-		AppendInstructionWords(m_annotations, words.data(), words.size());
+		AppendInstruction(m_annotations, annotation.opcode, id, annotation.operands);
 	}
 	return id;
 }
 
-uint32_t Builder::Constant(spv::Op opcode, uint32_t type,
-                           std::initializer_list<uint32_t> operands) {
-	return Constant(opcode, type, std::vector<uint32_t>(operands));
-}
-
-uint32_t Builder::Constant(spv::Op opcode, uint32_t type, const std::vector<uint32_t>& operands) {
-	std::vector<uint32_t> key;
-	key.reserve(operands.size() + 2u);
-	key.push_back(opcode);
-	key.push_back(type);
-	key.insert(key.end(), operands.begin(), operands.end());
+uint32_t Builder::DeclareConstant(spv::Op opcode, std::vector<uint32_t> key) {
 	if (const auto it = m_declaration_ids.find(key); it != m_declaration_ids.end()) {
 		return it->second;
 	}
 	const auto id = AllocateId();
+	AppendInstruction(m_declarations, opcode, key[1], id,
+	                  std::span<const uint32_t>(key).subspan(2));
 	m_declaration_ids.emplace(std::move(key), id);
-	std::vector<uint32_t> words {opcode, type, id};
-	words.insert(words.end(), operands.begin(), operands.end());
-	AppendInstructionWords(m_declarations, words.data(), words.size());
 	return id;
 }
 
@@ -142,7 +116,7 @@ uint32_t Builder::DefineGlobalVariable(uint32_t pointer_type, spv::StorageClass
 
 void Builder::DefineGlobalVariable(uint32_t id, uint32_t pointer_type,
                                    spv::StorageClass storage_class) {
-	AppendInstruction(m_declarations, spv::OpVariable, {pointer_type, id, storage_class});
+	AppendInstruction(m_declarations, spv::OpVariable, pointer_type, id, storage_class);
 }
 
 void Builder::AppendString(std::vector<uint32_t>& words, const char* text) {
@@ -161,27 +135,14 @@ void Builder::AppendString(std::vector<uint32_t>& words, const char* text) {
 	}
 }
 
-void Builder::AppendInstruction(std::vector<uint32_t>& section, spv::Op opcode,
-                                const std::vector<uint32_t>& operands) {
-	const uint32_t word_count = static_cast<uint32_t>(operands.size() + 1u);
-	section.push_back((word_count << spv::WordCountShift) | opcode);
-	section.insert(section.end(), operands.begin(), operands.end());
-}
-
-void Builder::AppendInstruction(std::vector<uint32_t>& section, spv::Op opcode,
-                                std::initializer_list<uint32_t> operands) {
-	const uint32_t word_count = static_cast<uint32_t>(operands.size() + 1u);
-	section.push_back((word_count << spv::WordCountShift) | opcode);
-	section.insert(section.end(), operands.begin(), operands.end());
-}
-
-void Builder::AddMemoryModel(std::initializer_list<uint32_t> operands) {
-	AppendInstruction(m_memory_model, spv::OpMemoryModel, operands);
+void Builder::AddMemoryModel(spv::AddressingModel addressing_model, spv::MemoryModel memory_model) {
+	AppendInstruction(m_memory_model, spv::OpMemoryModel, addressing_model, memory_model);
 }
 
 void Builder::AddEntryPoint(spv::ExecutionModel execution_model, uint32_t entry_point,
                             const char* name, const std::vector<uint32_t>& interfaces) {
-	std::vector<uint32_t> operands = {execution_model, entry_point};
+	std::vector<uint32_t> operands;
+	AppendOperands(operands, execution_model, entry_point);
 	AppendString(operands, name);
 	operands.insert(operands.end(), interfaces.begin(), interfaces.end());
 	if (m_version >= 0x00010400u) {
@@ -199,25 +160,13 @@ void Builder::AddEntryPoint(spv::ExecutionModel execution_model, uint32_t entry_
 	AppendInstruction(m_entry_points, spv::OpEntryPoint, operands);
 }
 
-void Builder::AddExecutionMode(std::initializer_list<uint32_t> operands) {
-	AppendInstruction(m_execution_modes, spv::OpExecutionMode, operands);
-}
-
 void Builder::AddName(uint32_t target, const char* name) {
 	std::vector<uint32_t> operands = {target};
 	AppendString(operands, name);
 	AppendInstruction(m_debug, spv::OpName, operands);
 }
 
-void Builder::AddAnnotation(std::initializer_list<uint32_t> words) {
-	AppendInstructionWords(m_annotations, words.begin(), words.size());
-}
-
-void Builder::AddFunction(std::initializer_list<uint32_t> words) {
-	AppendInstructionWords(m_functions, words.begin(), words.size());
-}
-
-void Builder::AddFunction(const std::vector<uint32_t>& words) {
+void Builder::AddFunction(std::span<const uint32_t> words) {
 	AppendInstructionWords(m_functions, words.data(), words.size());
 }
 

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

@@ -4,17 +4,21 @@
 #include "common/common.h"
 
 #include <initializer_list>
+#include <iterator>
 #include <map>
 #include <set>
+#include <span>
 #include <spirv/unified1/spirv.hpp>
 #include <string>
+#include <type_traits>
+#include <utility>
 #include <vector>
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv {
 
 struct TypeAnnotation {
 	spv::Op               opcode = spv::OpNop;
-	std::vector<uint32_t> operands;
+	std::vector<uint32_t> operands; // Already encoded operand words.
 };
 
 struct DeferredPhi {
@@ -32,33 +36,109 @@ public:
 	void     RequireCapability(spv::Capability capability);
 	void     RequireExtension(const char* name);
 	uint32_t Import(const char* name);
-	uint32_t Type(spv::Op opcode, std::initializer_list<uint32_t> operands = {});
-	uint32_t Type(spv::Op opcode, const std::vector<uint32_t>& operands);
-	uint32_t DecoratedType(spv::Op opcode, std::initializer_list<uint32_t> operands,
-	                       std::initializer_list<TypeAnnotation> annotations);
-	uint32_t Constant(spv::Op opcode, uint32_t type, std::initializer_list<uint32_t> operands = {});
-	uint32_t Constant(spv::Op opcode, uint32_t type, const std::vector<uint32_t>& operands);
+	template <typename... Args>
+	uint32_t Type(spv::Op opcode, const Args&... operands) {
+		return DeclareType(opcode, MakeTypeKey(opcode, operands...));
+	}
+
+	template <typename... Args>
+	uint32_t DecoratedType(spv::Op opcode, std::initializer_list<TypeAnnotation> annotations,
+	                       const Args&... operands) {
+		return DeclareDecoratedType(opcode, MakeTypeKey(opcode, operands...), annotations);
+	}
+
+	template <typename... Args>
+	uint32_t Constant(spv::Op opcode, uint32_t type, const Args&... operands) {
+		std::vector<uint32_t> key;
+		key.reserve(2u + (0u + ... + OperandWordCount(operands)));
+		AppendOperands(key, opcode, type, operands...);
+		return DeclareConstant(opcode, std::move(key));
+	}
+
 	uint32_t DefineGlobalVariable(uint32_t pointer_type, spv::StorageClass storage_class);
 	void DefineGlobalVariable(uint32_t id, uint32_t pointer_type, spv::StorageClass storage_class);
 
-	void        AddMemoryModel(std::initializer_list<uint32_t> operands);
+	void AddMemoryModel(spv::AddressingModel addressing_model, spv::MemoryModel memory_model);
 	void AddEntryPoint(spv::ExecutionModel execution_model, uint32_t entry_point, const char* name,
 	                   const std::vector<uint32_t>& interfaces);
-	void        AddExecutionMode(std::initializer_list<uint32_t> operands);
-	void        AddName(uint32_t target, const char* name);
-	void        AddAnnotation(std::initializer_list<uint32_t> words);
-	void        AddFunction(std::initializer_list<uint32_t> words);
-	void        AddFunction(const std::vector<uint32_t>& words);
+	template <typename... Args>
+	void AddExecutionMode(uint32_t entry_point, spv::ExecutionMode mode, const Args&... operands) {
+		AppendInstruction(m_execution_modes, spv::OpExecutionMode, entry_point, mode, operands...);
+	}
+
+	void AddName(uint32_t target, const char* name);
+	template <typename... Args>
+	void AddAnnotation(spv::Op opcode, const Args&... operands) {
+		AppendInstruction(m_annotations, opcode, operands...);
+	}
+
+	template <typename... Args>
+	void AddFunction(spv::Op opcode, const Args&... operands) {
+		AppendInstruction(m_functions, opcode, operands...);
+	}
+
+	void        AddFunction(std::span<const uint32_t> words);
 	DeferredPhi AddDeferredPhi(uint32_t type, uint32_t result, size_t incoming_count);
 	void        PatchDeferredPhi(DeferredPhi phi, size_t incoming, uint32_t value, uint32_t parent);
 
 	[[nodiscard]] std::vector<uint32_t> Build() const;
 
 private:
+	static void AppendOperand(std::vector<uint32_t>& words, uint32_t value) {
+		words.push_back(value);
+	}
+
+	static void AppendOperand(std::vector<uint32_t>& words, int32_t value) {
+		words.push_back(static_cast<uint32_t>(value));
+	}
+
+	template <typename T>
+	requires std::is_enum_v<T>
+	static void AppendOperand(std::vector<uint32_t>& words, T value) {
+		static_assert(sizeof(T) == sizeof(uint32_t));
+		words.push_back(static_cast<uint32_t>(value));
+	}
+
+	static void AppendOperand(std::vector<uint32_t>& words, std::span<const uint32_t> values) {
+		words.insert(words.end(), values.begin(), values.end());
+	}
+
+	template <typename... Args>
+	static void AppendOperands(std::vector<uint32_t>& words, const Args&... operands) {
+		(AppendOperand(words, operands), ...);
+	}
+
+	template <typename T>
+	static size_t OperandWordCount(const T& operand) {
+		if constexpr (std::is_integral_v<T> || std::is_enum_v<T>) {
+			return 1;
+		} else {
+			return std::size(operand);
+		}
+	}
+
+	template <typename... Args>
+	static std::vector<uint32_t> MakeTypeKey(spv::Op opcode, const Args&... operands) {
+		const auto operand_count = static_cast<uint32_t>((0u + ... + OperandWordCount(operands)));
+		std::vector<uint32_t> key;
+		key.reserve(2u + operand_count);
+		AppendOperands(key, opcode, operand_count, operands...);
+		return key;
+	}
+
+	template <typename... Args>
 	static void AppendInstruction(std::vector<uint32_t>& section, spv::Op opcode,
-	                              const std::vector<uint32_t>& operands);
-	static void AppendInstruction(std::vector<uint32_t>& section, spv::Op opcode,
-	                              std::initializer_list<uint32_t> operands);
+	                              const Args&... operands) {
+		const auto offset = section.size();
+		AppendOperands(section, opcode, operands...);
+		const auto word_count = static_cast<uint32_t>(section.size() - offset);
+		section[offset] |= word_count << spv::WordCountShift;
+	}
+
+	uint32_t    DeclareType(spv::Op opcode, std::vector<uint32_t> key);
+	uint32_t    DeclareDecoratedType(spv::Op opcode, std::vector<uint32_t> key,
+	                                 std::initializer_list<TypeAnnotation> annotations);
+	uint32_t    DeclareConstant(spv::Op opcode, std::vector<uint32_t> key);
 	static void AppendString(std::vector<uint32_t>& words, const char* text);
 
 	uint32_t                                  m_next_id = 1;

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

@@ -277,7 +277,8 @@ Emitter::SpirvRequirements Emitter::AnalyzeProgramRequirements(const IR::Program
 					break;
 				}
 				case IR::ValueOpcode::LaneId:
-					requirements.subgroup_local_invocation_id = true;
+					requirements.subgroup_local_invocation_id |=
+					    program.stage != ShaderType::TessellationControl;
 					break;
 				case IR::ValueOpcode::ImageQueryLod: requirements.compute_derivatives = true; break;
 				case IR::ValueOpcode::ImageGatherRaw:
@@ -306,8 +307,10 @@ std::vector<uint32_t> EmitProgram(const IR::Program& program,
 	using namespace Emitter;
 
 	if (program.stage != ShaderType::Compute && program.stage != ShaderType::Vertex &&
-	    program.stage != ShaderType::Pixel && program.stage != ShaderType::Mesh) {
-		Fail(program, "binary SPIR-V emitter supports compute, vertex, and pixel shaders");
+	    program.stage != ShaderType::Pixel && program.stage != ShaderType::Mesh &&
+	    program.stage != ShaderType::Local && program.stage != ShaderType::TessellationControl &&
+	    program.stage != ShaderType::TessellationEvaluation) {
+		Fail(program, "binary SPIR-V emitter received an unsupported shader stage");
 	}
 	if (!program.srt_plan_complete || !program.resource_tracking_complete ||
 	    !program.shader_info_complete || !program.binding_layout_complete) {

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

@@ -12,14 +12,14 @@ struct Pair {
 
 Pair ExtractPair(EmitterState& state, uint32_t value) {
 	Pair result {state.builder.AllocateId(), state.builder.AllocateId()};
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), result.low, value, 0});
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), result.high, value, 1});
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), result.low, value, 0);
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), result.high, value, 1);
 	return result;
 }
 
 uint32_t MakePair(EmitterState& state, uint32_t low, uint32_t high) {
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpCompositeConstruct, TypeU64(state), result, low, high});
+	state.builder.AddFunction(spv::OpCompositeConstruct, TypeU64(state), result, low, high);
 	return result;
 }
 
@@ -31,7 +31,7 @@ uint32_t CompareEqual64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_va
 }
 
 uint32_t CompareOrdered64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value,
-                          uint32_t high_compare, uint32_t low_compare) {
+                          spv::Op high_compare, spv::Op low_compare) {
 	const auto lhs         = ExtractPair(state, lhs_value);
 	const auto rhs         = ExtractPair(state, rhs_value);
 	const auto high_equal  = Binary(state, spv::OpIEqual, TypeBool(state), lhs.high, rhs.high);
@@ -51,10 +51,10 @@ uint32_t EmitMulHigh(EmitterState& state, uint32_t lhs, uint32_t rhs, bool signe
 		rhs_operand = Unary(state, spv::OpBitcast, TypeI32(state), rhs);
 	}
 	const auto extended = state.builder.AllocateId();
-	state.builder.AddFunction({signed_value ? spv::OpSMulExtended : spv::OpUMulExtended, pair_type,
-	                           extended, lhs_operand, rhs_operand});
+	state.builder.AddFunction(signed_value ? spv::OpSMulExtended : spv::OpUMulExtended, pair_type,
+	                          extended, lhs_operand, rhs_operand);
 	const auto high = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpCompositeExtract, operand_type, high, extended, 1});
+	state.builder.AddFunction(spv::OpCompositeExtract, operand_type, high, extended, 1);
 	return signed_value ? Unary(state, spv::OpBitcast, TypeU32(state), high) : high;
 }
 
@@ -96,7 +96,7 @@ uint32_t EmitShift64(EmitterState& state, spv::Op opcode, uint32_t value, uint32
 	                Select(state, TypeU32(state), at_least_32, fill, high));
 }
 
-uint32_t EmitConstantShift64(EmitterState& state, uint32_t opcode, uint32_t value, uint32_t shift) {
+uint32_t EmitConstantShift64(EmitterState& state, spv::Op opcode, uint32_t value, uint32_t shift) {
 	shift &= 63u;
 	if (shift == 0u) {
 		return value;
@@ -173,10 +173,10 @@ uint32_t EmitF32ToU32(EmitterState& state, uint32_t src, bool signed_value) {
 	const auto converted_raw = state.builder.AllocateId();
 	if (signed_value) {
 		const auto converted_i = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpConvertFToS, TypeI32(state), converted_i, trunc});
-		state.builder.AddFunction({spv::OpBitcast, TypeU32(state), converted_raw, converted_i});
+		state.builder.AddFunction(spv::OpConvertFToS, TypeI32(state), converted_i, trunc);
+		state.builder.AddFunction(spv::OpBitcast, TypeU32(state), converted_raw, converted_i);
 	} else {
-		state.builder.AddFunction({spv::OpConvertFToU, TypeU32(state), converted_raw, trunc});
+		state.builder.AddFunction(spv::OpConvertFToU, TypeU32(state), converted_raw, trunc);
 	}
 	const auto nan = EmitClassifyF32(state, src).nan;
 	if (signed_value) {
@@ -218,10 +218,10 @@ uint32_t EmitFindUMsb64(EmitterState& state, uint32_t value) {
 	const auto pair   = ExtractPair(state, value);
 	const auto high_i = state.builder.AllocateId();
 	const auto low_i  = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpExtInst, TypeI32(state), high_i, GlslStd450(state), GLSLstd450FindUMsb, pair.high});
-	state.builder.AddFunction(
-	    {spv::OpExtInst, TypeI32(state), low_i, GlslStd450(state), GLSLstd450FindUMsb, pair.low});
+	state.builder.AddFunction(spv::OpExtInst, TypeI32(state), high_i, GlslStd450(state),
+	                          GLSLstd450FindUMsb, pair.high);
+	state.builder.AddFunction(spv::OpExtInst, TypeI32(state), low_i, GlslStd450(state),
+	                          GLSLstd450FindUMsb, pair.low);
 	const auto high = Unary(state, spv::OpBitcast, TypeU32(state), high_i);
 	const auto low  = Unary(state, spv::OpBitcast, TypeU32(state), low_i);
 	const auto high_nonzero =
@@ -260,9 +260,9 @@ uint32_t EmitIAdd64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value)
 	const auto low_pair = state.builder.AllocateId();
 	const auto low      = state.builder.AllocateId();
 	const auto carry    = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpIAddCarry, TypeU32Pair(state), low_pair, lhs.low, rhs.low});
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), low, low_pair, 0});
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), carry, low_pair, 1});
+	state.builder.AddFunction(spv::OpIAddCarry, TypeU32Pair(state), low_pair, lhs.low, rhs.low);
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), low, low_pair, 0);
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), carry, low_pair, 1);
 	const auto high0 = Binary(state, spv::OpIAdd, TypeU32(state), lhs.high, rhs.high);
 	const auto high  = Binary(state, spv::OpIAdd, TypeU32(state), high0, carry);
 	return MakePair(state, low, high);
@@ -278,8 +278,8 @@ uint32_t EmitConvertU8U32(EmitterState& state, uint32_t arg0) {
 
 uint32_t EmitConvertF16F32(EmitterState& state, uint32_t arg0) {
 	const auto pair = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpCompositeConstruct, TypeF32Vector(state, 2), pair, arg0, ConstantF32(state, 0)});
+	state.builder.AddFunction(spv::OpCompositeConstruct, TypeF32Vector(state, 2), pair, arg0,
+	                          ConstantF32(state, 0));
 	return EmitPackHalf2x16(state, pair);
 }
 

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

@@ -4,35 +4,35 @@ namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
 uint32_t EmitAndConstant(EmitterState& state, uint32_t value, uint32_t mask) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), ret, value, ConstantU32(state, mask)});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), ret, value,
+	                          ConstantU32(state, mask));
 	return ret;
 }
 
 uint32_t EmitShiftRightConstant(EmitterState& state, uint32_t value, uint32_t shift) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpShiftRightLogical, TypeU32(state), ret, value, ConstantU32(state, shift)});
+	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), ret, value,
+	                          ConstantU32(state, shift));
 	return ret;
 }
 
-uint32_t EmitCompareU32Constant(EmitterState& state, uint32_t opcode, uint32_t value,
+uint32_t EmitCompareU32Constant(EmitterState& state, spv::Op opcode, uint32_t value,
                                 uint32_t constant) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, TypeBool(state), ret, value, ConstantU32(state, constant)});
+	state.builder.AddFunction(opcode, TypeBool(state), ret, value, ConstantU32(state, constant));
 	return ret;
 }
 
 uint32_t EmitSubConstantMinusU32(EmitterState& state, uint32_t constant, uint32_t value) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpISub, TypeU32(state), ret, ConstantU32(state, constant), value});
+	state.builder.AddFunction(spv::OpISub, TypeU32(state), ret, ConstantU32(state, constant),
+	                          value);
 	return ret;
 }
 
 uint32_t EmitF32ToF16RtzBits(EmitterState& state, uint32_t f32) {
 	const auto bits = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, f32});
+	state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, f32);
 
 	const auto sign = EmitAndConstant(state, EmitShiftRightConstant(state, bits, 16), 0x8000u);
 	const auto exp  = EmitAndConstant(state, EmitShiftRightConstant(state, bits, 23), 0xffu);
@@ -43,13 +43,12 @@ uint32_t EmitF32ToF16RtzBits(EmitterState& state, uint32_t f32) {
 	const auto normal_mant    = EmitShiftRightConstant(state, mant, 13);
 	const auto normal_payload = state.builder.AllocateId();
 	const auto normal         = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpISub, TypeU32(state), half_exp, exp, ConstantU32(state, 112)});
-	state.builder.AddFunction(
-	    {spv::OpShiftLeftLogical, TypeU32(state), normal_exp, half_exp, ConstantU32(state, 10)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseOr, TypeU32(state), normal_payload, normal_exp, normal_mant});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), normal, sign, normal_payload});
+	state.builder.AddFunction(spv::OpISub, TypeU32(state), half_exp, exp, ConstantU32(state, 112));
+	state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), normal_exp, half_exp,
+	                          ConstantU32(state, 10));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), normal_payload, normal_exp,
+	                          normal_mant);
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), normal, sign, normal_payload);
 
 	const auto mant_with_hidden = EmitOrU32(state, mant, ConstantU32(state, 0x00800000u));
 	const auto raw_sub_shift    = EmitSubConstantMinusU32(state, 126, exp);
@@ -61,9 +60,9 @@ uint32_t EmitF32ToF16RtzBits(EmitterState& state, uint32_t f32) {
 	    EmitSelectValueU32(state, exp_gt_112, ConstantU32(state, 14), sub_shift_low);
 	const auto sub_mant  = state.builder.AllocateId();
 	const auto subnormal = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpShiftRightLogical, TypeU32(state), sub_mant, mant_with_hidden, sub_shift});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), subnormal, sign, sub_mant});
+	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), sub_mant, mant_with_hidden,
+	                          sub_shift);
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), subnormal, sign, sub_mant);
 
 	const auto nan_payload =
 	    EmitOrU32(state, EmitShiftRightConstant(state, mant, 13), ConstantU32(state, 0x0200u));
@@ -86,18 +85,18 @@ uint32_t EmitF32ToF16RtzBits(EmitterState& state, uint32_t f32) {
 uint32_t EmitMinMaxU32Value(EmitterState& state, uint32_t lhs, uint32_t rhs, bool max_value) {
 	const auto cond = state.builder.AllocateId();
 	const auto ret  = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {max_value ? spv::OpUGreaterThan : spv::OpULessThan, TypeBool(state), cond, lhs, rhs});
-	state.builder.AddFunction({spv::OpSelect, TypeU32(state), ret, cond, lhs, rhs});
+	state.builder.AddFunction(max_value ? spv::OpUGreaterThan : spv::OpULessThan, TypeBool(state),
+	                          cond, lhs, rhs);
+	state.builder.AddFunction(spv::OpSelect, TypeU32(state), ret, cond, lhs, rhs);
 	return ret;
 }
 
 uint32_t EmitMinMaxI32Value(EmitterState& state, uint32_t lhs, uint32_t rhs, bool max_value) {
 	const auto cond = state.builder.AllocateId();
 	const auto ret  = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {max_value ? spv::OpSGreaterThan : spv::OpSLessThan, TypeBool(state), cond, lhs, rhs});
-	state.builder.AddFunction({spv::OpSelect, TypeU32(state), ret, cond, lhs, rhs});
+	state.builder.AddFunction(max_value ? spv::OpSGreaterThan : spv::OpSLessThan, TypeBool(state),
+	                          cond, lhs, rhs);
+	state.builder.AddFunction(spv::OpSelect, TypeU32(state), ret, cond, lhs, rhs);
 	return ret;
 }
 
@@ -184,8 +183,8 @@ uint32_t EmitMinMaxF32Value(EmitterState& state, uint32_t lhs, uint32_t rhs, boo
 	const auto rhs_class = EmitClassifyF32(state, rhs);
 
 	const auto numeric_cond = state.builder.AllocateId();
-	state.builder.AddFunction({max_value ? spv::OpFOrdGreaterThanEqual : spv::OpFOrdLessThan,
-	                           TypeBool(state), numeric_cond, lhs, rhs});
+	state.builder.AddFunction(max_value ? spv::OpFOrdGreaterThanEqual : spv::OpFOrdLessThan,
+	                          TypeBool(state), numeric_cond, lhs, rhs);
 	const auto ordered_bits =
 	    EmitSelectValueU32(state, numeric_cond, lhs_class.bits, rhs_class.bits);
 
@@ -209,18 +208,18 @@ uint32_t EmitFlushF32DenormToSignedZero(EmitterState& state, uint32_t value) {
 	const auto flush     = state.builder.AllocateId();
 	const auto selected  = state.builder.AllocateId();
 	const auto ret       = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, value});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), abs_bits, bits, ConstantU32(state, 0x7fffffffu)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), sign_bits, bits, ConstantU32(state, 0x80000000u)});
-	state.builder.AddFunction(
-	    {spv::OpINotEqual, TypeBool(state), non_zero, abs_bits, ConstantU32(state, 0)});
-	state.builder.AddFunction(
-	    {spv::OpULessThan, TypeBool(state), subnormal, abs_bits, ConstantU32(state, 0x00800000u)});
-	state.builder.AddFunction({spv::OpLogicalAnd, TypeBool(state), flush, non_zero, subnormal});
-	state.builder.AddFunction({spv::OpSelect, TypeU32(state), selected, flush, sign_bits, bits});
-	state.builder.AddFunction({spv::OpBitcast, TypeF32(state), ret, selected});
+	state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), abs_bits, bits,
+	                          ConstantU32(state, 0x7fffffffu));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), sign_bits, bits,
+	                          ConstantU32(state, 0x80000000u));
+	state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), non_zero, abs_bits,
+	                          ConstantU32(state, 0));
+	state.builder.AddFunction(spv::OpULessThan, TypeBool(state), subnormal, abs_bits,
+	                          ConstantU32(state, 0x00800000u));
+	state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), flush, non_zero, subnormal);
+	state.builder.AddFunction(spv::OpSelect, TypeU32(state), selected, flush, sign_bits, bits);
+	state.builder.AddFunction(spv::OpBitcast, TypeF32(state), ret, selected);
 	return ret;
 }
 
@@ -232,35 +231,35 @@ uint32_t EmitTrigCycleF32(EmitterState& state, uint32_t src, bool preserve_signe
 	const auto finite       = state.builder.AllocateId();
 	const auto large_finite = state.builder.AllocateId();
 	const auto reduced      = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpExtInst, TypeF32(state), fract, GlslStd450(state), GLSLstd450Fract, src});
-	state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, src});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), abs_bits, bits, ConstantU32(state, 0x7fffffffu)});
-	state.builder.AddFunction({spv::OpUGreaterThanEqual, TypeBool(state), large, abs_bits,
-	                           ConstantU32(state, 0x4b000000u)});
-	state.builder.AddFunction(
-	    {spv::OpULessThan, TypeBool(state), finite, abs_bits, ConstantU32(state, 0x7f800000u)});
-	state.builder.AddFunction({spv::OpLogicalAnd, TypeBool(state), large_finite, large, finite});
-	state.builder.AddFunction(
-	    {spv::OpSelect, TypeF32(state), reduced, large_finite, ConstantF32(state, 0), fract});
+	state.builder.AddFunction(spv::OpExtInst, TypeF32(state), fract, GlslStd450(state),
+	                          GLSLstd450Fract, src);
+	state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, src);
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), abs_bits, bits,
+	                          ConstantU32(state, 0x7fffffffu));
+	state.builder.AddFunction(spv::OpUGreaterThanEqual, TypeBool(state), large, abs_bits,
+	                          ConstantU32(state, 0x4b000000u));
+	state.builder.AddFunction(spv::OpULessThan, TypeBool(state), finite, abs_bits,
+	                          ConstantU32(state, 0x7f800000u));
+	state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), large_finite, large, finite);
+	state.builder.AddFunction(spv::OpSelect, TypeF32(state), reduced, large_finite,
+	                          ConstantF32(state, 0), fract);
 	if (!preserve_signed_zero) {
 		return reduced;
 	}
 	const auto zero = state.builder.AllocateId();
 	const auto ret  = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpIEqual, TypeBool(state), zero, abs_bits, ConstantU32(state, 0)});
-	state.builder.AddFunction({spv::OpSelect, TypeF32(state), ret, zero, src, reduced});
+	state.builder.AddFunction(spv::OpIEqual, TypeBool(state), zero, abs_bits,
+	                          ConstantU32(state, 0));
+	state.builder.AddFunction(spv::OpSelect, TypeF32(state), ret, zero, src, reduced);
 	return ret;
 }
 
 uint32_t EmitF16BitsToF32(EmitterState& state, uint32_t bits) {
 	const auto unpacked = state.builder.AllocateId();
 	const auto ret      = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpExtInst, TypeF32Vector(state, 2), unpacked, GlslStd450(state),
-	                           GLSLstd450UnpackHalf2x16, bits});
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeF32(state), ret, unpacked, 0});
+	state.builder.AddFunction(spv::OpExtInst, TypeF32Vector(state, 2), unpacked, GlslStd450(state),
+	                          GLSLstd450UnpackHalf2x16, bits);
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), ret, unpacked, 0);
 	return ret;
 }
 

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

@@ -82,8 +82,8 @@ uint32_t DescriptorElementPointer(EmitterState& state, uint32_t result_ptr_type,
 		ExitDescriptorBindingFailure(state, kind, resource, variable_name);
 	}
 	const auto pointer = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpAccessChain, result_ptr_type, pointer, variable_id,
-	                           ConstantU32(state, array_index)});
+	state.builder.AddFunction(spv::OpAccessChain, result_ptr_type, pointer, variable_id,
+	                          ConstantU32(state, array_index));
 	return pointer;
 }
 
@@ -109,8 +109,7 @@ uint32_t ImageScalarType(EmitterState& state, Prospero::TextureNumericClass nume
 
 uint32_t ImageVectorType(EmitterState& state, Prospero::TextureNumericClass numeric_class,
                          uint32_t components) {
-	return state.builder.Type(spv::OpTypeVector,
-	                          {ImageScalarType(state, numeric_class), components});
+	return state.builder.Type(spv::OpTypeVector, ImageScalarType(state, numeric_class), components);
 }
 
 uint32_t ImageType(EmitterState& state, const IR::ImageResource& image) {
@@ -131,10 +130,9 @@ uint32_t ImageType(EmitterState& state, const IR::ImageResource& image) {
 		EXIT("invalid image resource class");
 	}
 	const auto& info = ImageDimensionInfoFor(image.dimension);
-	return state.builder.Type(spv::OpTypeImage,
-	                          {ImageScalarType(state, image.numeric_class), info.spirv_dimension,
-	                           image.depth_compare ? 1u : 0u, info.arrayed, info.multisampled,
-	                           sampled, format});
+	return state.builder.Type(spv::OpTypeImage, ImageScalarType(state, image.numeric_class),
+	                          info.spirv_dimension, image.depth_compare ? 1u : 0u, info.arrayed,
+	                          info.multisampled, sampled, format);
 }
 
 uint32_t ImageViewSizeType(EmitterState& state, ImageDimension dimension) {
@@ -154,12 +152,12 @@ uint32_t LoadSampledImageDescriptor(EmitterState& state, uint32_t resource) {
 	const auto array_index  = ResourceForDescriptor(state, *kind, resource);
 	const auto variable     = state.image_variables[IR::ImageBindingIndex(*kind)];
 	const auto pointer_type = state.builder.Type(
-	    spv::OpTypePointer, {spv::StorageClassUniformConstant, ImageType(state, image_resource)});
+	    spv::OpTypePointer, spv::StorageClassUniformConstant, ImageType(state, image_resource));
 	const auto pointer =
 	    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({spv::OpLoad, ImageType(state, image_resource), image, pointer});
+	state.builder.AddFunction(spv::OpLoad, ImageType(state, image_resource), image, pointer);
 	return image;
 }
 
@@ -168,12 +166,12 @@ uint32_t LoadSamplerDescriptor(EmitterState& state, uint32_t sampler) {
 	    ResourceForDescriptor(state, IR::DescriptorBindingKind::Samplers, sampler);
 	const auto sampler_type = state.builder.Type(spv::OpTypeSampler);
 	const auto pointer_type =
-	    state.builder.Type(spv::OpTypePointer, {spv::StorageClassUniformConstant, sampler_type});
+	    state.builder.Type(spv::OpTypePointer, spv::StorageClassUniformConstant, sampler_type);
 	const auto pointer = DescriptorElementPointer(
 	    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({spv::OpLoad, sampler_type, sampler_id, pointer});
+	state.builder.AddFunction(spv::OpLoad, sampler_type, sampler_id, pointer);
 	return sampler_id;
 }
 
@@ -183,9 +181,8 @@ uint32_t MakeSampledImage(EmitterState& state, uint32_t resource, uint32_t sampl
 	const auto  sampler_id     = LoadSamplerDescriptor(state, sampler);
 	const auto  sampled_image = state.builder.AllocateId();
 	const auto  sampled_type =
-	    state.builder.Type(spv::OpTypeSampledImage, {ImageType(state, image_resource)});
-	state.builder.AddFunction(
-	    {spv::OpSampledImage, sampled_type, sampled_image, image, sampler_id});
+	    state.builder.Type(spv::OpTypeSampledImage, ImageType(state, image_resource));
+	state.builder.AddFunction(spv::OpSampledImage, sampled_type, sampled_image, image, sampler_id);
 	return sampled_image;
 }
 
@@ -194,9 +191,9 @@ uint32_t StorageImageDescriptorPointer(EmitterState& state, uint32_t 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 array_index  = ResourceForDescriptor(state, *kind, resource);
 	const auto pointer_type = state.builder.Type(
-	    spv::OpTypePointer, {spv::StorageClassUniformConstant, ImageType(state, image)});
+	    spv::OpTypePointer, spv::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");
@@ -214,18 +211,18 @@ void EmitStorageImageWrite(EmitterState& state, uint32_t resource, uint32_t mip_
 	const auto array_index = ResourceForDescriptor(state, *kind, resource);
 	const auto image_type  = ImageType(state, image);
 	const auto pointer_type =
-	    state.builder.Type(spv::OpTypePointer, {spv::StorageClassUniformConstant, image_type});
+	    state.builder.Type(spv::OpTypePointer, spv::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({spv::OpLoad, image_type, descriptor, pointer});
+		state.builder.AddFunction(spv::OpLoad, image_type, descriptor, pointer);
 		return descriptor;
 	};
 	if (image.mip_mode != IR::ImageMipMode::DynamicStorage) {
-		state.builder.AddFunction({spv::OpImageWrite, LoadAt(array_index), coord, texel});
+		state.builder.AddFunction(spv::OpImageWrite, LoadAt(array_index), coord, texel);
 		return;
 	}
 	if (image.mip_count == 0u) {
@@ -241,19 +238,22 @@ void EmitStorageImageWrite(EmitterState& state, uint32_t resource, uint32_t mip_
 		words.push_back(mip);
 		words.push_back(labels[mip]);
 	}
-	state.builder.AddFunction({spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone});
+	state.builder.AddFunction(spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone);
 	state.builder.AddFunction(words);
 	for (uint32_t mip = 0; mip < image.mip_count; mip++) {
 		EmitLabel(state, labels[mip]);
-		state.builder.AddFunction({spv::OpImageWrite, LoadAt(array_index + mip), coord, texel});
-		state.builder.AddFunction({spv::OpBranch, merge_label});
+		state.builder.AddFunction(spv::OpImageWrite, LoadAt(array_index + mip), coord, texel);
+		state.builder.AddFunction(spv::OpBranch, merge_label);
 	}
 	EmitLabel(state, merge_label);
 }
 
 spv::ExecutionModel ExecutionModelForStage(ShaderType stage) {
 	switch (stage) {
+		case ShaderType::Local:
 		case ShaderType::Vertex: return spv::ExecutionModelVertex;
+		case ShaderType::TessellationControl: return spv::ExecutionModelTessellationControl;
+		case ShaderType::TessellationEvaluation: return spv::ExecutionModelTessellationEvaluation;
 		case ShaderType::Mesh: return spv::ExecutionModelMeshEXT; // MeshEXT
 		case ShaderType::Pixel: return spv::ExecutionModelFragment;
 		default: return spv::ExecutionModelGLCompute;

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

@@ -45,28 +45,30 @@ uint32_t EmitBuiltinU32(EmitterState& state, IR::StageInputKind kind, uint32_t c
 	if (kind == IR::StageInputKind::FrontFacing) {
 		const auto value = state.builder.AllocateId();
 		const auto bits  = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpLoad, TypeBool(state), value, variable});
-		state.builder.AddFunction({spv::OpSelect, TypeU32(state), bits, value,
-		                           ConstantU32(state, 1), ConstantU32(state, 0)});
+		state.builder.AddFunction(spv::OpLoad, TypeBool(state), value, variable);
+		// PS5 initializes v_front_face with float +1.0/-1.0 bits.
+		state.builder.AddFunction(spv::OpSelect, TypeU32(state), bits, value,
+		                          ConstantU32(state, 0x3f800000u), ConstantU32(state, 0xbf800000u));
 		return bits;
 	}
 	if (kind == IR::StageInputKind::VertexIndex || kind == IR::StageInputKind::InstanceIndex ||
+	    kind == IR::StageInputKind::InvocationId || kind == IR::StageInputKind::PrimitiveId ||
 	    kind == IR::StageInputKind::Layer || kind == IR::StageInputKind::SampleId) {
 		const auto value = state.builder.AllocateId();
 		const auto bits  = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpLoad, TypeI32(state), value, variable});
-		state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, value});
+		state.builder.AddFunction(spv::OpLoad, TypeI32(state), value, variable);
+		state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
 		return bits;
 	}
-	if (kind == IR::StageInputKind::FragCoord) {
+	if (kind == IR::StageInputKind::FragCoord || kind == IR::StageInputKind::TessCoord) {
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
 		const auto bits    = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpAccessChain,
-		                           TypePointer(state, spv::StorageClassInput, TypeF32(state)),
-		                           pointer, variable, ConstantU32(state, component)});
-		state.builder.AddFunction({spv::OpLoad, TypeF32(state), value, pointer});
-		state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, value});
+		state.builder.AddFunction(spv::OpAccessChain,
+		                          TypePointer(state, spv::StorageClassInput, TypeF32(state)),
+		                          pointer, variable, ConstantU32(state, component));
+		state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
+		state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
 		return bits;
 	}
 	if (kind == IR::StageInputKind::BaryCoordSmooth ||
@@ -74,11 +76,11 @@ uint32_t EmitBuiltinU32(EmitterState& state, IR::StageInputKind kind, uint32_t c
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
 		const auto bits    = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpAccessChain,
-		                           TypePointer(state, spv::StorageClassInput, TypeF32(state)),
-		                           pointer, variable, ConstantU32(state, component + 1u)});
-		state.builder.AddFunction({spv::OpLoad, TypeF32(state), value, pointer});
-		state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, value});
+		state.builder.AddFunction(spv::OpAccessChain,
+		                          TypePointer(state, spv::StorageClassInput, TypeF32(state)),
+		                          pointer, variable, ConstantU32(state, component + 1u));
+		state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
+		state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
 		return bits;
 	}
 	return EmitInputComponentU32(state, kind, component);
@@ -99,37 +101,37 @@ uint32_t EmitDppWriteCondition(ValueEmitContext& ctx, const IR::DppMoveFlags& fl
 	const auto bank_ok    = state.builder.AllocateId();
 	const auto row_ok     = state.builder.AllocateId();
 	const auto masks_ok   = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpShiftRightLogical, TypeU32(state), bank_shift, lane, ConstantU32(state, 2)});
-	state.builder.AddFunction(
-	    {spv::OpShiftRightLogical, TypeU32(state), row_shift, lane, ConstantU32(state, 4)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), bank, bank_shift, ConstantU32(state, 3)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), row, row_shift, ConstantU32(state, 3)});
-	state.builder.AddFunction(
-	    {spv::OpShiftLeftLogical, TypeU32(state), bank_bit, ConstantU32(state, 1), bank});
-	state.builder.AddFunction(
-	    {spv::OpShiftLeftLogical, TypeU32(state), row_bit, ConstantU32(state, 1), row});
-	state.builder.AddFunction({spv::OpBitwiseAnd, TypeU32(state), bank_hit,
-	                           ConstantU32(state, flags.bank_mask), bank_bit});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), row_hit, ConstantU32(state, flags.row_mask), row_bit});
-	state.builder.AddFunction(
-	    {spv::OpINotEqual, TypeBool(state), bank_ok, bank_hit, ConstantU32(state, 0)});
-	state.builder.AddFunction(
-	    {spv::OpINotEqual, TypeBool(state), row_ok, row_hit, ConstantU32(state, 0)});
-	state.builder.AddFunction({spv::OpLogicalAnd, TypeBool(state), masks_ok, bank_ok, row_ok});
+	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), bank_shift, lane,
+	                          ConstantU32(state, 2));
+	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), row_shift, lane,
+	                          ConstantU32(state, 4));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), bank, bank_shift,
+	                          ConstantU32(state, 3));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), row, row_shift,
+	                          ConstantU32(state, 3));
+	state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), bank_bit,
+	                          ConstantU32(state, 1), bank);
+	state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), row_bit,
+	                          ConstantU32(state, 1), row);
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), bank_hit,
+	                          ConstantU32(state, flags.bank_mask), bank_bit);
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), row_hit,
+	                          ConstantU32(state, flags.row_mask), row_bit);
+	state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), bank_ok, bank_hit,
+	                          ConstantU32(state, 0));
+	state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), row_ok, row_hit,
+	                          ConstantU32(state, 0));
+	state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), masks_ok, bank_ok, row_ok);
 	uint32_t writable = masks_ok;
 	if (!flags.bound_control) {
 		const auto target  = EmitDppTargetLane(state, flags.control);
 		const auto bounded = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpLogicalAnd, TypeBool(state), bounded, writable, target.valid});
+		state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), bounded, writable,
+		                          target.valid);
 		writable = bounded;
 	}
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpLogicalAnd, TypeBool(state), result, exec, writable});
+	state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), result, exec, writable);
 	return result;
 }
 
@@ -138,17 +140,16 @@ uint32_t EmitAttribute(EmitterState& state, uint32_t attr, uint32_t chan) {
 	if (input == nullptr || input->variable_id == 0) {
 		return ConstantU32(state, 0);
 	}
-	if (state.program.stage == ShaderType::Vertex) {
+	if (state.program.stage == ShaderType::Vertex || state.program.stage == ShaderType::Local) {
 		return EmitVertexParameterComponentU32(state, *input, chan & 3u);
 	}
 	const auto load_per_vertex = [&](uint32_t vertex) {
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpAccessChain,
-		                           TypePointer(state, spv::StorageClassInput, TypeF32(state)),
-		                           pointer, input->variable_id, ConstantU32(state, vertex),
-		                           ConstantU32(state, chan & 3u)});
-		state.builder.AddFunction({spv::OpLoad, TypeF32(state), value, pointer});
+		state.builder.AddFunction(
+		    spv::OpAccessChain, TypePointer(state, spv::StorageClassInput, TypeF32(state)), pointer,
+		    input->variable_id, ConstantU32(state, vertex), ConstantU32(state, chan & 3u));
+		state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
 		return value;
 	};
 	if (input->per_vertex) {
@@ -161,31 +162,31 @@ uint32_t EmitAttribute(EmitterState& state, uint32_t attr, uint32_t chan) {
 			const auto pointer = state.builder.AllocateId();
 			const auto weight  = state.builder.AllocateId();
 			const auto product = state.builder.AllocateId();
-			state.builder.AddFunction({spv::OpAccessChain,
-			                           TypePointer(state, spv::StorageClassInput, TypeF32(state)),
-			                           pointer, barycentric, ConstantU32(state, vertex)});
-			state.builder.AddFunction({spv::OpLoad, TypeF32(state), weight, pointer});
-			state.builder.AddFunction(
-			    {spv::OpFMul, TypeF32(state), product, load_per_vertex(vertex), weight});
+			state.builder.AddFunction(spv::OpAccessChain,
+			                          TypePointer(state, spv::StorageClassInput, TypeF32(state)),
+			                          pointer, barycentric, ConstantU32(state, vertex));
+			state.builder.AddFunction(spv::OpLoad, TypeF32(state), weight, pointer);
+			state.builder.AddFunction(spv::OpFMul, TypeF32(state), product, load_per_vertex(vertex),
+			                          weight);
 			if (vertex == 0u) {
 				sum = product;
 			} else {
 				const auto next = state.builder.AllocateId();
-				state.builder.AddFunction({spv::OpFAdd, TypeF32(state), next, sum, product});
+				state.builder.AddFunction(spv::OpFAdd, TypeF32(state), next, sum, product);
 				sum = next;
 			}
 		}
 		const auto bits = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, sum});
+		state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, sum);
 		return bits;
 	}
 	const auto vector    = state.builder.AllocateId();
 	const auto component = state.builder.AllocateId();
 	const auto bits      = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpLoad, TypeF32Vector(state, 4), vector, input->variable_id});
-	state.builder.AddFunction(
-	    {spv::OpCompositeExtract, TypeF32(state), component, vector, chan & 3u});
-	state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, component});
+	state.builder.AddFunction(spv::OpLoad, TypeF32Vector(state, 4), vector, input->variable_id);
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), component, vector,
+	                          chan & 3u);
+	state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, component);
 	return bits;
 }
 
@@ -199,11 +200,10 @@ uint32_t EmitInterpolationParameter(ValueEmitContext& ctx, uint32_t attr, uint32
 	const auto load_vertex = [&](uint32_t vertex) {
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpAccessChain,
-		                           TypePointer(state, spv::StorageClassInput, TypeF32(state)),
-		                           pointer, input->variable_id, ConstantU32(state, vertex),
-		                           ConstantU32(state, chan & 3u)});
-		state.builder.AddFunction({spv::OpLoad, TypeF32(state), value, pointer});
+		state.builder.AddFunction(
+		    spv::OpAccessChain, TypePointer(state, spv::StorageClassInput, TypeF32(state)), pointer,
+		    input->variable_id, ConstantU32(state, vertex), ConstantU32(state, chan & 3u));
+		state.builder.AddFunction(spv::OpLoad, TypeF32(state), value, pointer);
 		return value;
 	};
 
@@ -211,11 +211,11 @@ uint32_t EmitInterpolationParameter(ValueEmitContext& ctx, uint32_t attr, uint32
 	uint32_t   value           = load_vertex(selected_vertex);
 	if (!PixelParameterIsCustom(state, attr) && mode < 2u) {
 		const auto delta = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpFSub, TypeF32(state), delta, value, load_vertex(0)});
+		state.builder.AddFunction(spv::OpFSub, TypeF32(state), delta, value, load_vertex(0));
 		value = delta;
 	}
 	const auto bits = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, value});
+	state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, value);
 	return bits;
 }
 
@@ -229,8 +229,8 @@ uint32_t MrtOutputMode(const EmitterState& state, const IR::ExportInfo& exp) {
 
 uint32_t ExportRawComponent(ValueEmitContext& ctx, uint32_t vector, uint32_t component) {
 	const auto value = ctx.state.builder.AllocateId();
-	ctx.state.builder.AddFunction(
-	    {spv::OpCompositeExtract, TypeU32(ctx.state), value, vector, component});
+	ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), value, vector,
+	                              component);
 	return value;
 }
 
@@ -248,22 +248,21 @@ uint32_t ExportVector(ValueEmitContext& ctx, uint32_t data, const IR::ExportInfo
 			}
 			const auto packed   = state.builder.AllocateId();
 			const auto unpacked = state.builder.AllocateId();
-			state.builder.AddFunction(
-			    {spv::OpCompositeExtract, TypeU32(state), packed, data, pair});
-			state.builder.AddFunction({spv::OpExtInst, TypeF32Vector(state, 2), unpacked,
-			                           GlslStd450(state), unpack, packed});
+			state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), packed, data, pair);
+			state.builder.AddFunction(spv::OpExtInst, TypeF32Vector(state, 2), unpacked,
+			                          GlslStd450(state), unpack, packed);
 			for (uint32_t lane = 0; lane < 2u; lane++) {
 				const auto component = pair * 2u + lane;
 				if (((exp.en >> component) & 1u) != 0u) {
 					f32[component] = state.builder.AllocateId();
-					state.builder.AddFunction(
-					    {spv::OpCompositeExtract, TypeF32(state), f32[component], unpacked, lane});
+					state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state),
+					                          f32[component], unpacked, lane);
 				}
 			}
 		}
 		const auto vector = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpCompositeConstruct, TypeF32Vector(state, 4), vector,
-		                           f32[0], f32[1], f32[2], f32[3]});
+		state.builder.AddFunction(spv::OpCompositeConstruct, TypeF32Vector(state, 4), vector,
+		                          f32[0], f32[1], f32[2], f32[3]);
 		return vector;
 	}
 	uint32_t raw[4] = {
@@ -284,9 +283,9 @@ uint32_t ExportVector(ValueEmitContext& ctx, uint32_t data, const IR::ExportInfo
 					continue;
 				}
 				raw[component] = state.builder.AllocateId();
-				state.builder.AddFunction({spv::OpBitFieldUExtract, TypeU32(state), raw[component],
-				                           packed, ConstantU32(state, lane * 16u),
-				                           ConstantU32(state, 16)});
+				state.builder.AddFunction(spv::OpBitFieldUExtract, TypeU32(state), raw[component],
+				                          packed, ConstantU32(state, lane * 16u),
+				                          ConstantU32(state, 16));
 			}
 		}
 	} else {
@@ -298,18 +297,18 @@ uint32_t ExportVector(ValueEmitContext& ctx, uint32_t data, const IR::ExportInfo
 	}
 	if (uint_output) {
 		const auto vector = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpCompositeConstruct, TypeU32Vector(state, 4), vector,
-		                           raw[0], raw[1], raw[2], raw[3]});
+		state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(state, 4), vector,
+		                          raw[0], raw[1], raw[2], raw[3]);
 		return vector;
 	}
 	uint32_t f32[4] {};
 	for (uint32_t component = 0; component < 4u; component++) {
 		f32[component] = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpBitcast, TypeF32(state), f32[component], raw[component]});
+		state.builder.AddFunction(spv::OpBitcast, TypeF32(state), f32[component], raw[component]);
 	}
 	const auto vector = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpCompositeConstruct, TypeF32Vector(state, 4), vector, f32[0],
-	                           f32[1], f32[2], f32[3]});
+	state.builder.AddFunction(spv::OpCompositeConstruct, TypeF32Vector(state, 4), vector, f32[0],
+	                          f32[1], f32[2], f32[3]);
 	return vector;
 }
 
@@ -325,19 +324,19 @@ void EmitAuxPositionExport(ValueEmitContext& ctx, uint32_t data, const IR::Expor
 			const auto raw = ExportRawComponent(ctx, data, component);
 			if (output.layer) {
 				const auto layer = state.builder.AllocateId();
-				state.builder.AddFunction(
-				    {spv::OpBitwiseAnd, TypeU32(state), layer, raw, ConstantU32(state, 0x7ffu)});
+				state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), layer, raw,
+				                          ConstantU32(state, 0x7ffu));
 				const auto pointer = state.program.stage == ShaderType::Mesh
 				                         ? MeshOutputPointer(state, IR::StageOutputKind::Layer)
 				                         : state.layer_variable;
-				state.builder.AddFunction({spv::OpStore, pointer, layer});
+				state.builder.AddFunction(spv::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({spv::OpBitFieldUExtract, TypeU32(state), viewport, raw,
-				                           ConstantU32(state, 16), ConstantU32(state, 4)});
-				state.builder.AddFunction({spv::OpStore, state.viewport_index_variable, viewport});
+				state.builder.AddFunction(spv::OpBitFieldUExtract, TypeU32(state), viewport, raw,
+				                          ConstantU32(state, 16), ConstantU32(state, 4));
+				state.builder.AddFunction(spv::OpStore, state.viewport_index_variable, viewport);
 			}
 			continue;
 		}
@@ -348,9 +347,9 @@ void EmitAuxPositionExport(ValueEmitContext& ctx, uint32_t data, const IR::Expor
 
 		const auto raw = ExportRawComponent(ctx, data, component);
 		const auto f32 = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpBitcast, TypeF32(state), f32, raw});
+		state.builder.AddFunction(spv::OpBitcast, TypeF32(state), f32, raw);
 		if (output.point_size) {
-			state.builder.AddFunction({spv::OpStore, state.point_size_variable, f32});
+			state.builder.AddFunction(spv::OpStore, state.point_size_variable, f32);
 			continue;
 		}
 		auto StoreDistance = [&](uint32_t variable, uint32_t index) {
@@ -358,10 +357,10 @@ void EmitAuxPositionExport(ValueEmitContext& ctx, uint32_t data, const IR::Expor
 				return;
 			}
 			const auto pointer = state.builder.AllocateId();
-			state.builder.AddFunction({spv::OpAccessChain,
-			                           TypePointer(state, spv::StorageClassOutput, TypeF32(state)),
-			                           pointer, variable, ConstantU32(state, index)});
-			state.builder.AddFunction({spv::OpStore, pointer, f32});
+			state.builder.AddFunction(spv::OpAccessChain,
+			                          TypePointer(state, spv::StorageClassOutput, TypeF32(state)),
+			                          pointer, variable, ConstantU32(state, index));
+			state.builder.AddFunction(spv::OpStore, pointer, f32);
 		};
 		StoreDistance(state.clip_distance_variable, output.clip_distance);
 		StoreDistance(state.cull_distance_variable, output.cull_distance);
@@ -374,14 +373,14 @@ uint32_t ConvertClipCoordinate(EmitterState& state, uint32_t coordinate, float s
 	const auto biased  = state.builder.AllocateId();
 	const auto divided = state.builder.AllocateId();
 	const auto ndc     = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpFMul, TypeF32(state), window, coordinate, ConstantF32Value(state, scale)});
-	state.builder.AddFunction(
-	    {spv::OpFAdd, TypeF32(state), biased, window, ConstantF32Value(state, offset)});
-	state.builder.AddFunction(
-	    {spv::OpFDiv, TypeF32(state), divided, biased, ConstantF32Value(state, half_extent)});
-	state.builder.AddFunction(
-	    {spv::OpFSub, TypeF32(state), ndc, divided, ConstantF32Value(state, 1.0f)});
+	state.builder.AddFunction(spv::OpFMul, TypeF32(state), window, coordinate,
+	                          ConstantF32Value(state, scale));
+	state.builder.AddFunction(spv::OpFAdd, TypeF32(state), biased, window,
+	                          ConstantF32Value(state, offset));
+	state.builder.AddFunction(spv::OpFDiv, TypeF32(state), divided, biased,
+	                          ConstantF32Value(state, half_extent));
+	state.builder.AddFunction(spv::OpFSub, TypeF32(state), ndc, divided,
+	                          ConstantF32Value(state, 1.0f));
 	return ndc;
 }
 
@@ -390,16 +389,16 @@ uint32_t ConvertPositionToClipSpace(EmitterState& state, uint32_t position) {
 	uint32_t    components[4] {};
 	for (uint32_t i = 0; i < 4; i++) {
 		components[i] = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpCompositeExtract, TypeF32(state), components[i], position, i});
+		state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), components[i], position,
+		                          i);
 	}
 	components[0] = ConvertClipCoordinate(state, components[0], transform.scale[0],
 	                                      transform.offset[0], transform.half_extent[0]);
 	components[1] = ConvertClipCoordinate(state, components[1], transform.scale[1],
 	                                      transform.offset[1], transform.half_extent[1]);
 	const auto converted = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpCompositeConstruct, TypeF32Vector(state, 4), converted,
-	                           components[0], components[1], components[2], components[3]});
+	state.builder.AddFunction(spv::OpCompositeConstruct, TypeF32Vector(state, 4), converted,
+	                          components[0], components[1], components[2], components[3]);
 	return converted;
 }
 
@@ -411,17 +410,17 @@ uint32_t EmitWqmU64(EmitterState& state, uint32_t value) {
 	const auto merged_two  = state.builder.AllocateId();
 	const auto quad_bits   = state.builder.AllocateId();
 	const auto result      = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpShiftRightLogical, TypeU64(state), shifted_one, value,
-	                           ConstantU64(state, 0x0000000100000001ull)});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU64(state), merged_one, value, shifted_one});
-	state.builder.AddFunction({spv::OpShiftRightLogical, TypeU64(state), shifted_two, merged_one,
-	                           ConstantU64(state, 0x0000000200000002ull)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseOr, TypeU64(state), merged_two, merged_one, shifted_two});
-	state.builder.AddFunction({spv::OpBitwiseAnd, TypeU64(state), quad_bits, merged_two,
-	                           ConstantU64(state, 0x1111111111111111ull)});
-	state.builder.AddFunction({spv::OpIMul, TypeU64(state), result, quad_bits,
-	                           ConstantU64(state, 0x0000000f0000000full)});
+	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU64(state), shifted_one, value,
+	                          ConstantU64(state, 0x0000000100000001ull));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU64(state), merged_one, value, shifted_one);
+	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU64(state), shifted_two, merged_one,
+	                          ConstantU64(state, 0x0000000200000002ull));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU64(state), merged_two, merged_one,
+	                          shifted_two);
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU64(state), quad_bits, merged_two,
+	                          ConstantU64(state, 0x1111111111111111ull));
+	state.builder.AddFunction(spv::OpIMul, TypeU64(state), result, quad_bits,
+	                          ConstantU64(state, 0x0000000f0000000full));
 	return result;
 }
 
@@ -432,9 +431,9 @@ void EmitSetAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
 	if (state.program.stage == ShaderType::Pixel && exp.vm && state.requirements.pixel_valid_mask &&
 	    state.pixel_valid_mask_variable != 0) {
 		const auto value = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpSelect, TypeU32(state), value, exec,
-		                           ConstantU32(state, 1), ConstantU32(state, 0)});
-		state.builder.AddFunction({spv::OpStore, state.pixel_valid_mask_variable, value});
+		state.builder.AddFunction(spv::OpSelect, TypeU32(state), value, exec, ConstantU32(state, 1),
+		                          ConstantU32(state, 0));
+		state.builder.AddFunction(spv::OpStore, state.pixel_valid_mask_variable, value);
 	}
 	if (exp.kind == IR::ExportTargetKind::Null || exp.en == 0u) {
 		return;
@@ -443,8 +442,8 @@ void EmitSetAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
 		const auto data = ctx.Arg(inst, 0);
 		if (exp.kind == IR::ExportTargetKind::Primitive) {
 			if (state.program.stage == ShaderType::Mesh) {
-				state.builder.AddFunction(
-				    {spv::OpStore, MeshPrimitivePointer(state), ExportRawComponent(ctx, data, 0)});
+				state.builder.AddFunction(spv::OpStore, MeshPrimitivePointer(state),
+				                          ExportRawComponent(ctx, data, 0));
 			}
 			return;
 		}
@@ -456,19 +455,18 @@ void EmitSetAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
 			if ((exp.en & 1u) != 0u && state.depth_variable != 0) {
 				const auto raw = ExportRawComponent(ctx, data, 0);
 				const auto f32 = state.builder.AllocateId();
-				state.builder.AddFunction({spv::OpBitcast, TypeF32(state), f32, raw});
-				state.builder.AddFunction({spv::OpStore, state.depth_variable, f32});
+				state.builder.AddFunction(spv::OpBitcast, TypeF32(state), f32, raw);
+				state.builder.AddFunction(spv::OpStore, state.depth_variable, f32);
 			}
 			if ((exp.en & 4u) != 0u && state.sample_mask_variable != 0) {
 				const auto raw     = ExportRawComponent(ctx, data, 2);
 				const auto value   = state.builder.AllocateId();
 				const auto pointer = state.builder.AllocateId();
-				state.builder.AddFunction({spv::OpBitcast, TypeI32(state), value, raw});
+				state.builder.AddFunction(spv::OpBitcast, TypeI32(state), value, raw);
 				state.builder.AddFunction(
-				    {spv::OpAccessChain,
-				     TypePointer(state, spv::StorageClassOutput, TypeI32(state)), pointer,
-				     state.sample_mask_variable, ConstantU32(state, 0)});
-				state.builder.AddFunction({spv::OpStore, pointer, value});
+				    spv::OpAccessChain, TypePointer(state, spv::StorageClassOutput, TypeI32(state)),
+				    pointer, state.sample_mask_variable, ConstantU32(state, 0));
+				state.builder.AddFunction(spv::OpStore, pointer, value);
 			}
 			return;
 		}
@@ -485,9 +483,9 @@ void EmitSetAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
 			const auto mapping = state.input_info.pixel->target_export_mapping[exp.index];
 			if (!mapping.IsIdentity()) {
 				const auto mapped = state.builder.AllocateId();
-				state.builder.AddFunction({spv::OpVectorShuffle, vector_type, mapped, value, value,
-				                           mapping.Map(0), mapping.Map(1), mapping.Map(2),
-				                           mapping.Map(3)});
+				state.builder.AddFunction(spv::OpVectorShuffle, vector_type, mapped, value, value,
+				                          mapping.Map(0), mapping.Map(1), mapping.Map(2),
+				                          mapping.Map(3));
 				value = mapped;
 			}
 		}
@@ -499,17 +497,17 @@ void EmitSetAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
 			const auto kind = exp.kind == IR::ExportTargetKind::Position
 			                      ? IR::StageOutputKind::Position
 			                      : IR::StageOutputKind::Parameter;
-			state.builder.AddFunction(
-			    {spv::OpStore, MeshOutputPointer(state, kind, exp.index), value});
+			state.builder.AddFunction(spv::OpStore, MeshOutputPointer(state, kind, exp.index),
+			                          value);
 		} else if (exp.kind == IR::ExportTargetKind::Position) {
 			const auto pointer = state.builder.AllocateId();
 			state.builder.AddFunction(
-			    {spv::OpAccessChain,
-			     TypePointer(state, spv::StorageClassOutput, TypeF32Vector(state, 4)), pointer,
-			     variable, ConstantU32(state, 0)});
-			state.builder.AddFunction({spv::OpStore, pointer, value});
+			    spv::OpAccessChain,
+			    TypePointer(state, spv::StorageClassOutput, TypeF32Vector(state, 4)), pointer,
+			    variable, ConstantU32(state, 0));
+			state.builder.AddFunction(spv::OpStore, pointer, value);
 		} else {
-			state.builder.AddFunction({spv::OpStore, variable, value});
+			state.builder.AddFunction(spv::OpStore, variable, value);
 		}
 	});
 }
@@ -521,11 +519,19 @@ uint32_t EmitIdentity(ValueEmitContext&, uint32_t value) {
 void EmitVoid(ValueEmitContext&) {}
 
 void EmitBarrier(EmitterState& state) {
-	const auto semantics =
-	    spv::MemorySemanticsAcquireReleaseMask | spv::MemorySemanticsWorkgroupMemoryMask;
-	state.builder.AddFunction({spv::OpControlBarrier, ConstantU32(state, spv::ScopeWorkgroup),
-	                           ConstantU32(state, spv::ScopeWorkgroup),
-	                           ConstantU32(state, semantics)});
+	const auto tessellation = state.program.stage == ShaderType::TessellationControl;
+	const auto memory_scope = tessellation ? spv::ScopeInvocation : spv::ScopeWorkgroup;
+	const auto semantics    = tessellation ? spv::MemorySemanticsMaskNone
+	                                       : spv::MemorySemanticsAcquireReleaseMask |
+	                                             spv::MemorySemanticsWorkgroupMemoryMask;
+	state.builder.AddFunction(spv::OpControlBarrier, ConstantU32(state, spv::ScopeWorkgroup),
+	                          ConstantU32(state, memory_scope), ConstantU32(state, semantics));
+}
+
+uint32_t EmitLaneId(EmitterState& state) {
+	return state.program.stage == ShaderType::TessellationControl
+	           ? EmitBuiltinU32(state, IR::StageInputKind::InvocationId, 0)
+	           : EmitSubgroupLocalInvocationId(state);
 }
 
 uint32_t EmitMeshDrawParameter(ValueEmitContext& ctx, const IR::Inst& inst) {
@@ -536,10 +542,10 @@ uint32_t EmitMeshDrawParameter(ValueEmitContext& ctx, const IR::Inst& inst) {
 		ctx.Fail(inst, "invalid mesh draw parameter");
 	}
 	const auto pointer = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpAccessChain, TypePushConstantElementPointer(state), pointer,
-	                           state.push_constant_variable, ConstantU32(state, 0),
-	                           ConstantU32(state, index)});
-	state.builder.AddFunction({spv::OpLoad, TypeU32(state), result, pointer});
+	state.builder.AddFunction(spv::OpAccessChain, TypePushConstantElementPointer(state), pointer,
+	                          state.push_constant_variable, ConstantU32(state, 0),
+	                          ConstantU32(state, index));
+	state.builder.AddFunction(spv::OpLoad, TypeU32(state), result, pointer);
 	return result;
 }
 
@@ -559,7 +565,7 @@ uint32_t EmitGetBuiltin(ValueEmitContext& ctx, IR::Value kind, IR::Value index)
 
 uint32_t EmitUndefU1(EmitterState& state, const IR::Inst& inst) {
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpUndef, TypeId(state, inst.GetType()), result});
+	state.builder.AddFunction(spv::OpUndef, TypeId(state, inst.GetType()), result);
 	return result;
 }
 
@@ -574,8 +580,8 @@ uint32_t EmitDppMoveU32(ValueEmitContext& ctx, const IR::Inst& inst) {
 	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(
-	    {spv::OpLogicalAnd, TypeBool(state), can_fetch, target.valid, source_active});
+	state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), can_fetch, target.valid,
+	                          source_active);
 	return EmitNative<spv::OpSelect, IR::Type::U32>(ctx.state, can_fetch, shuffled,
 	                                                ConstantU32(state, 0));
 }
@@ -604,8 +610,8 @@ uint32_t EmitReadLane(ValueEmitContext& ctx, const IR::Inst& inst) {
 uint32_t EmitWriteLane(ValueEmitContext& ctx, const IR::Inst& inst) {
 	auto&      state = ctx.state;
 	const auto hit   = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpIEqual, TypeBool(state), hit,
-	                           EmitSubgroupLocalInvocationId(state), ctx.Arg(inst, 2)});
+	state.builder.AddFunction(spv::OpIEqual, TypeBool(state), hit,
+	                          EmitSubgroupLocalInvocationId(state), ctx.Arg(inst, 2));
 	return EmitNative<spv::OpSelect, IR::Type::U32>(ctx.state, hit, ctx.Arg(inst, 1),
 	                                                ctx.Arg(inst, 0));
 }
@@ -624,35 +630,35 @@ uint32_t EmitPermlane16U32(ValueEmitContext& ctx, const IR::Inst& inst) {
 	const auto shifted   = state.builder.AllocateId();
 	const auto index     = state.builder.AllocateId();
 	const auto target    = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), row, subid, ConstantU32(state, 0xfffffff0u)});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), row, subid,
+	                          ConstantU32(state, 0xfffffff0u));
 	if (flags.x16) {
-		state.builder.AddFunction(
-		    {spv::OpBitwiseXor, TypeU32(state), row_value, row, ConstantU32(state, 16)});
+		state.builder.AddFunction(spv::OpBitwiseXor, TypeU32(state), row_value, row,
+		                          ConstantU32(state, 16));
 	} else {
-		state.builder.AddFunction({spv::OpCopyObject, TypeU32(state), row_value, row});
+		state.builder.AddFunction(spv::OpCopyObject, TypeU32(state), row_value, row);
 	}
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), lane, subid, ConstantU32(state, 15)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), lane8, lane, ConstantU32(state, 7)});
-	state.builder.AddFunction(
-	    {spv::OpShiftLeftLogical, TypeU32(state), shift, lane8, ConstantU32(state, 2)});
-	state.builder.AddFunction(
-	    {spv::OpUGreaterThanEqual, TypeBool(state), upper, lane, ConstantU32(state, 8)});
-	state.builder.AddFunction(
-	    {spv::OpSelect, TypeU32(state), selected, upper, ctx.Arg(inst, 2), ctx.Arg(inst, 1)});
-	state.builder.AddFunction({spv::OpShiftRightLogical, TypeU32(state), shifted, selected, shift});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), index, shifted, ConstantU32(state, 15)});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), target, row_value, index});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
+	                          ConstantU32(state, 15));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane8, lane,
+	                          ConstantU32(state, 7));
+	state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), shift, lane8,
+	                          ConstantU32(state, 2));
+	state.builder.AddFunction(spv::OpUGreaterThanEqual, TypeBool(state), upper, lane,
+	                          ConstantU32(state, 8));
+	state.builder.AddFunction(spv::OpSelect, TypeU32(state), selected, upper, ctx.Arg(inst, 2),
+	                          ctx.Arg(inst, 1));
+	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), shifted, selected, shift);
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), index, shifted,
+	                          ConstantU32(state, 15));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, row_value, index);
 	const auto shuffled = ctx.Shuffle(inst, 0, target);
 	uint32_t   result   = shuffled;
 	if (!flags.fetch_inactive) {
 		const auto source_exec = ctx.Shuffle(inst, 3, target);
 		result                 = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpSelect, TypeU32(state), result, source_exec, shuffled, ConstantU32(state, 0)});
+		state.builder.AddFunction(spv::OpSelect, TypeU32(state), result, source_exec, shuffled,
+		                          ConstantU32(state, 0));
 	}
 	return result;
 }

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

@@ -9,17 +9,17 @@ DppTargetLane EmitDppQuadPermTargetLane(EmitterState& state, uint32_t subid, uin
 	const auto selected0 = state.builder.AllocateId();
 	const auto selected  = state.builder.AllocateId();
 	const auto target    = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), quad_base, subid, ConstantU32(state, 0xfffffffcu)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), lane, subid, ConstantU32(state, 3)});
-	state.builder.AddFunction(
-	    {spv::OpShiftLeftLogical, TypeU32(state), shift, lane, ConstantU32(state, 1)});
-	state.builder.AddFunction(
-	    {spv::OpShiftRightLogical, TypeU32(state), selected0, ConstantU32(state, control), shift});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), selected, selected0, ConstantU32(state, 3)});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), target, quad_base, selected});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), quad_base, subid,
+	                          ConstantU32(state, 0xfffffffcu));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
+	                          ConstantU32(state, 3));
+	state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), shift, lane,
+	                          ConstantU32(state, 1));
+	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), selected0,
+	                          ConstantU32(state, control), shift);
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), selected, selected0,
+	                          ConstantU32(state, 3));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, quad_base, selected);
 	return {target, ConstantBool(state, true)};
 }
 
@@ -30,22 +30,22 @@ DppTargetLane EmitDppRowShiftTargetLane(EmitterState& state, uint32_t subid, uin
 	const auto lane_shifted = state.builder.AllocateId();
 	const auto target       = state.builder.AllocateId();
 	const auto valid        = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), row, subid, ConstantU32(state, 0xfffffff0u)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), lane, subid, ConstantU32(state, 15)});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), row, subid,
+	                          ConstantU32(state, 0xfffffff0u));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
+	                          ConstantU32(state, 15));
 	if (left) {
-		state.builder.AddFunction(
-		    {spv::OpIAdd, TypeU32(state), lane_shifted, lane, ConstantU32(state, amount)});
-		state.builder.AddFunction(
-		    {spv::OpULessThan, TypeBool(state), valid, lane, ConstantU32(state, 16u - amount)});
+		state.builder.AddFunction(spv::OpIAdd, TypeU32(state), lane_shifted, lane,
+		                          ConstantU32(state, amount));
+		state.builder.AddFunction(spv::OpULessThan, TypeBool(state), valid, lane,
+		                          ConstantU32(state, 16u - amount));
 	} else {
-		state.builder.AddFunction(
-		    {spv::OpISub, TypeU32(state), lane_shifted, lane, ConstantU32(state, amount)});
-		state.builder.AddFunction(
-		    {spv::OpUGreaterThanEqual, TypeBool(state), valid, lane, ConstantU32(state, amount)});
+		state.builder.AddFunction(spv::OpISub, TypeU32(state), lane_shifted, lane,
+		                          ConstantU32(state, amount));
+		state.builder.AddFunction(spv::OpUGreaterThanEqual, TypeBool(state), valid, lane,
+		                          ConstantU32(state, amount));
 	}
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), target, row, lane_shifted});
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, row, lane_shifted);
 	return {target, valid};
 }
 
@@ -58,18 +58,17 @@ DppTargetLane EmitDppRowRotateRightTargetLane(EmitterState& state, uint32_t subi
 	const auto plus     = state.builder.AllocateId();
 	const auto selected = state.builder.AllocateId();
 	const auto target   = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), row, subid, ConstantU32(state, 0xfffffff0u)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), lane, subid, ConstantU32(state, 15)});
-	state.builder.AddFunction(
-	    {spv::OpUGreaterThanEqual, TypeBool(state), in_high, lane, ConstantU32(state, amount)});
-	state.builder.AddFunction(
-	    {spv::OpISub, TypeU32(state), minus, lane, ConstantU32(state, amount)});
-	state.builder.AddFunction(
-	    {spv::OpIAdd, TypeU32(state), plus, lane, ConstantU32(state, 16u - amount)});
-	state.builder.AddFunction({spv::OpSelect, TypeU32(state), selected, in_high, minus, plus});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), target, row, selected});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), row, subid,
+	                          ConstantU32(state, 0xfffffff0u));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
+	                          ConstantU32(state, 15));
+	state.builder.AddFunction(spv::OpUGreaterThanEqual, TypeBool(state), in_high, lane,
+	                          ConstantU32(state, amount));
+	state.builder.AddFunction(spv::OpISub, TypeU32(state), minus, lane, ConstantU32(state, amount));
+	state.builder.AddFunction(spv::OpIAdd, TypeU32(state), plus, lane,
+	                          ConstantU32(state, 16u - amount));
+	state.builder.AddFunction(spv::OpSelect, TypeU32(state), selected, in_high, minus, plus);
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, row, selected);
 	return {target, ConstantBool(state, true)};
 }
 
@@ -80,13 +79,13 @@ DppTargetLane EmitDppMirrorTargetLane(EmitterState& state, uint32_t subid, bool
 	const auto lane      = state.builder.AllocateId();
 	const auto mirrored  = state.builder.AllocateId();
 	const auto target    = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), base, subid, ConstantU32(state, base_mask)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), lane, subid, ConstantU32(state, lane_mask)});
-	state.builder.AddFunction(
-	    {spv::OpISub, TypeU32(state), mirrored, ConstantU32(state, lane_mask), lane});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), target, base, mirrored});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), base, subid,
+	                          ConstantU32(state, base_mask));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
+	                          ConstantU32(state, lane_mask));
+	state.builder.AddFunction(spv::OpISub, TypeU32(state), mirrored, ConstantU32(state, lane_mask),
+	                          lane);
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, base, mirrored);
 	return {target, ConstantBool(state, true)};
 }
 
@@ -112,8 +111,8 @@ DppTargetLane EmitDppTargetLane(EmitterState& state, uint32_t control) {
 	}
 	if (control >= 0x160u && control <= 0x16fu) {
 		const auto target = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpBitwiseXor, TypeU32(state), target, subid, ConstantU32(state, control & 0xfu)});
+		state.builder.AddFunction(spv::OpBitwiseXor, TypeU32(state), target, subid,
+		                          ConstantU32(state, control & 0xfu));
 		return {target, ConstantBool(state, true)};
 	}
 	return {subid, ConstantBool(state, true)};
@@ -124,8 +123,8 @@ uint32_t EmitSubgroupLocalInvocationId(EmitterState& state) {
 		EXIT("SubgroupLocalInvocationId was not declared before SPIR-V function emission\n");
 	}
 	const auto value = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpLoad, TypeU32(state), value, state.subgroup_local_invocation_id_variable});
+	state.builder.AddFunction(spv::OpLoad, TypeU32(state), value,
+	                          state.subgroup_local_invocation_id_variable);
 	return state.lane_half == 0 ? value : EmitAddU32(state, value, ConstantU32(state, 32));
 }
 
@@ -154,10 +153,10 @@ uint32_t EmitInputComponentU32(EmitterState& state, IR::StageInputKind kind, uin
 	}
 	const auto pointer = state.builder.AllocateId();
 	const auto value   = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpAccessChain,
-	                           TypePointer(state, spv::StorageClassInput, TypeU32(state)), pointer,
-	                           variable, ConstantU32(state, component)});
-	state.builder.AddFunction({spv::OpLoad, TypeU32(state), value, pointer});
+	state.builder.AddFunction(spv::OpAccessChain,
+	                          TypePointer(state, spv::StorageClassInput, TypeU32(state)), pointer,
+	                          variable, ConstantU32(state, component));
+	state.builder.AddFunction(spv::OpLoad, TypeU32(state), value, pointer);
 	return value;
 }
 
@@ -167,7 +166,7 @@ uint32_t EmitLocalInvocationIndex(EmitterState& state) {
 		return ConstantU32(state, 0);
 	}
 	const auto value = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpLoad, TypeU32(state), value, variable});
+	state.builder.AddFunction(spv::OpLoad, TypeU32(state), value, variable);
 	if (state.lane_count == 2) {
 		const auto wave_base =
 		    EmitBinaryU32(state, spv::OpBitwiseAnd, value, ConstantU32(state, ~31u));
@@ -184,13 +183,13 @@ uint32_t EmitVertexParameterComponentU32(EmitterState& state, const InputBinding
 	const auto scalar_type = VertexParameterScalarType(state, kind);
 	uint32_t   raw         = state.builder.AllocateId();
 	if (count == 1u) {
-		state.builder.AddFunction({spv::OpLoad, scalar_type, raw, input.variable_id});
+		state.builder.AddFunction(spv::OpLoad, scalar_type, raw, input.variable_id);
 	} else {
 		const auto pointer_type = TypePointer(state, spv::StorageClassInput, scalar_type);
 		const auto pointer      = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpAccessChain, pointer_type, pointer, input.variable_id,
-		                           ConstantU32(state, component)});
-		state.builder.AddFunction({spv::OpLoad, scalar_type, raw, pointer});
+		state.builder.AddFunction(spv::OpAccessChain, pointer_type, pointer, input.variable_id,
+		                          ConstantU32(state, component));
+		state.builder.AddFunction(spv::OpLoad, scalar_type, raw, pointer);
 	}
 
 	if (kind == VertexInputScalarKind::Uint) {
@@ -198,49 +197,48 @@ uint32_t EmitVertexParameterComponentU32(EmitterState& state, const InputBinding
 	}
 
 	const auto bits = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpBitcast, TypeU32(state), bits, raw});
+	state.builder.AddFunction(spv::OpBitcast, TypeU32(state), bits, raw);
 	return bits;
 }
 
 uint32_t EmitSubgroupLaneActiveBool(EmitterState& state, uint32_t lane) {
 	const auto active_ballot = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpGroupNonUniformBallot, TypeU32Vector(state, 4), active_ballot,
-	                           ConstantU32(state, spv::ScopeSubgroup), ConstantBool(state, true)});
+	state.builder.AddFunction(spv::OpGroupNonUniformBallot, TypeU32Vector(state, 4), active_ballot,
+	                          ConstantU32(state, spv::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({spv::OpCompositeExtract, TypeU32(state), low, active_ballot, 0});
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), low, active_ballot, 0);
 	uint32_t mask = low;
 	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();
-		state.builder.AddFunction(
-		    {spv::OpCompositeExtract, TypeU32(state), high, active_ballot, 1});
-		state.builder.AddFunction(
-		    {spv::OpUGreaterThanEqual, TypeBool(state), in_high, lane, ConstantU32(state, 32)});
-		state.builder.AddFunction({spv::OpSelect, TypeU32(state), selected, in_high, high, low});
+		state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), high, active_ballot, 1);
+		state.builder.AddFunction(spv::OpUGreaterThanEqual, TypeBool(state), in_high, lane,
+		                          ConstantU32(state, 32));
+		state.builder.AddFunction(spv::OpSelect, TypeU32(state), selected, in_high, high, low);
 		mask = selected;
 	}
 
 	const auto lane_low = state.builder.AllocateId();
 	const auto bit      = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), lane_low, lane, ConstantU32(state, 31)});
-	state.builder.AddFunction(
-	    {spv::OpShiftLeftLogical, TypeU32(state), bit, ConstantU32(state, 1), lane_low});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane_low, lane,
+	                          ConstantU32(state, 31));
+	state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), bit, ConstantU32(state, 1),
+	                          lane_low);
 
 	const auto hit      = state.builder.AllocateId();
 	const auto active   = state.builder.AllocateId();
 	const auto in_range = state.builder.AllocateId();
 	const auto ret      = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpBitwiseAnd, TypeU32(state), hit, mask, bit});
-	state.builder.AddFunction(
-	    {spv::OpINotEqual, TypeBool(state), active, hit, ConstantU32(state, 0)});
-	state.builder.AddFunction({spv::OpULessThan, TypeBool(state), in_range, lane,
-	                           ConstantU32(state, state.program.wave_size)});
-	state.builder.AddFunction({spv::OpLogicalAnd, TypeBool(state), ret, active, in_range});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), hit, mask, bit);
+	state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), active, hit,
+	                          ConstantU32(state, 0));
+	state.builder.AddFunction(spv::OpULessThan, TypeBool(state), in_range, lane,
+	                          ConstantU32(state, state.program.wave_size));
+	state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), ret, active, in_range);
 	return ret;
 }
 

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

@@ -90,14 +90,14 @@ uint32_t CubeAxis(EmitterState& state, uint32_t value) {
 
 uint32_t CubeLayer(EmitterState& state, uint32_t value) {
 	const auto guest = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpConvertFToU, TypeU32(state), guest, value});
+	state.builder.AddFunction(spv::OpConvertFToU, TypeU32(state), guest, value);
 	const auto padding = Binary(
 	    state, spv::OpShiftLeftLogical, TypeU32(state),
 	    Binary(state, spv::OpShiftRightLogical, TypeU32(state), guest, ConstantU32(state, 3)),
 	    ConstantU32(state, 1));
 	const auto host   = Binary(state, spv::OpISub, TypeU32(state), guest, padding);
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpConvertUToF, TypeF32(state), result, host});
+	state.builder.AddFunction(spv::OpConvertUToF, TypeF32(state), result, host);
 	return result;
 }
 
@@ -118,11 +118,11 @@ uint32_t CoordF32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::In
 		             ? AddressF32(ctx, mem, address, first + 2u)
 		             : ZeroF32(ctx.state);
 		if (cube) z = CubeLayer(ctx.state, z);
-		ctx.state.builder.AddFunction(
-		    {spv::OpCompositeConstruct, TypeF32Vector(ctx.state, 3), result, x, y, z});
+		ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeF32Vector(ctx.state, 3),
+		                              result, x, y, z);
 	} else {
-		ctx.state.builder.AddFunction(
-		    {spv::OpCompositeConstruct, TypeF32Vector(ctx.state, 2), result, x, y});
+		ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeF32Vector(ctx.state, 2),
+		                              result, x, y);
 	}
 	return result;
 }
@@ -138,11 +138,11 @@ uint32_t CoordU32(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR::In
 	if (components == 3u) {
 		const auto z = mem.image_address_components > 2u ? AddressU32(ctx, mem, address, 2)
 		                                                 : ConstantU32(ctx.state, 0);
-		ctx.state.builder.AddFunction(
-		    {spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 3), result, x, y, z});
+		ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 3),
+		                              result, x, y, z);
 	} else {
-		ctx.state.builder.AddFunction(
-		    {spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 2), result, x, y});
+		ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 2),
+		                              result, x, y);
 	}
 	return result;
 }
@@ -193,9 +193,9 @@ uint32_t ResultVector(ValueEmitContext& ctx, uint32_t value,
 			if (dref) return value;
 			const auto component = gather ? index : DmaskComponent(mem.dmask, index);
 			const auto result    = ctx.state.builder.AllocateId();
-			ctx.state.builder.AddFunction({spv::OpCompositeExtract,
-			                               ImageScalarType(ctx.state, value_class), result, value,
-			                               component});
+			ctx.state.builder.AddFunction(spv::OpCompositeExtract,
+			                              ImageScalarType(ctx.state, value_class), result, value,
+			                              component);
 			return result;
 		};
 		for (uint32_t word = 0; word < mem.data_dwords; word++) {
@@ -218,17 +218,16 @@ uint32_t ResultVector(ValueEmitContext& ctx, uint32_t value,
 				                  ConstantU32(ctx.state, 16u)));
 			} else {
 				const auto pair = ctx.state.builder.AllocateId();
-				ctx.state.builder.AddFunction(
-				    {spv::OpCompositeConstruct, TypeF32Vector(ctx.state, 2), pair, low, high});
+				ctx.state.builder.AddFunction(spv::OpCompositeConstruct,
+				                              TypeF32Vector(ctx.state, 2), pair, low, high);
 				packed[word] = ctx.state.builder.AllocateId();
-				ctx.state.builder.AddFunction({spv::OpExtInst, TypeU32(ctx.state), packed[word],
-				                               GlslStd450(ctx.state), GLSLstd450PackHalf2x16,
-				                               pair});
+				ctx.state.builder.AddFunction(spv::OpExtInst, TypeU32(ctx.state), packed[word],
+				                              GlslStd450(ctx.state), GLSLstd450PackHalf2x16, pair);
 			}
 		}
 		const auto result = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction({spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4),
-		                               result, packed[0], packed[1], packed[2], packed[3]});
+		ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4),
+		                              result, packed[0], packed[1], packed[2], packed[3]);
 		return result;
 	}
 	uint32_t component[4] {};
@@ -238,14 +237,13 @@ uint32_t ResultVector(ValueEmitContext& ctx, uint32_t value,
 			continue;
 		}
 		const auto scalar = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction({spv::OpCompositeExtract,
-		                               ImageScalarType(ctx.state, value_class), scalar, value,
-		                               index});
+		ctx.state.builder.AddFunction(
+		    spv::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({spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result,
-	                               component[0], component[1], component[2], component[3]});
+	ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result,
+	                              component[0], component[1], component[2], component[3]);
 	return result;
 }
 
@@ -256,12 +254,12 @@ uint32_t QueryDimensions(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
 	const auto  image     = LoadSampledImageDescriptor(ctx.state, mem.resource);
 	const auto  size      = ctx.state.builder.AllocateId();
 	if (info.multisampled != 0u) {
-		ctx.state.builder.AddFunction(
-		    {spv::OpImageQuerySize, ImageViewSizeType(ctx.state, dimension), size, image});
+		ctx.state.builder.AddFunction(spv::OpImageQuerySize,
+		                              ImageViewSizeType(ctx.state, dimension), size, image);
 	} else {
-		ctx.state.builder.AddFunction({spv::OpImageQuerySizeLod,
-		                               ImageViewSizeType(ctx.state, dimension), size, image,
-		                               AddressU32(ctx, mem, address, 0)});
+		ctx.state.builder.AddFunction(spv::OpImageQuerySizeLod,
+		                              ImageViewSizeType(ctx.state, dimension), size, image,
+		                              AddressU32(ctx, mem, address, 0));
 	}
 	const auto components = info.coordinate_components;
 	uint32_t   result[4]  = {ConstantU32(ctx.state, 0), ConstantU32(ctx.state, 0),
@@ -271,18 +269,18 @@ uint32_t QueryDimensions(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
 	} else {
 		for (uint32_t index = 0; index < components; index++) {
 			result[index] = ctx.state.builder.AllocateId();
-			ctx.state.builder.AddFunction(
-			    {spv::OpCompositeExtract, TypeU32(ctx.state), result[index], size, index});
+			ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state),
+			                              result[index], size, index);
 		}
 	}
 	if (info.multisampled == 0u) {
 		result[3] = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction(
-		    {spv::OpImageQueryLevels, TypeU32(ctx.state), result[3], image});
+		ctx.state.builder.AddFunction(spv::OpImageQueryLevels, TypeU32(ctx.state), result[3],
+		                              image);
 	}
 	const auto vector = ctx.state.builder.AllocateId();
-	ctx.state.builder.AddFunction({spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), vector,
-	                               result[0], result[1], result[2], result[3]});
+	ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), vector,
+	                              result[0], result[1], result[2], result[3]);
 	return vector;
 }
 
@@ -294,11 +292,11 @@ uint32_t PackedOffset(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR
 		if (components == 1u) return zero;
 		const auto result = ctx.state.builder.AllocateId();
 		if (components == 3u) {
-			ctx.state.builder.AddFunction(
-			    {spv::OpCompositeConstruct, TypeI32Vector(ctx.state, 3), result, zero, zero, zero});
+			ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeI32Vector(ctx.state, 3),
+			                              result, zero, zero, zero);
 		} else {
-			ctx.state.builder.AddFunction(
-			    {spv::OpCompositeConstruct, TypeI32Vector(ctx.state, 2), result, zero, zero});
+			ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeI32Vector(ctx.state, 2),
+			                              result, zero, zero);
 		}
 		return result;
 	}
@@ -307,18 +305,18 @@ uint32_t PackedOffset(ValueEmitContext& ctx, const IR::MemoryInfo& mem, const IR
 	uint32_t   values[3] = {zero, zero, zero};
 	for (uint32_t index = 0; index < components; index++) {
 		values[index] = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction({spv::OpBitFieldSExtract, TypeI32(ctx.state), values[index],
-		                               packed, ConstantU32(ctx.state, index * 8u),
-		                               ConstantU32(ctx.state, 6)});
+		ctx.state.builder.AddFunction(spv::OpBitFieldSExtract, TypeI32(ctx.state), values[index],
+		                              packed, ConstantU32(ctx.state, index * 8u),
+		                              ConstantU32(ctx.state, 6));
 	}
 	if (components == 1u) return values[0];
 	const auto result = ctx.state.builder.AllocateId();
 	if (components == 3u) {
-		ctx.state.builder.AddFunction({spv::OpCompositeConstruct, TypeI32Vector(ctx.state, 3),
-		                               result, values[0], values[1], values[2]});
+		ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeI32Vector(ctx.state, 3),
+		                              result, values[0], values[1], values[2]);
 	} else {
-		ctx.state.builder.AddFunction(
-		    {spv::OpCompositeConstruct, TypeI32Vector(ctx.state, 2), result, values[0], values[1]});
+		ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeI32Vector(ctx.state, 2),
+		                              result, values[0], values[1]);
 	}
 	return result;
 }
@@ -327,7 +325,7 @@ 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(spv::OpTypeArray, {element_type, count});
+	const auto array_type   = state.builder.Type(spv::OpTypeArray, element_type, count);
 	uint32_t   offsets[4] {};
 	for (uint32_t index = 0; index < 4u; index++) {
 		const auto x = ConstantI32(state, static_cast<int32_t>(index) - 1);
@@ -335,11 +333,11 @@ uint32_t HorizontalOffsets(EmitterState& state, ImageDimension dimension) {
 			offsets[index] = x;
 		} else {
 			offsets[index] = state.builder.Constant(
-			    spv::OpConstantComposite, TypeI32Vector(state, 2), {x, ConstantI32(state, 0)});
+			    spv::OpConstantComposite, TypeI32Vector(state, 2), x, ConstantI32(state, 0));
 		}
 	}
-	return state.builder.Constant(spv::OpConstantComposite, array_type,
-	                              {offsets[0], offsets[1], offsets[2], offsets[3]});
+	return state.builder.Constant(spv::OpConstantComposite, array_type, offsets[0], offsets[1],
+	                              offsets[2], offsets[3]);
 }
 
 uint32_t InverseSwizzle(uint32_t swizzle, uint32_t component) {
@@ -367,15 +365,15 @@ uint32_t UnpackImageTexel(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint
 	if (info.format == Prospero::BufferFormat::kInvalid) return texel;
 
 	const auto packed = ctx.state.builder.AllocateId();
-	ctx.state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(ctx.state), packed, texel, 0u});
+	ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), packed, texel, 0u);
 	uint32_t components[4] = {ConstantU32(ctx.state, 0), ConstantU32(ctx.state, 0),
 	                          ConstantU32(ctx.state, 0), ConstantU32(ctx.state, 0)};
 	for (uint32_t component = 0; component < info.component_count; component++) {
 		components[component] = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction({spv::OpBitFieldUExtract, TypeU32(ctx.state),
-		                               components[component], packed,
-		                               ConstantU32(ctx.state, info.component_bit_offset[component]),
-		                               ConstantU32(ctx.state, info.component_bits[component])});
+		ctx.state.builder.AddFunction(spv::OpBitFieldUExtract, TypeU32(ctx.state),
+		                              components[component], packed,
+		                              ConstantU32(ctx.state, info.component_bit_offset[component]),
+		                              ConstantU32(ctx.state, info.component_bits[component]));
 	}
 	for (uint32_t component = info.component_count; component < 4u; component++) {
 		components[component] = components[component % info.component_count];
@@ -394,8 +392,8 @@ uint32_t UnpackImageTexel(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint
 		}
 	}
 	const auto result = ctx.state.builder.AllocateId();
-	ctx.state.builder.AddFunction({spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result,
-	                               selected[0], selected[1], selected[2], selected[3]});
+	ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result,
+	                              selected[0], selected[1], selected[2], selected[3]);
 	return result;
 }
 
@@ -409,24 +407,24 @@ uint32_t UnpackImageGather(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uin
 	if (selector < 4u) {
 		const auto value = ConstantU32(ctx.state, selector == 1u ? 1u : 0u);
 		return ctx.state.builder.Constant(spv::OpConstantComposite, TypeU32Vector(ctx.state, 4),
-		                                  {value, value, value, value});
+		                                  value, value, value, value);
 	}
 
 	uint32_t values[4] {};
 	for (uint32_t lane = 0; lane < 4u; lane++) {
 		const auto packed = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction(
-		    {spv::OpCompositeExtract, TypeU32(ctx.state), packed, gathered, lane});
+		ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), packed, gathered,
+		                              lane);
 		values[lane]        = ctx.state.builder.AllocateId();
 		const auto physical = (selector - 4u) % info.component_count;
-		ctx.state.builder.AddFunction({spv::OpBitFieldUExtract, TypeU32(ctx.state), values[lane],
-		                               packed,
-		                               ConstantU32(ctx.state, info.component_bit_offset[physical]),
-		                               ConstantU32(ctx.state, info.component_bits[physical])});
+		ctx.state.builder.AddFunction(spv::OpBitFieldUExtract, TypeU32(ctx.state), values[lane],
+		                              packed,
+		                              ConstantU32(ctx.state, info.component_bit_offset[physical]),
+		                              ConstantU32(ctx.state, info.component_bits[physical]));
 	}
 	const auto result = ctx.state.builder.AllocateId();
-	ctx.state.builder.AddFunction({spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result,
-	                               values[0], values[1], values[2], values[3]});
+	ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result,
+	                              values[0], values[1], values[2], values[3]);
 	return result;
 }
 
@@ -436,16 +434,16 @@ uint32_t EmitOneDimensionalGatherLz(ValueEmitContext& ctx, const IR::MemoryInfo&
 	state.builder.RequireCapability(spv::CapabilityImageQuery);
 	const auto image = LoadSampledImageDescriptor(state, mem.resource);
 	const auto width = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpImageQuerySizeLod, TypeU32(state), width, image, ConstantU32(state, 0)});
+	state.builder.AddFunction(spv::OpImageQuerySizeLod, TypeU32(state), width, image,
+	                          ConstantU32(state, 0));
 	const auto width_f32 = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpConvertUToF, TypeF32(state), width_f32, width});
+	state.builder.AddFunction(spv::OpConvertUToF, TypeF32(state), width_f32, width);
 	const auto left = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpExtInst, TypeF32(state), left, GlslStd450(state),
-	                           GLSLstd450Floor,
-	                           Binary(state, spv::OpFSub, TypeF32(state),
-	                                  Binary(state, spv::OpFMul, TypeF32(state), coord, width_f32),
-	                                  ConstantF32(state, 0x3f000000u))});
+	state.builder.AddFunction(spv::OpExtInst, TypeF32(state), left, GlslStd450(state),
+	                          GLSLstd450Floor,
+	                          Binary(state, spv::OpFSub, TypeF32(state),
+	                                 Binary(state, spv::OpFMul, TypeF32(state), coord, width_f32),
+	                                 ConstantF32(state, 0x3f000000u)));
 
 	const auto sampled     = MakeSampledImage(state, mem.resource, mem.sampler);
 	const auto vector_type = ImageVectorType(state, numeric_class, 4);
@@ -462,15 +460,15 @@ uint32_t EmitOneDimensionalGatherLz(ValueEmitContext& ctx, const IR::MemoryInfo&
 		                  ConstantF32(state, index == 0u ? 0x3f000000u : 0x3fc00000u)),
 		           width_f32);
 		const auto texel = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpImageSampleExplicitLod, vector_type, texel, sampled,
-		                           sample_coord, spv::ImageOperandsLodMask, ZeroF32(state)});
+		state.builder.AddFunction(spv::OpImageSampleExplicitLod, vector_type, texel, sampled,
+		                          sample_coord, spv::ImageOperandsLodMask, ZeroF32(state));
 		values[index] = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpCompositeExtract, scalar_type, values[index], texel, component});
+		state.builder.AddFunction(spv::OpCompositeExtract, scalar_type, values[index], texel,
+		                          component);
 	}
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpCompositeConstruct, vector_type, result, values[0], values[1],
-	                           values[1], values[0]});
+	state.builder.AddFunction(spv::OpCompositeConstruct, vector_type, result, values[0], values[1],
+	                          values[1], values[0]);
 	return result;
 }
 
@@ -481,8 +479,8 @@ uint32_t PackImageTexel(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint32
 	auto packed = ConstantU32(ctx.state, 0u);
 	for (uint32_t component = 0; component < info.component_count; component++) {
 		const auto value = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction(
-		    {spv::OpCompositeExtract, TypeU32(ctx.state), value, texel, component});
+		ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), value, texel,
+		                              component);
 		const auto maximum =
 		    ConstantU32(ctx.state, info.component_bits[component] == 32u
 		                               ? UINT32_MAX
@@ -490,8 +488,8 @@ uint32_t PackImageTexel(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint32
 		const auto within =
 		    Binary(ctx.state, spv::OpULessThan, TypeBool(ctx.state), value, maximum);
 		const auto clamped = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction(
-		    {spv::OpSelect, TypeU32(ctx.state), clamped, within, value, maximum});
+		ctx.state.builder.AddFunction(spv::OpSelect, TypeU32(ctx.state), clamped, within, value,
+		                              maximum);
 		const auto shifted =
 		    info.component_bit_offset[component] == 0u
 		        ? clamped
@@ -501,8 +499,8 @@ uint32_t PackImageTexel(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint32
 	}
 	const auto zero   = ConstantU32(ctx.state, 0u);
 	const auto result = ctx.state.builder.AllocateId();
-	ctx.state.builder.AddFunction(
-	    {spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result, packed, zero, zero, zero});
+	ctx.state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(ctx.state, 4), result,
+	                              packed, zero, zero, zero);
 	return result;
 }
 
@@ -516,9 +514,8 @@ uint32_t StoreTexel(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint32_t d
 		if (source < 4u && ((dmask >> source) & 1u) != 0u) {
 			const auto packed_index = DmaskComponentIndex(dmask, source);
 			raw                     = ctx.state.builder.AllocateId();
-			ctx.state.builder.AddFunction(
-			    {spv::OpCompositeExtract, TypeU32(ctx.state), raw, data,
-			     mem.data_bits == 16u ? packed_index / 2u : packed_index});
+			ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), raw, data,
+			                              mem.data_bits == 16u ? packed_index / 2u : packed_index);
 			if (mem.data_bits == 16u) {
 				if ((packed_index & 1u) != 0u) {
 					raw = Binary(ctx.state, spv::OpShiftRightLogical, TypeU32(ctx.state), raw,
@@ -534,14 +531,14 @@ uint32_t StoreTexel(ValueEmitContext& ctx, const IR::MemoryInfo& mem, uint32_t d
 		                        : Unary(ctx.state, spv::OpBitcast, TypeF32(ctx.state), raw);
 	}
 	const auto texel = ctx.state.builder.AllocateId();
-	ctx.state.builder.AddFunction(
-	    {spv::OpCompositeConstruct,
-	     integer ? TypeU32Vector(ctx.state, 4) : TypeF32Vector(ctx.state, 4), texel, values[0],
-	     values[1], values[2], values[3]});
+	ctx.state.builder.AddFunction(spv::OpCompositeConstruct,
+	                              integer ? TypeU32Vector(ctx.state, 4)
+	                                      : TypeF32Vector(ctx.state, 4),
+	                              texel, values[0], values[1], values[2], values[3]);
 	return PackImageTexel(ctx, mem, texel);
 }
 
-uint32_t ImageAtomicOpcode(IR::ValueOpcode opcode) {
+spv::Op ImageAtomicOpcode(IR::ValueOpcode opcode) {
 	switch (opcode) {
 		case IR::ValueOpcode::ImageAtomicSwap32: return spv::OpAtomicExchange;
 		case IR::ValueOpcode::ImageAtomicIAdd32: return spv::OpAtomicIAdd;
@@ -550,7 +547,7 @@ uint32_t ImageAtomicOpcode(IR::ValueOpcode opcode) {
 		case IR::ValueOpcode::ImageAtomicAnd32: return spv::OpAtomicAnd;
 		case IR::ValueOpcode::ImageAtomicOr32: return spv::OpAtomicOr;
 		case IR::ValueOpcode::ImageAtomicXor32: return spv::OpAtomicXor;
-		default: return 0;
+		default: return spv::OpNop;
 	}
 }
 
@@ -576,19 +573,19 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		const auto sampled   = MakeSampledImage(state, mem.resource, mem.sampler);
 		const auto lod       = state.builder.AllocateId();
 		state.builder.AddFunction(
-		    {spv::OpImageQueryLod, TypeF32Vector(state, 2), lod, sampled,
-		     CoordF32(ctx, mem, *address, 0, ImageDimensionInfoFor(dimension).spatial_components)});
+		    spv::OpImageQueryLod, TypeF32Vector(state, 2), lod, sampled,
+		    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++) {
 			const auto component = state.builder.AllocateId();
-			state.builder.AddFunction(
-			    {spv::OpCompositeExtract, TypeF32(state), component, lod, index});
+			state.builder.AddFunction(spv::OpCompositeExtract, TypeF32(state), component, lod,
+			                          index);
 			values[index] = FloatBits(ctx, component);
 		}
 		const auto result = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpCompositeConstruct, TypeU32Vector(state, 4), result,
-		                           values[0], values[1], values[2], values[3]});
+		state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(state, 4), result,
+		                          values[0], values[1], values[2], values[3]);
 		ctx.Define(inst, result);
 		return;
 	}
@@ -607,14 +604,14 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			        const auto coord      = CoordU32(ctx, mem, *address, dimension);
 			        if (dimension_info.multisampled != 0u) {
 				        state.builder.AddFunction(
-				            {spv::OpImageFetch, ImageVectorType(state, numeric_class, 4), color,
-				             descriptor, coord, spv::ImageOperandsSampleMask,
-				             AddressU32(ctx, mem, *address, dimension_info.coordinate_components)});
+				            spv::OpImageFetch, ImageVectorType(state, numeric_class, 4), color,
+				            descriptor, coord, spv::ImageOperandsSampleMask,
+				            AddressU32(ctx, mem, *address, dimension_info.coordinate_components));
 			        } else {
-				        state.builder.AddFunction({spv::OpImageFetch,
-				                                   ImageVectorType(state, numeric_class, 4), color,
-				                                   descriptor, coord, spv::ImageOperandsLodMask,
-				                                   LodU32(ctx, mem, *address, dimension)});
+				        state.builder.AddFunction(spv::OpImageFetch,
+				                                  ImageVectorType(state, numeric_class, 4), color,
+				                                  descriptor, coord, spv::ImageOperandsLodMask,
+				                                  LodU32(ctx, mem, *address, dimension));
 			        }
 			        return ResultVector(ctx, UnpackImageTexel(ctx, mem, color), numeric_class,
 			                            false, mem);
@@ -711,7 +708,7 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		                          HasFlag(mem, Decoder::ImageSampleFlagLod) ||
 		                          HasFlag(mem, Decoder::ImageSampleFlagLevelZero) ||
 		                          state.program.stage != ShaderType::Pixel;
-		uint32_t   opcode       = spv::OpImageSampleImplicitLod;
+		auto       opcode       = spv::OpImageSampleImplicitLod;
 		if (explicit_lod) {
 			opcode = dref ? spv::OpImageSampleDrefExplicitLod : spv::OpImageSampleExplicitLod;
 		} else if (dref) {
@@ -748,15 +745,15 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		const auto EmitSample = [&](uint32_t 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};
+			std::vector<uint32_t> sample_operands {result_type, sample, sampled, coord};
 			if (dref) {
-				words.push_back(dref_value);
+				sample_operands.push_back(dref_value);
 			}
 			if (operand_mask != 0u) {
-				words.push_back(operand_mask);
-				words.insert(words.end(), operands.begin(), operands.end());
+				sample_operands.push_back(operand_mask);
+				sample_operands.insert(sample_operands.end(), operands.begin(), operands.end());
 			}
-			state.builder.AddFunction(words);
+			state.builder.AddFunction(opcode, sample_operands);
 			return sample;
 		};
 		if (image.indirect_root != mem.resource) {
@@ -785,11 +782,11 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		}
 		const auto LoadMapping = [&](uint32_t index) {
 			const auto pointer = state.builder.AllocateId();
-			state.builder.AddFunction({spv::OpAccessChain, TypeStorageBufferElementPointer(state),
-			                           pointer, state.flattened_srt_variable, ConstantU32(state, 0),
-			                           index});
+			state.builder.AddFunction(spv::OpAccessChain, TypeStorageBufferElementPointer(state),
+			                          pointer, state.flattened_srt_variable, ConstantU32(state, 0),
+			                          index);
 			const auto value = state.builder.AllocateId();
-			state.builder.AddFunction({spv::OpLoad, TypeU32(state), value, pointer});
+			state.builder.AddFunction(spv::OpLoad, TypeU32(state), value, pointer);
 			return value;
 		};
 		const auto mapping  = ConstantU32(state, image.indirect_mapping_offset);
@@ -802,9 +799,9 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			const auto mid    = Binary(state, spv::OpShiftRightLogical, TypeU32(state),
 			                           Binary(state, spv::OpIAdd, TypeU32(state), low, high),
 			                           ConstantU32(state, 1u));
-			const auto probe = state.builder.AllocateId();
-			state.builder.AddFunction(
-			    {spv::OpSelect, TypeU32(state), probe, active, mid, ConstantU32(state, 0u)});
+			const auto probe  = state.builder.AllocateId();
+			state.builder.AddFunction(spv::OpSelect, TypeU32(state), probe, active, mid,
+			                          ConstantU32(state, 0u));
 			const auto entry = Binary(state, spv::OpIAdd, TypeU32(state), mapping,
 			                          Binary(state, spv::OpIAdd, TypeU32(state),
 			                                 Binary(state, spv::OpShiftLeftLogical, TypeU32(state),
@@ -816,8 +813,8 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			const auto equal = Binary(state, spv::OpIEqual, TypeBool(state), mapped_key, key);
 			const auto match = Binary(state, spv::OpLogicalAnd, TypeBool(state), active, equal);
 			const auto next_selected = state.builder.AllocateId();
-			state.builder.AddFunction(
-			    {spv::OpSelect, TypeU32(state), next_selected, match, candidate, selected});
+			state.builder.AddFunction(spv::OpSelect, TypeU32(state), next_selected, match,
+			                          candidate, selected);
 			selected              = next_selected;
 			const auto less = Binary(state, spv::OpULessThan, TypeBool(state), mapped_key, key);
 			const auto take_upper = Binary(state, spv::OpLogicalAnd, TypeBool(state), active, less);
@@ -825,12 +822,12 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			                               Unary(state, spv::OpLogicalNot, TypeBool(state), less));
 			const auto next_low   = state.builder.AllocateId();
 			state.builder.AddFunction(
-			    {spv::OpSelect, TypeU32(state), next_low, take_upper,
-			     Binary(state, spv::OpIAdd, TypeU32(state), mid, ConstantU32(state, 1u)), low});
+			    spv::OpSelect, TypeU32(state), next_low, take_upper,
+			    Binary(state, spv::OpIAdd, TypeU32(state), mid, ConstantU32(state, 1u)), low);
 			low                  = next_low;
 			const auto next_high = state.builder.AllocateId();
-			state.builder.AddFunction(
-			    {spv::OpSelect, TypeU32(state), next_high, take_lower, mid, high});
+			state.builder.AddFunction(spv::OpSelect, TypeU32(state), next_high, take_lower, mid,
+			                          high);
 			high = next_high;
 		}
 		const auto            default_label = state.builder.AllocateId();
@@ -842,19 +839,19 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 			switch_words.push_back(candidate);
 			switch_words.push_back(labels[candidate - 1u]);
 		}
-		state.builder.AddFunction(
-		    {spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone});
+		state.builder.AddFunction(spv::OpSelectionMerge, merge_label,
+		                          spv::SelectionControlMaskNone);
 		state.builder.AddFunction(switch_words);
 		std::vector<uint32_t> phi_words {spv::OpPhi, result_type, state.builder.AllocateId()};
 		EmitLabel(state, default_label);
 		phi_words.push_back(EmitSample(image.indirect_resources[0]));
 		phi_words.push_back(default_label);
-		state.builder.AddFunction({spv::OpBranch, merge_label});
+		state.builder.AddFunction(spv::OpBranch, merge_label);
 		for (uint32_t candidate = 1; candidate < image.indirect_resources.size(); candidate++) {
 			EmitLabel(state, labels[candidate - 1u]);
 			phi_words.push_back(EmitSample(image.indirect_resources[candidate]));
 			phi_words.push_back(labels[candidate - 1u]);
-			state.builder.AddFunction({spv::OpBranch, merge_label});
+			state.builder.AddFunction(spv::OpBranch, merge_label);
 		}
 		EmitLabel(state, merge_label);
 		state.builder.AddFunction(phi_words);
@@ -866,21 +863,21 @@ void EmitImage(ValueEmitContext& ctx, const IR::Inst& inst) {
 		return;
 	}
 	const auto atomic_opcode = ImageAtomicOpcode(op);
-	if (atomic_opcode != 0u) {
+	if (atomic_opcode != spv::OpNop) {
 		const auto dimension = image.dimension;
 		ctx.Define(inst, EmitValueOrZeroIfCondition(state, ctx.Arg(inst, 3), [&]() {
-			           const auto pointer = state.builder.AllocateId();
+			           const auto pointer      = state.builder.AllocateId();
 			           const auto pointer_type = state.builder.Type(
-			               spv::OpTypePointer, {spv::StorageClassImage, TypeU32(state)});
-			           state.builder.AddFunction(
-			               {spv::OpImageTexelPointer, pointer_type, pointer,
-			                StorageImageDescriptorPointer(state, mem.resource),
-			                CoordU32(ctx, mem, *address, dimension), ConstantU32(state, 0)});
+			               spv::OpTypePointer, spv::StorageClassImage, TypeU32(state));
+			           state.builder.AddFunction(spv::OpImageTexelPointer, pointer_type, pointer,
+			                                     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, spv::ScopeDevice),
-			                                      ConstantU32(state, spv::MemorySemanticsMaskNone),
-			                                      ctx.Arg(inst, 2)});
+			           state.builder.AddFunction(atomic_opcode, TypeU32(state), old, pointer,
+			                                     ConstantU32(state, spv::ScopeDevice),
+			                                     ConstantU32(state, spv::MemorySemanticsMaskNone),
+			                                     ctx.Arg(inst, 2));
 			           EmitDeviceAtomicMemoryBarrier(state);
 			           return old;
 		           }));

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

@@ -155,6 +155,8 @@ inline constexpr auto EmitInstPrefetch = EmitVoid;
 void                  EmitBarrier(EmitterState& state);
 void                  EmitMeshAllocate(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitMeshDrawParameter(ValueEmitContext& ctx, const IR::Inst& inst);
+uint32_t              EmitGetTessellationAttribute(ValueEmitContext& ctx, const IR::Inst& inst);
+void                  EmitSetTessellationAttribute(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitGetUserData(EmitterState& state, IR::ScalarReg reg);
 uint32_t              EmitGetBuiltin(ValueEmitContext& ctx, IR::Value kind, IR::Value index);
 uint32_t              EmitUndefU1(EmitterState& state, const IR::Inst& inst);
@@ -165,7 +167,7 @@ inline constexpr auto EmitUndefU64 = EmitUndefU1;
 uint32_t              EmitDppMoveU32(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitDppUpdateU32(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitWqmU64(EmitterState& state, uint32_t value);
-inline constexpr auto EmitLaneId = EmitSubgroupLocalInvocationId;
+uint32_t              EmitLaneId(EmitterState& state);
 uint32_t              EmitBallot(ValueEmitContext& ctx, IR::Value predicate);
 uint32_t              EmitReadFirstLane(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t              EmitReadLane(ValueEmitContext& ctx, const IR::Inst& inst);
@@ -271,6 +273,7 @@ inline constexpr auto EmitImageAtomicOr32      = EmitImage;
 inline constexpr auto EmitImageAtomicXor32     = EmitImage;
 void                  EmitUnreachable(ValueEmitContext& ctx, const IR::Inst& inst);
 inline constexpr auto EmitPhi                        = EmitUnreachable;
+inline constexpr auto EmitTessellationBase           = EmitUnreachable;
 inline constexpr auto EmitGetThreadBitScalarRegister = EmitUnreachable;
 inline constexpr auto EmitSetThreadBitScalarRegister = EmitUnreachable;
 inline constexpr auto EmitGetScalarMaskTag           = EmitUnreachable;

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

@@ -80,6 +80,10 @@ struct EmitterState {
 	Builder                                          builder;
 	const IR::Program&                               program;
 	ShaderStageInputInfo                             input_info;
+	std::array<uint32_t, 6>                          tess_variables {};
+	uint32_t                                         tess_inner_variable = 0;
+	uint32_t                                         tess_patch_base     = 0;
+
 	const SpirvRequirements                          requirements;
 	uint32_t                                         lane_count              = 1;
 	uint32_t                                         lane_half               = 0;
@@ -151,7 +155,7 @@ uint32_t TypeU32ElementPointer(EmitterState& state, spv::StorageClass storage_cl
 
 inline void EmitLabel(EmitterState& state, uint32_t label) {
 	state.current_label = label;
-	state.builder.AddFunction({spv::OpLabel, label});
+	state.builder.AddFunction(spv::OpLabel, label);
 }
 
 uint32_t TypeId(EmitterState& state, IR::Type type);
@@ -160,7 +164,7 @@ uint32_t TypeId(EmitterState& state, IR::Type type);
 template <spv::Op opcode, IR::Type type, typename... Args>
 uint32_t EmitNative(EmitterState& state, Args... args) {
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, TypeId(state, type), result, args...});
+	state.builder.AddFunction(opcode, TypeId(state, type), result, args...);
 	return result;
 }
 
@@ -171,23 +175,23 @@ uint32_t EmitGlsl(EmitterState& state, Args... args) {
 	return EmitNative<spv::OpExtInst, type>(state, GlslStd450(state), opcode, args...);
 }
 
-inline uint32_t Unary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t value) {
+inline uint32_t Unary(EmitterState& state, spv::Op opcode, uint32_t type, uint32_t value) {
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, type, result, value});
+	state.builder.AddFunction(opcode, type, result, value);
 	return result;
 }
 
-inline uint32_t Binary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t lhs,
+inline uint32_t Binary(EmitterState& state, spv::Op opcode, uint32_t type, uint32_t lhs,
                        uint32_t rhs) {
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, type, result, lhs, rhs});
+	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({spv::OpSelect, type, result, condition, true_value, false_value});
+	state.builder.AddFunction(spv::OpSelect, type, result, condition, true_value, false_value);
 	return result;
 }
 
@@ -313,7 +317,11 @@ uint32_t ConstantU64(EmitterState& state, uint64_t value);
 
 uint32_t ConstantU32CompositeZero(EmitterState& state, uint32_t components);
 
+uint32_t DefineInterfaceVariable(EmitterState& state, uint32_t type, spv::StorageClass storage,
+                                 const char* name);
 void     DefineModule(EmitterState& state);
+void     DefineTessellationInterfaces(EmitterState& state);
+void     DefineTessellationExecutionModes(EmitterState& state);
 void     DefineMeshOutputs(EmitterState& state);
 void     EmitMeshEntryPoint(EmitterState& state);
 void     EmitMeshAllocate(ValueEmitContext& ctx, const IR::Inst& inst);
@@ -350,7 +358,7 @@ uint32_t EmitSubgroupLaneActiveBool(EmitterState& state, uint32_t lane);
 
 inline constexpr auto EmitAddU32 = EmitNative<spv::OpIAdd, IR::Type::U32, uint32_t, uint32_t>;
 
-uint32_t EmitBinaryU32(EmitterState& state, uint32_t opcode, uint32_t lhs, uint32_t rhs);
+uint32_t EmitBinaryU32(EmitterState& state, spv::Op opcode, uint32_t lhs, uint32_t rhs);
 
 uint32_t EmitShaderDataDwordLoad(EmitterState& state, uint32_t dword_index);
 
@@ -417,7 +425,7 @@ uint32_t EmitShiftRightConstant(EmitterState& state, uint32_t value, uint32_t sh
 
 inline constexpr auto EmitOrU32 = EmitNative<spv::OpBitwiseOr, IR::Type::U32, uint32_t, uint32_t>;
 
-uint32_t EmitCompareU32Constant(EmitterState& state, uint32_t opcode, uint32_t value,
+uint32_t EmitCompareU32Constant(EmitterState& state, spv::Op opcode, uint32_t value,
                                 uint32_t constant);
 
 uint32_t EmitSubConstantMinusU32(EmitterState& state, uint32_t constant, uint32_t value);
@@ -474,11 +482,11 @@ template <typename Fn>
 void EmitIfCondition(EmitterState& state, uint32_t condition, Fn&& fn) {
 	const auto then_label  = state.builder.AllocateId();
 	const auto merge_label = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone});
-	state.builder.AddFunction({spv::OpBranchConditional, condition, then_label, merge_label});
+	state.builder.AddFunction(spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone);
+	state.builder.AddFunction(spv::OpBranchConditional, condition, then_label, merge_label);
 	EmitLabel(state, then_label);
 	fn();
-	state.builder.AddFunction({spv::OpBranch, merge_label});
+	state.builder.AddFunction(spv::OpBranch, merge_label);
 	EmitLabel(state, merge_label);
 }
 
@@ -489,19 +497,19 @@ uint32_t EmitValueOrDefaultIfCondition(EmitterState& state, uint32_t condition,
 	const auto then_exit   = state.builder.AllocateId();
 	const auto else_label  = state.builder.AllocateId();
 	const auto merge_label = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone});
-	state.builder.AddFunction({spv::OpBranchConditional, condition, then_label, else_label});
+	state.builder.AddFunction(spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone);
+	state.builder.AddFunction(spv::OpBranchConditional, condition, then_label, else_label);
 	EmitLabel(state, then_label);
 	const auto then_value = fn();
-	state.builder.AddFunction({spv::OpBranch, then_exit});
+	state.builder.AddFunction(spv::OpBranch, then_exit);
 	EmitLabel(state, then_exit);
-	state.builder.AddFunction({spv::OpBranch, merge_label});
+	state.builder.AddFunction(spv::OpBranch, merge_label);
 	EmitLabel(state, else_label);
-	state.builder.AddFunction({spv::OpBranch, merge_label});
+	state.builder.AddFunction(spv::OpBranch, merge_label);
 	EmitLabel(state, merge_label);
 	const auto value = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpPhi, type, value, then_value, then_exit, default_value, else_label});
+	state.builder.AddFunction(spv::OpPhi, type, value, then_value, then_exit, default_value,
+	                          else_label);
 	return value;
 }
 
@@ -528,30 +536,29 @@ uint32_t AtomicUpdate(EmitterState& state, uint32_t pointer, IR::ResourceKind ki
 	const auto initial   = state.builder.AllocateId();
 	const auto observed  = state.builder.AllocateId();
 	const auto exchanged = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpBranch, preheader});
+	state.builder.AddFunction(spv::OpBranch, preheader);
 	EmitLabel(state, preheader);
-	state.builder.AddFunction({spv::OpAtomicLoad, TypeU32(state), initial, pointer,
-	                           ConstantU32(state, scope),
-	                           ConstantU32(state, spv::MemorySemanticsMaskNone)});
-	state.builder.AddFunction({spv::OpBranch, header});
+	state.builder.AddFunction(spv::OpAtomicLoad, TypeU32(state), initial, pointer,
+	                          ConstantU32(state, scope),
+	                          ConstantU32(state, spv::MemorySemanticsMaskNone));
+	state.builder.AddFunction(spv::OpBranch, header);
 	EmitLabel(state, header);
-	state.builder.AddFunction(
-	    {spv::OpPhi, TypeU32(state), observed, initial, preheader, exchanged, cont});
+	state.builder.AddFunction(spv::OpPhi, TypeU32(state), observed, initial, preheader, exchanged,
+	                          cont);
 	const auto next = desired(observed);
-	state.builder.AddFunction({spv::OpAtomicCompareExchange, TypeU32(state), exchanged, pointer,
-	                           ConstantU32(state, scope),
-	                           ConstantU32(state, spv::MemorySemanticsMaskNone),
-	                           ConstantU32(state, spv::MemorySemanticsMaskNone), next, observed});
+	state.builder.AddFunction(spv::OpAtomicCompareExchange, TypeU32(state), exchanged, pointer,
+	                          ConstantU32(state, scope),
+	                          ConstantU32(state, spv::MemorySemanticsMaskNone),
+	                          ConstantU32(state, spv::MemorySemanticsMaskNone), next, observed);
 	const auto success = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpIEqual, TypeBool(state), success, exchanged, observed});
-	state.builder.AddFunction({spv::OpLoopMerge, merge, cont, spv::LoopControlMaskNone});
-	state.builder.AddFunction({spv::OpBranchConditional, success, merge, cont});
+	state.builder.AddFunction(spv::OpIEqual, TypeBool(state), success, exchanged, observed);
+	state.builder.AddFunction(spv::OpLoopMerge, merge, cont, spv::LoopControlMaskNone);
+	state.builder.AddFunction(spv::OpBranchConditional, success, merge, cont);
 	EmitLabel(state, cont);
-	state.builder.AddFunction({spv::OpBranch, header});
+	state.builder.AddFunction(spv::OpBranch, header);
 	EmitLabel(state, merge);
-	state.builder.AddFunction(
-	    {spv::OpMemoryBarrier, ConstantU32(state, scope),
-	     ConstantU32(state, spv::MemorySemanticsAcquireReleaseMask | memory)});
+	state.builder.AddFunction(spv::OpMemoryBarrier, ConstantU32(state, scope),
+	                          ConstantU32(state, spv::MemorySemanticsAcquireReleaseMask | memory));
 	return observed;
 }
 

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

@@ -17,11 +17,11 @@ uint32_t EmitDsMaskedLaneRead(EmitterState& state, uint32_t source, uint32_t tar
 		target = Binary(state, spv::OpBitwiseAnd, TypeU32(state), target, ConstantU32(state, 31));
 	}
 	const auto shuffled = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpGroupNonUniformShuffle, TypeU32(state), shuffled,
-	                           ConstantU32(state, spv::ScopeSubgroup), source, target});
+	state.builder.AddFunction(spv::OpGroupNonUniformShuffle, TypeU32(state), shuffled,
+	                          ConstantU32(state, spv::ScopeSubgroup), source, target);
 	const auto source_exec = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpGroupNonUniformShuffle, TypeBool(state), source_exec,
-	                           ConstantU32(state, spv::ScopeSubgroup), exec, target});
+	state.builder.AddFunction(spv::OpGroupNonUniformShuffle, TypeBool(state), source_exec,
+	                          ConstantU32(state, spv::ScopeSubgroup), exec, target);
 	const auto source_active =
 	    AndCondition(state, source_exec, EmitSubgroupLaneActiveBool(state, target));
 	return Select(state, TypeU32(state), source_active, shuffled, ConstantU32(state, 0));
@@ -107,9 +107,9 @@ uint32_t ScratchByteAddress(ValueEmitContext& ctx, const IR::MemoryInfo& mem, ui
 }
 
 uint32_t ConstantDeviceAddress(EmitterState& state, uint64_t value) {
-	return state.builder.Constant(
-	    spv::OpConstant, TypeScalarU64(state),
-	    {static_cast<uint32_t>(value), static_cast<uint32_t>(value >> 32u)});
+	return state.builder.Constant(spv::OpConstant, TypeScalarU64(state),
+	                              static_cast<uint32_t>(value),
+	                              static_cast<uint32_t>(value >> 32u));
 }
 
 uint32_t DeviceAddressFromWords(EmitterState& state, uint32_t low, uint32_t high) {
@@ -153,8 +153,8 @@ uint32_t GuestAddress(ValueEmitContext& ctx, const IR::Inst& inst, const IR::Mem
 
 uint32_t FaultElementPointer(EmitterState& state, uint32_t index) {
 	const auto pointer = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpAccessChain, TypeStorageBufferElementPointer(state), pointer,
-	                           state.fault_buffer_variable, ConstantU32(state, 0), index});
+	state.builder.AddFunction(spv::OpAccessChain, TypeStorageBufferElementPointer(state), pointer,
+	                          state.fault_buffer_variable, ConstantU32(state, 0), index);
 	return pointer;
 }
 
@@ -166,16 +166,16 @@ void RecordBdaFault(EmitterState& state, uint32_t page) {
 	           Binary(state, spv::OpBitwiseAnd, TypeU32(state), page, ConstantU32(state, 31)));
 	const auto pointer = FaultElementPointer(state, word);
 	const auto value   = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpLoad, TypeU32(state), value, pointer});
-	state.builder.AddFunction(
-	    {spv::OpStore, pointer, Binary(state, spv::OpBitwiseOr, TypeU32(state), value, bit)});
+	state.builder.AddFunction(spv::OpLoad, TypeU32(state), value, pointer);
+	state.builder.AddFunction(spv::OpStore, pointer,
+	                          Binary(state, spv::OpBitwiseOr, TypeU32(state), value, bit));
 }
 
 uint32_t GetBdaPointer(ValueEmitContext& ctx, uint32_t address) {
 	auto&      state  = ctx.state;
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpFunctionCall, TypeScalarU64(state), result, state.bda_pointer_function, address});
+	state.builder.AddFunction(spv::OpFunctionCall, TypeScalarU64(state), result,
+	                          state.bda_pointer_function, address);
 	return result;
 }
 
@@ -186,11 +186,12 @@ uint32_t LoadBdaDword(ValueEmitContext& ctx, uint32_t address) {
 	    Binary(state, spv::OpINotEqual, TypeBool(state), bda, ConstantDeviceAddress(state, 0));
 	return EmitValueOrZeroIfCondition(state, present, [&]() {
 		const auto pointer = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpConvertUToPtr, TypePhysicalU32Pointer(state), pointer, bda});
-		const auto value = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpLoad, TypeU32(state), value, pointer,
-		                           spv::MemoryAccessAlignedMask, sizeof(uint32_t)});
+		state.builder.AddFunction(spv::OpConvertUToPtr, TypePhysicalU32Pointer(state), pointer,
+		                          bda);
+		const auto         value     = state.builder.AllocateId();
+		constexpr uint32_t alignment = sizeof(uint32_t);
+		state.builder.AddFunction(spv::OpLoad, TypeU32(state), value, pointer,
+		                          spv::MemoryAccessAlignedMask, alignment);
 		return value;
 	});
 }
@@ -306,7 +307,7 @@ uint32_t LoadWordInBounds(ValueEmitContext& ctx, const MemoryResourceAccess& res
                           uint32_t index) {
 	const auto value   = ctx.state.builder.AllocateId();
 	const auto pointer = EmitMemoryElementPointer(ctx.state, resource, index);
-	ctx.state.builder.AddFunction({spv::OpLoad, TypeU32(ctx.state), value, pointer});
+	ctx.state.builder.AddFunction(spv::OpLoad, TypeU32(ctx.state), value, pointer);
 	return value;
 }
 
@@ -401,11 +402,11 @@ uint32_t LoadFormattedComponent(ValueEmitContext& ctx, const IR::MemoryInfo& mem
 		const auto source_value =
 		    type == TypeI32(ctx.state) ? Unary(ctx.state, spv::OpBitcast, type, raw) : raw;
 		const auto extracted = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction(
-		    {IsSignedFormatComponent(info.type) ? spv::OpBitFieldSExtract : spv::OpBitFieldUExtract,
-		     type, extracted, source_value,
-		     ConstantU32(ctx.state, info.component_bit_offset[component]),
-		     ConstantU32(ctx.state, bits)});
+		ctx.state.builder.AddFunction(IsSignedFormatComponent(info.type) ? spv::OpBitFieldSExtract
+		                                                                 : spv::OpBitFieldUExtract,
+		                              type, extracted, source_value,
+		                              ConstantU32(ctx.state, info.component_bit_offset[component]),
+		                              ConstantU32(ctx.state, bits));
 		raw = type == TypeI32(ctx.state)
 		          ? Unary(ctx.state, spv::OpBitcast, TypeU32(ctx.state), extracted)
 		          : extracted;
@@ -465,8 +466,8 @@ void StoreSubwordInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
 	};
 	if (mem.kind == IR::ResourceKind::Scratch) {
 		const auto old = ctx.state.builder.AllocateId();
-		ctx.state.builder.AddFunction({spv::OpLoad, TypeU32(ctx.state), old, pointer});
-		ctx.state.builder.AddFunction({spv::OpStore, pointer, merge(old)});
+		ctx.state.builder.AddFunction(spv::OpLoad, TypeU32(ctx.state), old, pointer);
+		ctx.state.builder.AddFunction(spv::OpStore, pointer, merge(old));
 	} else {
 		AtomicUpdate(ctx.state, pointer, mem.kind, merge);
 	}
@@ -495,15 +496,15 @@ void StoreWordPrepared(ValueEmitContext& ctx, const IR::Inst& inst, const IR::Me
                        const MemoryResourceAccess& resource, uint32_t data) {
 	const auto index = EmitMemoryElementIndex(ctx.state, resource, DwordIndex(ctx, inst, mem));
 	EmitIfCondition(ctx.state, EmitMemoryElementInBounds(ctx.state, resource, index), [&]() {
-		ctx.state.builder.AddFunction(
-		    {spv::OpStore, EmitMemoryElementPointer(ctx.state, resource, index), data});
+		ctx.state.builder.AddFunction(spv::OpStore,
+		                              EmitMemoryElementPointer(ctx.state, resource, index), data);
 	});
 }
 
 void StoreWordInBounds(ValueEmitContext& ctx, const MemoryResourceAccess& resource, uint32_t index,
                        uint32_t data) {
-	ctx.state.builder.AddFunction(
-	    {spv::OpStore, EmitMemoryElementPointer(ctx.state, resource, index), data});
+	ctx.state.builder.AddFunction(spv::OpStore,
+	                              EmitMemoryElementPointer(ctx.state, resource, index), data);
 }
 
 void StoreWord(ValueEmitContext& ctx, const IR::Inst& inst, IR::MemoryInfo mem) {
@@ -538,7 +539,7 @@ void FormattedStore(ValueEmitContext& ctx, const IR::Inst& inst, const IR::Memor
 	});
 }
 
-uint32_t SpirvAtomicOpcode(IR::ValueOpcode opcode) {
+spv::Op SpirvAtomicOpcode(IR::ValueOpcode opcode) {
 	switch (opcode) {
 		case IR::ValueOpcode::BufferAtomicCmpSwap32: return spv::OpAtomicCompareExchange;
 		case IR::ValueOpcode::BufferAtomicSwap32:
@@ -563,7 +564,7 @@ uint32_t SpirvAtomicOpcode(IR::ValueOpcode opcode) {
 		case IR::ValueOpcode::SharedAtomicOr32: return spv::OpAtomicOr;
 		case IR::ValueOpcode::BufferAtomicXor32:
 		case IR::ValueOpcode::SharedAtomicXor32: return spv::OpAtomicXor;
-		default: return 0;
+		default: return spv::OpNop;
 	}
 }
 
@@ -574,15 +575,14 @@ uint32_t EmitAtomicOperation(ValueEmitContext& ctx, const IR::Inst& inst, uint32
 		const auto desired    = ctx.Arg(inst, inst.NumArgs() - 3);
 		const auto comparator = ctx.Arg(inst, inst.NumArgs() - 2);
 		ctx.state.builder.AddFunction(
-		    {spv::OpAtomicCompareExchange, TypeU32(ctx.state), old, pointer,
-		     ConstantU32(ctx.state, scope), ConstantU32(ctx.state, spv::MemorySemanticsMaskNone),
-		     ConstantU32(ctx.state, spv::MemorySemanticsMaskNone), desired, comparator});
+		    spv::OpAtomicCompareExchange, TypeU32(ctx.state), old, pointer,
+		    ConstantU32(ctx.state, scope), ConstantU32(ctx.state, spv::MemorySemanticsMaskNone),
+		    ConstantU32(ctx.state, spv::MemorySemanticsMaskNone), 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, spv::MemorySemanticsMaskNone),
-		                               value});
+		ctx.state.builder.AddFunction(SpirvAtomicOpcode(inst.GetOpcode()), TypeU32(ctx.state), old,
+		                              pointer, ConstantU32(ctx.state, scope),
+		                              ConstantU32(ctx.state, spv::MemorySemanticsMaskNone), value);
 	}
 	return old;
 }
@@ -782,8 +782,8 @@ void StoreWideBuffer(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t compo
 			EmitIfCondition(state, plan.in_bounds, [&]() {
 				for (uint32_t component = 0; component < components; component++) {
 					const auto data = state.builder.AllocateId();
-					state.builder.AddFunction(
-					    {spv::OpCompositeExtract, TypeU32(state), data, composite, component});
+					state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), data,
+					                          composite, component);
 					StoreFormattedInBounds(ctx, mem, plan, component, data);
 				}
 			});
@@ -791,8 +791,8 @@ void StoreWideBuffer(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t compo
 		}
 		for (uint32_t component = 0; component < components; component++) {
 			const auto data = state.builder.AllocateId();
-			state.builder.AddFunction(
-			    {spv::OpCompositeExtract, TypeU32(state), data, composite, component});
+			state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), data, composite,
+			                          component);
 			StoreWordPrepared(ctx, inst, RebaseRawComponent(mem, component), resource, data);
 		}
 	});
@@ -851,49 +851,49 @@ void DefineGetBdaPointer(EmitterState& state) {
 		return;
 	}
 	const auto type            = TypeScalarU64(state);
-	const auto function_type   = state.builder.Type(spv::OpTypeFunction, {type, type});
+	const auto function_type   = state.builder.Type(spv::OpTypeFunction, type, type);
 	state.bda_pointer_function = state.builder.AllocateId();
 	const auto address         = state.builder.AllocateId();
 	const auto entry_label     = state.builder.AllocateId();
 	state.builder.AddName(state.bda_pointer_function, "get_bda_pointer");
-	state.builder.AddFunction({spv::OpFunction, type, state.bda_pointer_function,
-	                           spv::FunctionControlMaskNone, function_type});
-	state.builder.AddFunction({spv::OpFunctionParameter, type, address});
+	state.builder.AddFunction(spv::OpFunction, type, state.bda_pointer_function,
+	                          spv::FunctionControlMaskNone, function_type);
+	state.builder.AddFunction(spv::OpFunctionParameter, type, address);
 	EmitLabel(state, entry_label);
 
 	const auto page64        = Binary(state, spv::OpShiftRightLogical, type, address,
 	                                  ConstantDeviceAddress(state, BufferCache::CACHING_PAGEBITS));
 	const auto page          = Unary(state, spv::OpUConvert, TypeU32(state), page64);
 	const auto entry_pointer = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpAccessChain, TypeStorageBufferU64ElementPointer(state),
-	                           entry_pointer, state.bda_pagetable_variable, ConstantU32(state, 0),
-	                           page});
+	state.builder.AddFunction(spv::OpAccessChain, TypeStorageBufferU64ElementPointer(state),
+	                          entry_pointer, state.bda_pagetable_variable, ConstantU32(state, 0),
+	                          page);
 	const auto base = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpLoad, type, base, entry_pointer});
+	state.builder.AddFunction(spv::OpLoad, type, base, entry_pointer);
 	const auto missing =
 	    Binary(state, spv::OpIEqual, TypeBool(state), base, ConstantDeviceAddress(state, 0));
 	const auto fault_label     = state.builder.AllocateId();
 	const auto available_label = state.builder.AllocateId();
 	const auto merge_label     = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone});
-	state.builder.AddFunction({spv::OpBranchConditional, missing, fault_label, available_label});
+	state.builder.AddFunction(spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone);
+	state.builder.AddFunction(spv::OpBranchConditional, missing, fault_label, available_label);
 
 	EmitLabel(state, fault_label);
 	RecordBdaFault(state, page);
-	state.builder.AddFunction({spv::OpBranch, merge_label});
+	state.builder.AddFunction(spv::OpBranch, merge_label);
 
 	EmitLabel(state, available_label);
 	const auto offset    = Binary(state, spv::OpBitwiseAnd, type, address,
 	                              ConstantDeviceAddress(state, BufferCache::CACHING_PAGESIZE - 1));
 	const auto available = Binary(state, spv::OpIAdd, type, base, offset);
-	state.builder.AddFunction({spv::OpBranch, merge_label});
+	state.builder.AddFunction(spv::OpBranch, merge_label);
 
 	EmitLabel(state, merge_label);
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpPhi, type, result, ConstantDeviceAddress(state, 0),
-	                           fault_label, available, available_label});
-	state.builder.AddFunction({spv::OpReturnValue, result});
-	state.builder.AddFunction({spv::OpFunctionEnd});
+	state.builder.AddFunction(spv::OpPhi, type, result, ConstantDeviceAddress(state, 0),
+	                          fault_label, available, available_label);
+	state.builder.AddFunction(spv::OpReturnValue, result);
+	state.builder.AddFunction(spv::OpFunctionEnd);
 }
 
 uint32_t EmitAtomic32(ValueEmitContext& ctx, const IR::Inst& inst) {
@@ -905,8 +905,8 @@ uint32_t EmitAtomic32(ValueEmitContext& ctx, const IR::Inst& inst) {
 		if (mem.kind == IR::ResourceKind::Lds) {
 			const auto semantics =
 			    spv::MemorySemanticsAcquireReleaseMask | spv::MemorySemanticsWorkgroupMemoryMask;
-			ctx.state.builder.AddFunction({spv::OpMemoryBarrier, ConstantU32(ctx.state, scope),
-			                               ConstantU32(ctx.state, semantics)});
+			ctx.state.builder.AddFunction(spv::OpMemoryBarrier, ConstantU32(ctx.state, scope),
+			                              ConstantU32(ctx.state, semantics));
 		} else {
 			EmitDeviceAtomicMemoryBarrier(ctx.state);
 		}
@@ -932,11 +932,11 @@ uint32_t EmitBufferAtomic64(ValueEmitContext& ctx, const IR::Inst& inst) {
 			                                 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, spv::ScopeDevice),
-			             ConstantU32(state, spv::MemorySemanticsMaskNone), value});
+			            SpirvAtomicOpcode(inst.GetOpcode()), TypeScalarU64(state), old,
+			            EmitStorageBufferElementPointer(state, resource, index,
+			                                            TypeStorageBufferU64ElementPointer(state)),
+			            ConstantU32(state, spv::ScopeDevice),
+			            ConstantU32(state, spv::MemorySemanticsMaskNone), value);
 			        EmitDeviceAtomicMemoryBarrier(state);
 			        return Unary(state, spv::OpBitcast, TypeU64(state), old);
 		        });
@@ -959,11 +959,11 @@ void EmitSharedFloatAtomic(ValueEmitContext& ctx, const IR::Inst& inst) {
 		const auto access = PrepareMemoryElement(ctx, mem, DwordIndex(ctx, inst, mem));
 		EmitIfCondition(
 		    ctx.state, EmitMemoryElementInBounds(ctx.state, access.resource, access.index), [&]() {
-			    ctx.state.builder.AddFunction(
-			        {spv::OpStore, ctx.scratch_u32_variable, ctx.Arg(inst, 1)});
+			    ctx.state.builder.AddFunction(spv::OpStore, ctx.scratch_u32_variable,
+			                                  ctx.Arg(inst, 1));
 			    const auto data = ctx.state.builder.AllocateId();
-			    ctx.state.builder.AddFunction(
-			        {spv::OpLoad, TypeU32(ctx.state), data, ctx.scratch_u32_variable});
+			    ctx.state.builder.AddFunction(spv::OpLoad, TypeU32(ctx.state), data,
+			                                  ctx.scratch_u32_variable);
 			    AtomicUpdate(
 			        ctx.state, EmitMemoryElementPointer(ctx.state, access.resource, access.index),
 			        mem.kind, [&](uint32_t old) {
@@ -1006,8 +1006,8 @@ uint32_t EmitAppendConsume(ValueEmitContext& ctx, const IR::Inst& inst) {
 	const auto ballot = ctx.Ballot(inst.Arg(1));
 	const auto low    = state.builder.AllocateId();
 	const auto high   = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), low, ballot, 0});
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), high, ballot, 1});
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), low, ballot, 0);
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), high, ballot, 1);
 	const auto count = Binary(state, spv::OpIAdd, TypeU32(state),
 	                          Unary(state, spv::OpBitCount, TypeU32(state), low),
 	                          Unary(state, spv::OpBitCount, TypeU32(state), high));
@@ -1033,17 +1033,17 @@ uint32_t EmitAppendConsume(ValueEmitContext& ctx, const IR::Inst& inst) {
 	                 AndCondition(state, storage_bounds, m0_bounds)));
 	const auto atomic = EmitValueOrZeroIfCondition(state, condition, [&]() {
 		const auto value = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {append ? spv::OpAtomicIAdd : spv::OpAtomicISub, TypeU32(state), value,
-		     EmitMemoryElementPointer(state, access, index),
-		     ConstantU32(state, mem.kind == IR::ResourceKind::Gds ? spv::ScopeDevice
-		                                                          : spv::ScopeWorkgroup),
-		     ConstantU32(state, spv::MemorySemanticsMaskNone), count});
+		state.builder.AddFunction(append ? spv::OpAtomicIAdd : spv::OpAtomicISub, TypeU32(state),
+		                          value, EmitMemoryElementPointer(state, access, index),
+		                          ConstantU32(state, mem.kind == IR::ResourceKind::Gds
+		                                                 ? spv::ScopeDevice
+		                                                 : spv::ScopeWorkgroup),
+		                          ConstantU32(state, spv::MemorySemanticsMaskNone), count);
 		return value;
 	});
 	const auto result = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpGroupNonUniformShuffle, TypeU32(state), result,
-	                           ConstantU32(state, spv::ScopeSubgroup), atomic, source_lane});
+	state.builder.AddFunction(spv::OpGroupNonUniformShuffle, TypeU32(state), result,
+	                          ConstantU32(state, spv::ScopeSubgroup), atomic, source_lane);
 	return result;
 }
 
@@ -1053,9 +1053,9 @@ uint32_t EmitReadConst(ValueEmitContext& ctx, const IR::Inst& inst) {
 		ctx.Fail(inst, "requires the flattened SRT descriptor");
 	}
 	const auto pointer = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpAccessChain, TypeStorageBufferElementPointer(state), pointer,
-	                           state.flattened_srt_variable, ConstantU32(state, 0),
-	                           ctx.Arg(inst, 1)});
+	state.builder.AddFunction(spv::OpAccessChain, TypeStorageBufferElementPointer(state), pointer,
+	                          state.flattened_srt_variable, ConstantU32(state, 0),
+	                          ctx.Arg(inst, 1));
 	return EmitNative<spv::OpLoad, IR::Type::U32>(state, pointer);
 }
 
@@ -1131,7 +1131,7 @@ uint32_t EmitSharedIncDec(ValueEmitContext& ctx, const IR::Inst& inst) {
 
 uint32_t EmitSwizzleU32(ValueEmitContext& ctx, const IR::Inst& inst) {
 	auto& state = ctx.state;
-	state.builder.AddFunction({spv::OpStore, ctx.scratch_u32_variable, ctx.Arg(inst, 0)});
+	state.builder.AddFunction(spv::OpStore, ctx.scratch_u32_variable, ctx.Arg(inst, 0));
 	const auto source = EmitNative<spv::OpLoad, IR::Type::U32>(state, ctx.scratch_u32_variable);
 	const auto target = EmitDsSwizzleTargetLane(state, EmitSubgroupLocalInvocationId(state),
 	                                            inst.Arg(1).IsImmediate() ? inst.Arg(1).U32() : 0);

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

@@ -7,27 +7,27 @@ uint32_t EmitShaderDataDwordLoad(EmitterState& state, uint32_t dword_index) {
 		dword_index += state.program.bindings.push_data_start_dword;
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpAccessChain, TypePushConstantElementPointer(state),
-		                           pointer, state.push_constant_variable, ConstantU32(state, 0),
-		                           ConstantU32(state, dword_index)});
-		state.builder.AddFunction({spv::OpLoad, TypeU32(state), value, pointer});
+		state.builder.AddFunction(spv::OpAccessChain, TypePushConstantElementPointer(state),
+		                          pointer, state.push_constant_variable, ConstantU32(state, 0),
+		                          ConstantU32(state, dword_index));
+		state.builder.AddFunction(spv::OpLoad, TypeU32(state), value, pointer);
 		return value;
 	}
 	if (state.shader_data_storage_variable != 0) {
 		const auto pointer = state.builder.AllocateId();
 		const auto value   = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpAccessChain, TypeStorageBufferElementPointer(state),
-		                           pointer, state.shader_data_storage_variable,
-		                           ConstantU32(state, 0), ConstantU32(state, dword_index)});
-		state.builder.AddFunction({spv::OpLoad, TypeU32(state), value, pointer});
+		state.builder.AddFunction(spv::OpAccessChain, TypeStorageBufferElementPointer(state),
+		                          pointer, state.shader_data_storage_variable,
+		                          ConstantU32(state, 0), ConstantU32(state, dword_index));
+		state.builder.AddFunction(spv::OpLoad, TypeU32(state), value, pointer);
 		return value;
 	}
 	return ConstantU32(state, 0);
 }
 
-uint32_t EmitBinaryU32(EmitterState& state, uint32_t opcode, uint32_t lhs, uint32_t rhs) {
+uint32_t EmitBinaryU32(EmitterState& state, spv::Op opcode, uint32_t lhs, uint32_t rhs) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction({opcode, TypeU32(state), ret, lhs, rhs});
+	state.builder.AddFunction(opcode, TypeU32(state), ret, lhs, rhs);
 	return ret;
 }
 
@@ -88,12 +88,12 @@ MemoryResourceAccess PrepareStorageBufferResourceAccess(EmitterState& state,
 	    ResourceForDescriptor(state, IR::DescriptorBindingKind::Buffers, mem.resource);
 	MemoryResourceAccess access {.kind = mem.kind};
 	access.object_pointer = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpAccessChain, pointer_type, access.object_pointer, variable,
-	                           ConstantU32(state, array_index)});
+	state.builder.AddFunction(spv::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(
-	    {spv::OpArrayLength, TypeU32(state), access.length, access.object_pointer, 0});
+	state.builder.AddFunction(spv::OpArrayLength, TypeU32(state), access.length,
+	                          access.object_pointer, 0);
 	return access;
 }
 
@@ -139,8 +139,8 @@ MemoryResourceAccess PrepareMemoryResourceAccess(EmitterState& state, const IR::
 		default: EXIT("unsupported memory resource kind: %u\n", static_cast<unsigned>(mem.kind));
 	}
 	access.length = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpArrayLength, TypeU32(state), access.length, access.object_pointer, 0});
+	state.builder.AddFunction(spv::OpArrayLength, TypeU32(state), access.length,
+	                          access.object_pointer, 0);
 	return access;
 }
 
@@ -152,7 +152,7 @@ uint32_t EmitMemoryElementIndex(EmitterState& state, const MemoryResourceAccess&
 uint32_t EmitMemoryElementInBounds(EmitterState& state, const MemoryResourceAccess& access,
                                    uint32_t index) {
 	const auto in_bounds = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpULessThan, TypeBool(state), in_bounds, index, access.length});
+	state.builder.AddFunction(spv::OpULessThan, TypeBool(state), in_bounds, index, access.length);
 	return in_bounds;
 }
 
@@ -165,8 +165,8 @@ uint32_t EmitMemoryElementPointer(EmitterState& state, const MemoryResourceAcces
 		    : ShaderWorkgroupInput(state.program.stage, state.input_info) != nullptr
 		        ? spv::StorageClassWorkgroup
 		        : spv::StorageClassFunction;
-		state.builder.AddFunction({spv::OpAccessChain, TypeU32ElementPointer(state, storage_class),
-		                           pointer, access.object_pointer, index});
+		state.builder.AddFunction(spv::OpAccessChain, TypeU32ElementPointer(state, storage_class),
+		                          pointer, access.object_pointer, index);
 		return pointer;
 	}
 	return EmitStorageBufferElementPointer(state, access, index,
@@ -177,14 +177,14 @@ uint32_t EmitStorageBufferElementPointer(EmitterState& state,
                                          const MemoryResourceAccess& access, uint32_t index,
                                          uint32_t pointer_type) {
 	const auto pointer = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpAccessChain, pointer_type, pointer, access.object_pointer,
-	                           ConstantU32(state, 0), index});
+	state.builder.AddFunction(spv::OpAccessChain, pointer_type, pointer, access.object_pointer,
+	                          ConstantU32(state, 0), index);
 	return pointer;
 }
 
 uint32_t EmitTBufferBitcastU32ToI32(EmitterState& state, uint32_t value) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpBitcast, TypeI32(state), ret, value});
+	state.builder.AddFunction(spv::OpBitcast, TypeI32(state), ret, value);
 	return ret;
 }
 
@@ -218,10 +218,10 @@ uint32_t EmitUFloatToF32Bits(EmitterState& state, uint32_t raw, uint32_t bits) {
 
 	const auto mantissa_f32 = state.builder.AllocateId();
 	const auto subnormal    = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpConvertUToF, TypeF32(state), mantissa_f32, mantissa});
+	state.builder.AddFunction(spv::OpConvertUToF, TypeF32(state), mantissa_f32, mantissa);
 	state.builder.AddFunction(
-	    {spv::OpFMul, TypeF32(state), subnormal, mantissa_f32,
-	     ConstantF32Value(state, std::ldexp(1.0f, 1 - 15 - static_cast<int>(mantissa_bits)))});
+	    spv::OpFMul, TypeF32(state), subnormal, mantissa_f32,
+	    ConstantF32Value(state, std::ldexp(1.0f, 1 - 15 - static_cast<int>(mantissa_bits))));
 
 	const auto zero_exp    = EmitCompareU32Constant(state, spv::OpIEqual, exponent, 0);
 	const auto special_exp = EmitCompareU32Constant(state, spv::OpIEqual, exponent, 31);
@@ -238,22 +238,22 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 		case Format::ComponentType::Sint: return raw;
 		case Format::ComponentType::Uscaled: {
 			const auto value = state.builder.AllocateId();
-			state.builder.AddFunction({spv::OpConvertUToF, TypeF32(state), value, raw});
+			state.builder.AddFunction(spv::OpConvertUToF, TypeF32(state), value, raw);
 			return EmitBitcastF32ToU32(state, value);
 		}
 		case Format::ComponentType::Sscaled: {
 			const auto signed_raw = EmitTBufferBitcastU32ToI32(state, raw);
 			const auto value      = state.builder.AllocateId();
-			state.builder.AddFunction({spv::OpConvertSToF, TypeF32(state), value, signed_raw});
+			state.builder.AddFunction(spv::OpConvertSToF, TypeF32(state), value, signed_raw);
 			return EmitBitcastF32ToU32(state, value);
 		}
 		case Format::ComponentType::Unorm: {
 			const auto value      = state.builder.AllocateId();
 			const auto normalized = state.builder.AllocateId();
 			const auto max_value  = static_cast<float>((1u << bits) - 1u);
-			state.builder.AddFunction({spv::OpConvertUToF, TypeF32(state), value, raw});
-			state.builder.AddFunction({spv::OpFDiv, TypeF32(state), normalized, value,
-			                           ConstantF32Value(state, max_value)});
+			state.builder.AddFunction(spv::OpConvertUToF, TypeF32(state), value, raw);
+			state.builder.AddFunction(spv::OpFDiv, TypeF32(state), normalized, value,
+			                          ConstantF32Value(state, max_value));
 			return EmitBitcastF32ToU32(state, normalized);
 		}
 		case Format::ComponentType::Snorm: {
@@ -262,11 +262,11 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 			const auto normalized = state.builder.AllocateId();
 			const auto clamped    = state.builder.AllocateId();
 			const auto max_value  = static_cast<float>((1u << (bits - 1u)) - 1u);
-			state.builder.AddFunction({spv::OpConvertSToF, TypeF32(state), value, signed_raw});
-			state.builder.AddFunction({spv::OpFDiv, TypeF32(state), normalized, value,
-			                           ConstantF32Value(state, max_value)});
-			state.builder.AddFunction({spv::OpExtInst, TypeF32(state), clamped, GlslStd450(state),
-			                           GLSLstd450FMax, normalized, ConstantF32Value(state, -1.0f)});
+			state.builder.AddFunction(spv::OpConvertSToF, TypeF32(state), value, signed_raw);
+			state.builder.AddFunction(spv::OpFDiv, TypeF32(state), normalized, value,
+			                          ConstantF32Value(state, max_value));
+			state.builder.AddFunction(spv::OpExtInst, TypeF32(state), clamped, GlslStd450(state),
+			                          GLSLstd450FMax, normalized, ConstantF32Value(state, -1.0f));
 			return EmitBitcastF32ToU32(state, clamped);
 		}
 		case Format::ComponentType::Float:
@@ -284,8 +284,8 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 void EmitDeviceAtomicMemoryBarrier(EmitterState& state) {
 	const auto semantics =
 	    spv::MemorySemanticsAcquireReleaseMask | spv::MemorySemanticsUniformMemoryMask;
-	state.builder.AddFunction({spv::OpMemoryBarrier, ConstantU32(state, spv::ScopeDevice),
-	                           ConstantU32(state, semantics)});
+	state.builder.AddFunction(spv::OpMemoryBarrier, ConstantU32(state, spv::ScopeDevice),
+	                          ConstantU32(state, semantics));
 }
 
 uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t source,
@@ -300,7 +300,7 @@ uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t
 		const auto negative = EmitCompareU32Constant(state, spv::OpINotEqual,
 		                                             EmitAndConstant(state, bits, 0x80000000u), 0u);
 		const auto negative_key = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpNot, TypeU32(state), negative_key, bits});
+		state.builder.AddFunction(spv::OpNot, TypeU32(state), negative_key, bits);
 		const auto positive_key =
 		    EmitBinaryU32(state, spv::OpBitwiseXor, bits, ConstantU32(state, 0x80000000u));
 		return OrderedBits {
@@ -313,8 +313,8 @@ uint32_t EmitFloatAtomicReplacement(EmitterState& state, uint32_t old, uint32_t
 	    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 ? spv::OpUGreaterThan : spv::OpULessThan, TypeBool(state),
-	                           compare, source_class.key, old_class.key});
+	state.builder.AddFunction(max_value ? spv::OpUGreaterThan : spv::OpULessThan, TypeBool(state),
+	                          compare, source_class.key, old_class.key);
 	return EmitSelectValueU32(
 	    state, EmitLogicalAndBool(state, EmitLogicalNotBool(state, unordered), compare), source,
 	    old);
@@ -342,17 +342,17 @@ uint32_t EmitDsSwizzleTargetLane(EmitterState& state, uint32_t subid, uint32_t c
 		const auto perm   = state.builder.AllocateId();
 		const auto base   = state.builder.AllocateId();
 		const auto target = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpBitwiseAnd, TypeU32(state), lane2, subid, ConstantU32(state, 3)});
-		state.builder.AddFunction(
-		    {spv::OpShiftLeftLogical, TypeU32(state), shift, lane2, ConstantU32(state, 1)});
-		state.builder.AddFunction(
-		    {spv::OpShiftRightLogical, TypeU32(state), perm0, ConstantU32(state, control), shift});
-		state.builder.AddFunction(
-		    {spv::OpBitwiseAnd, TypeU32(state), perm, perm0, ConstantU32(state, 3)});
-		state.builder.AddFunction(
-		    {spv::OpBitwiseAnd, TypeU32(state), base, subid, ConstantU32(state, 0xfffffffcu)});
-		state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), target, base, perm});
+		state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane2, subid,
+		                          ConstantU32(state, 3));
+		state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), shift, lane2,
+		                          ConstantU32(state, 1));
+		state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), perm0,
+		                          ConstantU32(state, control), shift);
+		state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), perm, perm0,
+		                          ConstantU32(state, 3));
+		state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), base, subid,
+		                          ConstantU32(state, 0xfffffffcu));
+		state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, base, perm);
 		return target;
 	}
 
@@ -362,17 +362,17 @@ uint32_t EmitDsSwizzleTargetLane(EmitterState& state, uint32_t subid, uint32_t c
 	const auto xored  = state.builder.AllocateId();
 	const auto base   = state.builder.AllocateId();
 	const auto target = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), lane, subid, ConstantU32(state, 31)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), masked, lane, ConstantU32(state, control & 0x1fu)});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), ored, masked,
-	                           ConstantU32(state, (control >> 5u) & 0x1fu)});
-	state.builder.AddFunction({spv::OpBitwiseXor, TypeU32(state), xored, ored,
-	                           ConstantU32(state, (control >> 10u) & 0x1fu)});
-	state.builder.AddFunction(
-	    {spv::OpBitwiseAnd, TypeU32(state), base, subid, ConstantU32(state, 0xffffffe0u)});
-	state.builder.AddFunction({spv::OpBitwiseOr, TypeU32(state), target, base, xored});
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
+	                          ConstantU32(state, 31));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), masked, lane,
+	                          ConstantU32(state, control & 0x1fu));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), ored, masked,
+	                          ConstantU32(state, (control >> 5u) & 0x1fu));
+	state.builder.AddFunction(spv::OpBitwiseXor, TypeU32(state), xored, ored,
+	                          ConstantU32(state, (control >> 10u) & 0x1fu));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), base, subid,
+	                          ConstantU32(state, 0xffffffe0u));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, base, xored);
 	return target;
 }
 

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

@@ -4,15 +4,15 @@ namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 namespace {
 
 uint32_t MeshArray(EmitterState& state, spv::StorageClass storage, uint32_t type, uint32_t count) {
-	const auto array = state.builder.Type(spv::OpTypeArray, {type, ConstantU32(state, count)});
+	const auto array = state.builder.Type(spv::OpTypeArray, type, ConstantU32(state, count));
 	return state.builder.DefineGlobalVariable(TypePointer(state, storage, array), storage);
 }
 
 uint32_t MeshElement(EmitterState& state, uint32_t variable, spv::StorageClass storage,
                      uint32_t type, uint32_t index) {
 	const auto pointer = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpAccessChain, TypePointer(state, storage, type), pointer, variable, index});
+	state.builder.AddFunction(spv::OpAccessChain, TypePointer(state, storage, type), pointer,
+	                          variable, index);
 	return pointer;
 }
 
@@ -20,7 +20,7 @@ uint32_t MeshLoad(EmitterState& state, uint32_t variable, spv::StorageClass stor
                   uint32_t index) {
 	const auto pointer = MeshElement(state, variable, storage, type, index);
 	const auto value   = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpLoad, type, value, pointer});
+	state.builder.AddFunction(spv::OpLoad, type, value, pointer);
 	return value;
 }
 
@@ -50,17 +50,17 @@ void DefineMeshOutputs(EmitterState& state) {
 		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(
-			    {spv::OpDecorate, output.variable_id, spv::DecorationLocation, output.location});
+			state.builder.AddAnnotation(spv::OpDecorate, output.variable_id,
+			                            spv::DecorationLocation, output.location);
 		} else {
-			state.builder.AddAnnotation(
-			    {spv::OpDecorate, output.variable_id, spv::DecorationBuiltIn,
-			     output.kind == IR::StageOutputKind::Layer ? spv::BuiltInLayer
-			                                               : spv::BuiltInPosition});
+			state.builder.AddAnnotation(spv::OpDecorate, output.variable_id, spv::DecorationBuiltIn,
+			                            output.kind == IR::StageOutputKind::Layer
+			                                ? spv::BuiltInLayer
+			                                : spv::BuiltInPosition);
 		}
 		if (output.kind == IR::StageOutputKind::Layer) {
-			state.builder.AddAnnotation({spv::OpDecorate, output.variable_id,
-			                             spv::DecorationPerPrimitiveEXT}); // PerPrimitiveEXT
+			state.builder.AddAnnotation(spv::OpDecorate, output.variable_id,
+			                            spv::DecorationPerPrimitiveEXT); // PerPrimitiveEXT
 		}
 	}
 	state.mesh_allocation = MeshArray(state, spv::StorageClassWorkgroup, TypeU32(state), 2);
@@ -73,11 +73,11 @@ void DefineMeshOutputs(EmitterState& state) {
 	state.interface_variables.push_back(state.mesh_primitives);
 	state.interface_variables.push_back(state.mesh_cull);
 	state.builder.AddAnnotation(
-	    {spv::OpDecorate, state.mesh_primitives, spv::DecorationBuiltIn,
-	     spv::BuiltInPrimitiveTriangleIndicesEXT}); // PrimitiveTriangleIndicesEXT
-	state.builder.AddAnnotation({spv::OpDecorate, state.mesh_cull, spv::DecorationBuiltIn,
-	                             spv::BuiltInCullPrimitiveEXT}); // CullPrimitiveEXT
-	state.builder.AddAnnotation({spv::OpDecorate, state.mesh_cull, spv::DecorationPerPrimitiveEXT});
+	    spv::OpDecorate, state.mesh_primitives, spv::DecorationBuiltIn,
+	    spv::BuiltInPrimitiveTriangleIndicesEXT); // PrimitiveTriangleIndicesEXT
+	state.builder.AddAnnotation(spv::OpDecorate, state.mesh_cull, spv::DecorationBuiltIn,
+	                            spv::BuiltInCullPrimitiveEXT); // CullPrimitiveEXT
+	state.builder.AddAnnotation(spv::OpDecorate, state.mesh_cull, spv::DecorationPerPrimitiveEXT);
 }
 
 uint32_t MeshOutputPointer(EmitterState& state, IR::StageOutputKind kind, uint32_t index) {
@@ -103,45 +103,45 @@ uint32_t MeshPrimitivePointer(EmitterState& state) {
 void EmitMeshAllocate(ValueEmitContext& ctx, const IR::Inst& inst) {
 	auto&      state = ctx.state;
 	const auto first = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpIEqual, TypeBool(state), first,
-	                           EmitLocalInvocationIndex(state), ConstantU32(state, 0)});
+	state.builder.AddFunction(spv::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({spv::OpBitFieldUExtract, TypeU32(state), value, allocation,
-			                           ConstantU32(state, field * 12u),
-			                           ConstantU32(state, field == 0 ? 10u : 11u)});
+			state.builder.AddFunction(spv::OpBitFieldUExtract, TypeU32(state), value, allocation,
+			                          ConstantU32(state, field * 12u),
+			                          ConstantU32(state, field == 0 ? 10u : 11u));
 			const auto pointer =
 			    MeshElement(state, state.mesh_allocation, spv::StorageClassWorkgroup,
 			                TypeU32(state), ConstantU32(state, field));
-			state.builder.AddFunction({spv::OpStore, pointer, value});
+			state.builder.AddFunction(spv::OpStore, pointer, value);
 		}
 	});
 }
 
 void EmitMeshEntryPoint(EmitterState& state) {
-	state.builder.AddFunction({spv::OpFunction, TypeVoid(state), state.main_func,
-	                           spv::FunctionControlMaskNone, TypeFunction(state)});
+	state.builder.AddFunction(spv::OpFunction, TypeVoid(state), state.main_func,
+	                          spv::FunctionControlMaskNone, TypeFunction(state));
 	EmitLabel(state, state.builder.AllocateId());
-	state.builder.AddFunction(
-	    {spv::OpFunctionCall, TypeVoid(state), state.builder.AllocateId(), state.mesh_guest_func});
+	state.builder.AddFunction(spv::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({spv::OpControlBarrier, ConstantU32(state, spv::ScopeWorkgroup),
-	                           ConstantU32(state, spv::ScopeWorkgroup),
-	                           ConstantU32(state, spv::MemorySemanticsAcquireReleaseMask |
-	                                                  spv::MemorySemanticsWorkgroupMemoryMask)});
+	state.builder.AddFunction(spv::OpControlBarrier, ConstantU32(state, spv::ScopeWorkgroup),
+	                          ConstantU32(state, spv::ScopeWorkgroup),
+	                          ConstantU32(state, spv::MemorySemanticsAcquireReleaseMask |
+	                                                 spv::MemorySemanticsWorkgroupMemoryMask));
 	const auto vertices   = MeshLoad(state, state.mesh_allocation, spv::StorageClassWorkgroup,
 	                                 TypeU32(state), ConstantU32(state, 0));
 	const auto primitives = MeshLoad(state, state.mesh_allocation, spv::StorageClassWorkgroup,
 	                                 TypeU32(state), ConstantU32(state, 1));
-	state.builder.AddFunction(
-	    {spv::OpSetMeshOutputsEXT, vertices, primitives}); // OpSetMeshOutputsEXT
+	state.builder.AddFunction(spv::OpSetMeshOutputsEXT, 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({spv::OpULessThan, TypeBool(state), is_vertex, index, vertices});
+		state.builder.AddFunction(spv::OpULessThan, TypeBool(state), is_vertex, index, vertices);
 		EmitIfCondition(state, is_vertex, [&] {
 			for (const auto& output: state.outputs) {
 				if (output.kind == IR::StageOutputKind::Layer) {
@@ -153,12 +153,12 @@ void EmitMeshEntryPoint(EmitterState& state) {
 				             ConstantU32(state, half));
 				const auto pointer =
 				    MeshElement(state, output.variable_id, spv::StorageClassOutput, type, index);
-				state.builder.AddFunction({spv::OpStore, pointer, value});
+				state.builder.AddFunction(spv::OpStore, pointer, value);
 			}
 		});
 		const auto is_primitive = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpULessThan, TypeBool(state), is_primitive, index, primitives});
+		state.builder.AddFunction(spv::OpULessThan, TypeBool(state), is_primitive, index,
+		                          primitives);
 		EmitIfCondition(state, is_primitive, [&] {
 			const auto packed = MeshLoad(state, state.mesh_primitive_data, spv::StorageClassPrivate,
 			                             TypeU32(state), ConstantU32(state, half));
@@ -166,25 +166,25 @@ void EmitMeshEntryPoint(EmitterState& state) {
 			for (uint32_t component = 0; component < 3; component++) {
 				vertex[component] = state.builder.AllocateId();
 				state.builder.AddFunction(
-				    {spv::OpBitFieldUExtract, TypeU32(state), vertex[component], packed,
-				     ConstantU32(state, component * 10u), ConstantU32(state, 10)});
+				    spv::OpBitFieldUExtract, TypeU32(state), vertex[component], packed,
+				    ConstantU32(state, component * 10u), ConstantU32(state, 10));
 			}
 			const auto triangle = state.builder.AllocateId();
-			state.builder.AddFunction({spv::OpCompositeConstruct, TypeU32Vector(state, 3), triangle,
-			                           vertex[0], vertex[1], vertex[2]});
+			state.builder.AddFunction(spv::OpCompositeConstruct, TypeU32Vector(state, 3), triangle,
+			                          vertex[0], vertex[1], vertex[2]);
 			const auto triangle_pointer =
 			    MeshElement(state, state.mesh_primitives, spv::StorageClassOutput,
 			                TypeU32Vector(state, 3), index);
-			state.builder.AddFunction({spv::OpStore, triangle_pointer, triangle});
+			state.builder.AddFunction(spv::OpStore, triangle_pointer, triangle);
 			const auto null_bit =
 			    EmitBinaryU32(state, spv::OpBitwiseAnd, packed, ConstantU32(state, 0x80000000u));
 			const auto culled = state.builder.AllocateId();
-			state.builder.AddFunction(
-			    {spv::OpINotEqual, TypeBool(state), culled, null_bit, ConstantU32(state, 0)});
-			state.builder.AddFunction({spv::OpStore,
-			                           MeshElement(state, state.mesh_cull, spv::StorageClassOutput,
-			                                       TypeBool(state), index),
-			                           culled});
+			state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), culled, null_bit,
+			                          ConstantU32(state, 0));
+			state.builder.AddFunction(spv::OpStore,
+			                          MeshElement(state, state.mesh_cull, spv::StorageClassOutput,
+			                                      TypeBool(state), index),
+			                          culled);
 			for (const auto& output: state.outputs) {
 				if (output.kind != IR::StageOutputKind::Layer) {
 					continue;
@@ -194,13 +194,13 @@ void EmitMeshEntryPoint(EmitterState& state) {
 				                            vertex[state.input_info.vertex->mesh.provoking_vertex]);
 				const auto pointer = MeshElement(state, output.variable_id, spv::StorageClassOutput,
 				                                 TypeU32(state), index);
-				state.builder.AddFunction({spv::OpStore, pointer, layer});
+				state.builder.AddFunction(spv::OpStore, pointer, layer);
 			}
 		});
 	}
 	state.lane_half = 0;
-	state.builder.AddFunction({spv::OpReturn});
-	state.builder.AddFunction({spv::OpFunctionEnd});
+	state.builder.AddFunction(spv::OpReturn);
+	state.builder.AddFunction(spv::OpFunctionEnd);
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter

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

@@ -13,11 +13,11 @@ uint32_t TypeBool(EmitterState& state) {
 }
 
 uint32_t TypeBoolVector(EmitterState& state, uint32_t components) {
-	return state.builder.Type(spv::OpTypeVector, {TypeBool(state), components});
+	return state.builder.Type(spv::OpTypeVector, TypeBool(state), components);
 }
 
 uint32_t TypeU32(EmitterState& state) {
-	return state.builder.Type(spv::OpTypeInt, {32, 0});
+	return state.builder.Type(spv::OpTypeInt, 32, 0);
 }
 
 uint32_t TypeU64(EmitterState& state) {
@@ -25,29 +25,29 @@ uint32_t TypeU64(EmitterState& state) {
 }
 
 uint32_t TypeScalarU64(EmitterState& state) {
-	return state.builder.Type(spv::OpTypeInt, {64, 0});
+	return state.builder.Type(spv::OpTypeInt, 64, 0);
 }
 
 uint32_t TypeU32Pair(EmitterState& state) {
 	const auto element = TypeU32(state);
-	return state.builder.Type(spv::OpTypeStruct, {element, element});
+	return state.builder.Type(spv::OpTypeStruct, element, element);
 }
 
 uint32_t TypeI32(EmitterState& state) {
-	return state.builder.Type(spv::OpTypeInt, {32, 1});
+	return state.builder.Type(spv::OpTypeInt, 32, 1);
 }
 
 uint32_t TypeI32Pair(EmitterState& state) {
 	const auto element = TypeI32(state);
-	return state.builder.Type(spv::OpTypeStruct, {element, element});
+	return state.builder.Type(spv::OpTypeStruct, element, element);
 }
 
 uint32_t TypeF32(EmitterState& state) {
-	return state.builder.Type(spv::OpTypeFloat, {32});
+	return state.builder.Type(spv::OpTypeFloat, 32);
 }
 
 uint32_t TypeU32Vector(EmitterState& state, uint32_t components) {
-	return state.builder.Type(spv::OpTypeVector, {TypeU32(state), components});
+	return state.builder.Type(spv::OpTypeVector, TypeU32(state), components);
 }
 
 uint32_t TypeU32Composite(EmitterState& state, uint32_t components) {
@@ -56,31 +56,32 @@ uint32_t TypeU32Composite(EmitterState& state, uint32_t components) {
 }
 
 uint32_t TypeI32Vector(EmitterState& state, uint32_t components) {
-	return state.builder.Type(spv::OpTypeVector, {TypeI32(state), components});
+	return state.builder.Type(spv::OpTypeVector, TypeI32(state), components);
 }
 
 uint32_t TypeF32Vector(EmitterState& state, uint32_t components) {
-	return state.builder.Type(spv::OpTypeVector, {TypeF32(state), components});
+	return state.builder.Type(spv::OpTypeVector, TypeF32(state), components);
 }
 
 uint32_t TypePointer(EmitterState& state, spv::StorageClass storage_class, uint32_t pointee) {
-	return state.builder.Type(spv::OpTypePointer, {storage_class, pointee});
+	return state.builder.Type(spv::OpTypePointer, storage_class, pointee);
 }
 
 uint32_t TypeFunction(EmitterState& state) {
-	return state.builder.Type(spv::OpTypeFunction, {TypeVoid(state)});
+	return state.builder.Type(spv::OpTypeFunction, TypeVoid(state));
 }
 
 uint32_t StorageRuntimeArrayType(EmitterState& state) {
 	return state.builder.DecoratedType(
-	    spv::OpTypeRuntimeArray, {TypeU32(state)},
-	    {{spv::OpDecorate, {spv::DecorationArrayStride, sizeof(uint32_t)}}});
+	    spv::OpTypeRuntimeArray,
+	    {{spv::OpDecorate, {spv::DecorationArrayStride, sizeof(uint32_t)}}}, TypeU32(state));
 }
 
 uint32_t StorageBufferType(EmitterState& state) {
-	return state.builder.DecoratedType(spv::OpTypeStruct, {StorageRuntimeArrayType(state)},
+	return state.builder.DecoratedType(spv::OpTypeStruct,
 	                                   {{spv::OpMemberDecorate, {0, spv::DecorationOffset, 0}},
-	                                    {spv::OpDecorate, {spv::DecorationBlock}}});
+	                                    {spv::OpDecorate, {spv::DecorationBlock}}},
+	                                   StorageRuntimeArrayType(state));
 }
 
 uint32_t TypeStorageBufferPointer(EmitterState& state) {
@@ -93,14 +94,15 @@ uint32_t TypeStorageBufferElementPointer(EmitterState& state) {
 
 uint32_t StorageU64RuntimeArrayType(EmitterState& state) {
 	return state.builder.DecoratedType(
-	    spv::OpTypeRuntimeArray, {TypeScalarU64(state)},
-	    {{spv::OpDecorate, {spv::DecorationArrayStride, sizeof(uint64_t)}}});
+	    spv::OpTypeRuntimeArray,
+	    {{spv::OpDecorate, {spv::DecorationArrayStride, sizeof(uint64_t)}}}, TypeScalarU64(state));
 }
 
 uint32_t StorageBufferU64Type(EmitterState& state) {
-	return state.builder.DecoratedType(spv::OpTypeStruct, {StorageU64RuntimeArrayType(state)},
+	return state.builder.DecoratedType(spv::OpTypeStruct,
 	                                   {{spv::OpMemberDecorate, {0, spv::DecorationOffset, 0}},
-	                                    {spv::OpDecorate, {spv::DecorationBlock}}});
+	                                    {spv::OpDecorate, {spv::DecorationBlock}}},
+	                                   StorageU64RuntimeArrayType(state));
 }
 
 uint32_t TypeStorageBufferU64Pointer(EmitterState& state) {
@@ -122,7 +124,7 @@ uint32_t TypePushConstantElementPointer(EmitterState& state) {
 uint32_t TypeU32ArrayPointer(EmitterState& state, spv::StorageClass storage_class,
                              uint32_t dwords) {
 	const auto count = ConstantU32(state, std::max(dwords, 1u));
-	const auto array = state.builder.Type(spv::OpTypeArray, {TypeU32(state), count});
+	const auto array = state.builder.Type(spv::OpTypeArray, TypeU32(state), count);
 	return TypePointer(state, storage_class, array);
 }
 
@@ -135,29 +137,31 @@ namespace {
 uint32_t PushConstantArrayType(EmitterState& state) {
 	const auto count = ConstantU32(state, IR::PushData::DwordCount);
 	return state.builder.DecoratedType(
-	    spv::OpTypeArray, {TypeU32(state), count},
-	    {{spv::OpDecorate, {spv::DecorationArrayStride, sizeof(uint32_t)}}});
+	    spv::OpTypeArray, {{spv::OpDecorate, {spv::DecorationArrayStride, sizeof(uint32_t)}}},
+	    TypeU32(state), count);
 }
 
 uint32_t PushConstantBlockType(EmitterState& state) {
-	return state.builder.DecoratedType(spv::OpTypeStruct, {PushConstantArrayType(state)},
+	return state.builder.DecoratedType(spv::OpTypeStruct,
 	                                   {{spv::OpMemberDecorate, {0, spv::DecorationOffset, 0}},
-	                                    {spv::OpDecorate, {spv::DecorationBlock}}});
+	                                    {spv::OpDecorate, {spv::DecorationBlock}}},
+	                                   PushConstantArrayType(state));
 }
 
 uint32_t PerVertexType(EmitterState& state) {
 	return state.builder.DecoratedType(
-	    spv::OpTypeStruct, {TypeF32Vector(state, 4)},
+	    spv::OpTypeStruct,
 	    {{spv::OpMemberDecorate, {0, spv::DecorationBuiltIn, spv::BuiltInPosition}},
-	     {spv::OpDecorate, {spv::DecorationBlock}}});
+	     {spv::OpDecorate, {spv::DecorationBlock}}},
+	    TypeF32Vector(state, 4));
 }
 
 uint32_t SampleMaskArrayType(EmitterState& state) {
-	return state.builder.Type(spv::OpTypeArray, {TypeI32(state), ConstantU32(state, 1)});
+	return state.builder.Type(spv::OpTypeArray, TypeI32(state), ConstantU32(state, 1));
 }
 
 uint32_t F32ArrayType(EmitterState& state, uint32_t count) {
-	return state.builder.Type(spv::OpTypeArray, {TypeF32(state), ConstantU32(state, count)});
+	return state.builder.Type(spv::OpTypeArray, TypeF32(state), ConstantU32(state, count));
 }
 
 void DefineDescriptors(EmitterState& state) {
@@ -174,16 +178,15 @@ void DefineDescriptors(EmitterState& state) {
 			const auto variable =
 			    state.builder.DefineGlobalVariable(TypePointer(state, storage, type), storage);
 			state.builder.AddName(variable, name);
-			state.builder.AddAnnotation(
-			    {spv::OpDecorate, variable, spv::DecorationDescriptorSet, 0});
-			state.builder.AddAnnotation({spv::OpDecorate, variable, spv::DecorationBinding,
-			                             IR::NativeBinding(state.program.stage, binding.kind)});
+			state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationDescriptorSet, 0);
+			state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationBinding,
+			                            IR::NativeBinding(state.program.stage, binding.kind));
 			return variable;
 		};
 		const auto ArrayType = [&](uint32_t type) {
 			return state.builder.Type(
-			    spv::OpTypeArray,
-			    {type, ConstantU32(state, static_cast<uint32_t>(binding.resources.size()))});
+			    spv::OpTypeArray, type,
+			    ConstantU32(state, static_cast<uint32_t>(binding.resources.size())));
 		};
 		switch (binding.kind) {
 			case IR::DescriptorBindingKind::Buffers:
@@ -192,10 +195,10 @@ void DefineDescriptors(EmitterState& state) {
 				if (state.requirements.buffer_int64_atomics) {
 					state.storage_buffer_u64_variable =
 					    Define(ArrayType(StorageBufferU64Type(state)), "buffers_u64");
-					state.builder.AddAnnotation(
-					    {spv::OpDecorate, state.storage_buffer_variable, spv::DecorationAliased});
-					state.builder.AddAnnotation({spv::OpDecorate, state.storage_buffer_u64_variable,
-					                             spv::DecorationAliased});
+					state.builder.AddAnnotation(spv::OpDecorate, state.storage_buffer_variable,
+					                            spv::DecorationAliased);
+					state.builder.AddAnnotation(spv::OpDecorate, state.storage_buffer_u64_variable,
+					                            spv::DecorationAliased);
 				}
 				break;
 			case IR::DescriptorBindingKind::BdaPagetable:
@@ -242,15 +245,15 @@ void DefineDescriptors(EmitterState& state) {
 } // namespace
 
 uint32_t ConstantU32(EmitterState& state, uint32_t value) {
-	return state.builder.Constant(spv::OpConstant, TypeU32(state), {value});
+	return state.builder.Constant(spv::OpConstant, TypeU32(state), value);
 }
 
 uint32_t ConstantI32(EmitterState& state, int32_t value) {
-	return state.builder.Constant(spv::OpConstant, TypeI32(state), {static_cast<uint32_t>(value)});
+	return state.builder.Constant(spv::OpConstant, TypeI32(state), static_cast<uint32_t>(value));
 }
 
 uint32_t ConstantF32(EmitterState& state, uint32_t bits) {
-	return state.builder.Constant(spv::OpConstant, TypeF32(state), {bits});
+	return state.builder.Constant(spv::OpConstant, TypeF32(state), bits);
 }
 
 uint32_t FloatBits(float value) {
@@ -270,8 +273,8 @@ uint32_t ConstantBool(EmitterState& state, bool value) {
 
 uint32_t ConstantU64(EmitterState& state, uint64_t value) {
 	return state.builder.Constant(spv::OpConstantComposite, TypeU64(state),
-	                              {ConstantU32(state, static_cast<uint32_t>(value)),
-	                               ConstantU32(state, static_cast<uint32_t>(value >> 32u))});
+	                              ConstantU32(state, static_cast<uint32_t>(value)),
+	                              ConstantU32(state, static_cast<uint32_t>(value >> 32u)));
 }
 
 uint32_t ConstantU32CompositeZero(EmitterState& state, uint32_t components) {
@@ -287,7 +290,8 @@ uint32_t GlslStd450(EmitterState& state) {
 }
 
 VertexInputScalarKind VertexParameterScalarKind(const EmitterState& state, uint32_t location) {
-	if (state.program.stage != ShaderType::Vertex || location >= ShaderVertexInputInfo::RES_MAX ||
+	if ((state.program.stage != ShaderType::Vertex && state.program.stage != ShaderType::Local) ||
+	    location >= ShaderVertexInputInfo::RES_MAX ||
 	    location >= static_cast<uint32_t>(state.input_info.vertex->resources_num)) {
 		return VertexInputScalarKind::Float;
 	}
@@ -330,8 +334,6 @@ uint32_t VertexParameterScalarType(EmitterState& state, VertexInputScalarKind ki
 	}
 }
 
-namespace {
-
 uint32_t DefineInterfaceVariable(EmitterState& state, uint32_t type, spv::StorageClass storage,
                                  const char* name) {
 	const auto variable =
@@ -341,9 +343,14 @@ uint32_t DefineInterfaceVariable(EmitterState& state, uint32_t type, spv::Storag
 	return variable;
 }
 
+namespace {
+
 uint32_t BuiltInForInput(IR::StageInputKind kind) {
 	switch (kind) {
 		case IR::StageInputKind::VertexIndex: return spv::BuiltInVertexIndex;
+		case IR::StageInputKind::InvocationId: return spv::BuiltInInvocationId;
+		case IR::StageInputKind::PrimitiveId: return spv::BuiltInPrimitiveId;
+		case IR::StageInputKind::TessCoord: return spv::BuiltInTessCoord;
 		case IR::StageInputKind::InstanceIndex: return spv::BuiltInInstanceIndex;
 		case IR::StageInputKind::FragCoord: return spv::BuiltInFragCoord;
 		case IR::StageInputKind::FrontFacing: return spv::BuiltInFrontFacing;
@@ -384,6 +391,8 @@ void DefineInputs(EmitterState& state) {
 		uint32_t type = TypeU32(state);
 		switch (input.kind) {
 			case IR::StageInputKind::VertexIndex:
+			case IR::StageInputKind::InvocationId:
+			case IR::StageInputKind::PrimitiveId:
 			case IR::StageInputKind::InstanceIndex:
 			case IR::StageInputKind::Layer:
 			case IR::StageInputKind::SampleId: type = TypeI32(state); break;
@@ -391,20 +400,22 @@ void DefineInputs(EmitterState& state) {
 			case IR::StageInputKind::LocalInvocationId:
 			case IR::StageInputKind::GlobalInvocationId: type = TypeU32Vector(state, 3); break;
 			case IR::StageInputKind::FragCoord: type = TypeF32Vector(state, 4); break;
+			case IR::StageInputKind::TessCoord:
 			case IR::StageInputKind::BaryCoordSmooth:
 			case IR::StageInputKind::BaryCoordNoPerspective: type = TypeF32Vector(state, 3); break;
 			case IR::StageInputKind::FrontFacing: type = TypeBool(state); break;
 			case IR::StageInputKind::Parameter:
-				if (state.program.stage == ShaderType::Vertex) {
+				if (state.program.stage == ShaderType::Vertex ||
+				    state.program.stage == ShaderType::Local) {
 					type = VertexParameterScalarType(
 					    state, VertexParameterScalarKind(state, input.location));
 					const auto components = VertexParameterComponentCount(input);
 					if (components > 1u) {
-						type = state.builder.Type(spv::OpTypeVector, {type, components});
+						type = state.builder.Type(spv::OpTypeVector, type, components);
 					}
 				} else if (input.per_vertex) {
-					type = state.builder.Type(spv::OpTypeArray,
-					                          {TypeF32Vector(state, 4), ConstantU32(state, 3)});
+					type = state.builder.Type(spv::OpTypeArray, TypeF32Vector(state, 4),
+					                          ConstantU32(state, 3));
 				} else {
 					type = TypeF32Vector(state, 4);
 				}
@@ -414,38 +425,37 @@ void DefineInputs(EmitterState& state) {
 		input.variable_id =
 		    DefineInterfaceVariable(state, type, spv::StorageClassInput, input.debug_name.c_str());
 		if (input.kind == IR::StageInputKind::Layer || input.kind == IR::StageInputKind::SampleId) {
-			state.builder.AddAnnotation({spv::OpDecorate, input.variable_id, spv::DecorationFlat});
+			state.builder.AddAnnotation(spv::OpDecorate, input.variable_id, spv::DecorationFlat);
 		}
 		if (input.kind == IR::StageInputKind::Parameter) {
 			const auto flat = PixelParameterIsFlat(state, input.location);
 			if (input.per_vertex) {
-				state.builder.AddAnnotation(
-				    {spv::OpDecorate, input.variable_id, spv::DecorationPerVertexKHR});
+				state.builder.AddAnnotation(spv::OpDecorate, input.variable_id,
+				                            spv::DecorationPerVertexKHR);
 			} else if (flat) {
-				state.builder.AddAnnotation(
-				    {spv::OpDecorate, input.variable_id, spv::DecorationFlat});
+				state.builder.AddAnnotation(spv::OpDecorate, input.variable_id,
+				                            spv::DecorationFlat);
 			}
 			if (state.program.stage == ShaderType::Pixel &&
 			    state.input_info.pixel->ps_no_perspective && !flat && !input.per_vertex) {
-				state.builder.AddAnnotation(
-				    {spv::OpDecorate, input.variable_id, spv::DecorationNoPerspective});
+				state.builder.AddAnnotation(spv::OpDecorate, input.variable_id,
+				                            spv::DecorationNoPerspective);
 			}
-			state.builder.AddAnnotation({spv::OpDecorate, input.variable_id,
-			                             spv::DecorationLocation,
-			                             PixelParameterLocation(state, input.location)});
+			state.builder.AddAnnotation(spv::OpDecorate, input.variable_id, spv::DecorationLocation,
+			                            PixelParameterLocation(state, input.location));
 		} else if (const auto builtin = BuiltInForInput(input.kind); builtin != UINT32_MAX) {
-			state.builder.AddAnnotation(
-			    {spv::OpDecorate, input.variable_id, spv::DecorationBuiltIn, builtin});
+			state.builder.AddAnnotation(spv::OpDecorate, input.variable_id, spv::DecorationBuiltIn,
+			                            builtin);
 		}
 	}
 	if (state.requirements.subgroup_local_invocation_id) {
 		const auto variable = DefineInterfaceVariable(state, TypeU32(state), spv::StorageClassInput,
 		                                              "gl_SubgroupInvocationID");
 		state.subgroup_local_invocation_id_variable = variable;
-		state.builder.AddAnnotation({spv::OpDecorate, variable, spv::DecorationBuiltIn,
-		                             spv::BuiltInSubgroupLocalInvocationId});
+		state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationBuiltIn,
+		                            spv::BuiltInSubgroupLocalInvocationId);
 		if (state.program.stage == ShaderType::Pixel) {
-			state.builder.AddAnnotation({spv::OpDecorate, variable, spv::DecorationFlat});
+			state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationFlat);
 		}
 	}
 }
@@ -470,8 +480,7 @@ void DefineOutputs(EmitterState& state) {
 	                         spv::BuiltIn builtin) {
 		if (variable == 0) {
 			variable = DefineInterfaceVariable(state, type, spv::StorageClassOutput, name);
-			state.builder.AddAnnotation(
-			    {spv::OpDecorate, variable, spv::DecorationBuiltIn, builtin});
+			state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationBuiltIn, builtin);
 		}
 		return variable;
 	};
@@ -527,8 +536,8 @@ void DefineOutputs(EmitterState& state) {
 				const auto type = uint_output ? TypeU32Vector(state, 4) : TypeF32Vector(state, 4);
 				binding.variable_id = DefineInterfaceVariable(state, type, spv::StorageClassOutput,
 				                                              binding.debug_name.c_str());
-				state.builder.AddAnnotation({spv::OpDecorate, binding.variable_id,
-				                             spv::DecorationLocation, binding.location});
+				state.builder.AddAnnotation(spv::OpDecorate, binding.variable_id,
+				                            spv::DecorationLocation, binding.location);
 				break;
 			}
 		}
@@ -542,6 +551,7 @@ void DefineModule(EmitterState& state) {
 	                                  state.program.info.outputs.size());
 	DefineInputs(state);
 	DefineOutputs(state);
+	DefineTessellationInterfaces(state);
 	DefineDescriptors(state);
 	if (state.requirements.function_lds) {
 		state.lds_variable = state.builder.AllocateId();
@@ -556,12 +566,16 @@ void DefineModule(EmitterState& state) {
 		state.mesh_guest_func = state.builder.AllocateId();
 		state.builder.RequireCapability(spv::CapabilityMeshShadingEXT); // MeshShadingEXT
 		state.builder.RequireExtension("SPV_EXT_mesh_shader");
-		state.builder.AddExecutionMode(
-		    {state.main_func, spv::ExecutionModeOutputTrianglesEXT}); // OutputTrianglesEXT
-		state.builder.AddExecutionMode({state.main_func, spv::ExecutionModeOutputVertices,
-		                                state.input_info.vertex->mesh.max_vertices});
-		state.builder.AddExecutionMode({state.main_func, spv::ExecutionModeOutputPrimitivesEXT,
-		                                state.input_info.vertex->mesh.max_primitives});
+		state.builder.AddExecutionMode(state.main_func,
+		                               spv::ExecutionModeOutputTrianglesEXT); // OutputTrianglesEXT
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeOutputVertices,
+		                               state.input_info.vertex->mesh.max_vertices);
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeOutputPrimitivesEXT,
+		                               state.input_info.vertex->mesh.max_primitives);
+	}
+	if (state.program.stage == ShaderType::TessellationControl ||
+	    state.program.stage == ShaderType::TessellationEvaluation) {
+		DefineTessellationExecutionModes(state);
 	}
 	state.entry_label = state.builder.AllocateId();
 
@@ -621,14 +635,14 @@ void DefineModule(EmitterState& state) {
 		state.builder.RequireExtension("SPV_KHR_fragment_shader_barycentric");
 	}
 	state.builder.RequireExtension("SPV_KHR_float_controls");
-	state.builder.AddMemoryModel({state.program.info.uses_dma
-	                                  ? spv::AddressingModelPhysicalStorageBuffer64
-	                                  : spv::AddressingModelLogical,
-	                              spv::MemoryModelGLSL450});
+	state.builder.AddMemoryModel(state.program.info.uses_dma
+	                                 ? spv::AddressingModelPhysicalStorageBuffer64
+	                                 : spv::AddressingModelLogical,
+	                             spv::MemoryModelGLSL450);
 	// 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, spv::ExecutionModeSignedZeroInfNanPreserve, 32u});
+	state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeSignedZeroInfNanPreserve,
+	                               32u);
 	if (const auto* cs = ShaderWorkgroupInput(state.program.stage, state.input_info)) {
 		uint32_t    local_x = state.requirements.compute_derivatives ? 2u : 1u;
 		uint32_t    local_y = state.requirements.compute_derivatives ? 2u : 1u;
@@ -644,23 +658,22 @@ void DefineModule(EmitterState& state) {
 				local_x = 2u;
 			}
 		}
-		state.builder.AddExecutionMode(
-		    {state.main_func, spv::ExecutionModeLocalSize, local_x, local_y, local_z});
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeLocalSize, local_x,
+		                               local_y, local_z);
 	}
 	if (state.program.stage == ShaderType::Pixel) {
-		state.builder.AddExecutionMode({state.main_func, spv::ExecutionModeOriginUpperLeft});
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeOriginUpperLeft);
 		if (state.depth_variable != 0) {
-			state.builder.AddExecutionMode({state.main_func, spv::ExecutionModeDepthReplacing});
+			state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeDepthReplacing);
 		}
 		if (state.input_info.pixel->ps_early_z && !state.input_info.pixel->ps_pixel_kill_enable &&
 		    !state.input_info.pixel->ps_depth_export_enable &&
 		    !state.input_info.pixel->ps_sample_mask_export_enable) {
-			state.builder.AddExecutionMode({state.main_func, spv::ExecutionModeEarlyFragmentTests});
+			state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeEarlyFragmentTests);
 		}
 	}
 	if (state.requirements.compute_derivatives && state.program.stage == ShaderType::Compute) {
-		state.builder.AddExecutionMode(
-		    {state.main_func, spv::ExecutionModeDerivativeGroupQuadsKHR});
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeDerivativeGroupQuadsKHR);
 	}
 	state.builder.AddName(state.main_func, "main");
 	if (state.requirements.function_lds) {

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

@@ -17,11 +17,11 @@ void EmitKillIfBoolFalse(EmitterState& state, uint32_t active) {
 	const auto kill_label  = state.builder.AllocateId();
 	const auto merge_label = state.builder.AllocateId();
 	const auto inactive    = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpLogicalNot, TypeBool(state), inactive, active});
-	state.builder.AddFunction({spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone});
-	state.builder.AddFunction({spv::OpBranchConditional, inactive, kill_label, merge_label});
+	state.builder.AddFunction(spv::OpLogicalNot, TypeBool(state), inactive, active);
+	state.builder.AddFunction(spv::OpSelectionMerge, merge_label, spv::SelectionControlMaskNone);
+	state.builder.AddFunction(spv::OpBranchConditional, inactive, kill_label, merge_label);
 	EmitLabel(state, kill_label);
-	state.builder.AddFunction({spv::OpKill});
+	state.builder.AddFunction(spv::OpKill);
 	EmitLabel(state, merge_label);
 }
 
@@ -32,10 +32,10 @@ void EmitKillIfPixelValidMaskInactive(EmitterState& state) {
 
 	const auto mask_value = state.builder.AllocateId();
 	const auto active     = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpLoad, TypeU32(state), mask_value, state.pixel_valid_mask_variable});
-	state.builder.AddFunction(
-	    {spv::OpINotEqual, TypeBool(state), active, mask_value, ConstantU32(state, 0)});
+	state.builder.AddFunction(spv::OpLoad, TypeU32(state), mask_value,
+	                          state.pixel_valid_mask_variable);
+	state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), active, mask_value,
+	                          ConstantU32(state, 0));
 	EmitKillIfBoolFalse(state, active);
 }
 
@@ -75,8 +75,8 @@ void StoreDispatcherPhiEdge(ValueEmitContext& ctx, const DispatcherFunctionState
 		}
 		for (size_t index = 0; index < phi.NumArgs(); index++) {
 			if (phi.PhiBlock(index) == from) {
-				ctx.state.builder.AddFunction(
-				    {spv::OpStore, dispatcher.spills[ctx.half].at(&phi), ctx.Def(phi.Arg(index))});
+				ctx.state.builder.AddFunction(spv::OpStore, dispatcher.spills[ctx.half].at(&phi),
+				                              ctx.Def(phi.Arg(index)));
 				break;
 			}
 		}
@@ -94,7 +94,7 @@ const IR::Block* TargetBlock(const IR::Program& program, uint32_t id) {
 
 void EmitReturn(ValueEmitContext& ctx) {
 	EmitKillIfPixelValidMaskInactive(ctx.state);
-	ctx.state.builder.AddFunction({spv::OpReturn});
+	ctx.state.builder.AddFunction(spv::OpReturn);
 }
 
 uint32_t BranchCondition(ValueEmitContext& ctx, const IR::BlockInfo& info) {
@@ -108,16 +108,16 @@ uint32_t BranchCondition(ValueEmitContext& ctx, const IR::BlockInfo& info) {
 	const auto low    = ctx.state.builder.AllocateId();
 	const auto high   = ctx.state.builder.AllocateId();
 	const auto result = ctx.state.builder.AllocateId();
-	ctx.state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(ctx.state), low, ballot, 0});
-	ctx.state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(ctx.state), high, ballot, 1});
+	ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), low, ballot, 0);
+	ctx.state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(ctx.state), high, ballot, 1);
 	const auto kind     = info.terminator.condition;
 	const bool zero     = kind == CFG::BranchCondition::ExecZero ||
 	                      kind == CFG::BranchCondition::VccZero ||
 	                      kind == CFG::BranchCondition::SccZero;
 	const auto combined =
 	    EmitBinaryU32(ctx.state, zero ? spv::OpBitwiseAnd : spv::OpBitwiseOr, low, high);
-	ctx.state.builder.AddFunction({zero ? spv::OpIEqual : spv::OpINotEqual, TypeBool(ctx.state),
-	                               result, combined, ConstantU32(ctx.state, zero ? ~0u : 0u)});
+	ctx.state.builder.AddFunction(zero ? spv::OpIEqual : spv::OpINotEqual, TypeBool(ctx.state),
+	                              result, combined, ConstantU32(ctx.state, zero ? ~0u : 0u));
 	return result;
 }
 
@@ -130,14 +130,14 @@ void EmitStructuredTerminator(ValueEmitContext& ctx, const IR::Block* block,
 			const auto* merge = TargetBlock(program, term.merge_block);
 			const auto* cont  = TargetBlock(program, term.continue_block);
 			if (merge != nullptr && cont != nullptr) {
-				ctx.state.builder.AddFunction({spv::OpLoopMerge, ctx.Label(merge), ctx.Label(cont),
-				                               spv::LoopControlMaskNone});
+				ctx.state.builder.AddFunction(spv::OpLoopMerge, ctx.Label(merge), ctx.Label(cont),
+				                              spv::LoopControlMaskNone);
 			}
 		} else if (term.kind == CFG::TerminatorKind::ConditionalBranch &&
 		           term.merge_block != UINT32_MAX) {
 			if (const auto* merge = TargetBlock(program, term.merge_block); merge != nullptr) {
-				ctx.state.builder.AddFunction(
-				    {spv::OpSelectionMerge, ctx.Label(merge), spv::SelectionControlMaskNone});
+				ctx.state.builder.AddFunction(spv::OpSelectionMerge, ctx.Label(merge),
+				                              spv::SelectionControlMaskNone);
 			}
 		}
 	};
@@ -150,7 +150,7 @@ void EmitStructuredTerminator(ValueEmitContext& ctx, const IR::Block* block,
 				return;
 			}
 			emit_merge();
-			ctx.state.builder.AddFunction({spv::OpBranch, ctx.Label(target)});
+			ctx.state.builder.AddFunction(spv::OpBranch, ctx.Label(target));
 			return;
 		}
 		case CFG::TerminatorKind::ConditionalBranch: {
@@ -162,8 +162,8 @@ void EmitStructuredTerminator(ValueEmitContext& ctx, const IR::Block* block,
 			}
 			const auto condition = BranchCondition(ctx, info);
 			emit_merge();
-			ctx.state.builder.AddFunction({spv::OpBranchConditional, condition,
-			                               ctx.Label(true_block), ctx.Label(false_block)});
+			ctx.state.builder.AddFunction(spv::OpBranchConditional, condition,
+			                              ctx.Label(true_block), ctx.Label(false_block));
 			return;
 		}
 		default: EmitReturn(ctx); return;
@@ -192,10 +192,9 @@ uint32_t EmitDispatcherNextPc(ValueEmitContext& ctx, const DispatcherFunctionSta
 			EmitDispatcherTarget(ctx, dispatcher, block, term.true_block);
 			EmitDispatcherTarget(ctx, dispatcher, block, term.false_block);
 			const auto selected = ctx.state.builder.AllocateId();
-			ctx.state.builder.AddFunction({spv::OpSelect, TypeU32(ctx.state), selected,
-			                               BranchCondition(ctx, info),
-			                               ConstantU32(ctx.state, term.true_block),
-			                               ConstantU32(ctx.state, term.false_block)});
+			ctx.state.builder.AddFunction(
+			    spv::OpSelect, TypeU32(ctx.state), selected, BranchCondition(ctx, info),
+			    ConstantU32(ctx.state, term.true_block), ConstantU32(ctx.state, term.false_block));
 			return selected;
 		}
 		case CFG::TerminatorKind::IndirectBranch: {
@@ -214,12 +213,10 @@ uint32_t EmitDispatcherNextPc(ValueEmitContext& ctx, const DispatcherFunctionSta
 				for (size_t index = 0; index < std::min(values.size(), targets.size()); index++) {
 					const auto match = ctx.state.builder.AllocateId();
 					const auto next  = ctx.state.builder.AllocateId();
-					ctx.state.builder.AddFunction({spv::OpIEqual, TypeBool(ctx.state), match,
-					                               selector,
-					                               ConstantU32(ctx.state, values[index])});
-					ctx.state.builder.AddFunction({spv::OpSelect, TypeU32(ctx.state), next, match,
-					                               ConstantU32(ctx.state, targets[index]),
-					                               selected});
+					ctx.state.builder.AddFunction(spv::OpIEqual, TypeBool(ctx.state), match,
+					                              selector, ConstantU32(ctx.state, values[index]));
+					ctx.state.builder.AddFunction(spv::OpSelect, TypeU32(ctx.state), next, match,
+					                              ConstantU32(ctx.state, targets[index]), selected);
 					selected = next;
 				}
 			}
@@ -307,15 +304,15 @@ void EmitDispatcherInstruction(ValueEmitContext& ctx, const DispatcherFunctionSt
 		if (type == 0) {
 			ctx.Fail(inst, "cannot be loaded by the dispatcher");
 		}
-		ctx.state.builder.AddFunction(
-		    {spv::OpLoad, type, ctx.Result(inst), dispatcher.spills[ctx.half].at(&inst)});
+		ctx.state.builder.AddFunction(spv::OpLoad, type, ctx.Result(inst),
+		                              dispatcher.spills[ctx.half].at(&inst));
 		return;
 	}
 	EmitDirectInstruction(ctx, inst);
 	if (const auto found = dispatcher.spills[ctx.half].find(&inst);
 	    found != dispatcher.spills[ctx.half].end()) {
-		ctx.state.builder.AddFunction(
-		    {spv::OpStore, found->second, ctx.Def(IR::Value(const_cast<IR::Inst*>(&inst)))});
+		ctx.state.builder.AddFunction(spv::OpStore, found->second,
+		                              ctx.Def(IR::Value(const_cast<IR::Inst*>(&inst))));
 	}
 }
 
@@ -362,7 +359,7 @@ void PatchStructuredPhis(ValueEmitContext& ctx, StructuredFunctionState& structu
 void EmitStructuredFunction(ValueEmitContext& ctx) {
 	const auto& program = ctx.state.program;
 	StructuredFunctionState structured;
-	ctx.state.builder.AddFunction({spv::OpBranch, ctx.Label(program.blocks.front())});
+	ctx.state.builder.AddFunction(spv::OpBranch, ctx.Label(program.blocks.front()));
 	for (size_t index = 0; index < program.blocks.size(); index++) {
 		const auto* block = program.blocks[index];
 		EmitBlock(ctx, block, [&](ValueEmitContext& lane, const IR::Inst& inst) {
@@ -377,31 +374,31 @@ void EmitStructuredFunction(ValueEmitContext& ctx) {
 void EmitDispatcherFunction(ValueEmitContext& ctx, const DispatcherFunctionState& dispatcher) {
 	auto&       state = ctx.state;
 	const auto* entry = state.program.blocks.front();
-	state.builder.AddFunction({spv::OpBranch, ctx.Label(entry)});
+	state.builder.AddFunction(spv::OpBranch, ctx.Label(entry));
 	EmitBlock(ctx, entry, [&](ValueEmitContext& lane, const IR::Inst& inst) {
 		EmitDispatcherInstruction(lane, dispatcher, inst);
 	});
 	const auto initial_pc =
 	    EmitDispatcherNextPc(ctx, dispatcher, entry, state.program.block_info.front());
 	const auto initial_parent = state.current_label;
-	state.builder.AddFunction({spv::OpBranch, dispatcher.header_label});
+	state.builder.AddFunction(spv::OpBranch, dispatcher.header_label);
 
 	EmitLabel(state, dispatcher.header_label);
 	const auto pc      = state.builder.AllocateId();
 	const auto next_pc = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpPhi, TypeU32(state), pc, initial_pc, initial_parent, next_pc,
-	                           dispatcher.continue_label});
+	state.builder.AddFunction(spv::OpPhi, TypeU32(state), pc, initial_pc, initial_parent, next_pc,
+	                          dispatcher.continue_label);
 	const auto done = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpIEqual, TypeBool(state), done, pc, ConstantU32(ctx.state, UINT32_MAX)});
-	state.builder.AddFunction({spv::OpLoopMerge, dispatcher.merge_label, dispatcher.continue_label,
-	                           spv::LoopControlMaskNone});
-	state.builder.AddFunction(
-	    {spv::OpBranchConditional, done, dispatcher.merge_label, dispatcher.select_label});
+	state.builder.AddFunction(spv::OpIEqual, TypeBool(state), done, pc,
+	                          ConstantU32(ctx.state, UINT32_MAX));
+	state.builder.AddFunction(spv::OpLoopMerge, dispatcher.merge_label, dispatcher.continue_label,
+	                          spv::LoopControlMaskNone);
+	state.builder.AddFunction(spv::OpBranchConditional, done, dispatcher.merge_label,
+	                          dispatcher.select_label);
 
 	EmitLabel(state, dispatcher.select_label);
-	state.builder.AddFunction(
-	    {spv::OpSelectionMerge, dispatcher.after_switch_label, spv::SelectionControlMaskNone});
+	state.builder.AddFunction(spv::OpSelectionMerge, dispatcher.after_switch_label,
+	                          spv::SelectionControlMaskNone);
 	std::vector<uint32_t> words {spv::OpSwitch, pc, dispatcher.after_switch_label};
 	for (size_t index = 1; index < state.program.blocks.size(); index++) {
 		words.push_back(state.program.block_info[index].id);
@@ -420,13 +417,13 @@ void EmitDispatcherFunction(ValueEmitContext& ctx, const DispatcherFunctionState
 		                                           state.program.block_info[index]);
 		next_pc_words.push_back(selected);
 		next_pc_words.push_back(state.current_label);
-		state.builder.AddFunction({spv::OpBranch, dispatcher.after_switch_label});
+		state.builder.AddFunction(spv::OpBranch, dispatcher.after_switch_label);
 	}
 	EmitLabel(state, dispatcher.after_switch_label);
 	state.builder.AddFunction(next_pc_words);
-	state.builder.AddFunction({spv::OpBranch, dispatcher.continue_label});
+	state.builder.AddFunction(spv::OpBranch, dispatcher.continue_label);
 	EmitLabel(state, dispatcher.continue_label);
-	state.builder.AddFunction({spv::OpBranch, dispatcher.header_label});
+	state.builder.AddFunction(spv::OpBranch, dispatcher.header_label);
 	EmitLabel(state, dispatcher.merge_label);
 	EmitReturn(ctx);
 }
@@ -478,7 +475,8 @@ uint32_t ValueEmitContext::Def(IR::Value value) {
 				return loaded->second.second;
 			}
 			const auto id = state.builder.AllocateId();
-			state.builder.AddFunction({spv::OpLoad, TypeId(state, inst->GetType()), id, found->second});
+			state.builder.AddFunction(spv::OpLoad, TypeId(state, inst->GetType()), id,
+			                          found->second);
 			dispatcher_block_loads.insert_or_assign(inst, std::pair {state.current_label, id});
 			return id;
 		}
@@ -498,9 +496,9 @@ uint32_t ValueEmitContext::Ballot(IR::Value predicate) {
 	const auto ballot_type = TypeU32Vector(state, 4);
 	const auto scope       = ConstantU32(state, spv::ScopeSubgroup);
 	const auto low         = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpGroupNonUniformBallot, ballot_type, low, scope,
-	     other_half == nullptr || half == 0 ? Def(predicate) : other_half->Def(predicate)});
+	state.builder.AddFunction(spv::OpGroupNonUniformBallot, ballot_type, low, scope,
+	                          other_half == nullptr || half == 0 ? Def(predicate)
+	                                                             : other_half->Def(predicate));
 	if (other_half == nullptr) {
 		return low;
 	}
@@ -508,20 +506,20 @@ uint32_t ValueEmitContext::Ballot(IR::Value predicate) {
 	const auto low_word  = state.builder.AllocateId();
 	const auto high_word = state.builder.AllocateId();
 	const auto ballot    = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpGroupNonUniformBallot, ballot_type, high, scope,
-	                           half == 1 ? Def(predicate) : other_half->Def(predicate)});
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), low_word, low, 0});
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), high_word, high, 0});
-	state.builder.AddFunction({spv::OpCompositeConstruct, ballot_type, ballot, low_word, high_word,
-	                           ConstantU32(state, 0), ConstantU32(state, 0)});
+	state.builder.AddFunction(spv::OpGroupNonUniformBallot, ballot_type, high, scope,
+	                          half == 1 ? Def(predicate) : other_half->Def(predicate));
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), low_word, low, 0);
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), high_word, high, 0);
+	state.builder.AddFunction(spv::OpCompositeConstruct, ballot_type, ballot, low_word, high_word,
+	                          ConstantU32(state, 0), ConstantU32(state, 0));
 	return ballot;
 }
 
 uint32_t ValueEmitContext::FirstLane(uint32_t ballot) {
 	if (other_half == nullptr) {
 		const auto result = state.builder.AllocateId();
-		state.builder.AddFunction({spv::OpGroupNonUniformBallotFindLSB, TypeU32(state), result,
-		                           ConstantU32(state, spv::ScopeSubgroup), ballot});
+		state.builder.AddFunction(spv::OpGroupNonUniformBallotFindLSB, TypeU32(state), result,
+		                          ConstantU32(state, spv::ScopeSubgroup), ballot);
 		return result;
 	}
 	const auto low        = state.builder.AllocateId();
@@ -530,16 +528,16 @@ uint32_t ValueEmitContext::FirstLane(uint32_t ballot) {
 	const auto high_first = state.builder.AllocateId();
 	const auto low_active = state.builder.AllocateId();
 	const auto result     = state.builder.AllocateId();
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), low, ballot, 0});
-	state.builder.AddFunction({spv::OpCompositeExtract, TypeU32(state), high, ballot, 1});
-	state.builder.AddFunction(
-	    {spv::OpExtInst, TypeU32(state), low_first, GlslStd450(state), GLSLstd450FindILsb, low});
-	state.builder.AddFunction(
-	    {spv::OpExtInst, TypeU32(state), high_first, GlslStd450(state), GLSLstd450FindILsb, high});
-	state.builder.AddFunction(
-	    {spv::OpINotEqual, TypeBool(state), low_active, low, ConstantU32(state, 0)});
-	state.builder.AddFunction({spv::OpSelect, TypeU32(state), result, low_active, low_first,
-	                           EmitAddU32(state, high_first, ConstantU32(state, 32))});
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), low, ballot, 0);
+	state.builder.AddFunction(spv::OpCompositeExtract, TypeU32(state), high, ballot, 1);
+	state.builder.AddFunction(spv::OpExtInst, TypeU32(state), low_first, GlslStd450(state),
+	                          GLSLstd450FindILsb, low);
+	state.builder.AddFunction(spv::OpExtInst, TypeU32(state), high_first, GlslStd450(state),
+	                          GLSLstd450FindILsb, high);
+	state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), low_active, low,
+	                          ConstantU32(state, 0));
+	state.builder.AddFunction(spv::OpSelect, TypeU32(state), result, low_active, low_first,
+	                          EmitAddU32(state, high_first, ConstantU32(state, 32)));
 	return result;
 }
 
@@ -548,8 +546,8 @@ uint32_t ValueEmitContext::Shuffle(const IR::Inst& inst, size_t index, uint32_t
 	const auto scope = ConstantU32(state, spv::ScopeSubgroup);
 	const auto low   = state.builder.AllocateId();
 	if (other_half == nullptr) {
-		state.builder.AddFunction(
-		    {spv::OpGroupNonUniformShuffle, type, low, scope, Arg(inst, index), lane});
+		state.builder.AddFunction(spv::OpGroupNonUniformShuffle, type, low, scope, Arg(inst, index),
+		                          lane);
 		return low;
 	}
 	const auto physical_lane =
@@ -557,15 +555,14 @@ uint32_t ValueEmitContext::Shuffle(const IR::Inst& inst, size_t index, uint32_t
 	const auto high          = state.builder.AllocateId();
 	const auto in_high       = state.builder.AllocateId();
 	const auto value         = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {spv::OpGroupNonUniformShuffle, type, low, scope, HalfArg(inst, index, 0), physical_lane});
-	state.builder.AddFunction(
-	    {spv::OpGroupNonUniformShuffle, type, high, scope, HalfArg(inst, index, 1), physical_lane});
-	state.builder.AddFunction(
-	    {spv::OpINotEqual, TypeBool(state), in_high,
-	     EmitBinaryU32(state, spv::OpBitwiseAnd, lane, ConstantU32(state, 32)),
-	     ConstantU32(state, 0)});
-	state.builder.AddFunction({spv::OpSelect, type, value, in_high, high, low});
+	state.builder.AddFunction(spv::OpGroupNonUniformShuffle, type, low, scope,
+	                          HalfArg(inst, index, 0), physical_lane);
+	state.builder.AddFunction(spv::OpGroupNonUniformShuffle, type, high, scope,
+	                          HalfArg(inst, index, 1), physical_lane);
+	state.builder.AddFunction(spv::OpINotEqual, TypeBool(state), in_high,
+	                          EmitBinaryU32(state, spv::OpBitwiseAnd, lane, ConstantU32(state, 32)),
+	                          ConstantU32(state, 0));
+	state.builder.AddFunction(spv::OpSelect, type, value, in_high, high, low);
 	return value;
 }
 
@@ -581,8 +578,8 @@ uint32_t ValueEmitContext::Result(const IR::Inst& inst) {
 uint32_t ValueEmitContext::Define(const IR::Inst& inst, uint32_t value) {
 	if (const auto found = definitions.find(&inst); found != definitions.end()) {
 		if (found->second != value) {
-			state.builder.AddFunction(
-			    {spv::OpCopyObject, TypeId(state, inst.GetType()), found->second, value});
+			state.builder.AddFunction(spv::OpCopyObject, TypeId(state, inst.GetType()),
+			                          found->second, value);
 		}
 		return found->second;
 	}
@@ -718,28 +715,28 @@ void EmitProgram(EmitterState& state) {
 			break;
 		}
 	}
-	state.builder.AddFunction({spv::OpFunction, TypeVoid(state),
-	                           state.mesh_guest_func != 0 ? state.mesh_guest_func : state.main_func,
-	                           spv::FunctionControlMaskNone, TypeFunction(state)});
+	state.builder.AddFunction(spv::OpFunction, TypeVoid(state),
+	                          state.mesh_guest_func != 0 ? state.mesh_guest_func : state.main_func,
+	                          spv::FunctionControlMaskNone, TypeFunction(state));
 	EmitLabel(state, state.entry_label);
 	if (state.requirements.function_lds) {
 		state.builder.AddFunction(
-		    {spv::OpVariable,
-		     TypeU32ArrayPointer(state, spv::StorageClassFunction, LdsDwordCount(state)),
-		     state.lds_variable, spv::StorageClassFunction});
+		    spv::OpVariable,
+		    TypeU32ArrayPointer(state, spv::StorageClassFunction, LdsDwordCount(state)),
+		    state.lds_variable, spv::StorageClassFunction);
 	}
 	if (state.requirements.function_scratch) {
 		for (uint32_t half = 0; half < state.lane_count; half++) {
-			state.builder.AddFunction({spv::OpVariable,
-			                           TypeU32ArrayPointer(state, spv::StorageClassFunction,
-			                                               state.program.scratch_dwords),
-			                           state.scratch_variable[half], spv::StorageClassFunction});
+			state.builder.AddFunction(
+			    spv::OpVariable,
+			    TypeU32ArrayPointer(state, spv::StorageClassFunction, state.program.scratch_dwords),
+			    state.scratch_variable[half], spv::StorageClassFunction);
 		}
 	}
 	if (state.pixel_valid_mask_variable != 0) {
-		state.builder.AddFunction({spv::OpVariable,
-		                           TypePointer(state, spv::StorageClassFunction, TypeU32(state)),
-		                           state.pixel_valid_mask_variable, spv::StorageClassFunction});
+		state.builder.AddFunction(spv::OpVariable,
+		                          TypePointer(state, spv::StorageClassFunction, TypeU32(state)),
+		                          state.pixel_valid_mask_variable, spv::StorageClassFunction);
 	}
 	for (uint32_t half = 0; half < state.lane_count; half++) {
 		auto& lane = half == 0 ? ctx : high;
@@ -748,27 +745,27 @@ void EmitProgram(EmitterState& state) {
 				for (const auto& inst: *block) {
 					if (const auto found = dispatcher->spills[half].find(&inst);
 					    found != dispatcher->spills[half].end()) {
-						state.builder.AddFunction({spv::OpVariable,
-						                           SpillPointerType(lane, inst.GetType()),
-						                           found->second, spv::StorageClassFunction});
+						state.builder.AddFunction(spv::OpVariable,
+						                          SpillPointerType(lane, inst.GetType()),
+						                          found->second, spv::StorageClassFunction);
 					}
 				}
 			}
 		}
 		if (lane.scratch_u32_variable != 0) {
-			state.builder.AddFunction(
-			    {spv::OpVariable, TypePointer(state, spv::StorageClassFunction, TypeU32(state)),
-			     lane.scratch_u32_variable, spv::StorageClassFunction});
+			state.builder.AddFunction(spv::OpVariable,
+			                          TypePointer(state, spv::StorageClassFunction, TypeU32(state)),
+			                          lane.scratch_u32_variable, spv::StorageClassFunction);
 		}
 	}
 	if (state.gds_variable != 0) {
 		state.gds_length = state.builder.AllocateId();
-		state.builder.AddFunction(
-		    {spv::OpArrayLength, TypeU32(state), state.gds_length, state.gds_variable, 0});
+		state.builder.AddFunction(spv::OpArrayLength, TypeU32(state), state.gds_length,
+		                          state.gds_variable, 0);
 	}
 	if (state.pixel_valid_mask_variable != 0) {
-		state.builder.AddFunction(
-		    {spv::OpStore, state.pixel_valid_mask_variable, ConstantU32(state, 1)});
+		state.builder.AddFunction(spv::OpStore, state.pixel_valid_mask_variable,
+		                          ConstantU32(state, 1));
 	}
 	EmitMemoryOffsets(state);
 	if (program.blocks.empty()) {
@@ -778,7 +775,7 @@ void EmitProgram(EmitterState& state) {
 	} else {
 		EmitStructuredFunction(ctx);
 	}
-	state.builder.AddFunction({spv::OpFunctionEnd});
+	state.builder.AddFunction(spv::OpFunctionEnd);
 	if (state.program.stage == ShaderType::Mesh) {
 		EmitMeshEntryPoint(state);
 	}

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

@@ -0,0 +1,161 @@
+#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInstructions.h"
+
+#include <algorithm>
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
+namespace {
+
+uint32_t TessellationPointer(ValueEmitContext& ctx, const IR::Inst& inst) {
+	auto& state          = ctx.state;
+	using Attribute      = IR::TessellationAttribute;
+	const auto  kind     = static_cast<Attribute>(inst.Arg(0).U32());
+	const auto& tess     = state.input_info.vertex->tess;
+	const auto  variable = state.tess_variables.at(static_cast<uint32_t>(kind));
+	EXIT_IF(variable == 0);
+	const auto input   = kind == Attribute::ControlInput || kind == Attribute::EvaluationInput;
+	const auto storage = input ? spv::StorageClassInput : spv::StorageClassOutput;
+	const auto pointer = state.builder.AllocateId();
+	if (kind == Attribute::Factor) {
+		EXIT_NOT_IMPLEMENTED(!inst.Arg(1).IsImmediate());
+		const auto index = inst.Arg(1).U32() / 4u;
+		const auto outer = index < 3u;
+		EXIT_NOT_IMPLEMENTED(index >= 4u);
+		state.builder.AddFunction(spv::OpAccessChain, TypePointer(state, storage, TypeF32(state)),
+		                          pointer, outer ? variable : state.tess_inner_variable,
+		                          ConstantU32(state, outer ? index : index - 3u));
+		return pointer;
+	}
+	auto address = ctx.Arg(inst, 1);
+	if (kind == Attribute::PatchOutput) {
+		address =
+		    EmitBinaryU32(state, spv::OpISub, address, ConstantU32(state, state.tess_patch_base));
+	}
+	const bool local  = kind == Attribute::LocalOutput || kind == Attribute::ControlInput;
+	const auto stride = local ? tess.ls_stride : tess.hs_stride;
+	const auto offset = kind == Attribute::PatchOutput ? address
+	                                                   : EmitBinaryU32(state, spv::OpUMod, address,
+	                                                                   ConstantU32(state, stride));
+	const auto attribute =
+	    EmitBinaryU32(state, spv::OpShiftRightLogical, offset, ConstantU32(state, 4));
+	const auto component =
+	    EmitBinaryU32(state, spv::OpBitwiseAnd,
+	                  EmitBinaryU32(state, spv::OpShiftRightLogical, offset, ConstantU32(state, 2)),
+	                  ConstantU32(state, 3));
+	const auto type = TypePointer(state, storage, TypeU32(state));
+	if (kind == Attribute::LocalOutput || kind == Attribute::PatchOutput) {
+		state.builder.AddFunction(spv::OpAccessChain, type, pointer, variable, attribute,
+		                          component);
+	} else {
+		const auto vertex =
+		    kind == Attribute::ControlOutput
+		        ? EmitLaneId(state)
+		        : EmitBinaryU32(state, spv::OpUDiv, address, ConstantU32(state, stride));
+		state.builder.AddFunction(spv::OpAccessChain, type, pointer, variable, vertex, attribute,
+		                          component);
+	}
+	return pointer;
+}
+
+} // namespace
+
+void DefineTessellationInterfaces(EmitterState& state) {
+	using Attribute = IR::TessellationAttribute;
+	std::array<bool, 6> used {};
+	uint32_t            patch_begin = UINT32_MAX, patch_end = 0;
+	for (const auto* block: state.program.blocks) {
+		for (const auto& inst: *block) {
+			if (inst.GetOpcode() != IR::ValueOpcode::GetTessellationAttribute &&
+			    inst.GetOpcode() != IR::ValueOpcode::SetTessellationAttribute)
+				continue;
+			const auto kind = inst.Arg(0).U32();
+			used.at(kind)   = true;
+			if (kind == static_cast<uint32_t>(Attribute::PatchOutput)) {
+				EXIT_NOT_IMPLEMENTED(!inst.Arg(1).IsImmediate());
+				patch_begin = std::min(patch_begin, inst.Arg(1).U32());
+				patch_end   = std::max(patch_end, inst.Arg(1).U32() + 4u);
+			}
+		}
+	}
+	if (std::ranges::none_of(used, [](bool value) { return value; })) return;
+	const auto& tess  = state.input_info.vertex->tess;
+	const auto  array = [&](uint32_t type, uint32_t count) {
+		return state.builder.Type(spv::OpTypeArray, type, ConstantU32(state, count));
+	};
+	for (uint32_t index = 0; index < used.size(); index++) {
+		if (!used[index]) continue;
+		const auto kind    = static_cast<Attribute>(index);
+		const bool input   = kind == Attribute::ControlInput || kind == Attribute::EvaluationInput;
+		const auto storage = input ? spv::StorageClassInput : spv::StorageClassOutput;
+		uint32_t   type;
+		if (kind == Attribute::Factor) {
+			type = array(TypeF32(state), 4u);
+		} else if (kind == Attribute::PatchOutput) {
+			state.tess_patch_base = patch_begin;
+			type = array(TypeU32Vector(state, 4), (patch_end - patch_begin + 15u) / 16u);
+		} else {
+			const bool local  = kind == Attribute::LocalOutput || kind == Attribute::ControlInput;
+			const auto stride = local ? tess.ls_stride : tess.hs_stride;
+			type              = array(TypeU32Vector(state, 4), (stride + 15u) / 16u);
+			if (kind != Attribute::LocalOutput) {
+				type = array(type, local ? tess.input_control_points : tess.output_control_points);
+			}
+		}
+		auto& variable = state.tess_variables[index];
+		variable       = DefineInterfaceVariable(state, type, storage, "tess_attributes");
+		if (kind == Attribute::Factor) {
+			state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationBuiltIn,
+			                            spv::BuiltInTessLevelOuter);
+			state.tess_inner_variable =
+			    DefineInterfaceVariable(state, array(TypeF32(state), 2u), storage, "tess_inner");
+			state.builder.AddAnnotation(spv::OpDecorate, state.tess_inner_variable,
+			                            spv::DecorationBuiltIn, spv::BuiltInTessLevelInner);
+			state.builder.AddAnnotation(spv::OpDecorate, state.tess_inner_variable,
+			                            spv::DecorationPatch);
+		} else {
+			state.builder.AddAnnotation(
+			    spv::OpDecorate, variable, spv::DecorationLocation,
+			    kind == Attribute::PatchOutput ? (tess.hs_stride + 15u) / 16u : 0u);
+		}
+		if (kind == Attribute::Factor || kind == Attribute::PatchOutput) {
+			state.builder.AddAnnotation(spv::OpDecorate, variable, spv::DecorationPatch);
+		}
+	}
+}
+
+void DefineTessellationExecutionModes(EmitterState& state) {
+	const auto& tess = state.input_info.vertex->tess;
+	EXIT_NOT_IMPLEMENTED(tess.domain != 1u || tess.partitioning != 2u ||
+	                     tess.output_topology != 2u);
+	state.builder.RequireCapability(spv::CapabilityTessellation);
+	if (state.program.stage == ShaderType::TessellationControl) {
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeOutputVertices,
+		                               tess.output_control_points);
+	} else {
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeTriangles);
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeSpacingFractionalOdd);
+		state.builder.AddExecutionMode(state.main_func, spv::ExecutionModeVertexOrderCw);
+	}
+}
+
+uint32_t EmitGetTessellationAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
+	return EmitValueOrZeroIfCondition(ctx.state, ctx.Arg(inst, 2), [&] {
+		const auto value = ctx.state.builder.AllocateId();
+		ctx.state.builder.AddFunction(spv::OpLoad, TypeU32(ctx.state), value,
+		                              TessellationPointer(ctx, inst));
+		return value;
+	});
+}
+
+void EmitSetTessellationAttribute(ValueEmitContext& ctx, const IR::Inst& inst) {
+	EmitIfCondition(ctx.state, ctx.Arg(inst, 3), [&] {
+		auto value = ctx.Arg(inst, 2);
+		if (inst.Arg(0).U32() == static_cast<uint32_t>(IR::TessellationAttribute::Factor)) {
+			const auto floating = ctx.state.builder.AllocateId();
+			ctx.state.builder.AddFunction(spv::OpBitcast, TypeF32(ctx.state), floating, value);
+			value = floating;
+		}
+		ctx.state.builder.AddFunction(spv::OpStore, TessellationPointer(ctx, inst), value);
+	});
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.cpp

@@ -381,6 +381,25 @@ void DecodeInstruction(std::span<const uint32_t> code, uint32_t word_index, Inst
 	}
 }
 
+Program DecodeFrontProgram(std::span<const uint32_t> front) {
+	Program  result;
+	uint32_t front_words = 0;
+	while (front_words < front.size()) {
+		auto& inst = result.instructions.emplace_back();
+		DecodeInstruction(front, front_words, inst);
+		front_words += inst.word_count;
+		if (inst.opcode == Opcode::S_SETPC_B64) {
+			EXIT_NOT_IMPLEMENTED(inst.src0.kind != OperandKind::Sgpr || inst.src0.reg != 6u);
+			break;
+		}
+		EXIT_NOT_IMPLEMENTED(inst.opcode == Opcode::S_ENDPGM);
+	}
+	EXIT_IF(result.instructions.empty() ||
+	        result.instructions.back().opcode != Opcode::S_SETPC_B64);
+	result.code = front.first(front_words);
+	return result;
+}
+
 void DecodeProgram(std::span<const uint32_t> code, Program& program) {
 	program.instructions.clear();
 	program.instructions.reserve(code.size());

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h

@@ -190,6 +190,7 @@ enum class Opcode {
 	V_CEIL_F16,
 	V_TRUNC_F16,
 	V_RNDNE_F16,
+	V_FRACT_F16,
 	V_SIN_F16,
 	V_COS_F16,
 	V_SIN_F32,
@@ -395,6 +396,7 @@ enum class Opcode {
 	V_CMP_EQ_U16,
 	V_CMP_LE_U16,
 	V_CMP_GT_U16,
+	V_CMPX_LT_U16,
 	V_CMPX_GT_U16,
 	V_CMP_NE_U16,
 	V_CMP_GE_U16,
@@ -720,6 +722,7 @@ struct Program {
 Family GetInstructionFamily(uint32_t word);
 // The output object must be freshly initialized.
 void DecodeInstruction(std::span<const uint32_t> code, uint32_t word_index, Instruction& inst);
+Program DecodeFrontProgram(std::span<const uint32_t> front);
 void DecodeProgram(std::span<const uint32_t> code, Program& program);
 bool IsConditionalBranch(Opcode opcode);
 bool IsDirectBranch(Opcode opcode);

src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp

@@ -139,6 +139,7 @@ constexpr OpcodeMap VOP1_OPCODE_LIST[] = {
     {0x5cu, Opcode::V_CEIL_F16},
     {0x5du, Opcode::V_TRUNC_F16},
     {0x5eu, Opcode::V_RNDNE_F16},
+    {0x5fu, Opcode::V_FRACT_F16},
     {0x60u, Opcode::V_SIN_F16},
     {0x61u, Opcode::V_COS_F16},
 };
@@ -192,6 +193,7 @@ constexpr OpcodeMap VOP3_ENCODED_VOP1_OPCODE_LIST[] = {
     {0x5cu, Opcode::V_CEIL_F16},
     {0x5du, Opcode::V_TRUNC_F16},
     {0x5eu, Opcode::V_RNDNE_F16},
+    {0x5fu, Opcode::V_FRACT_F16},
     {0x60u, Opcode::V_SIN_F16},
     {0x61u, Opcode::V_COS_F16},
 };
@@ -233,6 +235,7 @@ constexpr VopcOpcodeInfo VOPC_OPCODE_LIST[] = {
     {0xa9u, Opcode::V_CMP_LT_U16},         {0xaau, Opcode::V_CMP_EQ_U16},
     {0xabu, Opcode::V_CMP_LE_U16},         {0xacu, Opcode::V_CMP_GT_U16},
     {0xadu, Opcode::V_CMP_NE_U16},         {0xaeu, Opcode::V_CMP_GE_U16},
+    {0xb9u, Opcode::V_CMPX_LT_U16, false},
     {0xbcu, Opcode::V_CMPX_GT_U16},        {0xc0u, Opcode::V_CMP_F_U32},
     {0xc1u, Opcode::V_CMP_LT_U32},         {0xc2u, Opcode::V_CMP_EQ_U32},
     {0xc3u, Opcode::V_CMP_LE_U32},         {0xc4u, Opcode::V_CMP_GT_U32},
@@ -479,6 +482,7 @@ bool IsVop1FloatSourceOpcode(Opcode opcode) {
 		case Opcode::V_CEIL_F16:
 		case Opcode::V_TRUNC_F16:
 		case Opcode::V_RNDNE_F16:
+		case Opcode::V_FRACT_F16:
 		case Opcode::V_SIN_F16:
 		case Opcode::V_COS_F16:
 		case Opcode::V_SIN_F32:
@@ -548,6 +552,8 @@ constexpr Vop1SdwaRule VOP1_SDWA_RULES[] = {
      SdwaSelWords() | SdwaSelFull(), true},
     {Opcode::V_RNDNE_F16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
      SdwaSelWords() | SdwaSelFull(), true},
+    {Opcode::V_FRACT_F16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
+     SdwaSelWords() | SdwaSelFull(), true},
     {Opcode::V_SIN_F16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
      SdwaSelWords() | SdwaSelFull(), true},
     {Opcode::V_COS_F16, SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),
@@ -773,6 +779,7 @@ bool IsVop1FloatResultOpcode(Opcode opcode) {
 		case Opcode::V_CEIL_F16:
 		case Opcode::V_TRUNC_F16:
 		case Opcode::V_RNDNE_F16:
+		case Opcode::V_FRACT_F16:
 		case Opcode::V_SIN_F16:
 		case Opcode::V_COS_F16:
 		case Opcode::V_SIN_F32:
@@ -1434,6 +1441,7 @@ bool IsVopcCompareExec(Opcode opcode) {
 		case Opcode::V_CMPX_GE_U32:
 		case Opcode::V_CMPX_NE_I64:
 		case Opcode::V_CMPX_NE_U64:
+		case Opcode::V_CMPX_LT_U16:
 		case Opcode::V_CMPX_GT_U16:
 		case Opcode::V_CMPX_LT_F16:
 		case Opcode::V_CMPX_EQ_F16:

src/graphics/shader/recompiler/frontend/translate/Translate.cpp

@@ -898,11 +898,15 @@ void ValidateTranslateOptions(const TranslateOptions& options) {
 	if (options.wave_size != 32u && options.wave_size != 64u) {
 		EXIT("shader translation requires wave32 or wave64, got %u", options.wave_size);
 	}
-	if (options.embedded_fetch != nullptr && options.stage != ShaderType::Vertex) {
-		EXIT("embedded fetch requires the vertex shader stage");
+	if (options.embedded_fetch != nullptr && options.stage != ShaderType::Vertex &&
+	    options.stage != ShaderType::Local) {
+		EXIT("embedded fetch requires a vertex or local shader");
 	}
 	switch (options.stage) {
 		case ShaderType::Vertex:
+		case ShaderType::Local:
+		case ShaderType::TessellationControl:
+		case ShaderType::TessellationEvaluation:
 		case ShaderType::Mesh:
 			if (options.input_info.vertex == nullptr) {
 				EXIT("vertex shader translation has no vertex input metadata");
@@ -941,6 +945,9 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 	result.user_data_count     = options.user_data_count;
 	switch (options.stage) {
 		case ShaderType::Vertex:
+		case ShaderType::Local:
+		case ShaderType::TessellationControl:
+		case ShaderType::TessellationEvaluation:
 			result.scratch_dwords = options.input_info.vertex->scratch_size_dwords;
 			break;
 		case ShaderType::Mesh:
@@ -1164,6 +1171,42 @@ IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph&
 			entry_ir.SetVectorReg(
 			    static_cast<IR::VectorReg>(8),
 			    entry_ir.IAdd(draw(2), builtin(IR::StageInputKind::WorkgroupId, 1)));
+		} else if (options.stage == ShaderType::Local) {
+			entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(3), IR::U32(IR::Value(64u)));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(2),
+			                      builtin(IR::StageInputKind::VertexIndex));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(3), IR::U32(IR::Value(0u)));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(5),
+			                      builtin(IR::StageInputKind::InstanceIndex));
+		} else if (options.stage == ShaderType::TessellationControl) {
+			const auto& tess = options.input_info.vertex->tess;
+			entry_ir.SetScalarReg(
+			    static_cast<IR::ScalarReg>(2),
+			    IR::U32(entry_ir.Emit(IR::ValueOpcode::TessellationBase, {IR::Value(0u)})));
+			entry_ir.SetScalarReg(
+			    static_cast<IR::ScalarReg>(4),
+			    IR::U32(entry_ir.Emit(IR::ValueOpcode::TessellationBase, {IR::Value(1u)})));
+			entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(3),
+			                      IR::U32(IR::Value(0x81010000u | tess.input_control_points |
+			                                        (tess.output_control_points << 8u))));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(0),
+			                      builtin(IR::StageInputKind::PrimitiveId));
+			entry_ir.SetVectorReg(
+			    static_cast<IR::VectorReg>(1),
+			    entry_ir.ShiftLeftLogical(builtin(IR::StageInputKind::InvocationId),
+			                              IR::U32(IR::Value(8u))));
+		} else if (options.stage == ShaderType::TessellationEvaluation) {
+			entry_ir.SetScalarReg(static_cast<IR::ScalarReg>(3), IR::U32(IR::Value(64u)));
+			entry_ir.SetScalarReg(
+			    static_cast<IR::ScalarReg>(4),
+			    IR::U32(entry_ir.Emit(IR::ValueOpcode::TessellationBase, {IR::Value(0u)})));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(5),
+			                      builtin(IR::StageInputKind::TessCoord, 0));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(6),
+			                      builtin(IR::StageInputKind::TessCoord, 1));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(7), IR::U32(IR::Value(0u)));
+			entry_ir.SetVectorReg(static_cast<IR::VectorReg>(8),
+			                      builtin(IR::StageInputKind::PrimitiveId));
 		} else if (options.stage == ShaderType::Pixel) {
 			const auto* ps = options.input_info.pixel;
 			if (ps->ps_perspective_center_vgpr != UINT32_MAX) {

src/graphics/shader/recompiler/frontend/translate/Vector.cpp

@@ -138,6 +138,9 @@ bool Translator::EmitVector(const Decoder::Instruction& inst) {
 		case O::V_CMP_LT_U16:
 			EmitInteger16Compare(inst, IR::ValueOpcode::ULessThan32, false, false);
 			return true;
+		case O::V_CMPX_LT_U16:
+			EmitInteger16Compare(inst, IR::ValueOpcode::ULessThan32, false, true);
+			return true;
 		case O::V_CMP_LE_U16:
 			EmitInteger16Compare(inst, IR::ValueOpcode::ULessThanEqual32, false, false);
 			return true;
@@ -370,6 +373,7 @@ bool Translator::EmitVector(const Decoder::Instruction& inst) {
 		case O::V_CEIL_F16: return Float16Unary(inst, IR::ValueOpcode::FPCeil32, false);
 		case O::V_TRUNC_F16: return Float16Unary(inst, IR::ValueOpcode::FPTrunc32, false);
 		case O::V_RNDNE_F16: return Float16Unary(inst, IR::ValueOpcode::FPRoundEven32, false);
+		case O::V_FRACT_F16: return Float16Unary(inst, IR::ValueOpcode::FPFract32, false);
 		case O::V_SIN_F16: return Float16Trig(inst, IR::ValueOpcode::FPSin);
 		case O::V_COS_F16: return Float16Trig(inst, IR::ValueOpcode::FPCos);
 		case O::V_MIN3_F16: return Float16Ternary(inst, IR::ValueOpcode::FPMinTri32, false, false);

src/graphics/shader/recompiler/ir/ShaderIR.h

@@ -150,8 +150,20 @@ struct SampledResourcePair {
 	bool operator==(const SampledResourcePair& other) const = default;
 };
 
+enum class TessellationAttribute {
+	LocalOutput,
+	ControlInput,
+	ControlOutput,
+	EvaluationInput,
+	PatchOutput,
+	Factor
+};
+
 enum class StageInputKind {
 	VertexIndex,
+	InvocationId,
+	PrimitiveId,
+	TessCoord,
 	InstanceIndex,
 	FragCoord,
 	FrontFacing,
@@ -287,8 +299,12 @@ static_assert(sizeof(PushData) == 128);
 constexpr uint32_t NativePushConstantSize = sizeof(PushData);
 
 [[nodiscard]] constexpr uint32_t NativeBinding(ShaderType stage, DescriptorBindingKind kind) {
+	const uint32_t group = stage == ShaderType::Pixel                    ? 1u
+	                       : stage == ShaderType::TessellationControl    ? 2u
+	                       : stage == ShaderType::TessellationEvaluation ? 3u
+	                                                                     : 0u;
 	return static_cast<uint32_t>(kind) +
-	       (stage == ShaderType::Pixel ? static_cast<uint32_t>(DescriptorBindingKind::Count) : 0u);
+	       group * static_cast<uint32_t>(DescriptorBindingKind::Count);
 }
 
 [[nodiscard]] constexpr ImageResourceClass ImageBindingResourceClass(DescriptorBindingKind kind) {

src/graphics/shader/recompiler/ir/opcodes/ValueOpcodes.cpp

@@ -90,6 +90,7 @@ bool HasSideEffects(ValueOpcode opcode) {
 		case ValueOpcode::Reference:
 		case ValueOpcode::ReferenceU32:
 		case ValueOpcode::SetAttribute:
+		case ValueOpcode::SetTessellationAttribute:
 		case ValueOpcode::MeshAllocate:
 		case ValueOpcode::Barrier: return true;
 		default: return false;

src/graphics/shader/recompiler/ir/opcodes/ValueOpcodes.inc

@@ -10,6 +10,9 @@ VALUE_OPCODE(GetUserData, U32, ScalarReg)
 VALUE_OPCODE(GetShaderBase, U64)
 VALUE_OPCODE(MeshDrawParameter, U32, U32)
 VALUE_OPCODE(MeshAllocate, Void, U32)
+VALUE_OPCODE(TessellationBase, U32, U32)
+VALUE_OPCODE(GetTessellationAttribute, U32, U32, U32, U1)
+VALUE_OPCODE(SetTessellationAttribute, Void, U32, U32, U32, U1)
 VALUE_OPCODE(GetBuiltin, U32, U32, U32)
 VALUE_OPCODE(GetThreadBitScalarRegister, U1, ScalarReg)
 VALUE_OPCODE(SetThreadBitScalarRegister, Void, ScalarReg, U1)

src/graphics/shader/recompiler/ir/passes/ConstantPropagation.cpp

@@ -228,6 +228,15 @@ void FoldInstruction(Block& block, Block::iterator instruction,
 			const auto offset = Arg(inst, 1);
 			const auto count  = Arg(inst, 2);
 			auto* source = value.TryInstruction();
+			if (source != nullptr && source->GetOpcode() == ValueOpcode::ShiftLeftLogical32 &&
+			    IsImmediate(offset, Type::U32) && IsImmediate(count, Type::U32)) {
+				const auto shift = Arg(*source, 1);
+				if (IsImmediate(shift, Type::U32) && shift.U32() < 32u &&
+				    offset.U32() <= shift.U32() && count.U32() <= shift.U32() - offset.U32()) {
+					Replace(inst, Value(0u));
+					return;
+				}
+			}
 			if (source != nullptr && source->GetOpcode() == ValueOpcode::GetBuiltin &&
 			    source->Arg(0) == Value(static_cast<uint32_t>(StageInputKind::PackedAncillary)) &&
 			    IsImmediate(offset, Type::U32) && IsImmediate(count, Type::U32) && count.U32() != 0u) {

src/graphics/shader/recompiler/ir/passes/ShaderInfoCollection.cpp

@@ -43,6 +43,9 @@ void AddOutput(ShaderInfo& info, StageOutputKind kind, uint32_t index, uint32_t
 void ValidateOptions(const Program& program, ShaderStageInputInfo input_info) {
 	switch (program.stage) {
 		case ShaderType::Vertex:
+		case ShaderType::Local:
+		case ShaderType::TessellationControl:
+		case ShaderType::TessellationEvaluation:
 		case ShaderType::Mesh:
 			if (input_info.vertex == nullptr) {
 				return Fail("vertex shader has no input metadata");
@@ -81,7 +84,8 @@ void ValidateValueReferences(const Program& program, ShaderStageInputInfo input_
 					    !inst.Arg(1).IsImmediate() || inst.Arg(1).GetType() != Type::U32) {
 						return Fail("typed attribute reference is not constant");
 					}
-					if (program.stage == ShaderType::Vertex &&
+					if ((program.stage == ShaderType::Vertex ||
+					     program.stage == ShaderType::Local) &&
 					    (inst.Arg(1).U32() >= 4u ||
 					     inst.Arg(0).U32() >=
 					         static_cast<uint32_t>(input_info.vertex->resources_num))) {
@@ -124,6 +128,8 @@ void ValidateValueReferences(const Program& program, ShaderStageInputInfo input_
 							}
 							break;
 						case StageInputKind::VertexIndex:
+						case StageInputKind::InvocationId:
+						case StageInputKind::PrimitiveId:
 						case StageInputKind::InstanceIndex:
 						case StageInputKind::FrontFacing:
 						case StageInputKind::LocalInvocationIndex:
@@ -142,6 +148,7 @@ void ValidateValueReferences(const Program& program, ShaderStageInputInfo input_
 								return Fail("typed barycentric component is out of range");
 							}
 							break;
+						case StageInputKind::TessCoord:
 						case StageInputKind::WorkgroupId:
 						case StageInputKind::LocalInvocationId:
 						case StageInputKind::GlobalInvocationId:
@@ -161,8 +168,10 @@ void ValidateValueReferences(const Program& program, ShaderStageInputInfo input_
 					}
 					const auto& exp = program.export_info[index];
 					if (exp.kind == ExportTargetKind::Position && exp.index != 0 && exp.en != 0 &&
-					    program.stage != ShaderType::Vertex && program.stage != ShaderType::Mesh) {
-						return Fail("auxiliary position export requires the vertex or mesh stage");
+					    program.stage != ShaderType::Vertex && program.stage != ShaderType::Mesh &&
+					    program.stage != ShaderType::TessellationEvaluation) {
+						return Fail("auxiliary position export requires a vertex, mesh, or "
+						            "tessellation evaluation shader");
 					}
 					break;
 				}
@@ -249,6 +258,10 @@ void CollectComputeInputs(const ShaderComputeInputInfo* compute, ShaderInfo& inf
 void CollectBuiltinInputs(const Program& program, ShaderInfo& info) {
 	for (const auto* block: program.blocks) {
 		for (const auto& inst: *block) {
+			if (program.stage == ShaderType::TessellationControl &&
+			    inst.GetOpcode() == ValueOpcode::LaneId) {
+				AddInput(info, StageInputKind::InvocationId, 0, 1, "gl_InvocationID");
+			}
 			if (inst.GetOpcode() != ValueOpcode::GetBuiltin) {
 				continue;
 			}
@@ -260,6 +273,13 @@ void CollectBuiltinInputs(const Program& program, ShaderInfo& info) {
 				case StageInputKind::InstanceIndex:
 					AddInput(info, kind, 0, 1, "gl_InstanceIndex");
 					break;
+				case StageInputKind::InvocationId:
+					AddInput(info, kind, 0, 1, "gl_InvocationID");
+					break;
+				case StageInputKind::PrimitiveId:
+					AddInput(info, kind, 0, 1, "gl_PrimitiveID");
+					break;
+				case StageInputKind::TessCoord: AddInput(info, kind, 0, 3, "gl_TessCoord"); break;
 				case StageInputKind::FragCoord: AddInput(info, kind, 0, 4, "gl_FragCoord"); break;
 				case StageInputKind::FrontFacing:
 					AddInput(info, kind, 0, 1, "gl_FrontFacing");
@@ -380,7 +400,10 @@ void CollectShaderInfo(Program& program, ShaderStageInputInfo input_info) {
 		    });
 	    });
 	switch (program.stage) {
-		case ShaderType::Vertex: CollectVertexInputs(program, input_info.vertex, next); break;
+		case ShaderType::Vertex:
+		case ShaderType::Local: CollectVertexInputs(program, input_info.vertex, next); break;
+		case ShaderType::TessellationControl:
+		case ShaderType::TessellationEvaluation:
 		case ShaderType::Mesh: break;
 		case ShaderType::Pixel: CollectPixelInputs(program, input_info.pixel, next); break;
 		case ShaderType::Compute: CollectComputeInputs(input_info.compute, next); break;

src/graphics/shader/rectListShader.cpp

@@ -55,7 +55,7 @@ class RectListEmitter {
 public:
 	RectListEmitter(const std::vector<Parameter>& parameters_, spv::ExecutionModel model)
 	    : parameters(parameters_) {
-		builder.AddMemoryModel({spv::AddressingModelLogical, spv::MemoryModelGLSL450});
+		builder.AddMemoryModel(spv::AddressingModelLogical, spv::MemoryModelGLSL450);
 
 		void_type       = Type(spv::OpTypeVoid);
 		uint_type       = Type(spv::OpTypeInt, 32u, 0u);
@@ -65,9 +65,10 @@ public:
 		function_type   = Type(spv::OpTypeFunction, void_type);
 
 		per_vertex_type = builder.DecoratedType(
-		    spv::OpTypeStruct, {vec4_float_type},
+		    spv::OpTypeStruct,
 		    {{spv::OpMemberDecorate, {0u, spv::DecorationBuiltIn, spv::BuiltInPosition}},
-		     {spv::OpDecorate, {spv::DecorationBlock}}});
+		     {spv::OpDecorate, {spv::DecorationBlock}}},
+		    vec4_float_type);
 
 		ptr_input_vec4_float  = Pointer(spv::StorageClassInput, vec4_float_type);
 		ptr_output_vec4_float = Pointer(spv::StorageClassOutput, vec4_float_type);
@@ -188,11 +189,11 @@ public:
 private:
 	template <typename... Args>
 	uint32_t Type(spv::Op opcode, Args... operands) {
-		return builder.Type(opcode, {operands...});
+		return builder.Type(opcode, operands...);
 	}
 
 	uint32_t Constant(uint32_t type, uint32_t value) {
-		return builder.Constant(spv::OpConstant, type, {value});
+		return builder.Constant(spv::OpConstant, type, value);
 	}
 
 	uint32_t Pointer(spv::StorageClass storage, uint32_t type) {
@@ -206,20 +207,20 @@ private:
 	template <typename... Args>
 	uint32_t Result(spv::Op opcode, uint32_t type, Args... operands) {
 		const auto id = builder.AllocateId();
-		builder.AddFunction({opcode, type, id, operands...});
+		builder.AddFunction(opcode, type, id, operands...);
 		return id;
 	}
 
 	template <typename... Args>
 	uint32_t ResultWithoutType(spv::Op opcode, Args... operands) {
 		const auto id = builder.AllocateId();
-		builder.AddFunction({opcode, id, operands...});
+		builder.AddFunction(opcode, id, operands...);
 		return id;
 	}
 
 	template <typename... Args>
 	void Emit(spv::Op opcode, Args... operands) {
-		builder.AddFunction({opcode, operands...});
+		builder.AddFunction(opcode, operands...);
 	}
 
 	template <typename... Args>
@@ -242,7 +243,7 @@ private:
 	}
 
 	void Decorate(uint32_t target, spv::Decoration decoration, uint32_t value) {
-		builder.AddAnnotation({spv::OpDecorate, target, decoration, value});
+		builder.AddAnnotation(spv::OpDecorate, target, decoration, value);
 	}
 
 	void DefineEntry(spv::ExecutionModel model) {
@@ -250,11 +251,11 @@ private:
 		builder.RequireCapability(spv::CapabilityTessellation);
 		main = Result(spv::OpFunction, void_type, spv::FunctionControlMaskNone, function_type);
 		if (model == spv::ExecutionModelTessellationControl) {
-			builder.AddExecutionMode({main, spv::ExecutionModeOutputVertices, 4u});
+			builder.AddExecutionMode(main, spv::ExecutionModeOutputVertices, 4u);
 		} else {
-			builder.AddExecutionMode({main, spv::ExecutionModeQuads});
-			builder.AddExecutionMode({main, spv::ExecutionModeSpacingEqual});
-			builder.AddExecutionMode({main, spv::ExecutionModeVertexOrderCw});
+			builder.AddExecutionMode(main, spv::ExecutionModeQuads);
+			builder.AddExecutionMode(main, spv::ExecutionModeSpacingEqual);
+			builder.AddExecutionMode(main, spv::ExecutionModeVertexOrderCw);
 		}
 		DefineInputs(model);
 		DefineOutputs(model);
@@ -296,10 +297,10 @@ private:
 			gl_out     = AddInterface(spv::StorageClassOutput, Array(per_vertex_type, 4u));
 			tess_inner = AddInterface(spv::StorageClassOutput, Array(float_type, 2u));
 			Decorate(tess_inner, spv::DecorationBuiltIn, spv::BuiltInTessLevelInner);
-			builder.AddAnnotation({spv::OpDecorate, tess_inner, spv::DecorationPatch});
+			builder.AddAnnotation(spv::OpDecorate, tess_inner, spv::DecorationPatch);
 			tess_outer = AddInterface(spv::StorageClassOutput, Array(float_type, 4u));
 			Decorate(tess_outer, spv::DecorationBuiltIn, spv::BuiltInTessLevelOuter);
-			builder.AddAnnotation({spv::OpDecorate, tess_outer, spv::DecorationPatch});
+			builder.AddAnnotation(spv::OpDecorate, tess_outer, spv::DecorationPatch);
 		} else {
 			gl_out = AddInterface(spv::StorageClassOutput, per_vertex_type);
 		}

src/graphics/shader/shader.cpp

@@ -11,6 +11,8 @@
 #include "graphics/guest_gpu/graphicsRun.h"
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
+#include "graphics/shader/recompiler/ShaderRecompiler.h"
+#include "graphics/shader/recompiler/Tessellation.h"
 #include "graphics/shader/recompiler/frontend/decode/ShaderDecoder.h"
 #include "graphics/shader/shaderCompiler.h"
 #include "graphics/shader/shaderVertexMetadata.h"
@@ -511,33 +513,28 @@ static uint32_t ShaderCalcPsSystemInputBase(const HW::ShaderRegisters& regs) {
 	return reg;
 }
 
-static bool ShaderGetStaticInputInfoVS(const HW::VertexShaderInfo& regs,
-	                                   const HW::ShaderRegisters& sh,
-	                                   const ShaderMappedData& data,
-	                                   ShaderVertexInputInfo& info) {
+static bool ShaderGetStaticVertexInputInfo(uint64_t shader_addr, const HW::UserSgprInfo& user_sgpr,
+                                           uint32_t user_sgpr_num, const HW::ShaderRegisters& sh,
+                                           const ShaderMappedData& data,
+                                           ShaderVertexInputInfo&  info) {
 	KYTY_PROFILER_FUNCTION();
 
 	info = {};
 
 	info.pa_cl_vs_out_cntl = sh.m_paClVsOutCntl;
 
-	EXIT_NOT_IMPLEMENTED(regs.es_regs.data_addr == 0);
-
-	uint64_t                shader_addr   = regs.es_regs.data_addr;
-	const HW::UserSgprInfo& user_sgpr     = regs.gs_user_sgpr;
-	auto                    user_sgpr_num = regs.gs_regs.rsrc2.user_sgpr;
+	EXIT_NOT_IMPLEMENTED(shader_addr == 0);
 	info.scratch_size_dwords = data.scratch_size_dwords;
 
 	if (data.user_data == nullptr) {
-		LOGF("ShaderGetInputInfoVS(): no AGC user data for shader=0x%016" PRIx64 " es=0x%016" PRIx64
-		     " gs=0x%016" PRIx64 " user_sgpr_num=%u\n",
-		     shader_addr, regs.es_regs.data_addr, regs.gs_regs.data_addr,
-		     static_cast<uint32_t>(user_sgpr_num));
+		LOGF("ShaderGetVertexInputInfo(): no AGC user data for shader=0x%016" PRIx64
+		     " user_sgpr_num=%u\n",
+		     shader_addr, user_sgpr_num);
 	}
 	ShaderVertexMetadata metadata;
 	std::string          metadata_error;
 	if (!ShaderReadVertexMetadata(data, HW::UserSgprInfo::SGPRS_MAX, metadata, &metadata_error)) {
-		LOGF("ShaderGetInputInfoVS(): invalid AGC metadata shader=0x%016" PRIx64 ": %s\n",
+		LOGF("ShaderGetVertexInputInfo(): invalid AGC metadata shader=0x%016" PRIx64 ": %s\n",
 		     shader_addr, metadata_error.c_str());
 		return false;
 	}
@@ -556,7 +553,7 @@ static bool ShaderGetStaticInputInfoVS(const HW::VertexShaderInfo& regs,
 		    (static_cast<uint64_t>(user_sgpr.value[metadata.vertex_buffer_reg + 1]) << 32u));
 
 		if (attrib == nullptr || buffer == nullptr) {
-			LOGF("ShaderGetInputInfoVS(): null vertex table pointer shader=0x%016" PRIx64 "\n",
+			LOGF("ShaderGetVertexInputInfo(): null vertex table pointer shader=0x%016" PRIx64 "\n",
 			     shader_addr);
 			return false;
 		}
@@ -576,9 +573,11 @@ static void ShaderGetStaticInputInfoPS(
 	ps_info = {};
 	ps_info.scratch_size_dwords = data.scratch_size_dwords;
 
-	// SPI_PS_IN_CONTROL.NUM_INTERP occupies bits 5:0. Keep the remaining control
-	// flags in the hardware state and extract only the input count here.
+	// SPI_PS_IN_CONTROL: NUM_INTERP occupies bits 5:0 and PS_W32_EN is bit 15.
 	ps_info.input_num            = sh.ps_in_control & 0x3fu;
+	if ((sh.ps_in_control & 0x8000u) != 0) {
+		ps_info.wave_size = 32;
+	}
 	EXIT_NOT_IMPLEMENTED(ps_info.input_num > std::size(ps_info.interpolator_settings));
 	ps_info.ps_system_input_base = ShaderCalcPsSystemInputBase(sh);
 	const uint32_t active_inputs = sh.ps_input_ena & sh.ps_input_addr;
@@ -673,6 +672,12 @@ void BuildStageStaticKey(const ShaderVertexInputInfo& info, std::vector<uint32_t
 		                       mesh.primitives_per_group, mesh.vertices_per_group,
 		                       mesh.max_vertices, mesh.max_primitives, mesh.provoking_vertex});
 	}
+	key.push_back(info.tess.input_control_points);
+	if (info.tess.input_control_points != 0) {
+		const auto& tess = info.tess;
+		key.insert(key.end(), {tess.output_control_points, tess.ls_stride, tess.hs_stride,
+		                       tess.domain, tess.partitioning, tess.output_topology});
+	}
 
 	for (int i = 0; i < info.resources_num; i++) {
 		const auto& resource    = info.resources[i];
@@ -698,6 +703,7 @@ void BuildStageStaticKey(const ShaderPixelInputInfo& info, std::vector<uint32_t>
 	key.clear();
 	key.push_back(info.scratch_size_dwords);
 	key.push_back(info.input_num);
+	key.push_back(info.wave_size);
 	key.push_back(info.ps_system_input_base);
 	key.push_back(info.custom_interpolation_mask);
 	key.push_back(info.ps_perspective_center_vgpr);
@@ -750,7 +756,8 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
 	    GetDeclaredShaderHash(regs.es_regs.data_addr),
 	    std::span<const uint32_t>(regs.gs_user_sgpr.value, regs.gs_regs.rsrc2.user_sgpr), data);
 	if ((context.GetShaderStages() & 0x20u) == 0) {
-		if (!ShaderGetStaticInputInfoVS(regs, sh, data, info)) {
+		if (!ShaderGetStaticVertexInputInfo(regs.es_regs.data_addr, regs.gs_user_sgpr,
+		                                    regs.gs_regs.rsrc2.user_sgpr, sh, data, info)) {
 			EXIT("failed to prepare vertex shader program\n");
 		}
 		return params;
@@ -759,6 +766,7 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
 	// its user-data pointer in s0:s1.
 	params.user_data.insert(params.user_data.begin(), 8u, 0u);
 	info                     = {};
+	info.logical_stage       = ShaderType::Mesh;
 	info.pa_cl_vs_out_cntl   = sh.m_paClVsOutCntl;
 	auto& mesh               = info.mesh;
 	mesh.input_primitive     = static_cast<uint32_t>(user_config.GetPrimType());
@@ -806,6 +814,65 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
 	return params;
 }
 
+std::array<ShaderParams, 3>
+PrepareTessellationPrograms(const HW::VertexShaderInfo& regs, const HW::Context& context,
+                            std::array<ShaderVertexInputInfo, 3>& input_info) {
+	const auto& sh        = context.GetShaderRegisters();
+	const auto  local     = ShaderGetMappedData(regs.ls_regs.data_addr, "ShaderGetInputInfoLS():");
+	const auto  control   = ShaderGetMappedData(regs.hs_regs.data_addr, "ShaderGetInputInfoHS():");
+	const auto evaluation = ShaderGetMappedData(regs.es_regs.data_addr, "ShaderGetInputInfoTES():");
+	EXIT_NOT_IMPLEMENTED(local.type != Prospero::ShaderBinaryType::kHsFront ||
+	                     control.type != Prospero::ShaderBinaryType::kHsBack ||
+	                     evaluation.type != Prospero::ShaderBinaryType::kGs);
+	EXIT_NOT_IMPLEMENTED((context.GetShaderStages() & 0x4u) == 0 ||
+	                     (context.GetShaderStages() & 0x00600020u) != 0);
+	EXIT_IF(regs.hs_regs.user_data_addr == 0);
+
+	const auto local_users      = std::span(regs.hs_user_sgpr.value, regs.hs_regs.rsrc2.user_sgpr);
+	const auto evaluation_users = std::span(regs.gs_user_sgpr.value, regs.gs_regs.rsrc2.user_sgpr);
+	std::array<ShaderParams, 3> params {
+	    GetShaderParams(regs.ls_regs.data_addr, "ShaderRecompiler LS",
+	                    GetDeclaredShaderHash(regs.ls_regs.data_addr), local_users, local),
+	    GetShaderParams(regs.hs_regs.data_addr, "ShaderRecompiler HS",
+	                    GetDeclaredShaderHash(regs.hs_regs.data_addr), local_users, control),
+	    GetShaderParams(regs.es_regs.data_addr, "ShaderRecompiler TES",
+	                    GetDeclaredShaderHash(regs.es_regs.data_addr), evaluation_users,
+	                    evaluation),
+	};
+	// The fused HS back half receives its separate user-data address in s0:s1.
+	// RDNA2 reserves s0:s7 before the native HS user SGPRs.
+	params[1].user_data.insert(params[1].user_data.begin(), 8u, 0u);
+	params[1].user_data[0] = static_cast<uint32_t>(regs.hs_regs.user_data_addr);
+	params[1].user_data[1] = static_cast<uint32_t>(regs.hs_regs.user_data_addr >> 32u);
+
+	input_info = {};
+	if (!ShaderGetStaticVertexInputInfo(regs.ls_regs.data_addr, regs.hs_user_sgpr,
+	                                    regs.hs_regs.rsrc2.user_sgpr, sh, local, input_info[0])) {
+		EXIT("failed to prepare local shader program\n");
+	}
+	input_info[0].logical_stage       = ShaderType::Local;
+	input_info[1].logical_stage       = ShaderType::TessellationControl;
+	input_info[1].scratch_size_dwords = control.scratch_size_dwords;
+	input_info[2].logical_stage       = ShaderType::TessellationEvaluation;
+	input_info[2].scratch_size_dwords = evaluation.scratch_size_dwords;
+	input_info[2].pa_cl_vs_out_cntl   = sh.m_paClVsOutCntl;
+
+	ShaderTessellationInputInfo tess {
+	    .input_control_points  = (sh.m_vgtLsHsConfig >> 8u) & 0x3fu,
+	    .output_control_points = (sh.m_vgtLsHsConfig >> 14u) & 0x3fu,
+	    .domain                = sh.m_vgtTfParam & 0x3u,
+	    .partitioning          = (sh.m_vgtTfParam >> 2u) & 0x3u,
+	    .output_topology       = (sh.m_vgtTfParam >> 5u) & 0x3u,
+	};
+	EXIT_IF(tess.input_control_points == 0 || tess.input_control_points > 32 ||
+	        tess.output_control_points == 0 || tess.output_control_points > 32);
+	ShaderRecompiler::AnalyzeTessellationPrograms(params[0].code, params[1].code, tess);
+	for (auto& stage: input_info) {
+		stage.tess = tess;
+	}
+	return params;
+}
+
 ShaderParams PrepareProgram(
     const HW::PixelShaderInfo& regs, const HW::ShaderRegisters& sh,
     std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,

src/graphics/shader/shader.h

@@ -33,7 +33,17 @@ struct ComputeShaderInfo;
 struct ShaderRegisters;
 } // namespace HW
 
-enum class ShaderType { Unknown, Vertex, Pixel, Fetch, Compute, Mesh };
+enum class ShaderType {
+	Unknown,
+	Vertex,
+	Pixel,
+	Fetch,
+	Compute,
+	Mesh,
+	Local,
+	TessellationControl,
+	TessellationEvaluation
+};
 
 namespace ShaderRecompiler::IR {
 struct CompiledShaderInfo;
@@ -99,6 +109,16 @@ struct ShaderMeshInputInfo: ShaderWorkgroupInputInfo {
 	}
 };
 
+struct ShaderTessellationInputInfo {
+	uint32_t input_control_points  = 0;
+	uint32_t output_control_points = 0;
+	uint32_t ls_stride             = 0;
+	uint32_t hs_stride             = 0;
+	uint32_t domain                = 0;
+	uint32_t partitioning          = 0;
+	uint32_t output_topology       = 0;
+};
+
 struct ShaderVertexInputInfo {
 	static constexpr int RES_MAX = 32;
 
@@ -106,6 +126,7 @@ struct ShaderVertexInputInfo {
 	ShaderVertexDestination resources_dst[RES_MAX];
 	ShaderVertexInputBuffer buffers[RES_MAX];
 	ShaderStageRuntime      stage;
+	ShaderType                  logical_stage        = ShaderType::Vertex;
 	int                     resources_num       = 0;
 	int                     fetch_attrib_reg    = 0;
 	int                     fetch_buffer_reg    = 0;
@@ -114,6 +135,7 @@ struct ShaderVertexInputInfo {
 	uint32_t                pa_cl_vs_out_cntl    = 0;
 	ShaderClipSpaceTransform clip_space;
 	ShaderMeshInputInfo      mesh;
+	ShaderTessellationInputInfo tess;
 	bool                    fetch_external      = false;
 	bool                    fetch_embedded      = false;
 };
@@ -131,6 +153,7 @@ struct ShaderComputeInputInfo: ShaderWorkgroupInputInfo {
 struct ShaderPixelInputInfo {
 	uint32_t                                       interpolator_settings[32]    = {0};
 	uint32_t                                       input_num                    = 0;
+	uint32_t                                       wave_size                    = 64;
 	uint32_t                                       ps_system_input_base         = 0;
 	uint32_t                                       custom_interpolation_mask    = 0;
 	uint32_t                                       ps_perspective_center_vgpr   = UINT32_MAX;

src/graphics/shader/shaderCompiler.h

@@ -30,6 +30,9 @@ void BuildStageStaticKey(const ShaderComputeInputInfo& input_info, std::vector<u
 
 ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context& context,
                             const HW::UserConfig& user_config, ShaderVertexInputInfo& input_info);
+std::array<ShaderParams, 3>
+PrepareTessellationPrograms(const HW::VertexShaderInfo& regs, const HW::Context& context,
+                            std::array<ShaderVertexInputInfo, 3>& input_info);
 ShaderParams PrepareProgram(
     const HW::PixelShaderInfo& regs, const HW::ShaderRegisters& sh,
     std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,

src/kernel/fileSystem.cpp

@@ -458,6 +458,7 @@ int KYTY_SYSV_ABI KernelOpen(const char* path, int flags, uint16_t mode) {
 	file->real_name = g_mount_points->ResolvePath(file->name);
 
 	if (trunc && rw_mode == Common::File::Mode::Read) {
+		g_files->DeleteDescriptor(descriptor);
 		return KERNEL_ERROR_EACCES;
 	}
 

src/kernel/memoryAddressSpace.inc

@@ -346,7 +346,9 @@ public:
 				auto entry = std::prev(next);
 				if (vaddr + size <= entry->second.vaddr + entry->second.size) {
 					old = entry->second;
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 					m_maps.erase(entry);
+#endif
 				}
 			}
 		}
@@ -355,6 +357,12 @@ public:
 			return false;
 		}
 
+		const uint64_t left_size  = vaddr - old.vaddr;
+		const uint64_t right_addr = vaddr + size;
+		const uint64_t old_end    = old.vaddr + old.size;
+		const uint64_t right_size = old_end - right_addr;
+
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 		const bool exact_unmap = (vaddr == old.vaddr && size == old.size);
 		if (exact_unmap) {
 			const bool unmapped =
@@ -370,12 +378,6 @@ public:
 			return true;
 		}
 
-		const uint64_t left_size  = vaddr - old.vaddr;
-		const uint64_t right_addr = vaddr + size;
-		const uint64_t old_end    = old.vaddr + old.size;
-		const uint64_t right_size = old_end - right_addr;
-
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 		if (UnmapViewPreservePlaceholder(reinterpret_cast<void*>(old.vaddr), old.size)) {
 			bool ok = true;
 			if (left_size != 0 && left_size != old.size) {
@@ -418,39 +420,28 @@ public:
 		m_maps.emplace(old.vaddr, old);
 		return false;
 #else
-		if (!UnmapViewPreservePlaceholder(reinterpret_cast<void*>(old.vaddr), old.size)) {
-			std::lock_guard lock(m_mutex);
-			m_maps.emplace(old.vaddr, old);
+		// Replace only the requested range so concurrent accesses and GPU watch
+		// protections on the surviving mappings remain intact.
+		if (!UnmapViewPreservePlaceholder(reinterpret_cast<void*>(vaddr), size)) {
 			return false;
 		}
 
-		bool ok = true;
+		std::lock_guard lock(m_mutex);
+		m_maps.erase(old.vaddr);
 		if (left_size != 0) {
-			ok = MapExistingPlaceholderFixed(old.vaddr, left_size, old.backing_offset, old.mode) &&
-			     ok;
+			m_maps.emplace(old.vaddr,
+			               MapEntry {old.vaddr, left_size, old.backing_offset, old.mode});
 		}
 		if (right_size != 0) {
-			ok = MapExistingPlaceholderFixed(right_addr, right_size,
-			                                 old.backing_offset + (right_addr - old.vaddr),
-			                                 old.mode) &&
-			     ok;
-		}
-		if (!ok) {
-			if (left_size != 0 && Contains(old.vaddr, left_size)) {
-				EXIT_IF(!UnmapOne(old.vaddr, left_size));
-			}
-			if (right_size != 0 && Contains(right_addr, right_size)) {
-				EXIT_IF(!UnmapOne(right_addr, right_size));
-			}
-			EXIT_IF(
-			    !MapExistingPlaceholderFixed(old.vaddr, old.size, old.backing_offset, old.mode));
-			return false;
+			m_maps.emplace(right_addr,
+			               MapEntry {right_addr, right_size,
+			                         old.backing_offset + right_addr - old.vaddr, old.mode});
 		}
 
 		if (placeholder_preserved != nullptr) {
 			*placeholder_preserved = true;
 		}
-		return ok;
+		return true;
 #endif
 	}
 

src/kernel/pthread.cpp

@@ -3093,10 +3093,6 @@ Pthread PthreadSwapSelfForSignal(Pthread thread) {
 	return previous;
 }
 
-int PthreadGetUniqueId(Pthread thread) {
-	return thread != nullptr ? thread->unique_id : 0;
-}
-
 uint64_t PthreadGetHostThreadId(Pthread thread) {
 	return thread != nullptr ? thread->host_thread_id : 0;
 }

src/kernel/pthread.h

@@ -110,7 +110,6 @@ int KYTY_SYSV_ABI     PthreadRename(Pthread thread, const char* name);
 void KYTY_SYSV_ABI    PthreadYield();
 int KYTY_SYSV_ABI     PthreadGetthreadid();
 int KYTY_SYSV_ABI     PthreadGetaffinity(Pthread thread, KernelCpumask* mask);
-int                   PthreadGetUniqueId(Pthread thread);
 uint64_t              PthreadGetHostThreadId(Pthread thread);
 void                  PthreadWakeForSignal(Pthread thread);
 void                  PthreadQueuePendingSignal(Pthread thread, int signum);

src/kernel/syncOnAddress.cpp

@@ -47,11 +47,17 @@ struct WaitDeadline {
 	Clock::time_point end {};
 };
 
+[[nodiscard]] WaitDeadline MakeDeadline(std::chrono::nanoseconds timeout) {
+	const auto now = Clock::now();
+	const auto remaining = Clock::time_point::max() - now;
+	return {true, timeout >= remaining ? Clock::time_point::max() : now + timeout};
+}
+
 [[nodiscard]] WaitDeadline MakeDeadline(const uint32_t* timeout_micros) {
 	if (timeout_micros == nullptr) {
 		return {};
 	}
-	return {true, Clock::now() + std::chrono::microseconds(*timeout_micros)};
+	return MakeDeadline(std::chrono::microseconds(*timeout_micros));
 }
 
 [[nodiscard]] uint32_t GetWaitSliceMicros(const WaitDeadline& deadline, bool first_wait) {
@@ -65,7 +71,7 @@ struct WaitDeadline {
 	}
 
 	const auto remaining =
-	    std::chrono::duration_cast<std::chrono::microseconds>(deadline.end - now).count();
+	    std::chrono::ceil<std::chrono::microseconds>(deadline.end - now).count();
 	return static_cast<uint32_t>(std::min<int64_t>(remaining, SIGNAL_POLL_MICROS));
 }
 
@@ -78,9 +84,8 @@ void PollSignals(signal_poll_func_t signal_poll) {
 #if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
 
 template <typename T>
-int WaitLinux(volatile T* address, T expected, const uint32_t* timeout_micros,
+int WaitLinux(volatile T* address, T expected, const WaitDeadline& deadline,
               signal_poll_func_t signal_poll) {
-	const auto deadline   = MakeDeadline(timeout_micros);
 	bool       first_wait = true;
 
 	for (;;) {
@@ -185,11 +190,10 @@ void UnregisterPortableWaiter(volatile void*                               addre
 }
 
 template <typename T>
-int WaitPortable(volatile T* address, T expected, const uint32_t* timeout_micros,
+int WaitPortable(volatile T* address, T expected, const WaitDeadline& deadline,
                  signal_poll_func_t signal_poll) {
 	PortableWaiter waiter;
 	auto           entry      = RegisterPortableWaiter(address, &waiter);
-	const auto     deadline   = MakeDeadline(timeout_micros);
 	bool           first_wait = true;
 	int            result     = OK;
 
@@ -250,7 +254,7 @@ int WakePortable(volatile void* address, int32_t count) {
 }
 
 template <typename T>
-int WaitImpl(volatile T* address, T expected, const uint32_t* timeout_micros,
+int WaitImpl(volatile T* address, T expected, const WaitDeadline& deadline,
              signal_poll_func_t signal_poll) {
 	if (!IsValidWaitAddress(address)) {
 		return KERNEL_ERROR_EINVAL;
@@ -258,9 +262,9 @@ int WaitImpl(volatile T* address, T expected, const uint32_t* timeout_micros,
 
 	int result = OK;
 #if KYTY_PLATFORM == KYTY_PLATFORM_LINUX && !defined(__APPLE__)
-	result = WaitLinux(address, expected, timeout_micros, signal_poll);
+	result = WaitLinux(address, expected, deadline, signal_poll);
 #else
-	result = WaitPortable(address, expected, timeout_micros, signal_poll);
+	result = WaitPortable(address, expected, deadline, signal_poll);
 #endif
 	PollSignals(signal_poll);
 	return result;
@@ -270,12 +274,17 @@ int WaitImpl(volatile T* address, T expected, const uint32_t* timeout_micros,
 
 int Wait32(volatile uint32_t* address, uint32_t expected, const uint32_t* timeout_micros,
            signal_poll_func_t signal_poll) {
-	return WaitImpl(address, expected, timeout_micros, signal_poll);
+	return WaitImpl(address, expected, MakeDeadline(timeout_micros), signal_poll);
 }
 
 int Wait64(volatile uint64_t* address, uint64_t expected, const uint32_t* timeout_micros,
            signal_poll_func_t signal_poll) {
-	return WaitImpl(address, expected, timeout_micros, signal_poll);
+	return WaitImpl(address, expected, MakeDeadline(timeout_micros), signal_poll);
+}
+
+int Wait64(volatile uint64_t* address, uint64_t expected, std::chrono::nanoseconds timeout,
+           signal_poll_func_t signal_poll) {
+	return WaitImpl(address, expected, MakeDeadline(timeout), signal_poll);
 }
 
 int Wake(volatile void* address, int32_t count) {

src/kernel/syncOnAddress.h

@@ -4,6 +4,8 @@
 #include "common/abi.h"
 #include "common/common.h"
 
+#include <chrono>
+
 namespace Libs::LibKernel::SyncOnAddress {
 
 using signal_poll_func_t = void (*)();
@@ -12,6 +14,8 @@ int Wait32(volatile uint32_t* address, uint32_t expected, const uint32_t* timeou
            signal_poll_func_t signal_poll = nullptr);
 int Wait64(volatile uint64_t* address, uint64_t expected, const uint32_t* timeout_micros,
            signal_poll_func_t signal_poll = nullptr);
+int Wait64(volatile uint64_t* address, uint64_t expected, std::chrono::nanoseconds timeout,
+           signal_poll_func_t signal_poll = nullptr);
 int Wake(volatile void* address, int32_t count);
 
 } // namespace Libs::LibKernel::SyncOnAddress

src/launcher/forms/configuration_list_widget.ui

@@ -218,6 +218,11 @@
            <string>Firmware Version</string>
           </property>
          </column>
+         <column>
+          <property name="text">
+           <string>Size</string>
+          </property>
+         </column>
          <column>
           <property name="text">
            <string>Path</string>

src/launcher/src/configurationItem.cpp

@@ -5,12 +5,15 @@
 #include <QApplication>
 #include <QComboBox>
 #include <QDir>
+#include <QDirIterator>
 #include <QFileInfo>
 #include <QFont>
+#include <QFutureWatcher>
 #include <QHBoxLayout>
 #include <QIcon>
 #include <QLabel>
 #include <QLineEdit>
+#include <QLocale>
 #include <QSize>
 #include <QStringList>
 #include <QStyle>
@@ -18,6 +21,7 @@
 #include <QVersionNumber>
 #include <QWheelEvent>
 #include <QWidget>
+#include <QtConcurrentRun>
 
 namespace {
 
@@ -26,6 +30,7 @@ enum Column {
 	SerialColumn,
 	GameVersionColumn,
 	FirmwareVersionColumn,
+	SizeColumn,
 	PathColumn,
 	StatusColumn,
 	CommentsColumn,
@@ -155,6 +160,34 @@ ConfigurationItem::ConfigurationItem(std::unique_ptr<Configuration> info, QTreeW
 
 	Update();
 	SetRunning(false);
+
+	setData(SizeColumn, Qt::UserRole, qint64(-1));
+	setText(SizeColumn, QStringLiteral("\u2014"));
+	auto* watcher = new QFutureWatcher<qint64>(this);
+	connect(watcher, &QFutureWatcher<qint64>::finished, this, [this, watcher]() {
+		const auto bytes = watcher->result();
+		setData(SizeColumn, Qt::UserRole, bytes);
+		if (bytes >= 0) {
+			setText(SizeColumn,
+			        QLocale().formattedDataSize(bytes, 2, QLocale::DataSizeTraditionalFormat));
+		}
+		watcher->deleteLater();
+	});
+	watcher->setFuture(QtConcurrent::run([path = m_info->basedir]() -> qint64 {
+		if (path.isEmpty() || !QDir(path).exists()) {
+			return -1;
+		}
+		qint64       bytes = 0;
+		QDirIterator files(path, QDir::Files | QDir::Hidden | QDir::System | QDir::NoSymLinks,
+		                   QDirIterator::Subdirectories);
+		while (files.hasNext()) {
+			files.next();
+			if (files.fileInfo().isFile()) {
+				bytes += files.fileInfo().size();
+			}
+		}
+		return bytes;
+	}));
 }
 
 ConfigurationItem::~ConfigurationItem() = default;
@@ -193,6 +226,9 @@ bool ConfigurationItem::operator<(const QTreeWidgetItem& other) const {
 	const int column = treeWidget() != nullptr ? treeWidget()->sortColumn() : NameColumn;
 	switch (column) {
 		case NameColumn: return GetSortText(*m_info) < GetSortText(*other_item->m_info);
+		case SizeColumn:
+			return data(SizeColumn, Qt::UserRole).toLongLong() <
+			       other.data(SizeColumn, Qt::UserRole).toLongLong();
 		case StatusColumn:
 			return GetStatusText(m_info->game_status) <
 			       GetStatusText(other_item->m_info->game_status);

src/launcher/src/configurationListWidget.cpp

@@ -61,9 +61,10 @@ constexpr int GAME_NAME_COLUMN             = 0;
 constexpr int GAME_SERIAL_COLUMN           = 1;
 constexpr int GAME_VERSION_COLUMN          = 2;
 constexpr int GAME_FIRMWARE_VERSION_COLUMN = 3;
-constexpr int GAME_PATH_COLUMN             = 4;
-constexpr int GAME_STATUS_COLUMN           = 5;
-constexpr int GAME_COMMENT_COLUMN          = 6;
+constexpr int GAME_SIZE_COLUMN             = 4;
+constexpr int GAME_PATH_COLUMN             = 5;
+constexpr int GAME_STATUS_COLUMN           = 6;
+constexpr int GAME_COMMENT_COLUMN          = 7;
 
 static QString NormalizeGameDirectory(const QString& dir) {
 	const auto trimmed = dir.trimmed();
@@ -217,6 +218,7 @@ ConfigurationListWidget::ConfigurationListWidget(QWidget* parent)
 	m_ui->cfgs_list->setColumnWidth(GAME_SERIAL_COLUMN, 110);
 	m_ui->cfgs_list->setColumnWidth(GAME_VERSION_COLUMN, 120);
 	m_ui->cfgs_list->setColumnWidth(GAME_FIRMWARE_VERSION_COLUMN, 150);
+	m_ui->cfgs_list->setColumnWidth(GAME_SIZE_COLUMN, 100);
 	m_ui->cfgs_list->setColumnWidth(GAME_PATH_COLUMN, 320);
 	m_ui->cfgs_list->setColumnWidth(GAME_STATUS_COLUMN, 150);
 	m_ui->cfgs_list->setColumnWidth(GAME_COMMENT_COLUMN, 240);

src/libs/avPlayer.cpp

@@ -552,6 +552,7 @@ private:
 struct ReadyFrame {
 	std::unique_ptr<GuestBuffer> buffer;
 	AvPlayerFrameInfoEx          info {};
+	uint64_t                    timestamp_offset = 0;
 };
 
 class FileStreamer {
@@ -933,6 +934,7 @@ public:
 		}
 		current_video = std::move(*frame);
 		*out          = current_video->info;
+		RecordLoopBoundary(*current_video);
 		if (deliver_seek_frame) {
 			seek_video_frame_pending = false;
 		}
@@ -963,11 +965,25 @@ public:
 		std::memcpy(out->details.audio.language_code,
 		            current_audio->info.details.audio.language_code, 4);
 		last_audio_ts = out->time_stamp;
+		RecordLoopBoundary(*current_audio);
 		return true;
 	}
-	std::optional<int32_t> TakeWarning() { return warnings.TryPop(); }
+	std::optional<int32_t> TakeWarning() {
+		std::lock_guard lock(mutex);
+		if (pending_loop_warnings == 0) {
+			return std::nullopt;
+		}
+		--pending_loop_warnings;
+		return AVPLAYER_WARNING_LOOPING_BACK;
+	}
 
 private:
+	void RecordLoopBoundary(const ReadyFrame& frame) {
+		if (frame.timestamp_offset > last_output_loop_offset) {
+			last_output_loop_offset = frame.timestamp_offset;
+			++pending_loop_warnings;
+		}
+	}
 	void Close() {
 		Stop();
 		if (fmt != nullptr) {
@@ -1014,6 +1030,8 @@ private:
 		audio_done               = true;
 		seek_video_frame_pending = false;
 		last_audio_ts            = 0;
+		last_output_loop_offset  = 0;
+		pending_loop_warnings    = 0;
 		current_video            = std::move(delivered_video);
 		current_audio            = std::move(delivered_audio);
 		retired_video            = std::move(in_flight_video);
@@ -1220,7 +1238,6 @@ private:
 					if (avformat_seek_file(fmt, stream, INT64_MIN, 0, INT64_MAX, 0) < 0) {
 						break;
 					}
-					warnings.Push(AVPLAYER_WARNING_LOOPING_BACK);
 					continue;
 				}
 				if (result != AVERROR_EOF && !worker_stop) {
@@ -1340,6 +1357,7 @@ private:
 				return false;
 			}
 			ready.info.time_stamp += timestamp_offset;
+			ready.timestamp_offset = timestamp_offset;
 			frames.Push(std::move(ready));
 			av_frame_free(&frame);
 		}
@@ -1603,7 +1621,6 @@ private:
 	std::optional<ReadyFrame>                current_video;
 	std::optional<ReadyFrame>                current_audio;
 	std::deque<std::unique_ptr<GuestBuffer>> retired_video;
-	WorkQueue<int32_t>                       warnings;
 	LibKernel::Pthread                       demux_thread = nullptr;
 	LibKernel::Pthread                       video_thread = nullptr;
 	LibKernel::Pthread                       audio_thread = nullptr;
@@ -1623,6 +1640,8 @@ private:
 	uint32_t                                 sync_mode     = 0;
 	uint64_t                                 start_time_ms = 0;
 	uint64_t                                 last_audio_ts = 0;
+	uint64_t                                 last_output_loop_offset = 0;
+	uint32_t                                 pending_loop_warnings   = 0;
 	std::chrono::steady_clock::time_point    clock_start {};
 	std::chrono::steady_clock::time_point    pause_time {};
 	std::chrono::steady_clock::duration      paused_extra {};
@@ -1951,6 +1970,7 @@ Bool KYTY_SYSV_ABI AvPlayerGetVideoData(AvPlayerInternal* h, AvPlayerFrameInfo*
 	video_info->details.video.height       = ex.details.video.height;
 	video_info->details.video.aspect_ratio = ex.details.video.aspect_ratio;
 	std::memcpy(video_info->details.video.language_code, ex.details.video.language_code, 4);
+	pump_warnings(h);
 	return 1;
 }
 Bool KYTY_SYSV_ABI AvPlayerGetVideoDataEx(AvPlayerInternal* h, AvPlayerFrameInfoEx* video_info) {

src/libs/hmd2.h

@@ -0,0 +1,24 @@
+#ifndef EMULATOR_SRC_LIBS_HMD2_H_
+#define EMULATOR_SRC_LIBS_HMD2_H_
+
+#include <atomic>
+#include <cstdint>
+
+namespace Libs::Hmd2 {
+
+inline constexpr uint32_t PANEL_WIDTH  = 4000;
+inline constexpr uint32_t PANEL_HEIGHT = 2040;
+inline constexpr uint64_t OUTPUT_MODE_89_91HZ  = 0x2000c;
+inline constexpr uint64_t OUTPUT_MODE_119_88HZ = 0x2000f;
+inline constexpr int32_t ERROR_REPROJECTION_NOT_INITIALIZED = -1972240373; // 0x8a72000b
+
+struct ReprojectionState {
+	std::atomic<uint64_t> output_mode {0};
+	std::atomic<uint32_t> timing_us {3000};
+};
+
+[[nodiscard]] ReprojectionState* GetReprojectionState();
+
+} // namespace Libs::Hmd2
+
+#endif // EMULATOR_SRC_LIBS_HMD2_H_

src/libs/libHmd2.cpp

@@ -0,0 +1,239 @@
+#include "common/emulatorConfig.h"
+#include "graphics/guest_gpu/tile.h"
+#include "libs/agc.h"
+#include "libs/errno.h"
+#include "libs/hmd2.h"
+#include "libs/libs.h"
+#include "loader/symbolDatabase.h"
+
+#include <atomic>
+#include <cstddef>
+#include <cstdint>
+#include <mutex>
+#include <new>
+
+namespace Libs {
+
+LIB_VERSION("Hmd2", 1, "Hmd2", 1, 1);
+
+namespace Hmd2 {
+
+constexpr int32_t ERROR_ALREADY_INITIALIZED = -1972240383; // 0x8a720001
+constexpr int32_t ERROR_NOT_INITIALIZED     = -1972240382; // 0x8a720002
+constexpr int32_t ERROR_PARAMETER_NULL      = -1972240376; // 0x8a720008
+constexpr int32_t ERROR_REPROJECTION_ALREADY_INITIALIZED = -1972240372; // 0x8a72000c
+constexpr int32_t ERROR_REPROJECTION_IN_VR_MODE = -1972240356; // 0x8a72001c
+
+struct SceHmd2InitializeParam {
+	void*   reserved0;
+	uint8_t reserved[8];
+};
+
+struct SceHmd2DeviceInformation {
+	uint32_t status;
+	uint32_t reserve0;
+	struct {
+		struct {
+			uint32_t width;
+			uint32_t height;
+		} panelResolution;
+		struct {
+			uint16_t refreshRate90Hz;
+			uint16_t refreshRate120Hz;
+		} flipToDisplayLatency;
+	} deviceInfo;
+	uint8_t hmuMount;
+	uint8_t lensSeparationDistance;
+	uint8_t reserve1[2];
+	float   virtualImageDistance;
+};
+
+struct SceHmd2FieldOfView {
+	float tanOut;
+	float tanIn;
+	float tanTop;
+	float tanBottom;
+};
+
+struct SceHmd2ReprojectionInitializeParam {
+	void*    pReprojectionBuff;
+	void*    pDisplayBuff;
+	int32_t  threadPriority;
+	int32_t  cpuAffinityMask;
+	uint32_t pipeId;
+	uint32_t queueId;
+	uint32_t type;
+	void*    pSeeThroughBuff;
+	int32_t  reprojectionTiming;
+	uint32_t reserved[5];
+};
+
+static_assert(sizeof(SceHmd2InitializeParam) == 16);
+static_assert(offsetof(SceHmd2InitializeParam, reserved) == 8);
+static_assert(sizeof(SceHmd2DeviceInformation) == 28);
+static_assert(offsetof(SceHmd2DeviceInformation, deviceInfo) == 8);
+static_assert(offsetof(SceHmd2DeviceInformation, hmuMount) == 20);
+static_assert(offsetof(SceHmd2DeviceInformation, virtualImageDistance) == 24);
+static_assert(sizeof(SceHmd2FieldOfView) == 16);
+static_assert(alignof(SceHmd2FieldOfView) == 4);
+static_assert(sizeof(SceHmd2ReprojectionInitializeParam) == 72);
+static_assert(alignof(SceHmd2ReprojectionInitializeParam) == 8);
+static_assert(offsetof(SceHmd2ReprojectionInitializeParam, threadPriority) == 16);
+static_assert(offsetof(SceHmd2ReprojectionInitializeParam, type) == 32);
+static_assert(offsetof(SceHmd2ReprojectionInitializeParam, pSeeThroughBuff) == 40);
+static_assert(offsetof(SceHmd2ReprojectionInitializeParam, reprojectionTiming) == 48);
+static_assert(offsetof(SceHmd2ReprojectionInitializeParam, reserved) == 52);
+
+static std::atomic<bool> g_initialized = false;
+static std::mutex g_reprojection_mutex;
+static std::atomic<ReprojectionState*> g_reprojection_state {nullptr};
+
+ReprojectionState* GetReprojectionState() {
+	return g_reprojection_state.load(std::memory_order_acquire);
+}
+
+static Graphics::SizeAlign ReprojectionDisplaySize() {
+	// The host display target is one uncompressed tiled 10-bit panel surface.
+	Graphics::TileSizeAlign size {};
+	Graphics::TileGetTextureTotalSize(Graphics::Prospero::BufferFormat::k10_10_10_2UNorm,
+	                                  PANEL_WIDTH, PANEL_HEIGHT, 1, 1,
+	                                  Graphics::Prospero::TileMode::kRenderTarget, false, size);
+	EXIT_IF(size.size == 0 || size.align == 0);
+	return {size.size, size.align};
+}
+
+static int32_t KYTY_SYSV_ABI Hmd2Initialize(const SceHmd2InitializeParam* param) {
+	PRINT_NAME();
+	if (param == nullptr) {
+		return ERROR_PARAMETER_NULL;
+	}
+	if (g_initialized.exchange(true)) {
+		return ERROR_ALREADY_INITIALIZED;
+	}
+	LOGF("Hmd2: initialized, virtual headset=%s\n", Config::VrEnabled() ? "enabled" : "disabled");
+	return OK;
+}
+
+static int32_t KYTY_SYSV_ABI Hmd2GetDeviceInformation(SceHmd2DeviceInformation* info) {
+	PRINT_NAME();
+	if (!g_initialized.load()) {
+		return ERROR_NOT_INITIALIZED;
+	}
+	if (info == nullptr) {
+		return ERROR_PARAMETER_NULL;
+	}
+	*info = {};
+	if (Config::VrEnabled()) {
+		// Virtual headset defaults: READY, MOUNT, 63 mm lens separation.
+		info->hmuMount               = 1;
+		info->lensSeparationDistance = 63;
+		// Headset characteristics: 4000x2040 panel and 2 m virtual image distance.
+		info->deviceInfo.panelResolution = {PANEL_WIDTH, PANEL_HEIGHT};
+		info->virtualImageDistance       = 2.0f;
+		// Host presentation has no emulated headset scan-out delay; latency remains zero.
+	} else {
+		info->status = 2; // SCE_HMD2_DEVICE_STATUS_NOT_DETECTED
+	}
+	LOGF("Hmd2: device status=%u panel=%ux%u mounted=%u\n", info->status,
+	     info->deviceInfo.panelResolution.width, info->deviceInfo.panelResolution.height,
+	     info->hmuMount);
+	return OK;
+}
+
+static int32_t KYTY_SYSV_ABI Hmd2GetFieldOfViewWithoutHandle(SceHmd2FieldOfView* fov) {
+	PRINT_NAME();
+	if (!g_initialized.load()) {
+		return ERROR_NOT_INITIALIZED;
+	}
+	if (fov == nullptr) {
+		return ERROR_PARAMETER_NULL;
+	}
+	// Virtual ER15 field of view: 55/42.5/51.2/51.2 degrees.
+	// These tangents describe that geometry, not bit-exact firmware output.
+	*fov = {1.4281480312f, 0.9163311720f, 1.2437491417f, 1.2437491417f};
+	LOGF("Hmd2: FOV tangents out=%f in=%f top=%f bottom=%f\n", fov->tanOut, fov->tanIn,
+	     fov->tanTop, fov->tanBottom);
+	return OK;
+}
+
+static Graphics::SizeAlign KYTY_SYSV_ABI Hmd2ReprojectionQueryBufferSizeAlign() {
+	const Graphics::SizeAlign size {sizeof(ReprojectionState), alignof(ReprojectionState)};
+	LOGF("Hmd2: reprojection work size=%" PRIu64 " alignment=%zu\n", size.m_size, size.m_align);
+	return size;
+}
+
+static Graphics::SizeAlign KYTY_SYSV_ABI Hmd2ReprojectionQueryDisplayBufferSizeAlign() {
+	const auto size = ReprojectionDisplaySize();
+	LOGF("Hmd2: reprojection display size=%" PRIu64 " alignment=%zu\n", size.m_size,
+	     size.m_align);
+	return size;
+}
+
+static int32_t KYTY_SYSV_ABI
+Hmd2ReprojectionInitialize(const SceHmd2ReprojectionInitializeParam* param, void*) {
+	std::scoped_lock lock {g_reprojection_mutex};
+	if (GetReprojectionState() != nullptr) {
+		return ERROR_REPROJECTION_ALREADY_INITIALIZED;
+	}
+	if (param == nullptr || param->pReprojectionBuff == nullptr || param->pDisplayBuff == nullptr) {
+		return ERROR_PARAMETER_NULL;
+	}
+	EXIT_NOT_IMPLEMENTED(param->pSeeThroughBuff != nullptr);
+	const auto display_size = ReprojectionDisplaySize();
+	auto* state = new (param->pReprojectionBuff) ReprojectionState;
+	if (param->reprojectionTiming != 0) {
+		state->timing_us.store(static_cast<uint32_t>(param->reprojectionTiming),
+		                      std::memory_order_relaxed);
+	}
+	g_reprojection_state.store(state, std::memory_order_release);
+	LOGF("Hmd2: reprojection initialized work=%p display=%p size=%" PRIu64 " timing=%u\n",
+	     static_cast<void*>(state), param->pDisplayBuff, display_size.m_size,
+	     state->timing_us.load(std::memory_order_relaxed));
+	return OK;
+}
+
+static int32_t KYTY_SYSV_ABI Hmd2ReprojectionEnableVrMode(uint64_t output_mode) {
+	auto* state = GetReprojectionState();
+	if (state == nullptr) {
+		return ERROR_REPROJECTION_NOT_INITIALIZED;
+	}
+	if (output_mode != OUTPUT_MODE_89_91HZ && output_mode != OUTPUT_MODE_119_88HZ) {
+		return VideoOut::VIDEO_OUT_ERROR_INVALID_VALUE;
+	}
+	if (!Config::VrEnabled()) {
+		return VideoOut::VIDEO_OUT_ERROR_NO_DEVICE;
+	}
+	uint64_t expected = 0;
+	if (!state->output_mode.compare_exchange_strong(expected, output_mode)) {
+		return ERROR_REPROJECTION_IN_VR_MODE;
+	}
+	LOGF("Hmd2: VR output mode=0x%" PRIx64 "\n", output_mode);
+	return OK;
+}
+
+static int32_t KYTY_SYSV_ABI Hmd2ReprojectionSetTiming(int32_t timing) {
+	auto* state = GetReprojectionState();
+	if (state == nullptr) {
+		return ERROR_REPROJECTION_NOT_INITIALIZED;
+	}
+	EXIT_NOT_IMPLEMENTED(timing < 2000 || timing > 7000);
+	const auto rounded = static_cast<uint32_t>((timing + 99) / 100 * 100);
+	state->timing_us.store(rounded, std::memory_order_release);
+	LOGF("Hmd2: reprojection timing=%u us\n", rounded);
+	return OK;
+}
+
+} // namespace Hmd2
+
+LIB_DEFINE(InitHmd2_1) {
+	LIB_FUNC("c812oYs7Vsc", Hmd2::Hmd2Initialize);
+	LIB_FUNC("bIi4YUfSRys", Hmd2::Hmd2GetDeviceInformation);
+	LIB_FUNC("gF8+lvc7GuQ", Hmd2::Hmd2GetFieldOfViewWithoutHandle);
+	LIB_FUNC("U-CnbmeyYaA", Hmd2::Hmd2ReprojectionQueryBufferSizeAlign);
+	LIB_FUNC("-C2nkoEYOnU", Hmd2::Hmd2ReprojectionQueryDisplayBufferSizeAlign);
+	LIB_FUNC("C0rPwER-yxg", Hmd2::Hmd2ReprojectionInitialize);
+	LIB_FUNC("VVvFh51o20s", Hmd2::Hmd2ReprojectionEnableVrMode);
+	LIB_FUNC("FkQX7rjFomk", Hmd2::Hmd2ReprojectionSetTiming);
+}
+
+} // namespace Libs

src/libs/libJson2.cpp

@@ -1006,6 +1006,21 @@ static int32_t KYTY_SYSV_ABI JsonValueCount(const JsonValue* self) {
 	}
 }
 
+static void KYTY_SYSV_ABI JsonValueClearMethod(void* self) {
+	PRINT_NAME();
+
+	LOGF("\t self = 0x%016" PRIx64 "\n", reinterpret_cast<uint64_t>(self));
+
+	auto* value = reinterpret_cast<JsonValue*>(self);
+	if (value != nullptr) {
+		auto* parent    = value->parent;
+		auto* rootparam = value->rootparam;
+		JsonValueClear(value);
+		value->parent    = parent;
+		value->rootparam = rootparam;
+	}
+}
+
 LIB_DEFINE(InitNet_1_Json2) {
 	LIB_FUNC("-hJRce8wn1U", LibJson2::JsonMemAllocatorCtor);
 	LIB_FUNC("WSOuge5IsCg", LibJson2::JsonInitParameter2Ctor);
@@ -1076,6 +1091,7 @@ LIB_DEFINE(InitNet_1_Json2) {
 	LIB_FUNC("RBw+4NukeGQ", LibJson2::JsonValueCount);
 	LIB_FUNC("+drDFyAS6u4", LibJson2::JsonInitializerSetGlobalNullAccessCallback);
 	LIB_FUNC("00oCq0RwSAY", LibJson2::JsonInitializerSetGlobalNullAccessCallback);
+	LIB_FUNC("FIjXN2TkuTs", LibJson2::JsonValueClearMethod);
 }
 
 } // namespace LibJson2

src/libs/libKernel.cpp

@@ -1168,7 +1168,6 @@ static int KYTY_SYSV_ABI KernelRaiseException(Pthread thread, int signum) {
 			return KERNEL_ERROR_EINVAL;
 		}
 
-		Common::CondVar::SignalThread(PthreadGetUniqueId(thread));
 		PthreadWakeForSignal(thread);
 		CloseHandle(target_thread);
 		return OK;
@@ -1193,7 +1192,6 @@ static int KYTY_SYSV_ABI KernelRaiseException(Pthread thread, int signum) {
 			return KERNEL_ERROR_EINVAL;
 		}
 
-		Common::CondVar::SignalThread(PthreadGetUniqueId(thread));
 		PthreadWakeForSignal(thread);
 		WaitForSignalDispatch(thread, signum);
 		return OK;
@@ -2128,7 +2126,13 @@ int KYTY_SYSV_ABI UmtxOp(volatile void* address, int operation, uint64_t value,
 					*GetErrorAddr() = POSIX_EINVAL;
 					return -1;
 				}
-				EXIT("Valid _umtx_op timed waits are not implemented\n");
+				const auto max_ns = std::chrono::nanoseconds::max().count();
+				const auto timeout_ns = duration.tv_sec > (max_ns - duration.tv_nsec) / 1000000000
+				                            ? max_ns
+				                            : duration.tv_sec * 1000000000 + duration.tv_nsec;
+				return POSIX_CALL(LibKernel::SyncOnAddress::Wait64(
+				    static_cast<volatile uint64_t*>(address), value, std::chrono::nanoseconds(timeout_ns),
+				    LibKernel::KernelDispatchPendingSignalForCurrentThread));
 			}
 			return POSIX_CALL(LibKernel::SyncOnAddress::Wait64(
 			    static_cast<volatile uint64_t*>(address), value, nullptr,

src/libs/libs.cpp

@@ -82,6 +82,7 @@ LIB_DEFINE(InitDialog_1);
 LIB_DEFINE(InitFont_1);
 LIB_DEFINE(InitFontFt_1);
 LIB_DEFINE(InitAgcDriver_1);
+LIB_DEFINE(InitHmd2_1);
 LIB_DEFINE(InitLibKernel_1);
 LIB_DEFINE(InitNet_1);
 LIB_DEFINE(InitPad_1);
@@ -115,6 +116,7 @@ void InitAll(Loader::SymbolDatabase* s) {
 	InitFont_1(s);
 	InitFontFt_1(s);
 	InitAgcDriver_1(s);
+	InitHmd2_1(s);
 	InitLibKernel_1(s);
 	LibMouse::InitMouse_1(s);
 	LibKeyboard::InitKeyboard_1(s);

src/libs/network.cpp

@@ -14,6 +14,7 @@
 #endif
 #else
 #include <arpa/inet.h>
+#include <netdb.h>
 #include <netinet/tcp.h>
 #include <sys/socket.h>
 #include <sys/select.h>
@@ -1296,7 +1297,6 @@ int KYTY_SYSV_ABI NetResolverStartNtoa(int rid, const char* hostname, void* addr
 		return OK;
 	}
 
-#if defined(_WIN32)
 	if (!EnsureSocketBackend()) {
 		return NET_ERROR_ENETDOWN;
 	}
@@ -1311,7 +1311,13 @@ int KYTY_SYSV_ABI NetResolverStartNtoa(int rid, const char* hostname, void* addr
 		if (result != nullptr) {
 			freeaddrinfo(result);
 		}
-		return ret == EAI_NONAME ? NET_ERROR_RESOLVER_ENOHOST : NET_ERROR_RESOLVER_EINTERNAL;
+		if (ret == EAI_NONAME) {
+			return NET_ERROR_RESOLVER_ENOHOST;
+		}
+		if (ret == EAI_NODATA) {
+			return NET_ERROR_RESOLVER_ENORECORD;
+		}
+		return NET_ERROR_RESOLVER_EINTERNAL;
 	}
 
 	for (auto* ai = result; ai != nullptr; ai = ai->ai_next) {
@@ -1326,9 +1332,6 @@ int KYTY_SYSV_ABI NetResolverStartNtoa(int rid, const char* hostname, void* addr
 
 	freeaddrinfo(result);
 	return NET_ERROR_RESOLVER_ENORECORD;
-#else
-	return NET_ERROR_RESOLVER_ENOTIMPLEMENTED;
-#endif
 }
 
 int KYTY_SYSV_ABI NetInetPton(int af, const char* src, void* dst) {
@@ -2025,22 +2028,27 @@ int64_t KYTY_SYSV_ABI Recvfrom(int s, void* buf, uint64_t len, int flags, void*
 
 	const auto host_len = static_cast<SocketIoLength>(
 	    std::min<uint64_t>(len, std::numeric_limits<SocketIoLength>::max()));
-	int64_t result = 0;
+	sockaddr_storage host_addr {};
+	SocketLength     host_addrlen = sizeof(host_addr);
+	int64_t          result       = 0;
 	if (addr == nullptr) {
 		result = ::recv(socket, static_cast<char*>(buf), host_len, host_flags);
 	} else {
-		sockaddr_storage host_addr {};
-		SocketLength     host_addrlen = sizeof(host_addr);
 		result = ::recvfrom(socket, static_cast<char*>(buf), host_len, host_flags,
 		                    reinterpret_cast<sockaddr*>(&host_addr), &host_addrlen);
-		if (result >= 0 &&
-		    ConvertHostSockaddr(&host_addr, host_addrlen, addr, addrlen) != 0) {
-			return -1;
-		}
 	}
+#if defined(_WIN32)
+	if (result < 0 && WSAGetLastError() == WSAEMSGSIZE) {
+		// Winsock copied the datagram prefix; POSIX reports its length as success.
+		result = host_len;
+	}
+#endif
 	if (result < 0) {
 		return SetHostSocketError();
 	}
+	if (addr != nullptr && ConvertHostSockaddr(&host_addr, host_addrlen, addr, addrlen) != 0) {
+		return -1;
+	}
 
 	return result;
 }

src/main.cpp

@@ -54,6 +54,7 @@ static void PrintUsage() {
 	::printf(
 	    "  --gpu <index>                        Vulkan physical device index. Default: auto.\n");
 	::printf("  --fullscreen                         Run in borderless desktop fullscreen.\n");
+	::printf("  --vr                                 Enable the virtual VR headset.\n");
 	::printf("  --vblank-frequency <num>             Virtual vblank frequency. Default: 60.\n");
 	::printf("  --console-language <0-29>            Console language. Default: 1 (English US).\n");
 	::printf("  --vulkan-validation <true|false>     Enable Vulkan validation.\n");
@@ -161,6 +162,11 @@ static bool ParseArgs(int argc, char* argv[], RunOptions& options, bool& show_he
 			continue;
 		}
 
+		if (arg == "--vr") {
+			options.config.vr_enabled = true;
+			continue;
+		}
+
 		if (arg == "--playgo-hack") {
 			options.config.playgo_hack_enabled = true;
 			continue;

tests/SyncOnAddressTests.cpp

@@ -110,6 +110,38 @@ void TestValueChangeAndWake() {
   Check(result == OK, "a value change plus wake releases the waiter");
 }
 
+void TestNanosecondTimeout() {
+  uint64_t word = 0;
+  const auto start = std::chrono::steady_clock::now();
+  Check(Wait64(&word, 0, std::chrono::milliseconds(1)) ==
+            Libs::LibKernel::KERNEL_ERROR_ETIMEDOUT,
+        "64-bit timespec wait expires");
+  Check(std::chrono::steady_clock::now() - start >= std::chrono::milliseconds(1),
+        "timespec wait preserves its deadline");
+  Check(Wait64(&word, 0, std::chrono::nanoseconds(0)) ==
+            Libs::LibKernel::KERNEL_ERROR_ETIMEDOUT,
+        "zero timespec polls a matching value");
+  Check(Wait64(&word, 1, std::chrono::nanoseconds(0)) == OK,
+        "zero timespec succeeds on mismatch");
+
+  g_signal_poll_count.store(0, std::memory_order_relaxed);
+  std::atomic<bool> returned{false};
+  std::thread waker([&] {
+    while (g_signal_poll_count.load(std::memory_order_relaxed) < 2 &&
+           !returned.load(std::memory_order_acquire)) {
+      std::this_thread::yield();
+    }
+    Store(&word, UINT64_C(0x100000000));
+    Check(Wake(&word, 1) == OK, "wake releases a long timespec wait");
+  });
+  const int result = Wait64(&word, 0, std::chrono::nanoseconds::max(), CountSignalPoll);
+  returned.store(true, std::memory_order_release);
+  waker.join();
+  Check(result == OK, "long timespec wait does not overflow its deadline");
+  Check(g_signal_poll_count.load(std::memory_order_relaxed) > 1,
+        "timespec wait continues polling guest signals");
+}
+
 void TestWakeOneThenAll() {
   constexpr int WAITER_COUNT = 4;
   uint32_t word = 0;
@@ -260,6 +292,7 @@ int main() {
   TestMismatchReturnsImmediately();
   TestTimeout();
   TestValueChangeAndWake();
+  TestNanosecondTimeout();
   TestWakeOneThenAll();
   TestAddressesAreIsolated();
   TestCompareRegisterWakeRace();

tests/VirtualMemoryAllocationTests.cpp

@@ -20,6 +20,11 @@
 #include <string>
 #include <vector>
 
+#if defined(__linux__)
+#include <sys/uio.h>
+#include <unistd.h>
+#endif
+
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #ifndef NOMINMAX
 #define NOMINMAX
@@ -1163,6 +1168,43 @@ void TestDirectPartialProtectUnmapPreservesNeighbors() {
 	std::printf("[host]    %-48s ok\n", test);
 }
 
+#if defined(__linux__)
+void TestPartialUnmapPreservesHostPermissions() {
+	const char* test = "PartialUnmapPreservesHostPermissions";
+	const auto base = MapNamedFlexible(test, SceKernelPageSize * 3, SceKernelProtCpuRw,
+	                                   "unmap_host_permissions");
+	using Common::VirtualMemory::Mode;
+	Check(test, Libs::LibKernel::Memory::ProtectGuestHostMemory(base, SceKernelPageSize, Mode::Read),
+	      "could not protect the left survivor from writes");
+	Check(test,
+	      Libs::LibKernel::Memory::ProtectGuestHostMemory(base + SceKernelPageSize * 2,
+	                                                      SceKernelPageSize, Mode::NoAccess),
+	      "could not protect the right survivor from reads");
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMunmap(base + SceKernelPageSize, SceKernelPageSize),
+	        "KernelMunmap(middle)");
+
+	// Probe actual host access without taking a signal or changing the page protections.
+	uint64_t value = 0;
+	iovec local {&value, sizeof(value)};
+	iovec left {reinterpret_cast<void*>(base), sizeof(value)};
+	iovec right {reinterpret_cast<void*>(base + SceKernelPageSize * 2), sizeof(value)};
+	Check(test, process_vm_readv(getpid(), &local, 1, &left, 1, 0) == sizeof(value),
+	      "partial unmap removed read access to the left survivor");
+	Check(test, process_vm_writev(getpid(), &local, 1, &left, 1, 0) == -1,
+	      "partial unmap removed the left survivor's write protection");
+	Check(test, process_vm_readv(getpid(), &local, 1, &right, 1, 0) == -1,
+	      "partial unmap removed the right survivor's read protection");
+
+	CheckOk(test, Libs::LibKernel::Memory::KernelMunmap(base, SceKernelPageSize),
+	        "KernelMunmap(left cleanup)");
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMunmap(base + SceKernelPageSize * 2, SceKernelPageSize),
+	        "KernelMunmap(right cleanup)");
+	std::printf("[host]    %-48s ok\n", test);
+}
+#endif
+
 void TestDirectMapValidationBeforeOwnerMutation() {
 	const char* test    = "DirectMapValidationBeforeOwnerMutation";
 	int64_t     invalid = -1;
@@ -2529,6 +2571,9 @@ int main(int argc, char** argv) {
 	RunTest(TestMunmapAcrossAdjacentFlexibleMappings);
 	RunTest(TestDirectMapQueryOffsetAndPartialMunmap);
 	RunTest(TestDirectPartialProtectUnmapPreservesNeighbors);
+#if defined(__linux__)
+	RunTest(TestPartialUnmapPreservesHostPermissions);
+#endif
 	RunTest(TestDirectMapValidationBeforeOwnerMutation);
 	RunTest(TestDirectReleaseRollbackRestoresOwnerMapping);
 	RunTest(TestDirectReleaseContracts);

tests/shaderCfgTests.cpp

@@ -1183,37 +1183,90 @@ void TestNativeShaderResourceDependencies() {
 }
 
 void TestNormalizedImageContracts() {
+  // Captured PS5 256x256 atlas: each layer is a 0x56000-byte mip chain.
   ImageInfo container{};
-  container.data = {0x10000, 0x15000};
+  container.data = {0x241741000, 0x2198000};
   container.pixel_format = vk::Format::eR8G8B8A8Unorm;
   container.guest_format = Prospero::BufferFormat::k8_8_8_8UNorm;
   container.type = Prospero::ImageType::kColor2D;
-  container.extent = {64, 64, 1};
-  container.resources = {3, 4};
-  container.pitch = 64;
+  container.extent = {256, 256, 1};
+  container.resources = {9, 100};
+  container.pitch = 256;
   container.bytes_per_block = 4;
   container.samples = 1;
-  container.tile_mode = Prospero::TileMode::kStandard64KB;
-  container.mip_layout[0] = {0, 0x10000, 64, 64};
-  container.mip_layout[1] = {0x10000, 0x4000, 32, 32};
-  container.mip_layout[2] = {0x14000, 0x1000, 16, 16};
+  container.tile_mode = Prospero::TileMode::kStandard4KB;
+  container.mip_layout[0] = {0x16000, 0x1900000, 256, 256};
+  container.mip_layout[1] = {0x6000, 0x640000, 128, 128};
+  container.mip_layout[2] = {0x2000, 0x190000, 64, 64};
+  container.mip_layout[3] = {0x1000, 0x64000, 32, 32};
+  for (uint32_t level = 4; level < container.resources.levels; ++level) {
+    container.mip_layout[level] = {0, 0x64000, 32, 32};
+  }
 
   ImageInfo subresource = container;
-  subresource.data = {0x22000, 0x1000};
-  subresource.extent = {32, 32, 1};
+  subresource.data = {0x2417ad000, 0x40000};
   subresource.resources = {1, 1};
-  subresource.pitch = 32;
   subresource.mip_layout = {};
-  subresource.mip_layout[0] = {0, 0x1000, 32, 32};
+  subresource.mip_layout[0] = {0, 0x40000, 256, 256};
 
-  Check(container.BlockExtent() == vk::Extent2D{64, 64},
+  Check(container.BlockExtent() == vk::Extent2D{256, 256},
         "normalized image block extent changed");
   Check(subresource.IsCompatible(container),
         "normalized compatible image was rejected");
-  Check(subresource.MipOf(container) == 1,
-        "normalized mip lookup missed a subresource");
-  Check(subresource.SliceOf(container, 1) == 2,
-        "normalized slice lookup missed a subresource");
+  Check(subresource.MipOf(container) == 0 &&
+            subresource.SliceOf(container, 0) == 1,
+        "PS5 mip-chain stride did not resolve the captured second array layer");
+
+  auto smaller = subresource;
+  smaller.data = {0x241747000, 0x10000};
+  smaller.extent = {128, 128, 1};
+  smaller.pitch = 128;
+  smaller.mip_layout[0] = {0, 0x10000, 128, 128};
+  Check(smaller.MipOf(container) == 1 && smaller.SliceOf(container, 1) == 0,
+        "PS5 reversed mip ordering did not resolve the captured smaller mip");
+
+  auto boundary = subresource;
+  boundary.data.address = container.data.address + 99 * 0x56000 + 0x16000;
+  Check(boundary.SliceOf(container, 0) == 99,
+        "PS5 array lookup rejected the final complete layer");
+  boundary.data.address += 0x56000;
+  Check(boundary.MipOf(container) == -1,
+        "PS5 array lookup accepted a layer beyond the parent range");
+  boundary = subresource;
+  boundary.data.address += 0x1000;
+  Check(boundary.MipOf(container) == -1,
+        "PS5 array lookup accepted a misaligned layer address");
+  boundary = subresource;
+  boundary.resources.layers = 2;
+  boundary.data.size *= 2;
+  boundary.mip_layout[0].size *= 2;
+  Check(boundary.MipOf(container) == -1,
+        "PS5 array lookup accepted contiguous child layers with a different stride");
+  boundary = subresource;
+  boundary.mip_layout[0].pitch *= 2;
+  Check(boundary.MipOf(container) == -1,
+        "PS5 array lookup accepted a different storage pitch");
+
+  auto small_atlas = container;
+  small_atlas.data = {0x239001000, 0x64000};
+  small_atlas.extent = {32, 32, 1};
+  small_atlas.pitch = 32;
+  small_atlas.resources = {6, 50};
+  small_atlas.mip_layout[0] = {0x1000, 0x32000, 32, 32};
+  for (uint32_t level = 1; level < small_atlas.resources.levels; ++level) {
+    small_atlas.mip_layout[level] = {0, 0x32000, 32, 32};
+  }
+  auto small_child = subresource;
+  small_child.data = {0x23900e000, 0x1000};
+  small_child.extent = {32, 32, 1};
+  small_child.pitch = 32;
+  small_child.mip_layout[0] = {0, 0x1000, 32, 32};
+  Check(small_child.MipOf(small_atlas) == 0 &&
+            small_child.SliceOf(small_atlas, 0) == 6,
+        "PS5 array lookup doubled the captured small-atlas layer index");
+  small_child.data.address = 0x23900d000;
+  Check(small_child.MipOf(small_atlas) == -1,
+        "PS5 array lookup treated a packed tail block as the full-size mip");
 
   auto incompatible = subresource;
   incompatible.samples = 2;
@@ -4170,6 +4223,24 @@ void TestNewShaderDecoderArchitecture() {
             !sin_f16.src0.absolute,
         "decoder rejected the VOP1 SDWA V_SIN_F16 instruction");
 
+  std::vector<uint32_t> fract_f16_code(0x9b0u / 4u, EncodeSopp(0x00));
+  fract_f16_code.push_back(0x7e12bf07u);
+  fract_f16_code.push_back(EncodeSopp(0x01));
+  Instruction fract_f16;
+  ShaderRecompiler::Decoder::DecodeInstruction(fract_f16_code, 0x9b0u / 4u,
+                                               fract_f16);
+  Check(fract_f16.pc == 0x9b0u && fract_f16.family == Family::VOP1 &&
+            fract_f16.opcode == Opcode::V_FRACT_F16 &&
+            fract_f16.opcode_id == 0x5fu && fract_f16.word_count == 1u &&
+            fract_f16.dst.kind == OperandKind::Vgpr && fract_f16.dst.reg == 9u &&
+            fract_f16.src_count == 1u &&
+            fract_f16.src0.kind == OperandKind::Vgpr && fract_f16.src0.reg == 7u,
+        "decoder rejected captured V_FRACT_F16 at pc 0x9b0");
+  ShaderRecompiler::Decoder::Program fract_program;
+  ShaderRecompiler::Decoder::DecodeProgram(fract_f16_code, fract_program);
+  Check(!ShaderRecompiler::CFG::BuildGraph(fract_program).unsupported,
+        "captured V_FRACT_F16 still fails CFG construction");
+
   const uint32_t literal_code[] = {EncodeVop1(0x01, 2, 255u), 0x12345678u};
   Instruction literal;
   ShaderRecompiler::Decoder::DecodeInstruction(literal_code, 0u, literal);
@@ -4416,6 +4487,62 @@ void TestNewShaderRecompilerCapturedVopcSdwaCmpxClass() {
   CheckSpirvBinaryValidates(result.spirv);
 }
 
+void TestNewShaderRecompilerCapturedVopcSdwaCmpxLtU16() {
+  using namespace ShaderRecompiler;
+
+  const uint32_t shader[] = {
+      0x7d72d4f9u, 0x86060000u, // v_cmpx_lt_u16 v0, vcc_lo (SDWA)
+      EncodeSopp(0x01),
+  };
+  Decoder::Instruction decoded;
+  Decoder::DecodeInstruction(shader, 0u, decoded);
+  Check(decoded.family == Decoder::Family::VOPC &&
+            decoded.opcode == Decoder::Opcode::V_CMPX_LT_U16 &&
+            decoded.opcode_id == 0xb9u && decoded.word_count == 2u &&
+            decoded.raw[0] == shader[0] && decoded.raw[1] == shader[1] &&
+            decoded.dst.kind == Decoder::OperandKind::ExecLo &&
+            decoded.src_count == 2u &&
+            decoded.src0.kind == Decoder::OperandKind::Vgpr &&
+            decoded.src0.reg == 0u && decoded.src0.sdwa_sel == 6u &&
+            decoded.src1.kind == Decoder::OperandKind::VccLo &&
+            decoded.src1.sdwa_sel == 6u &&
+            !decoded.src0.sdwa_sext && !decoded.src1.sdwa_sext &&
+            !decoded.src0.negate && !decoded.src1.negate &&
+            !decoded.src0.absolute && !decoded.src1.absolute,
+        "decoder rejected captured SDWA V_CMPX_LT_U16 fields");
+
+  Decoder::Program program;
+  Decoder::DecodeProgram(shader, program);
+  Check(!CFG::BuildGraph(program).unsupported,
+        "captured SDWA V_CMPX_LT_U16 still fails CFG construction");
+  auto result = RecompileForTest(shader, MakeCompileOptions(ShaderType::Pixel));
+  Check(Common::ContainsStr(result.decoded_dump,
+                            "V_CMPX_LT_U16 exec_lo, v0, vcc_lo"),
+        "captured SDWA V_CMPX_LT_U16 is missing from decoded dump");
+  CheckSpirvBinaryValidates(result.spirv);
+
+  // Independently assembled gfx1030 compact and VOP3 encodings.
+  const uint32_t compact[] = {0x7d720300u};
+  const uint32_t vop3[] = {0xd4b9007eu, 0x00020300u};
+  for (auto words : {std::span<const uint32_t>(compact),
+                     std::span<const uint32_t>(vop3)}) {
+    Decoder::DecodeInstruction(words, 0u, decoded);
+    Check(decoded.opcode == Decoder::Opcode::V_CMPX_LT_U16 &&
+              decoded.dst.kind == Decoder::OperandKind::ExecLo &&
+              decoded.src0.kind == Decoder::OperandKind::Vgpr &&
+              decoded.src0.reg == 0u &&
+              decoded.src1.kind == Decoder::OperandKind::Vgpr &&
+              decoded.src1.reg == 1u,
+          "compact/VOP3 V_CMPX_LT_U16 does not decode to an EXEC compare");
+  }
+  const uint32_t dpp[] = {EncodeVopc(0xb9u, 250u, 1u), EncodeVop2Dpp(0u)};
+  Decoder::DecodeInstruction(dpp, 0u, decoded);
+  Check(decoded.opcode == Decoder::Opcode::UNSUPPORTED &&
+            Common::ContainsStr(decoded.unsupported_reason,
+                                "VOPC DPP modifier is not supported for opcode"),
+        "V_CMPX_LT_U16 accepted an unsupported DPP encoding");
+}
+
 void TestNewShaderRecompilerIrLookupMissFailsExplicitly() {
   namespace Decoder = ShaderRecompiler::Decoder;
   namespace CFG = ShaderRecompiler::CFG;
@@ -10025,11 +10152,120 @@ void TestNewShaderRecompilerAuxPositionExports() {
         "PA_CL_VS_OUT_CNTL is absent from the vertex shader cache key");
 }
 
+void TestTypedSpirvSerialization() {
+  // Parameters exercise runtime enums, which cannot narrow implicitly to a
+  // uint32_t initializer list when the enum uses the Windows signed ABI.
+  const auto check_serialization = [](spv::StorageClass storage_class,
+                                      spv::FunctionControlMask control,
+                                      spv::Decoration decoration,
+                                      GLSLstd450 glsl_opcode,
+                                      int32_t signed_literal,
+                                      uint32_t high_bit_literal) {
+    ShaderRecompiler::Spirv::Builder builder;
+    builder.RequireCapability(spv::CapabilityShader);
+    const auto glsl = builder.Import("GLSL.std.450");
+    builder.AddMemoryModel(spv::AddressingModelLogical, spv::MemoryModelGLSL450);
+    const auto void_type = builder.Type(spv::OpTypeVoid);
+    const std::vector<uint32_t> uint_operands{32u, 0u};
+    const auto uint_type = builder.Type(spv::OpTypeInt, uint_operands);
+    Check(uint_type == builder.Type(spv::OpTypeInt, 32u, 0u) &&
+              uint_type == builder.Type(spv::OpTypeInt,
+                                        std::span<const uint32_t>{uint_operands}),
+          "typed, vector and span type operands produced different declarations");
+    const auto float_type = builder.Type(spv::OpTypeFloat, 32u);
+    const auto vector_type = builder.Type(spv::OpTypeVector, uint_type, 2u);
+    const auto pointer_type =
+        builder.Type(spv::OpTypePointer, storage_class, uint_type);
+    const auto function_type = builder.Type(spv::OpTypeFunction, void_type);
+    const auto all_bits =
+        builder.Constant(spv::OpConstant, uint_type, signed_literal);
+    Check(all_bits == builder.Constant(spv::OpConstant, uint_type, UINT32_MAX),
+          "signed and unsigned constant operands produced different declarations");
+    const auto high_bit =
+        builder.Constant(spv::OpConstant, uint_type, high_bit_literal);
+    const auto one = builder.Constant(spv::OpConstant, float_type, 0x3f800000u);
+    const std::vector<uint32_t> components{all_bits, high_bit};
+    const auto constant_vector = builder.Constant(
+        spv::OpConstantComposite, vector_type,
+        std::span<const uint32_t>{components});
+    Check(constant_vector == builder.Constant(spv::OpConstantComposite,
+                                              vector_type, components),
+          "vector and span constant operands produced different declarations");
+    const auto variable =
+        builder.DefineGlobalVariable(pointer_type, storage_class);
+    const auto function = builder.AllocateId();
+    const auto label = builder.AllocateId();
+    const auto absolute = builder.AllocateId();
+    const auto constructed = builder.AllocateId();
+    const auto shuffled = builder.AllocateId();
+    const auto loaded = builder.AllocateId();
+    builder.AddEntryPoint(spv::ExecutionModelGLCompute, function, "main", {});
+    builder.AddExecutionMode(function, spv::ExecutionModeLocalSize, 1u, 2u, 3u);
+    builder.AddAnnotation(spv::OpDecorate, one, decoration);
+    builder.AddFunction(spv::OpFunction, void_type, function, control,
+                        function_type);
+    const std::vector<uint32_t> label_words{248u, label};
+    builder.AddFunction(label_words);
+    builder.AddFunction(spv::OpExtInst, float_type, absolute, glsl, glsl_opcode,
+                        one);
+    builder.AddFunction(spv::OpCompositeConstruct, vector_type, constructed,
+                        components);
+    const std::array<uint32_t, 2> selectors{1u, 2u};
+    builder.AddFunction(spv::OpVectorShuffle, vector_type, shuffled,
+                        constant_vector, constructed,
+                        std::span<const uint32_t>{selectors});
+    builder.AddFunction(spv::OpLoad, uint_type, loaded, variable,
+                        spv::MemoryAccessAlignedMask, 4u);
+    builder.AddFunction(std::vector<uint32_t>{});
+    const std::array<uint32_t, 1> return_words{253u};
+    builder.AddFunction(std::span<const uint32_t>{return_words});
+    builder.AddFunction(spv::OpFunctionEnd);
+
+    // Literal headers and enum values keep this independent of the builder's
+    // instruction word counts, operand conversion and range flattening.
+    const std::vector<uint32_t> expected{
+        0x07230203u, 0x00010300u, 0u, loaded + 1u, 0u,
+        0x00020011u, 1u, // OpCapability Shader
+        0x0006000bu, glsl, 0x4c534c47u, 0x6474732eu, 0x3035342eu, 0u,
+        0x0003000eu, 0u, 1u, // OpMemoryModel Logical GLSL450
+        0x0005000fu, 5u, function, 0x6e69616du, 0u,
+        0x00060010u, function, 17u, 1u, 2u, 3u,
+        0x00030047u, one, 0u, // OpDecorate RelaxedPrecision
+        0x00020013u, void_type,
+        0x00040015u, uint_type, 32u, 0u,
+        0x00030016u, float_type, 32u,
+        0x00040017u, vector_type, uint_type, 2u,
+        0x00040020u, pointer_type, 6u, uint_type,
+        0x00030021u, function_type, void_type,
+        0x0004002bu, uint_type, all_bits, 0xffffffffu,
+        0x0004002bu, uint_type, high_bit, 0x80000000u,
+        0x0004002bu, float_type, one, 0x3f800000u,
+        0x0005002cu, vector_type, constant_vector, all_bits, high_bit,
+        0x0004003bu, pointer_type, variable, 6u,
+        0x00050036u, void_type, function, 0u, function_type,
+        0x000200f8u, label,
+        0x0006000cu, float_type, absolute, glsl, 4u, one,
+        0x00050050u, vector_type, constructed, all_bits, high_bit,
+        0x0007004fu, vector_type, shuffled, constant_vector, constructed, 1u, 2u,
+        0x0006003du, uint_type, loaded, variable, 2u, 4u,
+        0x000100fdu,
+        0x00010038u,
+    };
+    const auto binary = builder.Build();
+    Check(binary == expected,
+          "typed SPIR-V serialization changed instruction words or operand order");
+    CheckSpirvBinaryValidates(binary);
+  };
+  check_serialization(spv::StorageClassPrivate, spv::FunctionControlMaskNone,
+                      spv::DecorationRelaxedPrecision, GLSLstd450FAbs, -1,
+                      0x80000000u);
+}
+
 void TestDeferredSpirvPhiPatching() {
   ShaderRecompiler::Spirv::Builder builder;
-  const auto type = builder.Type(spv::OpTypeInt, {32u, 0u});
-  const auto zero = builder.Constant(spv::OpConstant, type, {0u});
-  const auto one = builder.Constant(spv::OpConstant, type, {1u});
+  const auto type = builder.Type(spv::OpTypeInt, 32u, 0u);
+  const auto zero = builder.Constant(spv::OpConstant, type, 0u);
+  const auto one = builder.Constant(spv::OpConstant, type, 1u);
   const auto left = builder.AllocateId();
   const auto right = builder.AllocateId();
   const auto result = builder.AllocateId();
@@ -10080,35 +10316,36 @@ void TestDemandDrivenSpirvDeclarations() {
   using ShaderRecompiler::Spirv::Builder;
 
   Builder declarations;
-  const auto uint_type = declarations.Type(spv::OpTypeInt, {32u, 0u});
-  Check(uint_type == declarations.Type(spv::OpTypeInt, {32u, 0u}),
+  const auto uint_type = declarations.Type(spv::OpTypeInt, 32u, 0u);
+  Check(uint_type == declarations.Type(spv::OpTypeInt, 32u, 0u),
         "structural type interning returned different IDs");
-  const auto zero = declarations.Constant(spv::OpConstant, uint_type, {0u});
-  Check(zero == declarations.Constant(spv::OpConstant, uint_type, {0u}),
+  const auto zero = declarations.Constant(spv::OpConstant, uint_type, 0u);
+  Check(zero == declarations.Constant(spv::OpConstant, uint_type, 0u),
         "constant interning returned different IDs");
   Check(declarations.Import("GLSL.std.450") ==
             declarations.Import("GLSL.std.450"),
         "extended-instruction import was duplicated");
   declarations.RequireCapability(spv::CapabilityShader);
   declarations.RequireCapability(spv::CapabilityShader);
-  const auto plain_array =
-      declarations.Type(spv::OpTypeArray, {uint_type, zero});
+  const auto plain_array = declarations.Type(spv::OpTypeArray, uint_type, zero);
   const auto decorated_array = declarations.DecoratedType(
-      spv::OpTypeArray, {uint_type, zero},
-      {{spv::OpDecorate, {spv::DecorationArrayStride, sizeof(uint32_t)}}});
+      spv::OpTypeArray,
+      {{spv::OpDecorate, {spv::DecorationArrayStride, sizeof(uint32_t)}}},
+      uint_type, zero);
   Check(decorated_array ==
             declarations.DecoratedType(
-                spv::OpTypeArray, {uint_type, zero},
+                spv::OpTypeArray,
                 {{spv::OpDecorate,
-                  {spv::DecorationArrayStride, sizeof(uint32_t)}}}),
+                  {spv::DecorationArrayStride, sizeof(uint32_t)}}},
+                uint_type, zero),
         "decorated type interning returned different IDs");
   Check(plain_array != decorated_array,
         "decorated and undecorated aggregate types were aliased");
-  Check(declarations.DecoratedType(spv::OpTypeArray, {uint_type, zero}, {}) ==
+  Check(declarations.DecoratedType(spv::OpTypeArray, {}, uint_type, zero) ==
             plain_array,
         "empty decorated type request bypassed structural interning");
   const auto ptr = declarations.Type(spv::OpTypePointer,
-                                     {spv::StorageClassFunction, uint_type});
+                                     spv::StorageClassFunction, uint_type);
   Check(declarations.DefineGlobalVariable(ptr, spv::StorageClassFunction) !=
             declarations.DefineGlobalVariable(ptr, spv::StorageClassFunction),
         "distinct variables were structurally interned");
@@ -13020,6 +13257,7 @@ int main() {
   TestNativeShaderResourceDependencies();
   TestNormalizedImageContracts();
   TestSpirvRequirementsAnalysis();
+  TestTypedSpirvSerialization();
   TestDeferredSpirvPhiPatching();
   TestCompilerStageInputOwnership();
   TestSpirvEmissionOwnsRequirements();
@@ -13045,6 +13283,7 @@ int main() {
   TestImageAddressOperands();
   TestSopkCompareImmediateExtension();
   TestNewShaderRecompilerCapturedVopcSdwaCmpxClass();
+  TestNewShaderRecompilerCapturedVopcSdwaCmpxLtU16();
   TestNewShaderRecompilerIrLookupMissFailsExplicitly();
   TestNewShaderRecompilerRejectsDppOn64BitCompares();
   TestPsInputCountRegisterDecode();