shader: remove recompiler error-string propagation

* cli: add --gpu <index> to select a Vulkan physical device

Select the Vulkan physical device by enumeration index, defaulting to
automatic selection (-1). Invalid or out-of-range indices fall back to
automatic selection with a clear message, and the selected device is
logged.

Co-Authored-By: Claude <noreply@anthropic.com>

* cli: reject invalid --gpu index, translate comments, apply clang-format

An invalid --gpu index now prints an error and exits with code 1 (usage),
matching --user-id / --present-mode / --console-language. The automatic
fallback is kept only for out-of-range indices detected at runtime.
Comments translated to English and the touched files brought in line
with clang-format (pre-commit v22.1.3).

Co-Authored-By: Claude <noreply@anthropic.com>

* cli: simplify Vulkan device selection

---------

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>

src/common/emulatorConfig.cpp

@@ -47,6 +47,10 @@ PresentMode GetPresentMode() {
 	return g_config->present_mode;
 }
 
+int32_t GetGpuIndex() {
+	return g_config->gpu_index;
+}
+
 bool FullscreenEnabled() {
 	return g_config->fullscreen_enabled;
 }

src/common/emulatorConfig.h

@@ -48,6 +48,7 @@ struct ConfigOptions {
 	std::string            user_name                   = "Kyty";
 	int32_t                user_id                     = DEFAULT_USER_ID;
 	PresentMode            present_mode                = PresentMode::Fifo;
+	int32_t                gpu_index                   = -1;
 	bool                   fullscreen_enabled          = false;
 	uint32_t               vblank_frequency            = 60;
 	uint32_t               console_language            = DEFAULT_CONSOLE_LANGUAGE;
@@ -80,6 +81,7 @@ uint32_t GetScreenHeight();
 const std::string& GetUserName();
 int32_t  GetUserId();
 PresentMode GetPresentMode();
+int32_t GetGpuIndex();
 bool     FullscreenEnabled();
 uint32_t GetVblankFrequency();
 uint32_t GetConsoleLanguage();

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -2433,12 +2433,12 @@ KYTY_CP_OP_PARSER(CpOpIndirectShRegs) {
 		auto value          = indirect_buffer[1];
 
 		// Not sure if this is correct
-		if (raw_cmd_offset != cmd_offset) {
-			LOGF_COLOR(Log::Color::Red,
-			           "\t temporary: normalized indirect SH register offset 0x%08" PRIx32
-			           " -> 0x%08" PRIx32 "\n",
-			           raw_cmd_offset, cmd_offset);
-		}
+		// if (raw_cmd_offset != cmd_offset) {
+		// 	LOGF_COLOR(Log::Color::Red,
+		// 	           "\t temporary: normalized indirect SH register offset 0x%08" PRIx32
+		// 	           " -> 0x%08" PRIx32 "\n",
+		// 	           raw_cmd_offset, cmd_offset);
+		// }
 
 		// Not sure if this is correct
 		if (cmd_offset == Pm4::SH_NOP || raw_cmd_offset == 0xffffffffu) {
@@ -2494,12 +2494,12 @@ KYTY_CP_OP_PARSER(CpOpIndirectUcRegs) {
 		auto value          = indirect_buffer[1];
 
 		// Not sure if this is correct
-		if (raw_cmd_offset != cmd_offset) {
-			LOGF_COLOR(Log::Color::Red,
-			           "\t temporary: normalized indirect UC register offset 0x%08" PRIx32
-			           " -> 0x%08" PRIx32 "\n",
-			           raw_cmd_offset, cmd_offset);
-		}
+		// if (raw_cmd_offset != cmd_offset) {
+		// 	LOGF_COLOR(Log::Color::Red,
+		// 	           "\t temporary: normalized indirect UC register offset 0x%08" PRIx32
+		// 	           " -> 0x%08" PRIx32 "\n",
+		// 	           raw_cmd_offset, cmd_offset);
+		// }
 		if (HwUcTrySetFakeRegister(cmd_offset, value)) {
 			continue;
 		}
@@ -2855,12 +2855,12 @@ KYTY_CP_OP_PARSER(CpOpSetUconfigReg) {
 		cmd_offset = raw_cmd_offset & 0x0fffffffu;
 	}
 
-	if (raw_cmd_offset != cmd_offset) {
-		LOGF_COLOR(Log::Color::Red,
-		           "\t temporary: normalized UC register offset 0x%08" PRIx32 " -> 0x%08" PRIx32
-		           "\n",
-		           raw_cmd_offset, cmd_offset);
-	}
+	// if (raw_cmd_offset != cmd_offset) {
+	// 	LOGF_COLOR(Log::Color::Red,
+	// 	           "\t temporary: normalized UC register offset 0x%08" PRIx32 " -> 0x%08" PRIx32
+	// 	           "\n",
+	// 	           raw_cmd_offset, cmd_offset);
+	// }
 	if (HwUcTrySetFakeRegisterRange(cmd_offset, buffer + 1, KYTY_PM4_LEN(cmd_id) - 2u)) {
 		return KYTY_PM4_LEN(cmd_id) - 1u;
 	}

src/graphics/host_gpu/renderer/debug.cpp

@@ -827,7 +827,12 @@ static void EqaaPrint(const char* func, const HW::EqaaControl& c) {
 	     c.static_anchor_associations ? "true" : "false");
 }
 
-static void EqaaCheck(const HW::EqaaControl& c) {
+static void EqaaCheck(const HW::EqaaControl& c, const HW::AaConfig& cf) {
+	// The EQAA controls only take effect while multisampling is on, so a single-sample
+	// target leaves them inert and has nothing to warn about.
+	if (cf.msaa_num_samples == 0) {
+		return;
+	}
 	if (c.max_anchor_samples != 0 || c.ps_iter_samples != 0 || c.mask_export_num_samples != 0 ||
 	    c.alpha_to_mask_num_samples != 0 || c.high_quality_intersections ||
 	    c.incoherent_eqaa_reads || c.interpolate_comp_z || c.static_anchor_associations) {
@@ -1192,7 +1197,7 @@ void hw_check(const CommandBuffer& buffer) {
 	log_phase("blend");
 	BcCheck(bc, bclr, cc);
 	log_phase("eqaa");
-	EqaaCheck(eqaa);
+	EqaaCheck(eqaa, ac);
 	log_phase("aa");
 	AaCheck(aa, ac);
 	log_phase("done");

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

@@ -895,9 +895,9 @@ PreparedBindings RenderExecutor::PrepareBindings(const ShaderStageRuntime& runti
 	EXIT_IF(!runtime);
 	const auto& program  = *runtime.program;
 	const auto& snapshot = *runtime.resources;
-	std::string error;
-	if (!ShaderRecompiler::IR::ValidateResourceSpecialization(program, snapshot, &error)) {
-		EXIT("invalid native shader runtime snapshot: %s\n", error.c_str());
+	if (!ShaderRecompiler::IR::ValidateResourceSpecialization(program, snapshot)) {
+		EXIT("invalid native shader runtime snapshot: hash=0x%016" PRIx64 " stage=%u\n",
+		     program.shader_hash, static_cast<unsigned>(program.stage));
 	}
 
 	PreparedBindings prepared;

src/graphics/presentation/window/vulkanWindow.cpp

@@ -157,6 +157,10 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 	    });
 	EXIT_NOT_IMPLEMENTED(devices.empty());
 
+	if (Config::GetGpuIndex() >= 0) {
+		devices = {devices[Config::GetGpuIndex()]};
+	}
+
 	vk::PhysicalDevice  best_device       = nullptr;
 	uint32_t            best_queue_family = static_cast<uint32_t>(-1);
 	SurfaceCapabilities best_capabilities;

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -472,20 +472,14 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 
 } // namespace
 
-bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
-                  CompileResult& result, std::string* error) {
+CompileResult Recompile(std::span<const uint32_t> code, const CompileOptions& options) {
 	if (code.empty()) {
-		if (error != nullptr) {
-			*error = "invalid shader recompiler input";
-		}
-		return false;
+		EXIT("shader recompiler input is empty\n");
 	}
 	if (options.stage != ShaderType::Compute && options.stage != ShaderType::Vertex &&
 	    options.stage != ShaderType::Pixel) {
-		if (error != nullptr) {
-			*error = "shader recompiler supports compute, vertex, and pixel stages";
-		}
-		return false;
+		EXIT("shader recompiler received unsupported stage %u\n",
+		     static_cast<unsigned>(options.stage));
 	}
 
 	const auto compile_begin = std::chrono::steady_clock::now();
@@ -500,9 +494,7 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 	     static_cast<uint64_t>(code.size()));
 
 	Decoder::Program decoded;
-	if (!Decoder::DecodeProgram(code, decoded, error)) {
-		return false;
-	}
+	Decoder::DecodeProgram(code, decoded);
 	LOGF("%s phase end: stage=%s hash=0x%016" PRIx64 " decode instructions=%" PRIu64
 	     " elapsed_ms=%" PRIu64 "\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
@@ -516,12 +508,9 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 		}
 	}
 
-	CFG::Graph cfg;
 	LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " CFG BuildGraph\n", GetDumpLabel(options),
 	     StageName(options.stage), options.shader_hash);
-	if (!CFG::BuildGraph(decoded, cfg, error)) {
-		return false;
-	}
+	auto cfg = CFG::BuildGraph(decoded);
 	LOGF("%s phase end: stage=%s hash=0x%016" PRIx64 " CFG BuildGraph blocks=%" PRIu64
 	     " loops=%" PRIu64 " back_edges=%" PRIu64 " elapsed_ms=%" PRIu64 "\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
@@ -534,13 +523,12 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 		dispatcher_reason   = cfg.unsupported_reason;
 		LogDispatcherFallback(options, cfg, "build", dispatcher_reason);
 	} else {
-		std::string structure_error;
-		const auto  unstructured_cfg = cfg;
+		const auto unstructured_cfg = cfg;
 		LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " CFG Structurize\n",
 		     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
-		if (!CFG::Structurize(cfg, &structure_error)) {
+		if (!CFG::Structurize(cfg)) {
 			dispatcher_fallback      = true;
-			dispatcher_reason        = structure_error;
+			dispatcher_reason        = cfg.unsupported_reason;
 			const auto failure_kind  = cfg.failure_kind;
 			const auto failure_block = cfg.failure_block;
 			LogDispatcherFallback(options, cfg, "structurize", dispatcher_reason);
@@ -548,7 +536,7 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 			cfg.unsupported        = true;
 			cfg.failure_kind       = failure_kind;
 			cfg.failure_block      = failure_block;
-			cfg.unsupported_reason = structure_error;
+			cfg.unsupported_reason = dispatcher_reason;
 		} else {
 			LOGF("%s structured CFG success: blocks=%" PRIu64 "\n", GetDumpLabel(options),
 			     static_cast<uint64_t>(cfg.blocks.size()));
@@ -560,14 +548,28 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 		     static_cast<uint64_t>(cfg.natural_loops.size()), phase_ms());
 	}
 
-	IR::Program ir;
 	const ShaderVertexInputInfo*  vertex  = nullptr;
 	const ShaderPixelInputInfo*   pixel   = nullptr;
 	const ShaderComputeInputInfo* compute = nullptr;
 	switch (options.stage) {
-		case ShaderType::Vertex: vertex = options.input_info.vertex; break;
-		case ShaderType::Pixel: pixel = options.input_info.pixel; break;
-		case ShaderType::Compute: compute = options.input_info.compute; break;
+		case ShaderType::Vertex:
+			vertex = options.input_info.vertex;
+			if (vertex == nullptr) {
+				EXIT("vertex shader recompilation has no input metadata\n");
+			}
+			break;
+		case ShaderType::Pixel:
+			pixel = options.input_info.pixel;
+			if (pixel == nullptr) {
+				EXIT("pixel shader recompilation has no input metadata\n");
+			}
+			break;
+		case ShaderType::Compute:
+			compute = options.input_info.compute;
+			if (compute == nullptr) {
+				EXIT("compute shader recompilation has no input metadata\n");
+			}
+			break;
 		default: break;
 	}
 	EmbeddedFetchData embedded_fetch;
@@ -597,9 +599,7 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 	};
 	LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " IR TranslateProgram\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
-	if (!Frontend::TranslateProgram(decoded, cfg, translate_options, ir, error)) {
-		return false;
-	}
+	auto ir = Frontend::TranslateProgram(decoded, cfg, translate_options);
 	LOGF("%s phase end: stage=%s hash=0x%016" PRIx64 " IR TranslateProgram blocks=%" PRIu64
 	     " elapsed_ms=%" PRIu64 "\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
@@ -626,18 +626,9 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 			     lds_barriers.inserted_barriers);
 		}
 	}
-	std::string srt_error;
-	if (!IR::BuildSrtPlan(ir, &srt_error)) {
-		LOGF("%s SRT planning failed: %s\n", GetDumpLabel(options), srt_error.c_str());
-		if (error != nullptr) {
-			*error = std::move(srt_error);
-		}
-		return false;
-	}
+	IR::BuildSrtPlan(ir);
 	IR::EliminateDeadCode(ir.blocks);
-	if (!IR::TrackResources(ir, error)) {
-		return false;
-	}
+	IR::TrackResources(ir);
 	IR::EliminateDeadCode(ir.blocks);
 	if (options.stage == ShaderType::Vertex) {
 		ir.info.vertex_offset_sgpr = embedded_fetch.vertex_offset_sgpr;
@@ -665,27 +656,20 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 			runtime.read_memory = ReadZeroMemory;
 		}
 		runtime.userdata         = options.read_memory_data;
-		if (!IR::MaterializeResources(ir, runtime, resources, error)) {
-			return false;
+		if (!IR::MaterializeResources(ir, runtime, resources)) {
+			EXIT("shader resource materialization failed: stage=%s hash=0x%016" PRIx64 "\n",
+			     StageName(options.stage), options.shader_hash);
 		}
 	}
-	if (!IR::SpecializeResources(ir, resources, error)) {
-		return false;
-	}
+	IR::SpecializeResources(ir, resources);
 
 	IR::ShaderInfoOptions info_options;
 	info_options.vertex  = vertex;
 	info_options.pixel   = pixel;
 	info_options.compute = compute;
-	if (!IR::CollectShaderInfo(ir, info_options, error)) {
-		return false;
-	}
-	if (!IR::AllocateBindings(ir, options.push_constant_offset, error)) {
-		return false;
-	}
-	if (!Spirv::AnalyzeProgramRequirements(ir, error)) {
-		return false;
-	}
+	IR::CollectShaderInfo(ir, info_options);
+	IR::AllocateBindings(ir, options.push_constant_offset);
+	Spirv::AnalyzeProgramRequirements(ir);
 	std::string ir_dump;
 	if (options.dump_ir) {
 		ir_dump = MakeIrDump(cfg, ir);
@@ -694,36 +678,22 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 		}
 	}
 
-	std::vector<uint32_t> spirv;
-	std::string           emit_error;
 	LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " SPIR-V EmitProgram\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
-	if (!Spirv::EmitProgram(ir, resources, options.input_info, spirv, &emit_error)) {
-		LOGF("%s typed SPIR-V emit failed: %s\n", GetDumpLabel(options), emit_error.c_str());
-		if (dispatcher_fallback && error != nullptr) {
-			*error = fmt::format("dispatcher fallback failed after {}: {}",
-			                     dispatcher_reason.c_str(), emit_error.c_str());
-			LOGF("%s dispatcher fallback emit failed: %s\n", GetDumpLabel(options), error->c_str());
-		} else if (error != nullptr) {
-			*error = emit_error;
-		}
-		return false;
-	}
+	auto spirv = Spirv::EmitProgram(ir, resources, options.input_info);
 	LOGF("%s phase end: stage=%s hash=0x%016" PRIx64 " SPIR-V EmitProgram words=%" PRIu64
 	     " elapsed_ms=%" PRIu64 "\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
 	     static_cast<uint64_t>(spirv.size()), phase_ms());
+	CompileResult result;
 	result.spirv     = std::move(spirv);
 	result.program   = std::move(ir);
 	result.resources = std::move(resources);
 	if (options.dump_ir) {
 		result.decoded_dump = std::move(decoded_dump);
 		result.ir_dump      = std::move(ir_dump);
-	} else {
-		result.decoded_dump.clear();
-		result.ir_dump.clear();
 	}
-	return true;
+	return result;
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler

src/graphics/shader/recompiler/ShaderRecompiler.h

@@ -39,8 +39,8 @@ struct CompileResult {
 	IR::ResourceSnapshot  resources;
 };
 
-bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
-                  CompileResult& result, std::string* error);
+[[nodiscard]] CompileResult Recompile(std::span<const uint32_t> code,
+                                      const CompileOptions& options);
 
 } // namespace Libs::Graphics::ShaderRecompiler
 

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

@@ -1,19 +1,23 @@
 #include "graphics/shader/recompiler/backend/spirv/SpirvEmitter.h"
 
+#include "common/assert.h"
 #include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
-#include "graphics/shader/shader.h"
 
 #include <algorithm>
 #include <array>
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv {
 
-using ShaderError::Fail;
-
 namespace {
 
-bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
+[[noreturn]] void Fail(const IR::Program& program, const char* reason) {
+	EXIT("SPIR-V validation failed: hash=0x%016" PRIx64 " stage=%u reason=%s\n",
+	     program.shader_hash, static_cast<unsigned>(program.stage), reason);
+	std::abort();
+}
+
+void ValidateNativeProgram(const IR::Program& program) {
 	using Kind                                             = IR::DescriptorBindingKind;
 	constexpr auto                               KindCount = static_cast<size_t>(Kind::Count);
 	std::array<std::vector<uint32_t>, KindCount> expected;
@@ -36,13 +40,13 @@ bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
 	for (uint32_t i = 0; i < program.info.images.size(); i++) {
 		const auto kind = IR::DescriptorBindingForImage(program.info.images[i]);
 		if (!kind.has_value()) {
-			return Fail(error, "native shader plan has an invalid image class");
+			Fail(program, "native shader plan has an invalid image class");
 		}
 		present[static_cast<size_t>(*kind)] = true;
 		const auto dynamic = program.info.images[i].mip_mode == IR::ImageMipMode::DynamicStorage;
 		const auto count   = dynamic ? program.info.images[i].mip_count : 1u;
 		if (count == 0u || (!dynamic && program.info.images[i].mip_count != 1u)) {
-			return Fail(error, "native shader plan has an invalid image mip descriptor count");
+			Fail(program, "native shader plan has an invalid image mip descriptor count");
 		}
 		expected[static_cast<size_t>(*kind)].insert(expected[static_cast<size_t>(*kind)].end(),
 		                                            count, i);
@@ -58,11 +62,11 @@ bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
 			}
 			const auto index = inst.Flags<IR::MemoryFlags>().index;
 			if (index >= program.memory_info.size()) {
-				return Fail(error, "shared operation has invalid memory metadata");
+				Fail(program, "shared operation has invalid memory metadata");
 			}
 			const auto kind = program.memory_info[index].kind;
 			if (kind != IR::ResourceKind::Lds && kind != IR::ResourceKind::Gds) {
-				return Fail(error, "shared operation has invalid resource kind");
+				Fail(program, "shared operation has invalid resource kind");
 			}
 			uses_gds |= kind == IR::ResourceKind::Gds;
 		}
@@ -91,13 +95,13 @@ bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
 		const auto kind = static_cast<size_t>(binding.kind);
 		if (kind >= KindCount || seen[kind] || !present[kind] ||
 		    binding.resources != expected[kind]) {
-			return Fail(error, "native descriptor groups do not match shader topology");
+			Fail(program, "native descriptor groups do not match shader topology");
 		}
 		seen[kind] = true;
 	}
 	for (size_t i = 0; i < KindCount; i++) {
 		if (present[i] != seen[i]) {
-			return Fail(error, "native shader plan is missing a required descriptor group");
+			Fail(program, "native shader plan is missing a required descriptor group");
 		}
 	}
 	const auto has_user_storage = present[static_cast<size_t>(Kind::UserData)];
@@ -114,7 +118,7 @@ bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
 	    std::adjacent_find(program.bindings.user_data_registers.begin(),
 	                       program.bindings.user_data_registers.end()) !=
 	        program.bindings.user_data_registers.end()) {
-		return Fail(error, "native user-data layout is inconsistent");
+		Fail(program, "native user-data layout is inconsistent");
 	}
 
 	const auto planning_only_handle = [&](const IR::Inst& handle) {
@@ -139,7 +143,7 @@ bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
 						break;
 					}
 					if (dense >= program.info.buffers.size()) {
-						return Fail(error, "typed buffer handle has an invalid dense resource");
+						Fail(program, "typed buffer handle has an invalid dense resource");
 					}
 					break;
 				case IR::ValueOpcode::GetAddressResource:
@@ -147,29 +151,29 @@ bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
 						break;
 					}
 					if (inst.NumArgs() != 2 || !program.info.uses_dma) {
-						return Fail(error, "typed address handle has invalid DMA metadata");
+						Fail(program, "typed address handle has invalid DMA metadata");
 					}
 					break;
 				case IR::ValueOpcode::GetScratchResource:
 					if (inst.NumArgs() != 0 || program.scratch_dwords == 0) {
-						return Fail(error, "typed scratch handle has invalid shader metadata");
+						Fail(program, "typed scratch handle has invalid shader metadata");
 					}
 					break;
 				case IR::ValueOpcode::GetImageResource:
 					if (dense >= program.info.images.size()) {
-						return Fail(error, "typed image handle has an invalid dense resource");
+						Fail(program, "typed image handle has an invalid dense resource");
 					}
 					break;
 				case IR::ValueOpcode::GetSamplerResource:
 					if (dense >= program.info.samplers.size()) {
-						return Fail(error, "typed sampler handle has an invalid dense resource");
+						Fail(program, "typed sampler handle has an invalid dense resource");
 					}
 					break;
 				case IR::ValueOpcode::ReadConst: {
 					const auto slot = inst.Arg(1).Resolve();
 					if (!slot.IsImmediate() || slot.GetType() != IR::Type::U32 ||
 					    slot.U32() >= program.srt_reads.size()) {
-						return Fail(error, "flattened SRT read has an invalid dense slot");
+						Fail(program, "flattened SRT read has an invalid dense slot");
 					}
 					break;
 				}
@@ -177,12 +181,11 @@ bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
 			}
 		}
 	}
-	return true;
 }
 
 } // namespace
 
-bool AnalyzeProgramRequirements(IR::Program& program, std::string* error) {
+void AnalyzeProgramRequirements(IR::Program& program) {
 	program.spirv_requirements.reset();
 	IR::SpirvRequirements requirements {};
 	const auto MarkBallot = [&] { requirements.subgroup_ballot = true; };
@@ -192,30 +195,30 @@ bool AnalyzeProgramRequirements(IR::Program& program, std::string* error) {
 			if (address_access != IR::AddressAccess::None) {
 				const auto memory_index = inst.Flags<IR::MemoryFlags>().index;
 				if (memory_index >= program.memory_info.size()) {
-					return Fail(error, "address operation has invalid memory metadata");
+					Fail(program, "address operation has invalid memory metadata");
 				}
 				if (program.memory_info[memory_index].kind == IR::ResourceKind::Scratch) {
 					if (program.scratch_dwords == 0) {
-						return Fail(error, "scratch operation has no per-thread storage");
+						Fail(program, "scratch operation has no per-thread storage");
 					}
 					requirements.function_scratch = true;
 				} else if (address_access == IR::AddressAccess::Write) {
-					return Fail(error, "writable FLAT/GLOBAL addresses require GPU ownership tracking");
+					Fail(program, "writable FLAT/GLOBAL addresses require GPU ownership tracking");
 				}
 			}
 			if (IR::BufferAccessOf(inst.GetOpcode()) != IR::BufferAccess::None) {
 				const auto memory_index = inst.Flags<IR::MemoryFlags>().index;
 				if (memory_index >= program.memory_info.size()) {
-					return Fail(error, "buffer operation has invalid memory metadata");
+					Fail(program, "buffer operation has invalid memory metadata");
 				}
 				const auto& memory = program.memory_info[memory_index];
 				if (memory.kind == IR::ResourceKind::Buffer) {
 					if (memory.resource >= program.info.buffers.size()) {
-						return Fail(error, "buffer operation has invalid resource metadata");
+						Fail(program, "buffer operation has invalid resource metadata");
 					}
 					if ((program.info.buffers[memory.resource].packed_stride & (1u << 20u)) != 0u) {
 						if (program.stage != ShaderType::Compute) {
-							return Fail(error, "buffer ADD_TID is only valid for compute shaders");
+							Fail(program, "buffer ADD_TID is only valid for compute shaders");
 						}
 						requirements.subgroup_local_invocation_id = true;
 					}
@@ -225,11 +228,11 @@ bool AnalyzeProgramRequirements(IR::Program& program, std::string* error) {
 			if (shared_access != IR::SharedAccess::None) {
 				const auto index = inst.Flags<IR::MemoryFlags>().index;
 				if (index >= program.memory_info.size()) {
-					return Fail(error, "shared operation has invalid memory metadata");
+					Fail(program, "shared operation has invalid memory metadata");
 				}
 				const auto kind = program.memory_info[index].kind;
 				if (kind != IR::ResourceKind::Lds && kind != IR::ResourceKind::Gds) {
-					return Fail(error, "shared operation has invalid resource kind");
+					Fail(program, "shared operation has invalid resource kind");
 				}
 				if (program.stage != ShaderType::Compute && kind == IR::ResourceKind::Lds) {
 					requirements.function_lds = true;
@@ -283,7 +286,7 @@ bool AnalyzeProgramRequirements(IR::Program& program, std::string* error) {
 				case IR::ValueOpcode::SetAttribute: {
 					const auto index = inst.Flags<IR::ExportFlags>().index;
 					if (index >= program.export_info.size()) {
-						return Fail(error, "attribute export has invalid metadata");
+						Fail(program, "attribute export has invalid metadata");
 					}
 					if (program.stage == ShaderType::Pixel &&
 					    program.export_info[index].vm) {
@@ -296,37 +299,30 @@ bool AnalyzeProgramRequirements(IR::Program& program, std::string* error) {
 		}
 	}
 	program.spirv_requirements.emplace(requirements);
-	return true;
 }
 
-bool EmitProgram(const IR::Program& program, const IR::ResourceSnapshot& resources,
-                 ShaderStageInputInfo input_info, std::vector<uint32_t>& spirv,
-                 std::string* error) {
+std::vector<uint32_t> EmitProgram(const IR::Program& program,
+                                  const IR::ResourceSnapshot& resources,
+                                  ShaderStageInputInfo input_info) {
 	using namespace Emitter;
 
 	if (program.stage != ShaderType::Compute && program.stage != ShaderType::Vertex &&
 	    program.stage != ShaderType::Pixel) {
-		SetError(error, "binary SPIR-V emitter supports compute, vertex, and pixel shaders");
-		return false;
+		Fail(program, "binary SPIR-V emitter supports compute, vertex, and pixel shaders");
 	}
 	if (!program.srt_plan_complete || !program.resource_tracking_complete ||
 	    !program.shader_info_complete || !program.binding_layout_complete ||
 	    !program.spirv_requirements.has_value()) {
-		SetError(error, "SPIR-V emitter requires a fully planned native shader program");
-		return false;
+		Fail(program, "SPIR-V emitter requires a fully planned native shader program");
 	}
-	if (!IR::ValidateResourceSnapshot(program, resources, error)) {
-		return false;
+	if (!IR::ValidateResourceSnapshot(program, resources)) {
+		Fail(program, "resource snapshot is inconsistent with the native shader plan");
 	}
-	if (!IR::ValidateResourceSpecialization(program, resources, error)) {
-		return false;
-	}
-	if (!ValidateNativeProgram(program, error)) {
-		return false;
-	}
-	if (!IR::ValidateProgram(program, true, error)) {
-		return false;
+	if (!IR::ValidateResourceSpecialization(program, resources)) {
+		Fail(program, "resource specialization is inconsistent with the native shader plan");
 	}
+	ValidateNativeProgram(program);
+	IR::ValidateProgram(program, true);
 	EmitterState state(program, resources, input_info);
 	state.stage     = program.stage;
 	state.wave_size = program.wave_size;
@@ -337,17 +333,13 @@ bool EmitProgram(const IR::Program& program, const IR::ResourceSnapshot& resourc
 	AllocateInputVariables(state);
 	AllocateOutputVariables(state);
 	DefineModule(state);
-	if (!EmitProgram(state, program, error)) {
-		return false;
-	}
+	EmitProgram(state, program);
 
 	auto binary = state.builder.Build();
 	if (binary.empty()) {
-		SetError(error, "SPIR-V builder returned an empty module");
-		return false;
+		Fail(program, "SPIR-V builder returned an empty module");
 	}
-	spirv = std::move(binary);
-	return true;
+	return binary;
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::Spirv

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

@@ -9,10 +9,11 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv {
 
-bool AnalyzeProgramRequirements(IR::Program& program, std::string* error);
+void AnalyzeProgramRequirements(IR::Program& program);
 
-bool EmitProgram(const IR::Program& program, const IR::ResourceSnapshot& resources,
-                 ShaderStageInputInfo input_info, std::vector<uint32_t>& spirv, std::string* error);
+std::vector<uint32_t> EmitProgram(const IR::Program& program,
+                                  const IR::ResourceSnapshot& resources,
+                                  ShaderStageInputInfo input_info);
 
 } // namespace Libs::Graphics::ShaderRecompiler::Spirv
 

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

@@ -37,12 +37,6 @@ bool PixelParameterIsCustom(const EmitterState& state, uint32_t attr) {
 	       ShaderPixelParameterIsCustom(*state.input_info.pixel, attr);
 }
 
-void SetError(std::string* error, const char* message) {
-	if (error != nullptr) {
-		*error = message;
-	}
-}
-
 bool HasOutput(const std::vector<OutputBinding>& outputs, IR::StageOutputKind kind,
                uint32_t index) {
 	return std::any_of(outputs.begin(), outputs.end(), [kind, index](const OutputBinding& binding) {

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

@@ -439,7 +439,8 @@ struct ValueEmitContext {
 	const IR::MemoryInfo& Memory(const IR::Inst& inst) const;
 	const IR::ExportInfo& Export(const IR::Inst& inst) const;
 	uint32_t              Label(const IR::Block* block) const;
-	void                  Fail(const IR::Inst& inst, const char* reason);
+	[[noreturn]] void     Fail(const char* reason) const;
+	[[noreturn]] void     Fail(const IR::Inst& inst, const char* reason) const;
 
 	EmitterState&                                                      state;
 	const IR::Program&                                            program;
@@ -449,8 +450,6 @@ struct ValueEmitContext {
 	std::unordered_map<const IR::Inst*, std::pair<uint32_t, uint32_t>> dispatcher_block_loads;
 	const IR::Block*                                                   current_block = nullptr;
 	uint32_t                                                           scratch_u32_variable = 0;
-	bool                                                               failed               = false;
-	std::string                                                        error;
 };
 
 enum class VertexInputScalarKind { Float, Sint, Uint };
@@ -613,8 +612,6 @@ uint32_t VertexParameterVectorOrScalarType(EmitterState& state, VertexInputScala
 uint32_t VertexParameterInputPointerType(EmitterState& state, VertexInputScalarKind kind,
                                          uint32_t components);
 
-void SetError(std::string* error, const char* message);
-
 bool HasOutput(const std::vector<OutputBinding>& outputs, IR::StageOutputKind kind, uint32_t index);
 
 void CopyProgramInputsAndOutputs(EmitterState& state, const IR::Program& program);
@@ -864,7 +861,7 @@ bool EmitValueMemory(ValueEmitContext& ctx, const IR::Inst& inst);
 
 bool EmitValueImage(ValueEmitContext& ctx, const IR::Inst& inst);
 
-bool EmitProgram(EmitterState& state, const IR::Program& program, std::string* error);
+void EmitProgram(EmitterState& state, const IR::Program& program);
 
 void DefineGetBdaPointer(EmitterState& state);
 

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

@@ -1,8 +1,9 @@
 #include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
 
+#include "common/assert.h"
+
 #include <algorithm>
 #include <bit>
-#include <fmt/format.h>
 #include <functional>
 #include <optional>
 #include <unordered_set>
@@ -196,57 +197,50 @@ uint32_t EmitDispatcherNextPc(ValueEmitContext& ctx, const DispatcherFunctionSta
 	}
 }
 
-bool EmitDirectInstruction(ValueEmitContext& ctx, const IR::Inst& inst) {
+void EmitDirectInstruction(ValueEmitContext& ctx, const IR::Inst& inst) {
 	if (EmitValueFlow(ctx, inst) || EmitValueAlu(ctx, inst) || EmitValueMemory(ctx, inst) ||
 	    EmitValueImage(ctx, inst)) {
-		return true;
+		return;
 	}
 	ctx.Fail(inst, "has no direct SPIR-V emitter");
-	return false;
 }
 
-bool EmitStructuredInstruction(ValueEmitContext& ctx, StructuredFunctionState& structured,
+void EmitStructuredInstruction(ValueEmitContext& ctx, StructuredFunctionState& structured,
                                const IR::Inst& inst) {
 	if (inst.GetOpcode() == IR::ValueOpcode::Phi) {
 		const auto type = ctx.TypeId(inst.GetType());
 		if (type == 0 || inst.NumArgs() == 0) {
 			ctx.Fail(inst, "has no native SPIR-V representation");
-			return false;
 		}
 		for (size_t index = 0; index < inst.NumArgs(); index++) {
 			const auto* predecessor = inst.PhiBlock(index);
 			if (predecessor == nullptr || !ctx.labels.contains(predecessor)) {
 				ctx.Fail(inst, "has a predecessor outside the structured function");
-				return false;
 			}
 		}
 		structured.deferred_phis.push_back(
 		    {ctx.state.builder.AddDeferredPhi(type, ctx.Result(inst), inst.NumArgs()), &inst});
-		return true;
+		return;
 	}
-	return EmitDirectInstruction(ctx, inst);
+	EmitDirectInstruction(ctx, inst);
 }
 
-bool EmitDispatcherInstruction(ValueEmitContext& ctx, const DispatcherFunctionState& dispatcher,
+void EmitDispatcherInstruction(ValueEmitContext& ctx, const DispatcherFunctionState& dispatcher,
                                const IR::Inst& inst) {
 	if (inst.GetOpcode() == IR::ValueOpcode::Phi) {
 		const auto type = ctx.TypeId(inst.GetType());
 		if (type == 0) {
 			ctx.Fail(inst, "cannot be loaded by the dispatcher");
-			return false;
 		}
 		ctx.state.builder.AddFunction(
 		    {OpLoad, type, ctx.Result(inst), dispatcher.spills.at(&inst)});
-		return true;
-	}
-	if (!EmitDirectInstruction(ctx, inst)) {
-		return false;
+		return;
 	}
+	EmitDirectInstruction(ctx, inst);
 	if (const auto found = dispatcher.spills.find(&inst); found != dispatcher.spills.end()) {
 		ctx.state.builder.AddFunction(
 		    {OpStore, found->second, ctx.Def(IR::Value(const_cast<IR::Inst*>(&inst)))});
 	}
-	return true;
 }
 
 template <typename EmitInstruction>
@@ -258,14 +252,11 @@ void EmitBlock(ValueEmitContext& ctx, const IR::Block* block, EmitInstruction&&
 		if (inst.GetOpcode() == IR::ValueOpcode::Phi) {
 			if (emitted_non_phi) {
 				ctx.Fail(inst, "appears after a non-Phi instruction");
-				return;
 			}
 		} else {
 			emitted_non_phi = true;
 		}
-		if (!emit_instruction(inst)) {
-			return;
-		}
+		emit_instruction(inst);
 	}
 }
 
@@ -275,15 +266,10 @@ void PatchStructuredPhis(ValueEmitContext& ctx, StructuredFunctionState& structu
 			const auto* predecessor = deferred.instruction->PhiBlock(index);
 			const auto  found       = structured.block_exit_labels.find(predecessor);
 			if (found == structured.block_exit_labels.end()) {
-				ctx.failed = true;
-				ctx.error  = "typed Phi predecessor was not emitted";
-				return;
+				ctx.Fail(*deferred.instruction, "has a predecessor that was not emitted");
 			}
 			ctx.state.builder.PatchDeferredPhi(
 			    deferred.phi, index, ctx.Def(deferred.instruction->Arg(index)), found->second);
-			if (ctx.failed) {
-				return;
-			}
 		}
 	}
 }
@@ -291,20 +277,15 @@ void PatchStructuredPhis(ValueEmitContext& ctx, StructuredFunctionState& structu
 void EmitStructuredFunction(ValueEmitContext& ctx) {
 	StructuredFunctionState structured;
 	ctx.state.builder.AddFunction({OpBranch, ctx.Label(ctx.program.blocks.front())});
-	for (size_t index = 0; index < ctx.program.blocks.size() && !ctx.failed; index++) {
+	for (size_t index = 0; index < ctx.program.blocks.size(); index++) {
 		const auto* block = ctx.program.blocks[index];
 		EmitBlock(ctx, block, [&](const IR::Inst& inst) {
-			return EmitStructuredInstruction(ctx, structured, inst);
+			EmitStructuredInstruction(ctx, structured, inst);
 		});
-		if (ctx.failed) {
-			break;
-		}
 		structured.block_exit_labels.emplace(block, ctx.state.current_label);
 		EmitStructuredTerminator(ctx, block, ctx.program.block_info[index]);
 	}
-	if (!ctx.failed) {
-		PatchStructuredPhis(ctx, structured);
-	}
+	PatchStructuredPhis(ctx, structured);
 }
 
 void EmitDispatcherFunction(ValueEmitContext& ctx, const DispatcherFunctionState& dispatcher) {
@@ -312,11 +293,8 @@ void EmitDispatcherFunction(ValueEmitContext& ctx, const DispatcherFunctionState
 	const auto* entry = ctx.program.blocks.front();
 	state.builder.AddFunction({OpBranch, ctx.Label(entry)});
 	EmitBlock(ctx, entry, [&](const IR::Inst& inst) {
-		return EmitDispatcherInstruction(ctx, dispatcher, inst);
+		EmitDispatcherInstruction(ctx, dispatcher, inst);
 	});
-	if (ctx.failed) {
-		return;
-	}
 	const auto initial_pc =
 	    EmitDispatcherNextPc(ctx, dispatcher, entry, ctx.program.block_info.front());
 	const auto initial_parent = state.current_label;
@@ -347,22 +325,16 @@ void EmitDispatcherFunction(ValueEmitContext& ctx, const DispatcherFunctionState
 	std::vector<uint32_t> next_pc_words {OpPhi, TypeU32(state), next_pc,
 	                                     ConstantU32(state, UINT32_MAX), dispatcher.select_label};
 
-	for (size_t index = 1; index < ctx.program.blocks.size() && !ctx.failed; index++) {
+	for (size_t index = 1; index < ctx.program.blocks.size(); index++) {
 		EmitBlock(ctx, ctx.program.blocks[index], [&](const IR::Inst& inst) {
-			return EmitDispatcherInstruction(ctx, dispatcher, inst);
+			EmitDispatcherInstruction(ctx, dispatcher, inst);
 		});
-		if (ctx.failed) {
-			break;
-		}
 		const auto selected = EmitDispatcherNextPc(ctx, dispatcher, ctx.program.blocks[index],
 		                                           ctx.program.block_info[index]);
 		next_pc_words.push_back(selected);
 		next_pc_words.push_back(state.current_label);
 		state.builder.AddFunction({OpBranch, dispatcher.after_switch_label});
 	}
-	if (ctx.failed) {
-		return;
-	}
 	EmitLabel(state, dispatcher.after_switch_label);
 	state.builder.AddFunction(next_pc_words);
 	state.builder.AddFunction({OpBranch, dispatcher.continue_label});
@@ -408,9 +380,7 @@ uint32_t ValueEmitContext::Def(IR::Value value) {
 	}
 	const auto* inst = value.ResolveInstruction();
 	if (inst == nullptr) {
-		failed = true;
-		error  = "direct SPIR-V emitter received a non-value argument";
-		return ConstantU32(state, 0);
+		Fail("direct SPIR-V emitter received a non-value argument");
 	}
 	if (dispatcher_spills != nullptr && current_block != nullptr &&
 	    inst->Parent() != current_block) {
@@ -469,9 +439,7 @@ uint32_t ValueEmitContext::Define(const IR::Inst& inst, uint32_t value) {
 uint32_t ValueEmitContext::ResourceIndex(IR::Value value, IR::ValueOpcode opcode) {
 	const auto* inst = value.ResolveInstruction();
 	if (inst == nullptr || inst->GetOpcode() != opcode) {
-		failed = true;
-		error  = "typed resource handle has the wrong producer";
-		return 0;
+		Fail("typed resource handle has the wrong producer");
 	}
 	return inst->Flags<uint32_t>();
 }
@@ -479,9 +447,7 @@ uint32_t ValueEmitContext::ResourceIndex(IR::Value value, IR::ValueOpcode opcode
 const IR::Inst* ValueEmitContext::ImageAddress(IR::Value value) {
 	const auto* inst = value.ResolveInstruction();
 	if (inst == nullptr || inst->GetOpcode() != IR::ValueOpcode::MakeImageAddress) {
-		failed = true;
-		error  = "typed image address was not constructed by MakeImageAddress";
-		return nullptr;
+		Fail("typed image address was not constructed by MakeImageAddress");
 	}
 	return inst;
 }
@@ -498,12 +464,20 @@ uint32_t ValueEmitContext::Label(const IR::Block* block) const {
 	return labels.at(block);
 }
 
-void ValueEmitContext::Fail(const IR::Inst& inst, const char* reason) {
-	failed = true;
-	error  = fmt::format("typed opcode {} {}", IR::ValueOpcodeName(inst.GetOpcode()), reason);
+[[noreturn]] void ValueEmitContext::Fail(const char* reason) const {
+	EXIT("SPIR-V emission failed: hash=0x%016" PRIx64 " stage=%u reason=%s\n",
+	     program.shader_hash, static_cast<unsigned>(program.stage), reason);
+	std::abort();
 }
 
-bool EmitProgram(EmitterState& state, const IR::Program& program, std::string* error) {
+[[noreturn]] void ValueEmitContext::Fail(const IR::Inst& inst, const char* reason) const {
+	EXIT("SPIR-V emission failed: hash=0x%016" PRIx64 " stage=%u opcode=%s reason=%s\n",
+	     program.shader_hash, static_cast<unsigned>(program.stage),
+	     IR::ValueOpcodeName(inst.GetOpcode()), reason);
+	std::abort();
+}
+
+void EmitProgram(EmitterState& state, const IR::Program& program) {
 	ValueEmitContext                       ctx(state, program);
 	std::optional<DispatcherFunctionState> dispatcher;
 	if (state.stage == ShaderType::Pixel && state.requirements.pixel_valid_mask) {
@@ -562,10 +536,6 @@ bool EmitProgram(EmitterState& state, const IR::Program& program, std::string* e
 		dispatch.merge_label        = state.builder.AllocateId();
 		ctx.dispatcher_spills       = &dispatch.spills;
 	}
-	if (ctx.failed) {
-		SetError(error, ctx.error.c_str());
-		return false;
-	}
 	DefineGetBdaPointer(state);
 	for (const auto* block: program.blocks) {
 		if (std::ranges::any_of(*block, [](const IR::Inst& inst) {
@@ -630,11 +600,6 @@ bool EmitProgram(EmitterState& state, const IR::Program& program, std::string* e
 		EmitStructuredFunction(ctx);
 	}
 	state.builder.AddFunction({OpFunctionEnd});
-	if (ctx.failed) {
-		SetError(error, ctx.error.c_str());
-		return false;
-	}
-	return true;
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter

src/graphics/shader/recompiler/frontend/cfg/ShaderCFG.cpp

@@ -1,5 +1,7 @@
 #include "graphics/shader/recompiler/frontend/cfg/ShaderCFG.h"
 
+#include "common/assert.h"
+
 #include <algorithm>
 #include <fmt/format.h>
 #include <iterator>
@@ -24,19 +26,18 @@ struct SetpcTargetInfo {
 	std::vector<uint32_t> selector_target_pcs;
 };
 
-void SetError(std::string* error, const std::string& message) {
-	if (error != nullptr) {
-		*error = message;
-	}
-}
-
-void SetFailure(Graph& graph, FailureKind kind, uint32_t block_id, const std::string& message,
-                std::string* error) {
+void SetFailure(Graph& graph, FailureKind kind, uint32_t block_id, const std::string& message) {
 	graph.unsupported        = true;
 	graph.failure_kind       = kind;
 	graph.failure_block      = block_id;
 	graph.unsupported_reason = message;
-	SetError(error, message);
+}
+
+[[noreturn]] void ExitBuildFailure(Graph& graph, FailureKind kind, uint32_t block_id,
+                                   const std::string& message) {
+	SetFailure(graph, kind, block_id, message);
+	EXIT("shader CFG build failed: %s", message.c_str());
+	std::abort();
 }
 
 uint32_t InstructionEndPc(const Instruction& inst) {
@@ -1267,7 +1268,7 @@ bool MergeLeavesContainingLoop(const Graph& graph, uint32_t header, uint32_t mer
 	return false;
 }
 
-bool CanonicalizeNaturalLoops(Graph& graph, std::string* error) {
+bool CanonicalizeNaturalLoops(Graph& graph) {
 	const auto rewrite_budget = graph.blocks.size() * 2u + 16u;
 	for (size_t rewrite = 0; rewrite < rewrite_budget; rewrite++) {
 		bool changed = false;
@@ -1322,19 +1323,18 @@ bool CanonicalizeNaturalLoops(Graph& graph, std::string* error) {
 	}
 
 	SetFailure(graph, FailureKind::StructuredControlFlow, graph.entry_block,
-	           "CFG loop canonicalization exceeded rewrite budget", error);
+	           "CFG loop canonicalization exceeded rewrite budget");
 	return false;
 }
 
-bool IsolateSemanticLoopHeaders(Graph& graph, std::string* error) {
+bool IsolateSemanticLoopHeaders(Graph& graph) {
 	const auto isolation_budget = graph.natural_loops.size() + 1u;
 	for (size_t isolation = 0; isolation < isolation_budget; isolation++) {
-		const auto loop = std::find_if(graph.natural_loops.begin(), graph.natural_loops.end(),
-		                               [&](const auto& value) {
-			                               const auto* header = graph.FindBlock(value.header);
-			                               return header != nullptr &&
-			                                      header->inst_begin != header->inst_end;
-		                               });
+		const auto loop = std::find_if(
+		    graph.natural_loops.begin(), graph.natural_loops.end(), [&](const auto& value) {
+			    const auto* header = graph.FindBlock(value.header);
+			    return header != nullptr && header->inst_begin != header->inst_end;
+		    });
 		if (loop == graph.natural_loops.end()) {
 			return true;
 		}
@@ -1346,14 +1346,13 @@ bool IsolateSemanticLoopHeaders(Graph& graph, std::string* error) {
 		// displacing OpLoopMerge into a helper-created block.
 		if (!IsolateSemanticLoopHeader(graph, loop->header)) {
 			SetFailure(graph, FailureKind::StructuredControlFlow, loop->header,
-			           fmt::format("failed to isolate semantic loop header {}", loop->header),
-			           error);
+			           fmt::format("failed to isolate semantic loop header {}", loop->header));
 			return false;
 		}
 	}
 
 	SetFailure(graph, FailureKind::StructuredControlFlow, graph.entry_block,
-	           "CFG semantic loop-header isolation exceeded rewrite budget", error);
+	           "CFG semantic loop-header isolation exceeded rewrite budget");
 	return false;
 }
 
@@ -1453,7 +1452,7 @@ std::vector<uint32_t> SelectionRegion(const Graph& graph, const BasicBlock& head
 	return region;
 }
 
-bool SplitOneSelectionMerge(Graph& graph, std::string* error) {
+bool SplitOneSelectionMerge(Graph& graph) {
 	std::vector<uint32_t> loop_headers;
 	loop_headers.reserve(graph.natural_loops.size());
 	for (const auto& loop: graph.natural_loops) {
@@ -1501,8 +1500,7 @@ bool SplitOneSelectionMerge(Graph& graph, std::string* error) {
 			    graph, FailureKind::StructuredControlFlow, block_id,
 			    fmt::format("selection header block {} has externally entered region block {}; "
 			                "semantic block cloning is disabled",
-			                block_id, *external),
-			    error);
+			                block_id, *external));
 			return false;
 		}
 		const auto construct_blocks = DominatedBlocks(graph, block_id, merge);
@@ -1514,11 +1512,11 @@ bool SplitOneSelectionMerge(Graph& graph, std::string* error) {
 	return false;
 }
 
-bool SplitSharedMergeBlocks(Graph& graph, std::string* error) {
+bool SplitSharedMergeBlocks(Graph& graph) {
 	const auto original_block_count = static_cast<uint32_t>(graph.blocks.size());
 	const auto split_budget         = std::max<uint32_t>(16u, original_block_count * 4u);
 	for (uint32_t splits = 0; splits < split_budget; splits++) {
-		if (!SplitOneLoopMerge(graph) && !SplitOneSelectionMerge(graph, error)) {
+		if (!SplitOneLoopMerge(graph) && !SplitOneSelectionMerge(graph)) {
 			return !graph.unsupported;
 		}
 		RebuildPredecessors(graph);
@@ -1528,8 +1526,7 @@ bool SplitSharedMergeBlocks(Graph& graph, std::string* error) {
 	           fmt::format("CFG shared merge splitting exceeded budget: original_blocks={} "
 	                       "current_blocks={} split_budget={}",
 	                       original_block_count, static_cast<uint64_t>(graph.blocks.size()),
-	                       split_budget),
-	           error);
+	                       split_budget));
 	return false;
 }
 
@@ -1890,12 +1887,11 @@ bool RouteSharedSelectionArm(Graph& graph, uint32_t route_variable) {
 
 } // namespace
 
-bool BuildGraph(const Decoder::Program& program, Graph& graph, std::string* error) {
-	graph = {};
-
+Graph BuildGraph(const Decoder::Program& program) {
+	Graph graph;
 	if (program.instructions.empty()) {
-		SetError(error, "cannot build CFG for empty shader");
-		return false;
+		ExitBuildFailure(graph, FailureKind::InvalidLabel, UINT32_MAX,
+		                 "cannot build CFG for empty shader");
 	}
 
 	const auto first_pc = program.instructions.front().pc;
@@ -1905,11 +1901,10 @@ bool BuildGraph(const Decoder::Program& program, Graph& graph, std::string* erro
 	for (const auto& inst: program.instructions) {
 		instruction_pcs.insert(inst.pc);
 		if (inst.opcode == Opcode::UNSUPPORTED) {
-			SetFailure(graph, FailureKind::UnsupportedInstruction, UINT32_MAX,
-			           fmt::format("unsupported decoded instruction in CFG at pc 0x{:08x}: {}",
-			                       inst.pc, Decoder::InstructionToString(inst).c_str()),
-			           error);
-			return false;
+			ExitBuildFailure(
+			    graph, FailureKind::UnsupportedInstruction, UINT32_MAX,
+			    fmt::format("unsupported decoded instruction in CFG at pc 0x{:08x}: {}", inst.pc,
+			                Decoder::InstructionToString(inst).c_str()));
 		}
 	}
 
@@ -1923,11 +1918,9 @@ bool BuildGraph(const Decoder::Program& program, Graph& graph, std::string* erro
 		const auto  next_pc = InstructionEndPc(inst);
 		if (IsBranch(inst.opcode)) {
 			if (!IsValidTarget(inst.branch_target, instruction_pcs, first_pc, end_pc)) {
-				SetFailure(graph, FailureKind::InvalidBranchTarget, UINT32_MAX,
-				           fmt::format("branch at pc 0x{:08x} targets invalid pc 0x{:08x}", inst.pc,
-				                       inst.branch_target),
-				           error);
-				return false;
+				ExitBuildFailure(graph, FailureKind::InvalidBranchTarget, UINT32_MAX,
+				                 fmt::format("branch at pc 0x{:08x} targets invalid pc 0x{:08x}",
+				                             inst.pc, inst.branch_target));
 			}
 			labels.insert(inst.branch_target);
 			if (next_pc <= end_pc) {
@@ -1936,21 +1929,19 @@ bool BuildGraph(const Decoder::Program& program, Graph& graph, std::string* erro
 		} else if (inst.opcode == Opcode::S_SETPC_B64) {
 			SetpcTargetInfo target_info;
 			if (!ResolveSetpcTargets(program, i, target_info)) {
-				SetFailure(graph, FailureKind::InvalidBranchTarget, UINT32_MAX,
-				           fmt::format("unsupported dynamic S_SETPC_B64 at pc 0x{:08x}", inst.pc),
-				           error);
-				return false;
+				ExitBuildFailure(
+				    graph, FailureKind::InvalidBranchTarget, UINT32_MAX,
+				    fmt::format("unsupported dynamic S_SETPC_B64 at pc 0x{:08x}", inst.pc));
 			}
 			const auto& target_pcs = target_info.indirect
 			                             ? target_info.target_pcs
 			                             : std::vector<uint32_t> {target_info.target};
 			for (const auto target: target_pcs) {
 				if (!IsValidTarget(target, instruction_pcs, first_pc, end_pc)) {
-					SetFailure(graph, FailureKind::InvalidBranchTarget, UINT32_MAX,
-					           fmt::format("S_SETPC_B64 at pc 0x{:08x} targets invalid pc 0x{:08x}",
-					                       inst.pc, target),
-					           error);
-					return false;
+					ExitBuildFailure(
+					    graph, FailureKind::InvalidBranchTarget, UINT32_MAX,
+					    fmt::format("S_SETPC_B64 at pc 0x{:08x} targets invalid pc 0x{:08x}",
+					                inst.pc, target));
 				}
 				labels.insert(target);
 			}
@@ -1971,10 +1962,9 @@ bool BuildGraph(const Decoder::Program& program, Graph& graph, std::string* erro
 			continue;
 		}
 		if (start != end_pc && !instruction_pcs.contains(start)) {
-			SetFailure(graph, FailureKind::InvalidLabel, UINT32_MAX,
-			           fmt::format("CFG label does not start on an instruction: 0x{:08x}", start),
-			           error);
-			return false;
+			ExitBuildFailure(
+			    graph, FailureKind::InvalidLabel, UINT32_MAX,
+			    fmt::format("CFG label does not start on an instruction: 0x{:08x}", start));
 		}
 
 		BasicBlock block;
@@ -2046,11 +2036,10 @@ bool BuildGraph(const Decoder::Program& program, Graph& graph, std::string* erro
 			block.terminator.true_block = pc_to_block.at(last.branch_target);
 			const auto fallthrough      = pc_to_block.find(next_pc);
 			if (fallthrough == pc_to_block.end()) {
-				SetFailure(graph, FailureKind::MissingFallthrough, block.id,
-				           fmt::format("conditional branch at pc 0x{:08x} has no fallthrough block",
-				                       last.pc),
-				           error);
-				return false;
+				ExitBuildFailure(
+				    graph, FailureKind::MissingFallthrough, block.id,
+				    fmt::format("conditional branch at pc 0x{:08x} has no fallthrough block",
+				                last.pc));
 			}
 			block.terminator.false_block = fallthrough->second;
 		} else {
@@ -2131,27 +2120,26 @@ bool BuildGraph(const Decoder::Program& program, Graph& graph, std::string* erro
 		graph.unsupported = false;
 	}
 
-	return true;
+	return graph;
 }
 
 namespace {
 
-bool StructurizeImpl(Graph& graph, std::string* error) {
+bool StructurizeImpl(Graph& graph) {
 	if (graph.unsupported || graph.irreducible) {
 		if (graph.unsupported_reason.empty()) {
 			graph.unsupported_reason = "unsupported CFG";
 		}
-		SetError(error, graph.unsupported_reason);
 		return false;
 	}
 
-	if (!CanonicalizeNaturalLoops(graph, error)) {
+	if (!CanonicalizeNaturalLoops(graph)) {
 		return false;
 	}
-	if (!SplitSharedMergeBlocks(graph, error)) {
+	if (!SplitSharedMergeBlocks(graph)) {
 		return false;
 	}
-	if (!IsolateSemanticLoopHeaders(graph, error)) {
+	if (!IsolateSemanticLoopHeaders(graph)) {
 		return false;
 	}
 	ClearStructuredTerminators(graph);
@@ -2164,8 +2152,7 @@ bool StructurizeImpl(Graph& graph, std::string* error) {
 			    graph, FailureKind::StructuredControlFlow, header,
 			    fmt::format(
 			        "duplicate structured merge block {} for header {} (already used by header {})",
-			        merge, header, it->second),
-			    error);
+			        merge, header, it->second));
 			return false;
 		}
 		return true;
@@ -2176,16 +2163,14 @@ bool StructurizeImpl(Graph& graph, std::string* error) {
 		if (header == nullptr || loop.merge == UINT32_MAX || loop.continue_block == UINT32_MAX) {
 			SetFailure(
 			    graph, FailureKind::StructuredControlFlow, loop.header,
-			    fmt::format("loop at block {} has no structured merge/continue", loop.header),
-			    error);
+			    fmt::format("loop at block {} has no structured merge/continue", loop.header));
 			return false;
 		}
 		if (header->inst_begin != header->inst_end ||
 		    header->terminator.kind != TerminatorKind::Branch) {
-			SetFailure(graph, FailureKind::StructuredControlFlow, loop.header,
-			           fmt::format("loop header {} is not a dedicated empty control block",
-			                       loop.header),
-			           error);
+			SetFailure(
+			    graph, FailureKind::StructuredControlFlow, loop.header,
+			    fmt::format("loop header {} is not a dedicated empty control block", loop.header));
 			return false;
 		}
 		if (!reserve_merge_block(loop.header, loop.merge)) {
@@ -2208,8 +2193,7 @@ bool StructurizeImpl(Graph& graph, std::string* error) {
 		const auto merge = FindSelectionMerge(graph, block);
 		if (merge == UINT32_MAX) {
 			SetFailure(graph, FailureKind::StructuredControlFlow, block.id,
-			           fmt::format("conditional block {} has no structured merge", block.id),
-			           error);
+			           fmt::format("conditional block {} has no structured merge", block.id));
 			return false;
 		}
 
@@ -2224,16 +2208,15 @@ bool StructurizeImpl(Graph& graph, std::string* error) {
 
 } // namespace
 
-bool Structurize(Graph& graph, std::string* error) {
+bool Structurize(Graph& graph) {
 	Graph original = graph;
-	if (StructurizeImpl(graph, error)) {
+	if (StructurizeImpl(graph)) {
 		return true;
 	}
 
-	Graph       failed_graph = std::move(graph);
-	std::string failed_error = error != nullptr ? *error : std::string {};
-	graph                    = std::move(original);
-	const auto route_budget  = static_cast<uint32_t>(graph.blocks.size());
+	Graph failed_graph      = std::move(graph);
+	graph                   = std::move(original);
+	const auto route_budget = static_cast<uint32_t>(graph.blocks.size());
 	// Apply one route at a time and retry. Eagerly routing every matching diamond can
 	// rewrite unrelated selections that were already structurally valid.
 	for (uint32_t route_variable = 0; route_variable < route_budget; route_variable++) {
@@ -2241,18 +2224,12 @@ bool Structurize(Graph& graph, std::string* error) {
 			break;
 		}
 		Graph routed = graph;
-		if (error != nullptr) {
-			error->clear();
-		}
-		if (StructurizeImpl(routed, error)) {
+		if (StructurizeImpl(routed)) {
 			graph = std::move(routed);
 			return true;
 		}
 	}
 	graph = std::move(failed_graph);
-	if (error != nullptr) {
-		*error = std::move(failed_error);
-	}
 	return false;
 }
 

src/graphics/shader/recompiler/frontend/cfg/ShaderCFG.h

@@ -108,8 +108,8 @@ struct Graph {
 	uint32_t          FindNearestCommonPostDominator(uint32_t block_a, uint32_t block_b) const;
 };
 
-bool        BuildGraph(const Decoder::Program& program, Graph& graph, std::string* error);
-bool        Structurize(Graph& graph, std::string* error);
+Graph       BuildGraph(const Decoder::Program& program);
+bool        Structurize(Graph& graph);
 std::string BranchConditionToString(BranchCondition condition);
 std::string FailureKindToString(FailureKind kind);
 std::string GraphToString(const Graph& graph);

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

@@ -357,8 +357,8 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 
 	DecodeVectorGpr(vdata, inst.dst);
 	DecodeVectorGpr(vaddr, inst.src0);
-	DecodeScalarSource(srsrc * 4u, pc, inst.src1, nullptr);
-	DecodeScalarSource(ssamp * 4u, pc, inst.src2, nullptr);
+	DecodeScalarSource(srsrc * 4u, pc, inst.src1);
+	DecodeScalarSource(ssamp * 4u, pc, inst.src2);
 	inst.src_count = 3;
 }
 

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

@@ -241,11 +241,11 @@ void DecodeSmem(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 		MarkMemoryUnsupported(inst, Family::SMEM, opcode, "SMEM opcode is not implemented");
 	}
 
-	DecodeScalarDestination(sdst, pc, inst.dst, nullptr);
+	DecodeScalarDestination(sdst, pc, inst.dst);
 	// SMEM encodes SBASE in SGPR pairs. Scalar-buffer loads still use the same
 	// pair index; their descriptor operand consumes four SGPRs from that base.
-	DecodeScalarSource(sbase * 2u, pc, inst.src0, nullptr);
-	DecodeScalarSource(soffset, pc, inst.src1, nullptr);
+	DecodeScalarSource(sbase * 2u, pc, inst.src0);
+	DecodeScalarSource(soffset, pc, inst.src1);
 	inst.src_count = 2;
 }
 
@@ -278,8 +278,8 @@ void DecodeMubuf(uint32_t pc, std::span<const uint32_t> code, uint32_t word_inde
 
 	DecodeVectorGpr(vdata, inst.dst);
 	DecodeVectorGpr(vaddr, inst.src0);
-	DecodeScalarSource(srsrc * 4u, pc, inst.src1, nullptr);
-	DecodeScalarSource(soffset, pc, inst.src2, nullptr);
+	DecodeScalarSource(srsrc * 4u, pc, inst.src1);
+	DecodeScalarSource(soffset, pc, inst.src2);
 	inst.src_count = 3;
 }
 
@@ -316,8 +316,8 @@ void DecodeMtbuf(uint32_t pc, std::span<const uint32_t> code, uint32_t word_inde
 
 	DecodeVectorGpr(vdata, inst.dst);
 	DecodeVectorGpr(vaddr, inst.src0);
-	DecodeScalarSource(srsrc * 4u, pc, inst.src1, nullptr);
-	DecodeScalarSource(soffset, pc, inst.src2, nullptr);
+	DecodeScalarSource(srsrc * 4u, pc, inst.src1);
+	DecodeScalarSource(soffset, pc, inst.src2);
 	inst.src_count = 3;
 }
 
@@ -364,7 +364,7 @@ void DecodeFlat(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 		DecodeVectorGpr(addr + 1u, inst.src1);
 		inst.src_count = 2;
 	} else {
-		DecodeScalarSource(saddr, pc, inst.src1, nullptr);
+		DecodeScalarSource(saddr, pc, inst.src1);
 		inst.src_count = 2;
 	}
 }

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

@@ -105,21 +105,18 @@ constexpr auto SOPK_OPS = Detail::MakeOpcodeTable<0x20>(SOPK_OPCODE_LIST);
 constexpr auto SOPC_OPS = Detail::MakeOpcodeTable<0x80>(SOPC_OPCODE_LIST);
 constexpr auto SOPP_OPS = Detail::MakeOpcodeTable<0x80>(SOPP_OPCODE_LIST);
 
-bool DecodeBinarySources(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
-                         Instruction& inst, uint32_t ssrc0, uint32_t ssrc1, std::string* error) {
-	if (!DecodeScalarSource(ssrc0, pc, inst.src0, error) ||
-	    !DecodeScalarSource(ssrc1, pc, inst.src1, error)) {
-		return false;
-	}
+void DecodeBinarySources(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                         Instruction& inst, uint32_t ssrc0, uint32_t ssrc1) {
+	DecodeScalarSource(ssrc0, pc, inst.src0);
+	DecodeScalarSource(ssrc1, pc, inst.src1);
 	inst.src_count = 2;
 	ReadLiteralOperands(code, word_index, inst);
-	return true;
 }
 
 } // namespace
 
-bool DecodeSop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error) {
+void DecodeSop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst) {
 	const uint32_t word   = code[word_index];
 	const uint32_t opcode = (word >> 8u) & 0xffu;
 	const uint32_t ssrc0  = word & 0xffu;
@@ -134,35 +131,31 @@ bool DecodeSop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 
 	if (inst.opcode == Opcode::UNSUPPORTED) {
 		SetUnsupported(inst, Family::SOP1, opcode, "SOP1 opcode is not implemented");
-		return true;
+		return;
 	}
 
 	switch (inst.opcode) {
 		case Opcode::S_GETPC_B64:
 			inst.src_count = 0;
-			return DecodeScalarDestination(sdst, pc, inst.dst, error);
+			DecodeScalarDestination(sdst, pc, inst.dst);
+			return;
 		case Opcode::S_SETPC_B64:
 			inst.src_count = 1;
 			inst.dst.kind  = OperandKind::Null;
-			if (!DecodeScalarSource(ssrc0, pc, inst.src0, error)) {
-				return false;
-			}
+			DecodeScalarSource(ssrc0, pc, inst.src0);
 			ReadLiteralOperands(code, word_index, inst);
-			return true;
+			return;
 		default: break;
 	}
 
-	if (!DecodeScalarSource(ssrc0, pc, inst.src0, error) ||
-	    !DecodeScalarDestination(sdst, pc, inst.dst, error)) {
-		return false;
-	}
+	DecodeScalarSource(ssrc0, pc, inst.src0);
+	DecodeScalarDestination(sdst, pc, inst.dst);
 	inst.src_count = 1;
 	ReadLiteralOperands(code, word_index, inst);
-	return true;
 }
 
-bool DecodeSop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error) {
+void DecodeSop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst) {
 	const uint32_t word   = code[word_index];
 	const uint32_t opcode = (word >> 23u) & 0x7fu;
 	const uint32_t ssrc1  = (word >> 8u) & 0xffu;
@@ -178,17 +171,15 @@ bool DecodeSop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 
 	if (inst.opcode == Opcode::UNSUPPORTED) {
 		SetUnsupported(inst, Family::SOP2, opcode, "SOP2 opcode is not implemented");
-		return true;
+		return;
 	}
 
-	if (!DecodeScalarDestination(sdst, pc, inst.dst, error)) {
-		return false;
-	}
-	return DecodeBinarySources(pc, code, word_index, inst, ssrc0, ssrc1, error);
+	DecodeScalarDestination(sdst, pc, inst.dst);
+	DecodeBinarySources(pc, code, word_index, inst, ssrc0, ssrc1);
 }
 
-bool DecodeSopk(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error) {
+void DecodeSopk(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst) {
 	const uint32_t word   = code[word_index];
 	const uint32_t opcode = (word >> 23u) & 0x1fu;
 	const uint32_t sdst   = (word >> 16u) & 0x7fu;
@@ -207,18 +198,18 @@ bool DecodeSopk(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 
 	if (inst.opcode == Opcode::UNSUPPORTED) {
 		SetUnsupported(inst, Family::SOPK, opcode, "SOPK opcode is not implemented");
-		return true;
+		return;
 	}
 
 	switch (inst.opcode) {
-		case Opcode::S_MOVK_I32: return DecodeScalarDestination(sdst, pc, inst.dst, error);
+		case Opcode::S_MOVK_I32: DecodeScalarDestination(sdst, pc, inst.dst); return;
 		case Opcode::S_WAITCNT: {
 			const uint32_t waitcnt = word & 0xffffu;
 			inst.dst.kind          = OperandKind::Null;
 			inst.src0.signed_val   = static_cast<int32_t>(waitcnt);
 			inst.src0.value        = waitcnt;
 			inst.src_count         = 1;
-			return true;
+			return;
 		}
 		case Opcode::S_SETREG_B32:
 			inst.dst.kind        = OperandKind::Null;
@@ -226,26 +217,25 @@ bool DecodeSopk(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 			inst.src1.value      = word & 0xffffu;
 			inst.src1.signed_val = static_cast<int32_t>(imm);
 			inst.src_count       = 2;
-			return DecodeScalarSource(sdst, pc, inst.src0, error);
+			DecodeScalarSource(sdst, pc, inst.src0);
+			return;
 		default: break;
 	}
 
 	inst.src1 = inst.src0;
-	if (!DecodeScalarSource(sdst, pc, inst.src0, error)) {
-		return false;
-	}
+	DecodeScalarSource(sdst, pc, inst.src0);
 	if (inst.opcode == Opcode::S_ADD_I32 || inst.opcode == Opcode::S_MULK_I32) {
 		inst.src_count = 2;
-		return DecodeScalarDestination(sdst, pc, inst.dst, error);
+		DecodeScalarDestination(sdst, pc, inst.dst);
+		return;
 	}
 
 	inst.dst.kind  = OperandKind::Scc;
 	inst.src_count = 2;
-	return true;
 }
 
-bool DecodeSopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error) {
+void DecodeSopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst) {
 	const uint32_t word   = code[word_index];
 	const uint32_t ssrc1  = (word >> 8u) & 0xffu;
 	const uint32_t ssrc0  = word & 0xffu;
@@ -261,10 +251,10 @@ bool DecodeSopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 
 	if (inst.opcode == Opcode::UNSUPPORTED) {
 		SetUnsupported(inst, Family::SOPC, opcode, "SOPC opcode is not implemented");
-		return true;
+		return;
 	}
 
-	return DecodeBinarySources(pc, code, word_index, inst, ssrc0, ssrc1, error);
+	DecodeBinarySources(pc, code, word_index, inst, ssrc0, ssrc1);
 }
 
 void DecodeSopp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,

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

@@ -5,14 +5,14 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 
-bool DecodeSop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error);
-bool DecodeSop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error);
-bool DecodeSopk(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error);
-bool DecodeSopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error);
+void DecodeSop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst);
+void DecodeSop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst);
+void DecodeSopk(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst);
+void DecodeSopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst);
 void DecodeSopp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
                 Instruction& inst);
 

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

@@ -1,5 +1,6 @@
 #include "graphics/shader/recompiler/frontend/decode/ShaderDecoder.h"
 
+#include "common/assert.h"
 #include "graphics/shader/recompiler/frontend/decode/ExportOps.h"
 #include "graphics/shader/recompiler/frontend/decode/ImageOps.h"
 #include "graphics/shader/recompiler/frontend/decode/MemoryOps.h"
@@ -17,12 +18,6 @@ uint32_t FloatBits(float value) {
 	return std::bit_cast<uint32_t>(value);
 }
 
-void SetError(std::string* error, const char* message) {
-	if (error != nullptr) {
-		*error = message;
-	}
-}
-
 bool HasLiteral(const Instruction& inst) {
 	return inst.src0.kind == OperandKind::LiteralConstant ||
 	       inst.src1.kind == OperandKind::LiteralConstant ||
@@ -216,86 +211,76 @@ const char* ImageDimensionToString(ImageDimension dimension) {
 	}
 }
 
-bool DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand, std::string* error) {
+void DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand) {
 	operand = {};
 
 	if (code <= 105u) {
 		operand.kind = OperandKind::Sgpr;
 		operand.reg  = code;
-		return true;
+		return;
 	}
 	if (code >= 128u && code <= 192u) {
 		operand.kind       = OperandKind::IntegerInlineConstant;
 		operand.signed_val = static_cast<int32_t>(code - 128u);
 		operand.value      = static_cast<uint32_t>(operand.signed_val);
-		return true;
+		return;
 	}
 	if (code >= 193u && code <= 208u) {
 		operand.kind       = OperandKind::IntegerInlineConstant;
 		operand.signed_val = 192 - static_cast<int32_t>(code);
 		operand.value      = static_cast<uint32_t>(operand.signed_val);
-		return true;
+		return;
 	}
 	if (code >= 240u && code <= 247u) {
 		constexpr float values[] = {0.5f, -0.5f, 1.0f, -1.0f, 2.0f, -2.0f, 4.0f, -4.0f};
 		operand.kind             = OperandKind::FloatInlineConstant;
 		operand.float_val        = values[code - 240u];
 		operand.value            = FloatBits(operand.float_val);
-		return true;
+		return;
 	}
 	if (code >= 256u && code <= 511u) {
 		DecodeVectorGpr(code - 256u, operand);
-		return true;
+		return;
 	}
 
 	switch (code) {
-		case 106u: operand.kind = OperandKind::VccLo; return true;
-		case 107u: operand.kind = OperandKind::VccHi; return true;
-		case 124u: operand.kind = OperandKind::M0; return true;
-		case 125u: operand.kind = OperandKind::Null; return true;
-		case 126u: operand.kind = OperandKind::ExecLo; return true;
-		case 127u: operand.kind = OperandKind::ExecHi; return true;
-		case 239u: operand.kind = OperandKind::PopsExitingWaveId; return true;
+		case 106u: operand.kind = OperandKind::VccLo; return;
+		case 107u: operand.kind = OperandKind::VccHi; return;
+		case 124u: operand.kind = OperandKind::M0; return;
+		case 125u: operand.kind = OperandKind::Null; return;
+		case 126u: operand.kind = OperandKind::ExecLo; return;
+		case 127u: operand.kind = OperandKind::ExecHi; return;
+		case 239u: operand.kind = OperandKind::PopsExitingWaveId; return;
 		case 248u:
 			operand.kind      = OperandKind::FloatInlineConstant;
 			operand.float_val = 0.15915494309189535f;
 			operand.value     = FloatBits(operand.float_val);
-			return true;
-		case 251u: operand.kind = OperandKind::VccZ; return true;
-		case 252u: operand.kind = OperandKind::ExecZ; return true;
-		case 253u: operand.kind = OperandKind::Scc; return true;
-		case 255u: operand.kind = OperandKind::LiteralConstant; return true;
-		default:
-			if (error != nullptr) {
-				*error = fmt::format("unsupported scalar source operand 0x{:08x} at pc 0x{:08x}",
-				                     code, pc);
-			}
-			return false;
+			return;
+		case 251u: operand.kind = OperandKind::VccZ; return;
+		case 252u: operand.kind = OperandKind::ExecZ; return;
+		case 253u: operand.kind = OperandKind::Scc; return;
+		case 255u: operand.kind = OperandKind::LiteralConstant; return;
+		default: EXIT("unsupported scalar source operand 0x%08x at pc 0x%08x", code, pc);
 	}
 }
 
-bool DecodeScalarDestination(uint32_t code, uint32_t pc, Operand& operand, std::string* error) {
+void DecodeScalarDestination(uint32_t code, uint32_t pc, Operand& operand) {
 	operand = {};
 
 	if (code <= 105u) {
 		operand.kind = OperandKind::Sgpr;
 		operand.reg  = code;
-		return true;
+		return;
 	}
 
 	switch (code) {
-		case 106u: operand.kind = OperandKind::VccLo; return true;
-		case 107u: operand.kind = OperandKind::VccHi; return true;
-		case 124u: operand.kind = OperandKind::M0; return true;
-		case 125u: operand.kind = OperandKind::Null; return true;
-		case 126u: operand.kind = OperandKind::ExecLo; return true;
-		case 127u: operand.kind = OperandKind::ExecHi; return true;
-		default:
-			if (error != nullptr) {
-				*error = fmt::format(
-				    "unsupported scalar destination operand 0x{:08x} at pc 0x{:08x}", code, pc);
-			}
-			return false;
+		case 106u: operand.kind = OperandKind::VccLo; return;
+		case 107u: operand.kind = OperandKind::VccHi; return;
+		case 124u: operand.kind = OperandKind::M0; return;
+		case 125u: operand.kind = OperandKind::Null; return;
+		case 126u: operand.kind = OperandKind::ExecLo; return;
+		case 127u: operand.kind = OperandKind::ExecHi; return;
+		default: EXIT("unsupported scalar destination operand 0x%08x at pc 0x%08x", code, pc);
 	}
 }
 
@@ -368,39 +353,33 @@ Family GetInstructionFamily(uint32_t word) {
 	}
 }
 
-bool DecodeInstruction(std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                       std::string* error) {
+void DecodeInstruction(std::span<const uint32_t> code, uint32_t word_index, Instruction& inst) {
 	const uint32_t pc = word_index * sizeof(uint32_t);
 	switch (GetInstructionFamily(code[word_index])) {
-		case Family::SOP1: return DecodeSop1(pc, code, word_index, inst, error);
-		case Family::SOP2: return DecodeSop2(pc, code, word_index, inst, error);
-		case Family::SOPK: return DecodeSopk(pc, code, word_index, inst, error);
-		case Family::SOPC: return DecodeSopc(pc, code, word_index, inst, error);
-		case Family::SOPP: DecodeSopp(pc, code, word_index, inst); return true;
-		case Family::VOP1: return DecodeVop1(pc, code, word_index, inst, error);
-		case Family::VOP2: return DecodeVop2(pc, code, word_index, inst, error);
-		case Family::VOP3: return DecodeVop3(pc, code, word_index, inst, error);
-		case Family::VOP3P: return DecodeVop3p(pc, code, word_index, inst, error);
-		case Family::VOPC: return DecodeVopc(pc, code, word_index, inst, error);
-		case Family::VINTRP: DecodeVintrp(pc, code, word_index, inst); return true;
-		case Family::SMEM: DecodeSmem(pc, code, word_index, inst); return true;
-		case Family::MUBUF: DecodeMubuf(pc, code, word_index, inst); return true;
-		case Family::MTBUF: DecodeMtbuf(pc, code, word_index, inst); return true;
-		case Family::FLAT: DecodeFlat(pc, code, word_index, inst); return true;
-		case Family::DS: DecodeDs(pc, code, word_index, inst); return true;
-		case Family::MIMG: DecodeMimg(pc, code, word_index, inst); return true;
-		case Family::EXP: DecodeExp(pc, code, word_index, inst); return true;
+		case Family::SOP1: DecodeSop1(pc, code, word_index, inst); return;
+		case Family::SOP2: DecodeSop2(pc, code, word_index, inst); return;
+		case Family::SOPK: DecodeSopk(pc, code, word_index, inst); return;
+		case Family::SOPC: DecodeSopc(pc, code, word_index, inst); return;
+		case Family::SOPP: DecodeSopp(pc, code, word_index, inst); return;
+		case Family::VOP1: DecodeVop1(pc, code, word_index, inst); return;
+		case Family::VOP2: DecodeVop2(pc, code, word_index, inst); return;
+		case Family::VOP3: DecodeVop3(pc, code, word_index, inst); return;
+		case Family::VOP3P: DecodeVop3p(pc, code, word_index, inst); return;
+		case Family::VOPC: DecodeVopc(pc, code, word_index, inst); return;
+		case Family::VINTRP: DecodeVintrp(pc, code, word_index, inst); return;
+		case Family::SMEM: DecodeSmem(pc, code, word_index, inst); return;
+		case Family::MUBUF: DecodeMubuf(pc, code, word_index, inst); return;
+		case Family::MTBUF: DecodeMtbuf(pc, code, word_index, inst); return;
+		case Family::FLAT: DecodeFlat(pc, code, word_index, inst); return;
+		case Family::DS: DecodeDs(pc, code, word_index, inst); return;
+		case Family::MIMG: DecodeMimg(pc, code, word_index, inst); return;
+		case Family::EXP: DecodeExp(pc, code, word_index, inst); return;
 		default:
-			if (error != nullptr) {
-				*error =
-				    fmt::format("unknown RDNA2 instruction family at pc 0x{:08x}, raw=0x{:08x}", pc,
-				                code[word_index]);
-			}
-			return false;
+			EXIT("unknown RDNA2 instruction family at pc 0x%08x, raw=0x%08x", pc, code[word_index]);
 	}
 }
 
-bool DecodeProgram(std::span<const uint32_t> code, Program& program, std::string* error) {
+void DecodeProgram(std::span<const uint32_t> code, Program& program) {
 	program.instructions.clear();
 	program.instructions.reserve(code.size());
 	program.code = code;
@@ -408,10 +387,7 @@ bool DecodeProgram(std::span<const uint32_t> code, Program& program, std::string
 	std::vector<bool> branch_targets;
 	for (uint32_t word_index = 0; word_index < code.size();) {
 		program.instructions.emplace_back();
-		if (!DecodeInstruction(code, word_index, program.instructions.back(), error)) {
-			program.instructions.pop_back();
-			return false;
-		}
+		DecodeInstruction(code, word_index, program.instructions.back());
 
 		const auto& inst = program.instructions.back();
 		word_index += inst.word_count;
@@ -425,12 +401,11 @@ bool DecodeProgram(std::span<const uint32_t> code, Program& program, std::string
 		}
 		if (inst.opcode == Opcode::S_ENDPGM &&
 		    (word_index >= code.size() || branch_targets.empty() || !branch_targets[word_index])) {
-			return true;
+			return;
 		}
 	}
 
-	SetError(error, "shader decode reached the code boundary before S_ENDPGM");
-	return false;
+	EXIT("shader decode reached the code boundary before S_ENDPGM");
 }
 
 std::string OperandToString(const Operand& operand) {

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

@@ -351,6 +351,7 @@ enum class Opcode {
 	V_CMP_GE_I32,
 	V_CMP_T_I32,
 	V_CMP_CLASS_F32,
+	V_CMPX_CLASS_F32,
 	V_CMP_LT_I16,
 	V_CMP_EQ_I16,
 	V_CMP_LE_I16,
@@ -631,6 +632,7 @@ struct Operand {
 	uint32_t dpp_ctrl           = 0;
 	uint32_t dpp_row_mask       = 0xf;
 	uint32_t dpp_bank_mask      = 0xf;
+	bool     explicit_sdwa_dst  = false;
 	bool     sdwa_sext          = false;
 	bool     dpp_fetch_inactive = false;
 	bool     dpp_bound_ctrl     = false;
@@ -702,12 +704,11 @@ struct Program {
 // Code spans are trusted to contain complete instructions, valid branch targets, and 32-bit PCs.
 Family GetInstructionFamily(uint32_t word);
 // The output object must be freshly initialized.
-bool DecodeInstruction(std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                       std::string* error);
-bool DecodeProgram(std::span<const uint32_t> code, Program& program, std::string* error);
+void DecodeInstruction(std::span<const uint32_t> code, uint32_t word_index, Instruction& inst);
+void DecodeProgram(std::span<const uint32_t> code, Program& program);
 
-bool DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand, std::string* error);
-bool DecodeScalarDestination(uint32_t code, uint32_t pc, Operand& operand, std::string* error);
+void DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand);
+void DecodeScalarDestination(uint32_t code, uint32_t pc, Operand& operand);
 void DecodeVectorGpr(uint32_t reg, Operand& operand);
 void ReadLiteralOperands(std::span<const uint32_t> code, uint32_t word_index, Instruction& inst);
 void SetRawWords(Instruction& inst, std::span<const uint32_t> code, uint32_t word_index,

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

@@ -5,16 +5,16 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 
-bool DecodeVop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error);
-bool DecodeVop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error);
-bool DecodeVopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error);
-bool DecodeVop3(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index, Instruction& inst,
-                std::string* error);
-bool DecodeVop3p(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
-                 Instruction& inst, std::string* error);
+void DecodeVop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst);
+void DecodeVop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst);
+void DecodeVopc(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst);
+void DecodeVop3(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                Instruction& inst);
+void DecodeVop3p(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
+                 Instruction& inst);
 void DecodeVintrp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
                   Instruction& inst);
 

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

@@ -1,9 +1,8 @@
+#include "common/assert.h"
 #include "graphics/shader/recompiler/frontend/translate/Translator.h"
-#include "graphics/shader/shader.h"
 
 #include <algorithm>
 #include <array>
-#include <fmt/format.h>
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
 
@@ -47,14 +46,13 @@ IR::ExportFlags Translator::AddExportInfo(const Decoder::Instruction& inst) {
 	return {.index = index, .pc = inst.pc};
 }
 
-bool Translator::TranslateEmbeddedFetch(const Decoder::Instruction& inst, uint32_t attribute,
+void Translator::TranslateEmbeddedFetch(const Decoder::Instruction& inst, uint32_t attribute,
                                         uint32_t component_count) {
 	for (uint32_t component = 0; component < component_count; component++) {
 		const auto value =
 		    ir.Emit(IR::ValueOpcode::GetAttribute, {IR::Value(attribute), IR::Value(component)});
 		WriteOperand(OffsetOperand(inst.dst, component), value);
 	}
-	return true;
 }
 
 void Translator::V_INTERP_P1_F32() {}
@@ -65,25 +63,20 @@ void Translator::V_INTERP_P2_F32(const Decoder::Instruction& inst) {
 	WriteOperand(inst.dst, value);
 }
 
-bool Translator::V_INTERP_MOV_F32(const Decoder::Instruction& inst, std::string* error) {
+void Translator::V_INTERP_MOV_F32(const Decoder::Instruction& inst) {
 	if (inst.src0.value >= 3u) {
-		return ShaderError::Fail(error,
-		                         fmt::format("v_interp_mov_f32 mode {} is reserved at pc 0x{:08x}",
-		                                     inst.src0.value, inst.pc));
+		EXIT("v_interp_mov_f32 mode %u is reserved at pc 0x%08x", inst.src0.value, inst.pc);
 	}
 	const auto value = ir.Emit(
 	    IR::ValueOpcode::GetInterpolationParameter,
 	    {IR::Value(inst.src1.value), IR::Value(inst.src2.value), IR::Value(inst.src0.value)});
 	WriteOperand(inst.dst, value);
-	return true;
 }
 
-bool Translator::EXP(const Decoder::Instruction& inst, std::string* error) {
+void Translator::EXP(const Decoder::Instruction& inst) {
 	uint32_t index = 0;
 	if (ExportTargetKindFromTarget(inst.exp.target, index) == IR::ExportTargetKind::Unknown) {
-		return ShaderError::Fail(error,
-		                         fmt::format("unsupported EXP target 0x{:02x} at pc 0x{:08x}",
-		                                     inst.exp.target, inst.pc));
+		EXIT("unsupported EXP target 0x%02x at pc 0x%08x", inst.exp.target, inst.pc);
 	}
 	std::array<IR::Value, 4> components {IR::Value(0u), IR::Value(0u), IR::Value(0u),
 	                                     IR::Value(0u)};
@@ -93,14 +86,13 @@ bool Translator::EXP(const Decoder::Instruction& inst, std::string* error) {
 	const auto data = ir.Emit(IR::ValueOpcode::CompositeConstructU32x4,
 	                          {components[0], components[1], components[2], components[3]});
 	ir.Emit(IR::ValueOpcode::SetAttribute, {data, ir.GetExec()}, AddExportInfo(inst));
-	return true;
 }
 
-bool Translator::EmitInterpolation(const Decoder::Instruction& inst, std::string* error) {
+bool Translator::EmitInterpolation(const Decoder::Instruction& inst) {
 	switch (inst.opcode) {
 		case Decoder::Opcode::V_INTERP_P1_F32: V_INTERP_P1_F32(); return true;
 		case Decoder::Opcode::V_INTERP_P2_F32: V_INTERP_P2_F32(inst); return true;
-		case Decoder::Opcode::V_INTERP_MOV_F32: return V_INTERP_MOV_F32(inst, error);
+		case Decoder::Opcode::V_INTERP_MOV_F32: V_INTERP_MOV_F32(inst); return true;
 		default: return false;
 	}
 }

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

@@ -12,6 +12,7 @@ void Translator::EmitCompareResult(const Decoder::Instruction& inst, IR::U1 valu
 	if (cmpx) {
 		const auto mask = BallotMask(masked);
 		ir.SetExec(masked);
+		ir.SetExecMaskTag(IR::U1(IR::Value(true)));
 		ir.SetExecLo(mask[0]);
 		ir.SetExecHi(mask[1]);
 		return;
@@ -55,11 +56,11 @@ void Translator::EmitFloatOrderedCompare(const Decoder::Instruction& inst, bool
 	EmitCompareResult(inst, ordered ? ir.LogicalNot(unordered) : unordered, false, false);
 }
 
-void Translator::V_CMP_CLASS_F32(const Decoder::Instruction& inst) {
+void Translator::EmitFloatClassCompare(const Decoder::Instruction& inst, bool cmpx) {
 	const auto value = ReadOperand(inst.src0, IR::Type::F32);
 	const auto mask  = ReadOperand(inst.src1, IR::Type::U32);
 	EmitCompareResult(inst, IR::U1(ir.Emit(IR::ValueOpcode::FPCmpClass32, {value, mask})), false,
-	                  false);
+	                  cmpx);
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::Frontend

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

@@ -1,3 +1,4 @@
+#include "common/assert.h"
 #include "graphics/shader/recompiler/frontend/translate/Translator.h"
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
@@ -23,18 +24,28 @@ Decoder::Operand ConditionOperand(Decoder::OperandKind kind) {
 
 void Translator::S_SAVEEXEC(const Decoder::Instruction& inst, IR::ValueOpcode operation,
                             bool negate_exec, bool negate_source, bool write_64) {
+	if (write_64) {
+		const auto exec_operand  = ConditionOperand(Decoder::OperandKind::ExecLo);
+		const auto old           = ReadScalarU64(exec_operand);
+		const auto bit_operation = operation == IR::ValueOpcode::LogicalAnd
+		                               ? IR::ValueOpcode::BitwiseAnd32
+		                               : IR::ValueOpcode::BitwiseOr32;
+		const auto result = BinaryScalarU64(old, ReadScalarU64(inst.src0), operation, bit_operation,
+		                                    negate_exec, negate_source, false);
+		WriteScalarU64(inst.dst, old);
+		WriteScalarU64(exec_operand, result);
+		ir.SetScc(ScalarU64NonZero(result));
+		return;
+	}
 	const auto old    = ir.GetExec();
 	const auto src    = ReadMask(inst.src0);
 	const auto lhs    = negate_exec ? ir.LogicalNot(old) : old;
 	const auto rhs    = negate_source ? ir.LogicalNot(src) : src;
 	const auto result = IR::U1(ir.Emit(operation, {lhs, rhs}));
-	if (write_64) {
-		WriteMask64(inst.dst, old);
-	} else {
-		WriteMask(inst.dst, old);
-	}
+	WriteMask(inst.dst, old);
 	const auto mask = BallotMask(result);
 	ir.SetExec(result);
+	ir.SetExecMaskTag(IR::U1(IR::Value(true)));
 	ir.SetExecLo(mask[0]);
 	ir.SetExecHi(mask[1]);
 	ir.SetScc(result);
@@ -153,12 +164,12 @@ void Translator::S_INST_PREFETCH() {
 	ir.Emit(IR::ValueOpcode::InstPrefetch);
 }
 
-bool Translator::S_GETPC_B64(const Decoder::Instruction& inst, std::string* error) {
+void Translator::S_GETPC_B64(const Decoder::Instruction& inst) {
 	const auto base    = IR::U64(ir.Emit(IR::ValueOpcode::GetShaderBase));
 	const auto address = IR::U64(
 	    ir.Emit(IR::ValueOpcode::IAdd64, {base, IR::Value(static_cast<uint64_t>(inst.pc) + 4u)}));
 	if (inst.dst.kind == Decoder::OperandKind::Null) {
-		return true;
+		return;
 	}
 	auto high = PlainOperand(inst.dst);
 	switch (inst.dst.kind) {
@@ -169,25 +180,19 @@ bool Translator::S_GETPC_B64(const Decoder::Instruction& inst, std::string* erro
 				high.kind = Decoder::OperandKind::VccLo;
 				high.reg  = 0u;
 			} else {
-				if (error != nullptr) {
-					*error = "S_GETPC_B64 destination does not name a valid scalar pair";
-				}
-				return false;
+				EXIT("S_GETPC_B64 destination does not name a valid scalar pair at pc 0x%08x",
+				     inst.pc);
 			}
 			break;
 		case Decoder::OperandKind::VccLo: high.kind = Decoder::OperandKind::VccHi; break;
 		case Decoder::OperandKind::M0: high.kind = Decoder::OperandKind::Null; break;
 		case Decoder::OperandKind::ExecLo: high.kind = Decoder::OperandKind::ExecHi; break;
 		default:
-			if (error != nullptr) {
-				*error = "S_GETPC_B64 destination does not name a valid scalar pair";
-			}
-			return false;
+			EXIT("S_GETPC_B64 destination does not name a valid scalar pair at pc 0x%08x", inst.pc);
 	}
 	const auto words = ExtractU64(address);
 	WriteOperand(inst.dst, words[0]);
 	WriteOperand(high, words[1]);
-	return true;
 }
 
 void Translator::S_CSELECT_B32(const Decoder::Instruction& inst) {
@@ -196,30 +201,21 @@ void Translator::S_CSELECT_B32(const Decoder::Instruction& inst) {
 }
 
 void Translator::S_CSELECT_B64(const Decoder::Instruction& inst) {
-	const auto condition     = ir.GetScc();
-	const auto lhs           = ReadU32Pair(inst.src0);
-	const auto rhs           = ReadU32Pair(inst.src1);
-	const auto selected_mask = IR::U1(
-	    ir.Emit(IR::ValueOpcode::SelectU1, {condition, ReadMask(inst.src0), ReadMask(inst.src1)}));
-	const auto selected_mask_valid =
-	    IR::U1(ir.Emit(IR::ValueOpcode::SelectU1,
-	                   {condition, ReadMaskValid(inst.src0), ReadMaskValid(inst.src1)}));
-	if (IsExecOrVcc(inst.dst)) {
-		WriteMask64(inst.dst, selected_mask);
-		return;
-	}
-	WriteU32Pair(inst.dst,
-	             {ir.Select(condition, lhs[0], rhs[0]), ir.Select(condition, lhs[1], rhs[1])});
-	if (inst.dst.kind == Decoder::OperandKind::Sgpr) {
-		const auto dst = static_cast<IR::ScalarReg>(inst.dst.reg);
-		ir.SetThreadBitScalarReg(dst, selected_mask);
-		ir.SetScalarMaskTag(dst, selected_mask_valid);
-	}
+	const auto condition = ir.GetScc();
+	const auto lhs       = ReadScalarU64(inst.src0);
+	const auto rhs       = ReadScalarU64(inst.src1);
+	WriteScalarU64(inst.dst, SelectScalarU64(condition, lhs, rhs));
 }
 
 void Translator::MOV_B32(const Decoder::Instruction& inst, bool apply_float_modifiers) {
+	if (inst.src0.kind == Decoder::OperandKind::Sgpr &&
+	    inst.dst.kind == Decoder::OperandKind::Sgpr && inst.src0.reg == inst.dst.reg) {
+		return;
+	}
 	if (IsExecOrVcc(inst.src0) && IsExecOrVcc(inst.dst)) {
-		WriteMask(inst.dst, ReadMask(inst.src0));
+		if (inst.src0.kind != inst.dst.kind) {
+			WriteRawU32(inst.dst, ReadRawU32(inst.src0));
+		}
 	} else if (apply_float_modifiers && (inst.src0.negate || inst.src0.absolute)) {
 		WriteOperand(DestinationOperand(inst), ReadOperand(inst.src0, IR::Type::F32));
 	} else {
@@ -228,67 +224,28 @@ void Translator::MOV_B32(const Decoder::Instruction& inst, bool apply_float_modi
 }
 
 void Translator::S_MOV_B64(const Decoder::Instruction& inst) {
-	if (IsExecOrVcc(inst.dst) || IsExecOrVcc(inst.src0)) {
-		WriteMask64(inst.dst, ReadMask(inst.src0));
-		return;
-	}
-	const bool scalar_copy =
-	    inst.dst.kind == Decoder::OperandKind::Sgpr && inst.src0.kind == Decoder::OperandKind::Sgpr;
-	IR::U1 source_mask;
-	IR::U1 source_mask_valid;
-	if (scalar_copy) {
-		source_mask       = ir.GetThreadBitScalarReg(static_cast<IR::ScalarReg>(inst.src0.reg));
-		source_mask_valid = ir.GetScalarMaskTag(static_cast<IR::ScalarReg>(inst.src0.reg));
-	}
-	WriteU32Pair(inst.dst, ReadU32Pair(inst.src0));
-	if (scalar_copy) {
-		ir.SetThreadBitScalarReg(static_cast<IR::ScalarReg>(inst.dst.reg), source_mask);
-		ir.SetScalarMaskTag(static_cast<IR::ScalarReg>(inst.dst.reg), source_mask_valid);
-	}
+	WriteScalarU64(inst.dst, ReadScalarU64(inst.src0));
 }
 
 void Translator::S_WQM_B64(const Decoder::Instruction& inst) {
-	if (!IsExecOrVcc(inst.dst) && !IsExecOrVcc(inst.src0)) {
-		const auto mask_valid = ReadMaskValid(inst.src0);
-		const auto invocation_result =
-		    IR::U1(ir.Emit(IR::ValueOpcode::WqmMask, {ReadMask(inst.src0)}));
-		const auto result =
-		    IR::U64(ir.Emit(IR::ValueOpcode::WqmU64, {ReadOperand(inst.src0, IR::Type::U64)}));
-		WriteOperand(DestinationOperand(inst), result);
-		if (inst.dst.kind == Decoder::OperandKind::Sgpr) {
-			const auto dst = static_cast<IR::ScalarReg>(inst.dst.reg);
-			ir.SetThreadBitScalarReg(dst, invocation_result);
-			ir.SetScalarMaskTag(dst, mask_valid);
-			const auto raw_nonzero =
-			    IR::U1(ir.Emit(IR::ValueOpcode::INotEqual64, {result, IR::Value(uint64_t {0})}));
-			ir.SetScc(IR::U1(
-			    ir.Emit(IR::ValueOpcode::SelectU1, {mask_valid, invocation_result, raw_nonzero})));
-		} else {
-			ir.SetScc(
-			    IR::U1(ir.Emit(IR::ValueOpcode::INotEqual64, {result, IR::Value(uint64_t {0})})));
-		}
-		return;
-	}
-	const auto result = IR::U1(ir.Emit(IR::ValueOpcode::WqmMask, {ReadMask(inst.src0)}));
-	WriteMask64(inst.dst, result);
-	ir.SetScc(result);
+	const auto source = ReadScalarU64(inst.src0);
+	const auto raw    = ExtractU64(
+	    IR::U64(ir.Emit(IR::ValueOpcode::WqmU64, {ir.ConstructU64(source.raw[0], source.raw[1])})));
+	const auto invocation = IR::U1(ir.Emit(IR::ValueOpcode::WqmMask, {source.invocation}));
+	const auto result     = MakeScalarU64Result(raw, invocation, source.mask_valid);
+	WriteScalarU64(inst.dst, result);
+	ir.SetScc(ScalarU64NonZero(result));
 }
 
-bool Translator::V_MOVRELS_B32(const Decoder::Instruction& inst, std::string* error) {
+void Translator::V_MOVRELS_B32(const Decoder::Instruction& inst) {
 	if (inst.dst.kind != Decoder::OperandKind::Vgpr ||
 	    inst.src0.kind != Decoder::OperandKind::Vgpr) {
-		if (error != nullptr) {
-			*error = "V_MOVRELS_B32 requires VGPR source and destination";
-		}
-		return false;
+		EXIT("V_MOVRELS_B32 requires VGPR source and destination at pc 0x%08x", inst.pc);
 	}
 	if (inst.dst.sdwa_sel != 6u || inst.dst.omod != 0u || inst.dst.clamp ||
 	    inst.src0.sdwa_sel != 6u || inst.src0.sdwa_sext || inst.src0.negate || inst.src0.absolute ||
 	    inst.src0.dpp) {
-		if (error != nullptr) {
-			*error = "V_MOVRELS_B32 modifiers are not implemented";
-		}
-		return false;
+		EXIT("V_MOVRELS_B32 modifiers are not implemented at pc 0x%08x", inst.pc);
 	}
 	const auto base     = inst.src0.reg;
 	const auto m0       = ir.BitwiseAnd(ReadU32(ConditionOperand(Decoder::OperandKind::M0)),
@@ -299,23 +256,16 @@ bool Translator::V_MOVRELS_B32(const Decoder::Instruction& inst, std::string* er
 		selected = ir.Select(match, ir.GetVectorReg(static_cast<IR::VectorReg>(index)), selected);
 	}
 	WriteOperand(DestinationOperand(inst), selected);
-	return true;
 }
 
-bool Translator::V_MOVRELD_B32(const Decoder::Instruction& inst, std::string* error) {
+void Translator::V_MOVRELD_B32(const Decoder::Instruction& inst) {
 	if (inst.dst.kind != Decoder::OperandKind::Vgpr) {
-		if (error != nullptr) {
-			*error = "V_MOVRELD_B32 requires VGPR destination";
-		}
-		return false;
+		EXIT("V_MOVRELD_B32 requires VGPR destination at pc 0x%08x", inst.pc);
 	}
 	if (inst.dst.sdwa_sel != 6u || inst.dst.omod != 0u || inst.dst.clamp ||
 	    inst.src0.sdwa_sel != 6u || inst.src0.sdwa_sext || inst.src0.negate || inst.src0.absolute ||
 	    inst.src0.dpp) {
-		if (error != nullptr) {
-			*error = "V_MOVRELD_B32 modifiers are not implemented";
-		}
-		return false;
+		EXIT("V_MOVRELD_B32 modifiers are not implemented at pc 0x%08x", inst.pc);
 	}
 	const auto base  = inst.dst.reg;
 	const auto value = ReadU32(inst.src0);
@@ -327,7 +277,6 @@ bool Translator::V_MOVRELD_B32(const Decoder::Instruction& inst, std::string* er
 		const auto write = ir.LogicalAnd(ir.GetExec(), match);
 		ir.SetVectorReg(reg, ir.Select(write, value, ir.GetVectorReg(reg)));
 	}
-	return true;
 }
 
 void Translator::V_READFIRSTLANE_B32(const Decoder::Instruction& inst) {

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

@@ -141,8 +141,10 @@ void Translator::V_CVT_OFF_F32_I4(const Decoder::Instruction& inst) {
 }
 
 void Translator::V_CVT_PKRTZ_F16_F32(const Decoder::Instruction& inst) {
-	const auto lhs = ReadOperand(SourceAt(inst, 0), IR::Type::F32);
-	const auto rhs = ReadOperand(SourceAt(inst, 1), IR::Type::F32);
+	const auto lhs = ApplyF32ResultModifiers(
+	    inst.dst, IR::F32(ReadOperand(SourceAt(inst, 0), IR::Type::F32)));
+	const auto rhs = ApplyF32ResultModifiers(
+	    inst.dst, IR::F32(ReadOperand(SourceAt(inst, 1), IR::Type::F32)));
 	WriteOperand(DestinationOperand(inst), ir.Emit(IR::ValueOpcode::PackFloat2x16Rtz, {lhs, rhs}));
 }
 

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

@@ -1,23 +1,18 @@
+#include "common/assert.h"
 #include "graphics/shader/recompiler/frontend/translate/Translator.h"
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
 
-bool Translator::TranslateInstruction(const Decoder::Instruction& inst, std::string* error) {
+void Translator::TranslateInstruction(const Decoder::Instruction& inst) {
 	current_opcode = inst.opcode;
 	current_pc     = inst.pc;
 
 	switch (inst.opcode) {
 		case Decoder::Opcode::UNKNOWN:
 		case Decoder::Opcode::COUNT:
-			if (error != nullptr) {
-				*error = "decoded opcode has no IR translation";
-			}
-			return false;
+			EXIT("decoded opcode has no IR translation at pc 0x%08x", inst.pc);
 		case Decoder::Opcode::UNSUPPORTED:
-			if (error != nullptr) {
-				*error = "unsupported decoded instruction: " + Decoder::InstructionToString(inst);
-			}
-			return false;
+			EXIT("unsupported decoded instruction: %s", Decoder::InstructionToString(inst).c_str());
 		default: break;
 	}
 
@@ -27,27 +22,30 @@ bool Translator::TranslateInstruction(const Decoder::Instruction& inst, std::str
 		case Decoder::Family::SOP2:
 		case Decoder::Family::SOPK:
 		case Decoder::Family::SOPC:
-		case Decoder::Family::SOPP: translated = EmitScalar(inst, error); break;
+		case Decoder::Family::SOPP: translated = EmitScalar(inst); break;
 		case Decoder::Family::VOP1:
 		case Decoder::Family::VOP2:
 		case Decoder::Family::VOP3:
 		case Decoder::Family::VOP3P:
-		case Decoder::Family::VOPC: translated = EmitVector(inst, error); break;
+		case Decoder::Family::VOPC: translated = EmitVector(inst); break;
 		case Decoder::Family::SMEM:
 		case Decoder::Family::MUBUF:
 		case Decoder::Family::MTBUF:
 		case Decoder::Family::FLAT:
 		case Decoder::Family::DS:
-		case Decoder::Family::MIMG: translated = EmitMemory(inst, error); break;
-		case Decoder::Family::VINTRP: translated = EmitInterpolation(inst, error); break;
-		case Decoder::Family::EXP: translated = EXP(inst, error); break;
+		case Decoder::Family::MIMG: translated = EmitMemory(inst); break;
+		case Decoder::Family::VINTRP: translated = EmitInterpolation(inst); break;
+		case Decoder::Family::EXP:
+			EXP(inst);
+			translated = true;
+			break;
 		default: break;
 	}
 
-	if (!translated && error != nullptr && error->empty()) {
-		*error = "decoded opcode has no IR translation";
+	if (!translated) {
+		EXIT("opcode %s at pc 0x%08x has no IR translation",
+		     Decoder::InstructionToString(inst).c_str(), inst.pc);
 	}
-	return translated;
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::Frontend

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

@@ -77,68 +77,15 @@ bool Translator::PackedInteger16MinMax(const Decoder::Instruction& inst, IR::Val
 	return true;
 }
 
-IR::U1 Translator::U64MaskBinary(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
-                                 bool negate_rhs, bool negate_result) {
-	const auto lhs = ReadMask(inst.src0);
-	auto       rhs = ReadMask(inst.src1);
-	if (negate_rhs) {
-		rhs = ir.LogicalNot(rhs);
-	}
-	auto result = IR::U1(ir.Emit(opcode, {lhs, rhs}));
-	return negate_result ? ir.LogicalNot(result) : result;
-}
-
 bool Translator::S_U64_MASK(const Decoder::Instruction& inst, IR::ValueOpcode logical_opcode,
                             IR::ValueOpcode bit_opcode, bool negate_rhs, bool negate_result,
                             bool unary) {
-	const auto invocation_result =
-	    unary ? ir.LogicalNot(ReadMask(inst.src0))
-	          : U64MaskBinary(inst, logical_opcode, negate_rhs, negate_result);
-	const auto is_exec_or_vcc = [](const Decoder::Operand& operand) {
-		switch (operand.kind) {
-			case Decoder::OperandKind::ExecLo:
-			case Decoder::OperandKind::ExecHi:
-			case Decoder::OperandKind::VccLo:
-			case Decoder::OperandKind::VccHi: return true;
-			default: return false;
-		}
-	};
-	if (is_exec_or_vcc(inst.dst) || is_exec_or_vcc(inst.src0) ||
-	    (inst.src_count > 1u && is_exec_or_vcc(inst.src1))) {
-		WriteMask64(inst.dst, invocation_result);
-		ir.SetScc(invocation_result);
-		return true;
-	}
-
-	const auto lhs        = ReadU32Pair(inst.src0);
-	auto       mask_valid = ReadMaskValid(inst.src0);
-	if (inst.src_count > 1u) {
-		mask_valid = ir.LogicalAnd(mask_valid, ReadMaskValid(inst.src1));
-	}
-	std::array<IR::U32, 2> result;
-	if (unary) {
-		result = {ir.BitwiseNot(lhs[0]), ir.BitwiseNot(lhs[1])};
-	} else {
-		const auto rhs = ReadU32Pair(inst.src1);
-		for (uint32_t component = 0; component < 2u; component++) {
-			auto value = ir.Emit(
-			    bit_opcode, {lhs[component], negate_rhs ? IR::Value(ir.BitwiseNot(rhs[component]))
-			                                            : IR::Value(rhs[component])});
-			result[component] = negate_result ? ir.BitwiseNot(IR::U32(value)) : IR::U32(value);
-		}
-	}
-	WriteU32Pair(inst.dst, result);
-	if (inst.dst.kind == Decoder::OperandKind::Sgpr) {
-		const auto dst = static_cast<IR::ScalarReg>(inst.dst.reg);
-		ir.SetThreadBitScalarReg(dst, invocation_result);
-		ir.SetScalarMaskTag(dst, mask_valid);
-		const auto raw_nonzero =
-		    ir.INotEqual(ir.BitwiseOr(result[0], result[1]), IR::U32(IR::Value(0u)));
-		ir.SetScc(IR::U1(
-		    ir.Emit(IR::ValueOpcode::SelectU1, {mask_valid, invocation_result, raw_nonzero})));
-	} else {
-		ir.SetScc(ir.INotEqual(ir.BitwiseOr(result[0], result[1]), IR::U32(IR::Value(0u))));
-	}
+	const auto lhs    = ReadScalarU64(inst.src0);
+	const auto result = unary ? NotScalarU64(lhs)
+	                          : BinaryScalarU64(lhs, ReadScalarU64(inst.src1), logical_opcode,
+	                                            bit_opcode, false, negate_rhs, negate_result);
+	WriteScalarU64(inst.dst, result);
+	ir.SetScc(ScalarU64NonZero(result));
 	return true;
 }
 

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

@@ -1,9 +1,7 @@
-#include "common/magicEnum.h"
 #include "graphics/shader/recompiler/frontend/translate/Translator.h"
 
 #include <algorithm>
 #include <array>
-#include <fmt/format.h>
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
 
@@ -37,6 +35,7 @@ Decoder::Operand OffsetDecodedRegister(const Decoder::Operand& operand, uint32_t
 	result.dpp_ctrl           = 0;
 	result.dpp_row_mask       = 0xf;
 	result.dpp_bank_mask      = 0xf;
+	result.explicit_sdwa_dst  = false;
 	result.dpp_fetch_inactive = false;
 	result.dpp_bound_ctrl     = false;
 	result.dpp                = false;
@@ -56,11 +55,14 @@ uint32_t ResourceIndexFromOperand(const Decoder::Operand& operand) {
 	}
 }
 
-uint32_t RawScalarLoadBase(const Decoder::Operand& operand) {
-	if (operand.kind == Decoder::OperandKind::Sgpr) {
-		return operand.reg;
+Decoder::Operand RawScalarLoadBase(const Decoder::Operand& operand) {
+	if (operand.kind != Decoder::OperandKind::M0) {
+		return operand;
 	}
-	return operand.kind == Decoder::OperandKind::VccLo ? 106u : 0u;
+	auto result = operand;
+	// The compressed SBASE value 62 decodes as code 124, but means NULL for S_LOAD.
+	result.kind = Decoder::OperandKind::Null;
+	return result;
 }
 
 ResourceKind FlatSegmentResourceKind(uint32_t segment) {
@@ -320,8 +322,9 @@ Translator::AddressOperands Translator::ReadAddressOperands(const Decoder::Instr
 	return {GetAddressResource(low, high), low, high};
 }
 
-IR::Value Translator::GetScalarAddressResource(uint32_t base) {
-	return GetAddressResource(ReadScalarCode(base), ReadScalarCode(base + 1u));
+IR::Value Translator::GetScalarAddressResource(const Decoder::Operand& base) {
+	const auto raw = ReadU32Pair(base);
+	return GetAddressResource(raw[0], raw[1]);
 }
 
 IR::Value Translator::GetImageResource(const IR::MemoryInfo& memory) {
@@ -885,7 +888,7 @@ bool Translator::DS_BPERMUTE_B32(const Decoder::Instruction& inst) {
 	return true;
 }
 
-bool Translator::EmitMemory(const Decoder::Instruction& inst, std::string* error) {
+bool Translator::EmitMemory(const Decoder::Instruction& inst) {
 	switch (inst.opcode) {
 		case Decoder::Opcode::S_LOAD_DWORD:
 		case Decoder::Opcode::S_LOAD_DWORDX2:
@@ -1075,12 +1078,7 @@ bool Translator::EmitMemory(const Decoder::Instruction& inst, std::string* error
 		case Decoder::Opcode::DS_WRITE_B64:
 		case Decoder::Opcode::DS_WRITE_B96:
 		case Decoder::Opcode::DS_WRITE_B128: return DS_WRITE(inst);
-		default:
-			if (error != nullptr) {
-				*error = fmt::format("memory-family opcode has no specialized IR translation: {}",
-				                     magic_enum::enum_name(inst.opcode));
-			}
-			return false;
+		default: return false;
 	}
 }
 

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

@@ -2,14 +2,14 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
 
-bool Translator::EmitScalar(const Decoder::Instruction& inst, std::string* error) {
+bool Translator::EmitScalar(const Decoder::Instruction& inst) {
 	using O = Decoder::Opcode;
 	switch (inst.opcode) {
 		case O::S_MOV_B32:
 		case O::S_MOVK_I32: MOV_B32(inst, false); return true;
 		case O::S_MOV_B64: S_MOV_B64(inst); return true;
 		case O::S_WQM_B64: S_WQM_B64(inst); return true;
-		case O::S_GETPC_B64: return S_GETPC_B64(inst, error);
+		case O::S_GETPC_B64: S_GETPC_B64(inst); return true;
 		case O::S_SETPC_B64: return true;
 		case O::S_CSELECT_B32: S_CSELECT_B32(inst); return true;
 		case O::S_CSELECT_B64: S_CSELECT_B64(inst); return true;

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

@@ -1,16 +1,14 @@
+#include "common/assert.h"
 #include "graphics/shader/recompiler/frontend/translate/Translator.h"
 #include "graphics/shader/shader.h"
 
 #include <algorithm>
 #include <array>
-#include <fmt/format.h>
 #include <unordered_map>
 #include <utility>
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
 
-using ShaderError::Fail;
-
 Decoder::Operand Translator::SourceAt(const Decoder::Instruction& inst, uint32_t index) {
 	switch (index) {
 		case 0: return inst.src0;
@@ -104,6 +102,7 @@ Decoder::Operand Translator::PlainOperand(const Decoder::Operand& operand) {
 	result.dpp_ctrl           = 0;
 	result.dpp_row_mask       = 0xf;
 	result.dpp_bank_mask      = 0xf;
+	result.explicit_sdwa_dst  = false;
 	result.dpp_fetch_inactive = false;
 	result.dpp_bound_ctrl     = false;
 	result.dpp                = false;
@@ -152,11 +151,8 @@ IR::U32 Translator::ReadScalarCode(uint32_t code) {
 		case 106u: return ir.GetVccLo();
 		case 107u: return ir.GetVccHi();
 		case 124u: return ir.GetM0();
-		case 126u:
-		case 127u: {
-			const auto mask = BallotMask(ir.GetExec());
-			return mask[code - 126u];
-		}
+		case 126u: return ir.GetExecLo();
+		case 127u: return ir.GetExecHi();
 		default: return IR::U32(IR::Value(0u));
 	}
 }
@@ -316,19 +312,23 @@ void Translator::WriteRawU32(const Decoder::Operand& operand, IR::U32 value) {
 		case Decoder::OperandKind::VccLo:
 			ir.SetVccLo(IR::U32(value));
 			ir.SetVcc(ThreadBit(IR::U32(value)));
+			ir.SetVccMaskValidTag(IR::U1(IR::Value(false)));
 			break;
 		case Decoder::OperandKind::VccHi:
 			ir.SetVccHi(IR::U32(value));
 			ir.SetVcc(ThreadBit(ir.GetVccLo()));
+			ir.SetVccMaskValidTag(IR::U1(IR::Value(false)));
 			break;
 		case Decoder::OperandKind::M0: ir.SetM0(IR::U32(value)); break;
 		case Decoder::OperandKind::ExecLo:
 			ir.SetExecLo(IR::U32(value));
 			ir.SetExec(ThreadBit(IR::U32(value)));
+			ir.SetExecMaskTag(IR::U1(IR::Value(false)));
 			break;
 		case Decoder::OperandKind::ExecHi:
 			ir.SetExecHi(IR::U32(value));
 			ir.SetExec(ThreadBit(ir.GetExecLo()));
+			ir.SetExecMaskTag(IR::U1(IR::Value(false)));
 			break;
 		case Decoder::OperandKind::Scc:
 			ir.SetScc(ir.INotEqual(value, IR::U32(IR::Value(0u))));
@@ -373,6 +373,7 @@ void Translator::WriteOperand(const Decoder::Operand& operand, IR::Value value)
 			case Decoder::OperandKind::ExecHi: {
 				const auto mask = BallotMask(IR::U1(value));
 				ir.SetExec(IR::U1(value));
+				ir.SetExecMaskTag(IR::U1(IR::Value(true)));
 				ir.SetExecLo(mask[0]);
 				ir.SetExecHi(mask[1]);
 				return;
@@ -381,6 +382,7 @@ void Translator::WriteOperand(const Decoder::Operand& operand, IR::Value value)
 			case Decoder::OperandKind::VccHi: {
 				const auto mask = BallotMask(IR::U1(value));
 				ir.SetVcc(IR::U1(value));
+				ir.SetVccMaskValidTag(IR::U1(IR::Value(true)));
 				ir.SetVccLo(mask[0]);
 				ir.SetVccHi(mask[1]);
 				return;
@@ -392,12 +394,12 @@ void Translator::WriteOperand(const Decoder::Operand& operand, IR::Value value)
 		}
 	}
 	if (type == IR::Type::U16) {
-		WriteRawU32(operand, IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16, {IR::U16(value)})));
+		Write16Bits(operand, IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16, {IR::U16(value)})));
 		return;
 	}
 	if (type == IR::Type::F16) {
 		const auto bits = IR::U16(ir.Emit(IR::ValueOpcode::BitCastU16F16, {value}));
-		WriteRawU32(operand, IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16, {bits})));
+		Write16Bits(operand, IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16, {bits})));
 		return;
 	}
 	if (type == IR::Type::U64) {
@@ -417,38 +419,32 @@ IR::U32 Translator::PackHalf2x16(IR::F32 low, IR::F32 high) {
 	return IR::U32(ir.Emit(IR::ValueOpcode::PackHalf2x16, {pair}));
 }
 
+void Translator::Write16Bits(const Decoder::Operand& operand, IR::U32 value) {
+	auto destination = operand;
+	// Native GFX10 16-bit results preserve the unselected half. Plain/DPP destinations use the
+	// low half; native VOP3 may already select either half. Explicit SDWA retains encoded DST_U.
+	if (!destination.explicit_sdwa_dst) {
+		if (destination.sdwa_sel == 6u) {
+			destination.sdwa_sel = 4u;
+		}
+		destination.sdwa_dst_unused = 2u;
+	}
+	destination.omod   = 0u;
+	destination.op_sel = false;
+	destination.clamp  = false;
+	WriteRawU32(destination, ir.BitwiseAnd(value, IR::U32(IR::Value(0xffffu))));
+}
+
 void Translator::WriteF16(const Decoder::Operand& operand, IR::F32 value) {
 	value           = ApplyF32ResultModifiers(operand, value);
 	const auto half = IR::F16(ir.Emit(IR::ValueOpcode::ConvertF16F32, {value}));
 	const auto bits = IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16,
 	                                  {IR::U16(ir.Emit(IR::ValueOpcode::BitCastU16F16, {half}))}));
-	auto       merged = bits;
-	if (operand.sdwa_sel == 4u || operand.sdwa_sel == 5u) {
-		merged = IR::U32(ir.Emit(IR::ValueOpcode::BitFieldInsert,
-		                         {ReadRawU32(PlainOperand(operand)), bits,
-		                          IR::Value(operand.sdwa_sel == 5u ? 16u : 0u), IR::Value(16u)}));
-	}
-	auto raw     = operand;
-	raw.sdwa_sel = 6u;
-	raw.omod     = 0u;
-	raw.op_sel   = false;
-	raw.clamp    = false;
-	WriteRawU32(raw, merged);
+	Write16Bits(operand, bits);
 }
 
 void Translator::WriteU16(const Decoder::Operand& operand, IR::U32 value) {
-	auto merged = ir.BitwiseAnd(value, IR::U32(IR::Value(0xffffu)));
-	if (operand.sdwa_sel == 4u || operand.sdwa_sel == 5u) {
-		merged = IR::U32(ir.Emit(IR::ValueOpcode::BitFieldInsert,
-		                         {ReadRawU32(PlainOperand(operand)), merged,
-		                          IR::Value(operand.sdwa_sel == 5u ? 16u : 0u), IR::Value(16u)}));
-	}
-	auto raw     = operand;
-	raw.sdwa_sel = 6u;
-	raw.omod     = 0u;
-	raw.op_sel   = false;
-	raw.clamp    = false;
-	WriteRawU32(raw, merged);
+	Write16Bits(operand, value);
 }
 
 IR::U32 Translator::ReadU32(const Decoder::Operand& operand) {
@@ -587,11 +583,13 @@ void Translator::WriteU32Pair(const Decoder::Operand&       operand,
 	switch (operand.kind) {
 		case Decoder::OperandKind::ExecLo:
 			ir.SetExec(thread_bit);
+			ir.SetExecMaskTag(IR::U1(IR::Value(false)));
 			ir.SetExecLo(value[0]);
 			ir.SetExecHi(value[1]);
 			return;
 		case Decoder::OperandKind::VccLo:
 			ir.SetVcc(thread_bit);
+			ir.SetVccMaskValidTag(IR::U1(IR::Value(false)));
 			ir.SetVccLo(value[0]);
 			ir.SetVccHi(value[1]);
 			return;
@@ -602,6 +600,87 @@ void Translator::WriteU32Pair(const Decoder::Operand&       operand,
 	WriteRawU32(OffsetOperand(operand, 1), value[1]);
 }
 
+Translator::ScalarU64 Translator::ReadScalarU64(const Decoder::Operand& operand) {
+	return {ReadU32Pair(operand), ReadMask(operand), ReadMaskValid(operand)};
+}
+
+Translator::ScalarU64 Translator::MakeScalarU64Mask(IR::U1 value) {
+	return {BallotMask(value), value, IR::U1(IR::Value(true))};
+}
+
+Translator::ScalarU64 Translator::MakeScalarU64Result(const std::array<IR::U32, 2>& raw,
+                                                      IR::U1 invocation, IR::U1 mask_valid) {
+	return {raw,
+	        IR::U1(ir.Emit(IR::ValueOpcode::SelectU1, {mask_valid, invocation, ThreadBit(raw[0])})),
+	        mask_valid};
+}
+
+Translator::ScalarU64 Translator::SelectScalarU64(IR::U1 condition, const ScalarU64& lhs,
+                                                  const ScalarU64& rhs) {
+	return {
+	    {ir.Select(condition, lhs.raw[0], rhs.raw[0]),
+	     ir.Select(condition, lhs.raw[1], rhs.raw[1])},
+	    IR::U1(ir.Emit(IR::ValueOpcode::SelectU1, {condition, lhs.invocation, rhs.invocation})),
+	    IR::U1(ir.Emit(IR::ValueOpcode::SelectU1, {condition, lhs.mask_valid, rhs.mask_valid}))};
+}
+
+Translator::ScalarU64 Translator::NotScalarU64(const ScalarU64& value) {
+	return MakeScalarU64Result({ir.BitwiseNot(value.raw[0]), ir.BitwiseNot(value.raw[1])},
+	                           ir.LogicalNot(value.invocation), value.mask_valid);
+}
+
+Translator::ScalarU64 Translator::BinaryScalarU64(const ScalarU64& lhs, const ScalarU64& rhs,
+                                                  IR::ValueOpcode logical_opcode,
+                                                  IR::ValueOpcode bit_opcode, bool negate_lhs,
+                                                  bool negate_rhs, bool negate_result) {
+	auto lhs_mask = negate_lhs ? ir.LogicalNot(lhs.invocation) : lhs.invocation;
+	auto rhs_mask = negate_rhs ? ir.LogicalNot(rhs.invocation) : rhs.invocation;
+	auto mask     = IR::U1(ir.Emit(logical_opcode, {lhs_mask, rhs_mask}));
+	if (negate_result) {
+		mask = ir.LogicalNot(mask);
+	}
+	std::array<IR::U32, 2> raw;
+	for (uint32_t component = 0; component < raw.size(); component++) {
+		auto lhs_raw   = negate_lhs ? ir.BitwiseNot(lhs.raw[component]) : lhs.raw[component];
+		auto rhs_raw   = negate_rhs ? ir.BitwiseNot(rhs.raw[component]) : rhs.raw[component];
+		raw[component] = IR::U32(ir.Emit(bit_opcode, {lhs_raw, rhs_raw}));
+		if (negate_result) {
+			raw[component] = ir.BitwiseNot(raw[component]);
+		}
+	}
+	return MakeScalarU64Result(raw, mask, ir.LogicalAnd(lhs.mask_valid, rhs.mask_valid));
+}
+
+void Translator::WriteScalarU64(const Decoder::Operand& operand, const ScalarU64& value) {
+	WriteU32Pair(operand, value.raw);
+	switch (operand.kind) {
+		case Decoder::OperandKind::ExecLo:
+		case Decoder::OperandKind::ExecHi:
+			ir.SetExec(value.invocation);
+			ir.SetExecMaskTag(value.mask_valid);
+			break;
+		case Decoder::OperandKind::VccLo:
+		case Decoder::OperandKind::VccHi:
+			ir.SetVcc(value.invocation);
+			ir.SetVccMaskValidTag(value.mask_valid);
+			break;
+		case Decoder::OperandKind::Sgpr: {
+			const auto reg = static_cast<IR::ScalarReg>(operand.reg);
+			ir.SetThreadBitScalarReg(reg, value.invocation);
+			ir.SetScalarMaskTag(reg, value.mask_valid);
+			break;
+		}
+		default: break;
+	}
+}
+
+IR::U1 Translator::ScalarU64NonZero(const ScalarU64& value) {
+	const auto raw_nonzero =
+	    ir.INotEqual(ir.BitwiseOr(value.raw[0], value.raw[1]), IR::U32(IR::Value(0u)));
+	return IR::U1(
+	    ir.Emit(IR::ValueOpcode::SelectU1, {value.mask_valid, value.invocation, raw_nonzero}));
+}
+
 IR::U1 Translator::ThreadBit(IR::U32 low) {
 	return ir.INotEqual(low, IR::U32(IR::Value(0u)));
 }
@@ -657,9 +736,9 @@ IR::U1 Translator::ReadMaskValid(const Decoder::Operand& operand) {
 		case Decoder::OperandKind::Sgpr:
 			return ir.GetScalarMaskTag(static_cast<IR::ScalarReg>(operand.reg));
 		case Decoder::OperandKind::ExecLo:
-		case Decoder::OperandKind::ExecHi:
+		case Decoder::OperandKind::ExecHi: return ir.GetExecMaskTag();
 		case Decoder::OperandKind::VccLo:
-		case Decoder::OperandKind::VccHi:
+		case Decoder::OperandKind::VccHi: return ir.GetVccMaskValidTag();
 		case Decoder::OperandKind::VccZ:
 		case Decoder::OperandKind::ExecZ:
 		case Decoder::OperandKind::Scc: return IR::U1(IR::Value(true));
@@ -692,6 +771,7 @@ void Translator::WriteMask(const Decoder::Operand& operand, IR::U1 value) {
 		case Decoder::OperandKind::ExecHi: {
 			const auto mask = BallotMask(value);
 			ir.SetExec(value);
+			ir.SetExecMaskTag(IR::U1(IR::Value(true)));
 			ir.SetExecLo(mask[0]);
 			ir.SetExecHi(mask[1]);
 			return;
@@ -700,6 +780,7 @@ void Translator::WriteMask(const Decoder::Operand& operand, IR::U1 value) {
 		case Decoder::OperandKind::VccHi: {
 			const auto mask = BallotMask(value);
 			ir.SetVcc(value);
+			ir.SetVccMaskValidTag(IR::U1(IR::Value(true)));
 			ir.SetVccLo(mask[0]);
 			ir.SetVccHi(mask[1]);
 			return;
@@ -713,25 +794,7 @@ void Translator::WriteMask(const Decoder::Operand& operand, IR::U1 value) {
 }
 
 void Translator::WriteMask64(const Decoder::Operand& operand, IR::U1 value) {
-	if (operand.kind != Decoder::OperandKind::Sgpr) {
-		WriteMask(operand, value);
-		return;
-	}
-	const auto reg  = static_cast<IR::ScalarReg>(operand.reg);
-	const auto mask = BallotMask(value);
-	ir.SetThreadBitScalarReg(reg, value);
-	ir.SetScalarMaskTag(reg, IR::U1(IR::Value(true)));
-	if (IR::RegIndex(reg) > 0u) {
-		ir.SetScalarMaskTag(static_cast<IR::ScalarReg>(IR::RegIndex(reg) - 1u),
-		                    IR::U1(IR::Value(false)));
-	}
-	ir.SetScalarReg(reg, mask[0]);
-	if (IR::RegIndex(reg) + 1u < IR::NumScalarRegs) {
-		const auto high = static_cast<IR::ScalarReg>(IR::RegIndex(reg) + 1u);
-		ir.SetScalarReg(high, mask[1]);
-		ir.SetThreadBitScalarReg(high, IR::U1(IR::Value(false)));
-		ir.SetScalarMaskTag(high, IR::U1(IR::Value(false)));
-	}
+	WriteScalarU64(operand, MakeScalarU64Mask(value));
 }
 
 void Translator::WriteCompareResult(const Decoder::Operand& operand, IR::U1 value) {
@@ -745,11 +808,10 @@ void Translator::WriteCompareResult(const Decoder::Operand& operand, IR::U1 valu
 	WriteMask(operand, ir.LogicalAnd(ir.GetExec(), value));
 }
 
-bool Translator::AddBranchCondition(const CFG::BasicBlock& source, IR::BlockInfo& info,
-                                    std::string* error) {
+void Translator::AddBranchCondition(const CFG::BasicBlock& source, IR::BlockInfo& info) {
 	if (source.terminator.goto_value >= 0) {
 		if (source.terminator.goto_variable == UINT32_MAX) {
-			return Fail(error, fmt::format("block {} sets an invalid goto variable", source.id));
+			EXIT("block %u sets an invalid goto variable", source.id);
 		}
 		ir.SetGotoVariable(source.terminator.goto_variable,
 		                   IR::U1(IR::Value(source.terminator.goto_value != 0)));
@@ -761,13 +823,13 @@ bool Translator::AddBranchCondition(const CFG::BasicBlock& source, IR::BlockInfo
 			info.indirect_target =
 			    ir.GetScalarReg(static_cast<IR::ScalarReg>(source.terminator.indirect_pc_sgpr));
 		} else {
-			return Fail(error, fmt::format("block {} has no indirect branch selector", source.id));
+			EXIT("block %u has no indirect branch selector", source.id);
 		}
 		ir.Emit(IR::ValueOpcode::ReferenceU32, {info.indirect_target});
-		return true;
+		return;
 	}
 	if (source.terminator.kind != CFG::TerminatorKind::ConditionalBranch) {
-		return true;
+		return;
 	}
 	// EXEC and VCC are invocation-local Boolean masks. Branching on that Boolean lets inactive
 	// invocations leave the region without reconstructing a host-subgroup mask.
@@ -782,17 +844,15 @@ bool Translator::AddBranchCondition(const CFG::BasicBlock& source, IR::BlockInfo
 		case CFG::BranchCondition::ExecNonZero: condition = ir.GetExec(); break;
 		case CFG::BranchCondition::GotoVariable:
 			if (source.terminator.goto_variable == UINT32_MAX) {
-				return Fail(error,
-				            fmt::format("block {} reads an invalid goto variable", source.id));
+				EXIT("block %u reads an invalid goto variable", source.id);
 			}
 			condition = ir.GetGotoVariable(source.terminator.goto_variable);
 			break;
 		case CFG::BranchCondition::Unknown:
-			return Fail(error, fmt::format("block {} has an unknown branch condition", source.id));
+			EXIT("block %u has an unknown branch condition", source.id);
 	}
 	info.condition = condition;
 	ir.Emit(IR::ValueOpcode::Reference, {condition});
-	return true;
 }
 
 namespace {
@@ -914,47 +974,42 @@ void IncludeInstructionVectorRegisters(const Decoder::Instruction& inst, uint32_
 	}
 }
 
-bool ValidateTranslateOptions(const TranslateOptions& options, std::string* error) {
+void ValidateTranslateOptions(const TranslateOptions& options) {
 	if (options.wave_size != 32u && options.wave_size != 64u) {
-		return Fail(error, "shader translation requires wave32 or wave64");
+		EXIT("shader translation requires wave32 or wave64, got %u", options.wave_size);
 	}
 	switch (options.stage) {
 		case ShaderType::Vertex:
-			return options.vertex != nullptr ||
-			       Fail(error, "vertex shader translation has no vertex input metadata");
+			if (options.vertex == nullptr) {
+				EXIT("vertex shader translation has no vertex input metadata");
+			}
+			return;
 		case ShaderType::Pixel:
-			return options.pixel != nullptr ||
-			       Fail(error, "pixel shader translation has no pixel input metadata");
+			if (options.pixel == nullptr) {
+				EXIT("pixel shader translation has no pixel input metadata");
+			}
+			return;
 		case ShaderType::Compute:
-			return options.compute != nullptr ||
-			       Fail(error, "compute shader translation has no compute input metadata");
-		default: return Fail(error, "shader translation has an unsupported stage");
+			if (options.compute == nullptr) {
+				EXIT("compute shader translation has no compute input metadata");
+			}
+			return;
+		default:
+			EXIT("shader translation has unsupported stage %u",
+			     static_cast<uint32_t>(options.stage));
 	}
 }
 
 } // namespace
 
-bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
-                      const TranslateOptions& options, IR::Program& result, std::string* error) {
-	result = {};
-	struct ResetOnFailure {
-		IR::Program& program;
-		bool         complete = false;
-
-		~ResetOnFailure() {
-			if (!complete) {
-				program = {};
-			}
-		}
-	} reset_on_failure {result};
-
-	if (!ValidateTranslateOptions(options, error)) {
-		return false;
-	}
+IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
+                             const TranslateOptions& options) {
+	ValidateTranslateOptions(options);
 	if (cfg.blocks.empty()) {
-		return Fail(error, "cannot translate an empty CFG");
+		EXIT("cannot translate an empty CFG");
 	}
 
+	IR::Program result;
 	result.stage               = options.stage;
 	result.wave_size           = options.wave_size;
 	result.shader_hash         = options.shader_hash;
@@ -972,8 +1027,8 @@ bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
 	for (const auto& cfg_block: cfg.blocks) {
 		for (uint32_t index = cfg_block.inst_begin; index < cfg_block.inst_end; index++) {
 			if (index >= decoded.instructions.size()) {
-				return Fail(error,
-				            "CFG block references an instruction outside the decoded program");
+				EXIT("CFG block %u references instruction %u outside decoded program of size %zu",
+				     cfg_block.id, index, decoded.instructions.size());
 			}
 			const auto& instruction = decoded.instructions[index];
 			if (IsCodeTableLoad(cfg, instruction.pc)) {
@@ -990,7 +1045,7 @@ bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
 		                    return block.id;
 	                    })->id;
 	if (max_id == UINT32_MAX) {
-		return Fail(error, "cannot allocate a typed entry block id");
+		EXIT("cannot allocate a typed entry block id");
 	}
 	CFG::Terminator terminator;
 	terminator.kind       = CFG::TerminatorKind::Branch;
@@ -1004,7 +1059,7 @@ bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
 	block_indices.reserve(cfg.blocks.size());
 	for (const auto& source_block: cfg.blocks) {
 		if (!block_indices.emplace(source_block.id, result.blocks.size()).second) {
-			return Fail(error, fmt::format("CFG contains duplicate block id {}", source_block.id));
+			EXIT("CFG contains duplicate block id %u", source_block.id);
 		}
 		result.block_storage.push_back(std::make_unique<IR::Block>());
 		result.blocks.push_back(result.block_storage.back().get());
@@ -1016,8 +1071,7 @@ bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
 		for (const auto successor: source_block.successors) {
 			const auto target = block_indices.find(successor);
 			if (target == block_indices.end()) {
-				return Fail(error, fmt::format("CFG block {} has unknown successor {}",
-				                               source_block.id, successor));
+				EXIT("CFG block %u has unknown successor %u", source_block.id, successor);
 			}
 			result.blocks[source_index]->AddBranch(result.blocks[target->second]);
 		}
@@ -1041,6 +1095,7 @@ bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
 			entry_ir.SetScalarMaskTag(reg, IR::U1(IR::Value(false)));
 		}
 		entry_ir.SetExec(IR::U1(IR::Value(true)));
+		entry_ir.SetExecMaskTag(IR::U1(IR::Value(true)));
 		entry_ir.SetExecLo(IR::U32(IR::Value(1u)));
 		entry_ir.SetExecHi(IR::U32(IR::Value(0u)));
 		if (options.stage == ShaderType::Compute) {
@@ -1115,7 +1170,7 @@ bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
 		}
 	}
 	for (const auto& cfg_block: cfg.blocks) {
-		const auto         typed_index = block_indices.at(cfg_block.id);
+		const auto typed_index = block_indices.at(cfg_block.id);
 		Translator translator(result, result.blocks[typed_index], vector_limit, options.wave_size);
 		for (uint32_t index = cfg_block.inst_begin; index < cfg_block.inst_end; index++) {
 			const auto& instruction = decoded.instructions[index];
@@ -1131,44 +1186,27 @@ bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
 			    instruction.data_dwords == embedded->components &&
 			    instruction.dst.kind == Decoder::OperandKind::Vgpr) {
 				if (options.vertex == nullptr) {
-					return Fail(error, "embedded vertex fetch plan has no vertex input metadata");
+					EXIT("embedded vertex fetch plan has no vertex input metadata");
 				}
 				const auto resource = ResolveEmbeddedFetchResource(*options.vertex, *embedded);
 				if (resource < 0 || resource >= options.vertex->resources_num) {
-					return Fail(
-					    error,
-					    fmt::format(
-					        "embedded vertex fetch at 0x{:08x} has no resource for attribute {}",
-					        instruction.pc, embedded->attrib_id));
+					EXIT("embedded vertex fetch at 0x%08x has no resource for attribute %d",
+					     instruction.pc, embedded->attrib_id);
 				}
 				result.info.vertex_fetch_components[static_cast<size_t>(resource)] =
 				    static_cast<uint8_t>(std::max<uint32_t>(
 				        result.info.vertex_fetch_components[static_cast<size_t>(resource)],
 				        embedded->components));
-				if (!translator.TranslateEmbeddedFetch(instruction, static_cast<uint32_t>(resource),
-				                                       embedded->components)) {
-					return false;
-				}
+				translator.TranslateEmbeddedFetch(instruction, static_cast<uint32_t>(resource),
+				                                  embedded->components);
 				continue;
 			}
-			if (!translator.TranslateInstruction(instruction, error)) {
-				if (error == nullptr || error->empty()) {
-					return Fail(error, fmt::format("opcode {} at 0x{:08x} has no IR translation",
-					                               magic_enum::enum_name(instruction.opcode),
-					                               instruction.pc));
-				}
-				return false;
-			}
-		}
-		if (!translator.AddBranchCondition(cfg_block, result.block_info[typed_index], error)) {
-			return false;
+			translator.TranslateInstruction(instruction);
 		}
+		translator.AddBranchCondition(cfg_block, result.block_info[typed_index]);
 	}
-	if (!IR::ValidateProgram(result, false, error)) {
-		return false;
-	}
-	reset_on_failure.complete = true;
-	return true;
+	IR::ValidateProgram(result, false);
+	return result;
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::Frontend

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

@@ -38,7 +38,7 @@ struct TranslateOptions {
 	const EmbeddedFetchPlan*      embedded_fetch = nullptr;
 };
 
-bool TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
-                      const TranslateOptions& options, IR::Program& result, std::string* error);
+IR::Program TranslateProgram(const Decoder::Program& decoded, const CFG::Graph& cfg,
+                             const TranslateOptions& options);
 
 } // namespace Libs::Graphics::ShaderRecompiler::Frontend

src/graphics/shader/recompiler/frontend/translate/Translator.h

@@ -13,12 +13,20 @@ public:
 	    : program(program), ir(block), current_vector_limit(vector_limit),
 	      current_wave_size(wave_size) {}
 
-	bool TranslateInstruction(const Decoder::Instruction& inst, std::string* error);
-	bool TranslateEmbeddedFetch(const Decoder::Instruction& inst, uint32_t attribute,
+	void TranslateInstruction(const Decoder::Instruction& inst);
+	void TranslateEmbeddedFetch(const Decoder::Instruction& inst, uint32_t attribute,
 	                            uint32_t component_count);
-	bool AddBranchCondition(const CFG::BasicBlock& source, IR::BlockInfo& info, std::string* error);
+	void AddBranchCondition(const CFG::BasicBlock& source, IR::BlockInfo& info);
 
 private:
+	struct ScalarU64 {
+		std::array<IR::U32, 2> raw;
+		// Raw words and Kyty's per-invocation mask model remain independent.
+		// mask_valid selects invocation for logical consumers and SCC.
+		IR::U1 invocation;
+		IR::U1 mask_valid;
+	};
+
 	Decoder::Operand SourceAt(const Decoder::Instruction& inst, uint32_t index);
 	Decoder::Operand DestinationOperand(const Decoder::Instruction& inst);
 	Decoder::Operand OffsetOperand(const Decoder::Operand& operand, uint32_t offset);
@@ -34,6 +42,7 @@ private:
 	IR::F32                ApplyF32ResultModifiers(const Decoder::Operand& operand, IR::F32 value);
 	void                   WriteOperand(const Decoder::Operand& operand, IR::Value value);
 	IR::U32                PackHalf2x16(IR::F32 low, IR::F32 high);
+	void                   Write16Bits(const Decoder::Operand& operand, IR::U32 value);
 	void                   WriteF16(const Decoder::Operand& operand, IR::F32 value);
 	void                   WriteU16(const Decoder::Operand& operand, IR::U32 value);
 	IR::U32                ReadU32(const Decoder::Operand& operand);
@@ -47,14 +56,25 @@ private:
 	IR::U32 ReadU16AsU32(const Decoder::Operand& operand, bool sign_extend);
 	IR::U32 ReadF16LaneBits(const Decoder::Operand& operand, bool high_lane);
 	std::array<IR::U32, 2> ExtractU64(IR::U64 value);
-	void    WriteU32Pair(const Decoder::Operand& operand, const std::array<IR::U32, 2>& value);
-	IR::U1  ReadCondition(const Decoder::Operand& operand);
-	IR::U32 ConditionBit(const Decoder::Operand& operand);
-	IR::U1  ReadMask(const Decoder::Operand& operand);
-	IR::U1  ReadMaskValid(const Decoder::Operand& operand);
-	void    WriteMask(const Decoder::Operand& operand, IR::U1 value);
-	void    WriteMask64(const Decoder::Operand& operand, IR::U1 value);
-	void    WriteCompareResult(const Decoder::Operand& operand, IR::U1 value);
+	void      WriteU32Pair(const Decoder::Operand& operand, const std::array<IR::U32, 2>& value);
+	ScalarU64 ReadScalarU64(const Decoder::Operand& operand);
+	ScalarU64 MakeScalarU64Mask(IR::U1 value);
+	ScalarU64 MakeScalarU64Result(const std::array<IR::U32, 2>& raw, IR::U1 invocation,
+	                              IR::U1 mask_valid);
+	ScalarU64 SelectScalarU64(IR::U1 condition, const ScalarU64& lhs, const ScalarU64& rhs);
+	ScalarU64 NotScalarU64(const ScalarU64& value);
+	ScalarU64 BinaryScalarU64(const ScalarU64& lhs, const ScalarU64& rhs,
+	                          IR::ValueOpcode logical_opcode, IR::ValueOpcode bit_opcode,
+	                          bool negate_lhs, bool negate_rhs, bool negate_result);
+	void      WriteScalarU64(const Decoder::Operand& operand, const ScalarU64& value);
+	IR::U1    ScalarU64NonZero(const ScalarU64& value);
+	IR::U1    ReadCondition(const Decoder::Operand& operand);
+	IR::U32   ConditionBit(const Decoder::Operand& operand);
+	IR::U1    ReadMask(const Decoder::Operand& operand);
+	IR::U1    ReadMaskValid(const Decoder::Operand& operand);
+	void      WriteMask(const Decoder::Operand& operand, IR::U1 value);
+	void      WriteMask64(const Decoder::Operand& operand, IR::U1 value);
+	void      WriteCompareResult(const Decoder::Operand& operand, IR::U1 value);
 
 	IR::MemoryFlags AddMemoryInfo(const IR::MemoryInfo& memory, uint32_t pc);
 	IR::ExportFlags AddExportInfo(const Decoder::Instruction& inst);
@@ -72,7 +92,7 @@ private:
 	IR::U32         GetResourceDword(uint32_t index, uint32_t dword);
 	IR::Value       GetBufferResource(const IR::MemoryInfo& memory);
 	IR::Value       GetAddressResource(IR::Value low, IR::Value high);
-	IR::Value       GetScalarAddressResource(uint32_t base);
+	IR::Value       GetScalarAddressResource(const Decoder::Operand& base);
 	IR::Value       GetImageResource(const IR::MemoryInfo& memory);
 	IR::Value       GetSamplerResource(const IR::MemoryInfo& memory);
 	IR::Value     MakeImageAddress(const Decoder::Instruction& inst, const Decoder::Operand& base);
@@ -127,7 +147,7 @@ private:
 	void EmitFloatCompare(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool half,
 	                      bool cmpx);
 	void EmitFloatOrderedCompare(const Decoder::Instruction& inst, bool ordered);
-	void V_CMP_CLASS_F32(const Decoder::Instruction& inst);
+	void EmitFloatClassCompare(const Decoder::Instruction& inst, bool cmpx);
 	void V_CVT_F32_UBYTE(const Decoder::Instruction& inst, uint32_t byte_index);
 	void V_CVT_F32_U32(const Decoder::Instruction& inst);
 	void V_CVT_F32_I32(const Decoder::Instruction& inst);
@@ -174,28 +194,26 @@ private:
 	bool PackedInteger16MinMax(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool sign);
 	bool S_U64_MASK(const Decoder::Instruction& inst, IR::ValueOpcode logical_opcode,
 	                IR::ValueOpcode bit_opcode, bool negate_rhs, bool negate_result, bool unary);
-	IR::U1  U64MaskBinary(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool negate_rhs,
-	                      bool negate_result);
-	bool    SimpleInteger(const Decoder::Instruction& inst, IR::ValueOpcode opcode, IR::Type type,
-	                      bool reverse, bool mask_shift_count, bool update_scc);
-	bool    ComposedIntegerBinary(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
-	                              bool negate_rhs, bool negate_result, bool update_scc);
-	bool    V_AND_OR_B32(const Decoder::Instruction& inst);
-	bool    V_OR3_B32(const Decoder::Instruction& inst);
-	bool    V_XOR3_B32(const Decoder::Instruction& inst);
-	bool    S_FF1_I32_B64(const Decoder::Instruction& inst);
-	bool    V_FFBH_32(const Decoder::Instruction& inst, bool sign);
-	bool    S_FLBIT_I32_B64(const Decoder::Instruction& inst);
-	bool    Integer24(const Decoder::Instruction& inst, bool sign, bool addend);
-	bool    V_MAD_U64_U32(const Decoder::Instruction& inst);
-	bool    V_SAD_U32(const Decoder::Instruction& inst);
-	bool    V_ADD3_U32(const Decoder::Instruction& inst);
-	bool    S_BITSET_B32(const Decoder::Instruction& inst, bool set);
-	bool    V_BCNT_U32_B32(const Decoder::Instruction& inst);
-	bool    V_MBCNT_U32_B32(const Decoder::Instruction& inst, bool low);
-	bool    S_BITREPLICATE_B64_B32(const Decoder::Instruction& inst);
-	bool    S_QUADMASK_B64(const Decoder::Instruction& inst);
-	bool    BFM_B32(const Decoder::Instruction& inst);
+	bool SimpleInteger(const Decoder::Instruction& inst, IR::ValueOpcode opcode, IR::Type type,
+	                   bool reverse, bool mask_shift_count, bool update_scc);
+	bool ComposedIntegerBinary(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
+	                           bool negate_rhs, bool negate_result, bool update_scc);
+	bool V_AND_OR_B32(const Decoder::Instruction& inst);
+	bool V_OR3_B32(const Decoder::Instruction& inst);
+	bool V_XOR3_B32(const Decoder::Instruction& inst);
+	bool S_FF1_I32_B64(const Decoder::Instruction& inst);
+	bool V_FFBH_32(const Decoder::Instruction& inst, bool sign);
+	bool S_FLBIT_I32_B64(const Decoder::Instruction& inst);
+	bool Integer24(const Decoder::Instruction& inst, bool sign, bool addend);
+	bool V_MAD_U64_U32(const Decoder::Instruction& inst);
+	bool V_SAD_U32(const Decoder::Instruction& inst);
+	bool V_ADD3_U32(const Decoder::Instruction& inst);
+	bool S_BITSET_B32(const Decoder::Instruction& inst, bool set);
+	bool V_BCNT_U32_B32(const Decoder::Instruction& inst);
+	bool V_MBCNT_U32_B32(const Decoder::Instruction& inst, bool low);
+	bool S_BITREPLICATE_B64_B32(const Decoder::Instruction& inst);
+	bool S_QUADMASK_B64(const Decoder::Instruction& inst);
+	bool BFM_B32(const Decoder::Instruction& inst);
 	IR::U32 RightMask32(IR::U32 count);
 	IR::U64 RightMask64(IR::U32 count);
 	bool    S_BFM_B64(const Decoder::Instruction& inst);
@@ -228,27 +246,27 @@ private:
 	void S_SENDMSG();
 	void S_TTRACEDATA();
 	void S_INST_PREFETCH();
-	bool S_GETPC_B64(const Decoder::Instruction& inst, std::string* error);
+	void S_GETPC_B64(const Decoder::Instruction& inst);
 	void S_CSELECT_B32(const Decoder::Instruction& inst);
 	void S_CSELECT_B64(const Decoder::Instruction& inst);
 	void MOV_B32(const Decoder::Instruction& inst, bool apply_float_modifiers);
 	void S_MOV_B64(const Decoder::Instruction& inst);
 	void S_WQM_B64(const Decoder::Instruction& inst);
-	bool V_MOVRELS_B32(const Decoder::Instruction& inst, std::string* error);
-	bool V_MOVRELD_B32(const Decoder::Instruction& inst, std::string* error);
+	void V_MOVRELS_B32(const Decoder::Instruction& inst);
+	void V_MOVRELD_B32(const Decoder::Instruction& inst);
 	void V_READFIRSTLANE_B32(const Decoder::Instruction& inst);
 	void V_READLANE_B32(const Decoder::Instruction& inst);
 	void V_WRITELANE_B32(const Decoder::Instruction& inst);
 	void V_PERMLANE16_B32(const Decoder::Instruction& inst, bool x16);
 	void V_INTERP_P1_F32();
 	void V_INTERP_P2_F32(const Decoder::Instruction& inst);
-	bool V_INTERP_MOV_F32(const Decoder::Instruction& inst, std::string* error);
-	bool EXP(const Decoder::Instruction& inst, std::string* error);
+	void V_INTERP_MOV_F32(const Decoder::Instruction& inst);
+	void EXP(const Decoder::Instruction& inst);
 
-	bool EmitScalar(const Decoder::Instruction& inst, std::string* error);
-	bool EmitVector(const Decoder::Instruction& inst, std::string* error);
-	bool EmitInterpolation(const Decoder::Instruction& inst, std::string* error);
-	bool EmitMemory(const Decoder::Instruction& inst, std::string* error);
+	bool EmitScalar(const Decoder::Instruction& inst);
+	bool EmitVector(const Decoder::Instruction& inst);
+	bool EmitInterpolation(const Decoder::Instruction& inst);
+	bool EmitMemory(const Decoder::Instruction& inst);
 
 	IR::Program&    program;
 	IR::IREmitter   ir;

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

@@ -2,7 +2,7 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
 
-bool Translator::EmitVector(const Decoder::Instruction& inst, std::string* error) {
+bool Translator::EmitVector(const Decoder::Instruction& inst) {
 	using O = Decoder::Opcode;
 	switch (inst.opcode) {
 		case O::V_NOP: return true;
@@ -13,8 +13,8 @@ bool Translator::EmitVector(const Decoder::Instruction& inst, std::string* error
 		case O::V_SUBREV_CO_CI_U32: SUBB_U32(inst, true); return true;
 
 		case O::V_MOV_B32: MOV_B32(inst, true); return true;
-		case O::V_MOVRELS_B32: return V_MOVRELS_B32(inst, error);
-		case O::V_MOVRELD_B32: return V_MOVRELD_B32(inst, error);
+		case O::V_MOVRELS_B32: V_MOVRELS_B32(inst); return true;
+		case O::V_MOVRELD_B32: V_MOVRELD_B32(inst); return true;
 		case O::V_READFIRSTLANE_B32: V_READFIRSTLANE_B32(inst); return true;
 		case O::V_READLANE_B32: V_READLANE_B32(inst); return true;
 		case O::V_WRITELANE_B32: V_WRITELANE_B32(inst); return true;
@@ -271,7 +271,8 @@ bool Translator::EmitVector(const Decoder::Instruction& inst, std::string* error
 			return true;
 		case O::V_CMP_O_F32: EmitFloatOrderedCompare(inst, true); return true;
 		case O::V_CMP_U_F32: EmitFloatOrderedCompare(inst, false); return true;
-		case O::V_CMP_CLASS_F32: V_CMP_CLASS_F32(inst); return true;
+		case O::V_CMP_CLASS_F32: EmitFloatClassCompare(inst, false); return true;
+		case O::V_CMPX_CLASS_F32: EmitFloatClassCompare(inst, true); return true;
 
 		case O::V_CVT_F32_UBYTE0: V_CVT_F32_UBYTE(inst, 0); return true;
 		case O::V_CVT_F32_UBYTE1: V_CVT_F32_UBYTE(inst, 1); return true;

src/graphics/shader/recompiler/ir/IREmitter.cpp

@@ -76,6 +76,14 @@ void IREmitter::SetExec(U1 value) {
 	Emit(ValueOpcode::SetExec, {value});
 }
 
+U1 IREmitter::GetExecMaskTag() {
+	return U1(Emit(ValueOpcode::GetExecMaskTag));
+}
+
+void IREmitter::SetExecMaskTag(U1 value) {
+	Emit(ValueOpcode::SetExecMaskTag, {value});
+}
+
 U32 IREmitter::GetExecLo() {
 	return U32(Emit(ValueOpcode::GetExecLo));
 }
@@ -100,6 +108,14 @@ void IREmitter::SetVcc(U1 value) {
 	Emit(ValueOpcode::SetVcc, {value});
 }
 
+U1 IREmitter::GetVccMaskValidTag() {
+	return U1(Emit(ValueOpcode::GetVccMaskValidTag));
+}
+
+void IREmitter::SetVccMaskValidTag(U1 value) {
+	Emit(ValueOpcode::SetVccMaskValidTag, {value});
+}
+
 U32 IREmitter::GetVccLo() {
 	return U32(Emit(ValueOpcode::GetVccLo));
 }

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

@@ -39,12 +39,16 @@ public:
 	void SetScc(U1 value);
 	U1   GetExec();
 	void SetExec(U1 value);
+	U1   GetExecMaskTag();
+	void SetExecMaskTag(U1 value);
 	U32  GetExecLo();
 	void SetExecLo(U32 value);
 	U32  GetExecHi();
 	void SetExecHi(U32 value);
 	U1   GetVcc();
 	void SetVcc(U1 value);
+	U1   GetVccMaskValidTag();
+	void SetVccMaskValidTag(U1 value);
 	U32  GetVccLo();
 	void SetVccLo(U32 value);
 	U32  GetVccHi();

src/graphics/shader/recompiler/ir/Program.cpp

@@ -1,7 +1,6 @@
+#include "common/assert.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 
-#include "graphics/shader/shader.h"
-
 #include <fmt/format.h>
 #include <map>
 #include <new>
@@ -11,10 +10,13 @@
 
 namespace Libs::Graphics::ShaderRecompiler::IR {
 
-using ShaderError::Fail;
-
 namespace {
 
+[[noreturn]] void Fail(std::string_view message) {
+	EXIT("shader IR validation failed: %s", std::string(message).c_str());
+	std::abort();
+}
+
 bool IsRegisterStatePseudo(ValueOpcode opcode) {
 	switch (opcode) {
 		case ValueOpcode::GetThreadBitScalarRegister:
@@ -31,12 +33,16 @@ bool IsRegisterStatePseudo(ValueOpcode opcode) {
 		case ValueOpcode::SetScc:
 		case ValueOpcode::GetExec:
 		case ValueOpcode::SetExec:
+		case ValueOpcode::GetExecMaskTag:
+		case ValueOpcode::SetExecMaskTag:
 		case ValueOpcode::GetExecLo:
 		case ValueOpcode::SetExecLo:
 		case ValueOpcode::GetExecHi:
 		case ValueOpcode::SetExecHi:
 		case ValueOpcode::GetVcc:
 		case ValueOpcode::SetVcc:
+		case ValueOpcode::GetVccMaskValidTag:
+		case ValueOpcode::SetVccMaskValidTag:
 		case ValueOpcode::GetVccLo:
 		case ValueOpcode::SetVccLo:
 		case ValueOpcode::GetVccHi:
@@ -149,10 +155,10 @@ Value ResolveInvariantPhi(const Program& program, Value value) {
 	return invariant;
 }
 
-bool ValidateProgram(const Program& program, bool require_ssa, std::string* error) {
+void ValidateProgram(const Program& program, bool require_ssa) {
 	if (program.blocks.size() != program.block_info.size() ||
 	    program.blocks.size() != program.block_storage.size()) {
-		return Fail(error, "value IR block storage is inconsistent");
+		return Fail("value IR block storage is inconsistent");
 	}
 	std::unordered_set<const Block*>           blocks;
 	std::unordered_map<const Block*, size_t>   block_indices;
@@ -162,26 +168,26 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 		const auto* block = program.blocks[block_index];
 		if (block == nullptr || program.block_storage[block_index] == nullptr ||
 		    block != program.block_storage[block_index].get()) {
-			return Fail(error, "value IR block pointer is inconsistent");
+			return Fail("value IR block pointer is inconsistent");
 		}
 		if (!blocks.insert(block).second) {
-			return Fail(error, "value IR block pointer is duplicated");
+			return Fail("value IR block pointer is duplicated");
 		}
 		block_indices.emplace(block, block_index);
 		if (program.block_info[block_index].id == UINT32_MAX) {
-			return Fail(error, "value IR block uses the reserved exit id");
+			return Fail("value IR block uses the reserved exit id");
 		}
 		if (!blocks_by_id.emplace(program.block_info[block_index].id, block).second) {
-			return Fail(error, "value IR block id is duplicated");
+			return Fail("value IR block id is duplicated");
 		}
 		for (const auto& inst: *block) {
 			if (!instructions.insert(&inst).second) {
-				return Fail(error, "value IR instruction is duplicated");
+				return Fail("value IR instruction is duplicated");
 			}
 		}
 	}
 	if (!program.blocks.empty() && !program.blocks.front()->ImmPredecessors().empty()) {
-		return Fail(error, "value IR entry block has a predecessor");
+		return Fail("value IR entry block has a predecessor");
 	}
 
 	for (size_t block_index = 0; block_index < program.blocks.size(); block_index++) {
@@ -189,28 +195,28 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 		std::unordered_set<const Block*> predecessors;
 		for (const auto* predecessor: block->ImmPredecessors()) {
 			if (predecessor == nullptr || !blocks.contains(predecessor)) {
-				return Fail(error, "value IR block has a foreign predecessor");
+				return Fail("value IR block has a foreign predecessor");
 			}
 			if (!predecessors.insert(predecessor).second) {
-				return Fail(error, "value IR block predecessor is duplicated");
+				return Fail("value IR block predecessor is duplicated");
 			}
 			if (std::ranges::find(predecessor->ImmSuccessors(), block) ==
 			    predecessor->ImmSuccessors().end()) {
-				return Fail(error, "value IR predecessor edge is not reciprocal");
+				return Fail("value IR predecessor edge is not reciprocal");
 			}
 		}
 
 		std::unordered_set<const Block*> successors;
 		for (const auto* successor: block->ImmSuccessors()) {
 			if (successor == nullptr || !blocks.contains(successor)) {
-				return Fail(error, "value IR block has a foreign successor");
+				return Fail("value IR block has a foreign successor");
 			}
 			if (!successors.insert(successor).second) {
-				return Fail(error, "value IR block successor is duplicated");
+				return Fail("value IR block successor is duplicated");
 			}
 			if (std::ranges::find(successor->ImmPredecessors(), block) ==
 			    successor->ImmPredecessors().end()) {
-				return Fail(error, "value IR successor edge is not reciprocal");
+				return Fail("value IR successor edge is not reciprocal");
 			}
 		}
 
@@ -234,41 +240,40 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 		switch (terminator.kind) {
 			case CFG::TerminatorKind::Branch:
 				if (!add_target(terminator.true_block)) {
-					return Fail(error, "value IR branch target is missing");
+					return Fail("value IR branch target is missing");
 				}
 				break;
 			case CFG::TerminatorKind::ConditionalBranch:
 				if (!add_target(terminator.true_block) || !add_target(terminator.false_block)) {
-					return Fail(error, "value IR conditional branch target is missing");
+					return Fail("value IR conditional branch target is missing");
 				}
 				if (!validate_control_value(program.block_info[block_index].condition, Type::U1)) {
-					return Fail(error, "value IR conditional branch condition is invalid");
+					return Fail("value IR conditional branch condition is invalid");
 				}
 				break;
 			case CFG::TerminatorKind::IndirectBranch: {
 				if (!validate_control_value(program.block_info[block_index].indirect_target,
 				                            Type::U32)) {
-					return Fail(error, "value IR indirect branch selector is invalid");
+					return Fail("value IR indirect branch selector is invalid");
 				}
 				std::unordered_set<uint32_t> indirect_targets;
 				for (const auto target: terminator.indirect_targets) {
 					if (!indirect_targets.insert(target).second) {
-						return Fail(error, "value IR indirect branch target is duplicated");
+						return Fail("value IR indirect branch target is duplicated");
 					}
 					if (!add_target(target)) {
-						return Fail(error, "value IR indirect branch target is missing");
+						return Fail("value IR indirect branch target is missing");
 					}
 				}
 				if (terminator.indirect_selector_values.size() !=
 				    terminator.indirect_selector_targets.size()) {
-					return Fail(error, "value IR indirect selector table is inconsistent");
+					return Fail("value IR indirect selector table is inconsistent");
 				}
 				for (const auto target: terminator.indirect_selector_targets) {
 					const auto found = blocks_by_id.find(target);
 					if (found == blocks_by_id.end() ||
 					    !expected_successors.contains(found->second)) {
-						return Fail(error,
-						            "value IR indirect selector target is not a CFG successor");
+						return Fail("value IR indirect selector target is not a CFG successor");
 					}
 				}
 				break;
@@ -280,73 +285,73 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 		     !blocks_by_id.contains(terminator.merge_block)) ||
 		    (terminator.continue_block != UINT32_MAX &&
 		     !blocks_by_id.contains(terminator.continue_block))) {
-			return Fail(error, "value IR structured control target is missing");
+			return Fail("value IR structured control target is missing");
 		}
 		if (successors != expected_successors) {
-			return Fail(error, "value IR terminator and successor graph disagree");
+			return Fail("value IR terminator and successor graph disagree");
 		}
 
 		bool saw_non_phi = false;
 		for (const auto& inst: *block) {
 			if (inst.Parent() != block) {
-				return Fail(error, "value IR instruction has the wrong parent block");
+				return Fail("value IR instruction has the wrong parent block");
 			}
 			if (inst.GetOpcode() == ValueOpcode::Phi) {
 				if (saw_non_phi) {
-					return Fail(error, "value IR Phi appears after a non-Phi instruction");
+					return Fail("value IR Phi appears after a non-Phi instruction");
 				}
 				if (inst.NumPhiBlocks() != inst.NumArgs()) {
-					return Fail(error, "value IR Phi parent table is inconsistent");
+					return Fail("value IR Phi parent table is inconsistent");
 				}
 				if (inst.NumArgs() == 0) {
-					return Fail(error, "value IR Phi has no incoming values");
+					return Fail("value IR Phi has no incoming values");
 				}
 				std::unordered_set<const Block*> incoming_blocks;
 				for (size_t arg_index = 0; arg_index < inst.NumArgs(); arg_index++) {
 					const auto* predecessor = inst.PhiBlock(arg_index);
 					if (predecessor == nullptr || !blocks.contains(predecessor) ||
 					    !predecessors.contains(predecessor)) {
-						return Fail(error, "value IR Phi has a foreign or non-predecessor parent");
+						return Fail("value IR Phi has a foreign or non-predecessor parent");
 					}
 					if (!incoming_blocks.insert(predecessor).second) {
-						return Fail(error, "value IR Phi parent is duplicated");
+						return Fail("value IR Phi parent is duplicated");
 					}
 					if (inst.Arg(arg_index).GetType() != inst.GetType()) {
-						return Fail(error, "value IR Phi incoming type does not match its result");
+						return Fail("value IR Phi incoming type does not match its result");
 					}
 				}
 				if (incoming_blocks != predecessors) {
-					return Fail(error, "value IR Phi does not cover every predecessor");
+					return Fail("value IR Phi does not cover every predecessor");
 				}
 			} else {
 				saw_non_phi = true;
 			}
 			if (require_ssa && IsRegisterStatePseudo(inst.GetOpcode())) {
-				return Fail(error, fmt::format("register-state pseudo {} survived SSA rewrite",
-				                               ValueOpcodeName(inst.GetOpcode())));
+				return Fail(fmt::format("register-state pseudo {} survived SSA rewrite",
+				                        ValueOpcodeName(inst.GetOpcode())));
 			}
 			if (inst.GetOpcode() != ValueOpcode::Phi && inst.GetOpcode() != ValueOpcode::Identity &&
 			    inst.GetType() == Type::Opaque) {
-				return Fail(error, fmt::format("untyped opcode {} survived translation",
-				                               ValueOpcodeName(inst.GetOpcode())));
+				return Fail(fmt::format("untyped opcode {} survived translation",
+				                        ValueOpcodeName(inst.GetOpcode())));
 			}
 			const bool fixed_signature =
 			    inst.GetOpcode() != ValueOpcode::Phi && inst.GetOpcode() != ValueOpcode::Identity;
 			if (fixed_signature && inst.NumArgs() != NumArgsOf(inst.GetOpcode())) {
-				return Fail(error, fmt::format("{} has {} arguments, expected {}",
-				                               ValueOpcodeName(inst.GetOpcode()), inst.NumArgs(),
-				                               NumArgsOf(inst.GetOpcode())));
+				return Fail(fmt::format("{} has {} arguments, expected {}",
+				                        ValueOpcodeName(inst.GetOpcode()), inst.NumArgs(),
+				                        NumArgsOf(inst.GetOpcode())));
 			}
 			if (inst.GetOpcode() == ValueOpcode::ReadConstBuffer) {
 				const auto memory_index = inst.Flags<MemoryFlags>().index;
 				if (memory_index >= program.memory_info.size()) {
-					return Fail(error, fmt::format("{} has an invalid memory-info index",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has an invalid memory-info index",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				const auto& memory = program.memory_info[memory_index];
 				if (memory.kind != ResourceKind::ScalarBuffer) {
-					return Fail(error, fmt::format("{} has an invalid scalar-memory resource kind",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has an invalid scalar-memory resource kind",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				const bool valid_group_width =
 				    memory.component_count == 1u || memory.component_count == 2u ||
@@ -354,23 +359,23 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 				    memory.component_count == 16u;
 				if (memory.data_bits != 32u || memory.data_dwords != 1u || !valid_group_width ||
 				    memory.component_index >= memory.component_count) {
-					return Fail(error, fmt::format("{} has inconsistent scalar-memory metadata",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has inconsistent scalar-memory metadata",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 			}
 			const auto address_info = AddressOpcodeInfoOf(inst.GetOpcode());
 			if (address_info.access != AddressAccess::None) {
 				const auto memory_index = inst.Flags<MemoryFlags>().index;
 				if (memory_index >= program.memory_info.size()) {
-					return Fail(error, fmt::format("{} has an invalid memory-info index",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has an invalid memory-info index",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				const auto& memory = program.memory_info[memory_index];
 				if (!IsAddressResourceKind(memory.kind) ||
 				    (memory.kind == ResourceKind::ScalarAddress &&
 				     inst.GetOpcode() != ValueOpcode::LoadAddressU32)) {
-					return Fail(error, fmt::format("{} has an invalid address resource kind",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has an invalid address resource kind",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				const bool scalar_address = memory.kind == ResourceKind::ScalarAddress;
 				const bool valid_group_width =
@@ -382,50 +387,50 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 				if (memory.data_bits != address_info.data_bits || memory.data_dwords != 1u ||
 				    !valid_group_width || memory.component_index >= memory.component_count ||
 				    memory.sampler != 0u) {
-					return Fail(error, fmt::format("{} has inconsistent address-memory metadata",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has inconsistent address-memory metadata",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 			}
 			const auto buffer_components = BufferComponentCount(inst.GetOpcode());
 			if (buffer_components != 0u) {
 				const auto memory_index = inst.Flags<MemoryFlags>().index;
 				if (memory_index >= program.memory_info.size()) {
-					return Fail(error, fmt::format("{} has an invalid memory-info index",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has an invalid memory-info index",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				const auto& memory = program.memory_info[memory_index];
 				if (memory.kind != ResourceKind::Buffer &&
 				    memory.kind != ResourceKind::ScalarBuffer) {
-					return Fail(error, fmt::format("{} has a non-buffer resource kind",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has a non-buffer resource kind",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				if (buffer_components > 1u &&
 				    (memory.kind != ResourceKind::Buffer || memory.data_bits != 32u ||
 				     memory.data_dwords != buffer_components || memory.component_index != 0u)) {
-					return Fail(error, fmt::format("{} has inconsistent native-wide metadata",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has inconsistent native-wide metadata",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				if (buffer_components == 1u &&
 				    (inst.GetOpcode() == ValueOpcode::LoadBufferU32 ||
 				     inst.GetOpcode() == ValueOpcode::StoreBufferU32) &&
 				    memory.data_dwords != 1u) {
-					return Fail(error, fmt::format("{} retains scalar-sibling width metadata",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} retains scalar-sibling width metadata",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 			}
 			const auto shared_components = SharedComponentCount(inst.GetOpcode());
 			if (shared_components != 0u) {
 				const auto memory_index = inst.Flags<MemoryFlags>().index;
 				if (memory_index >= program.memory_info.size()) {
-					return Fail(error, fmt::format("{} has an invalid memory-info index",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has an invalid memory-info index",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				const auto& memory = program.memory_info[memory_index];
 				if ((memory.kind != ResourceKind::Lds && memory.kind != ResourceKind::Gds) ||
 				    memory.resource != 0u || memory.sampler != 0u || memory.component_count == 0u ||
 				    memory.component_index >= memory.component_count) {
-					return Fail(error, fmt::format("{} has invalid shared-memory metadata",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has invalid shared-memory metadata",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				uint32_t expected_bits = 32u;
 				if (inst.GetOpcode() == ValueOpcode::LoadSharedU8 ||
@@ -437,26 +442,26 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 				}
 				if (memory.data_bits != expected_bits || memory.data_dwords != shared_components ||
 				    (shared_components > 1u && memory.component_index != 0u)) {
-					return Fail(error, fmt::format("{} has inconsistent shared-memory width",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has inconsistent shared-memory width",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 			}
 			const auto image_info = ImageOpcodeInfoOf(inst.GetOpcode());
 			if (image_info.access != ImageAccess::None) {
 				const auto memory_index = inst.Flags<MemoryFlags>().index;
 				if (memory_index >= program.memory_info.size()) {
-					return Fail(error, fmt::format("{} has an invalid memory-info index",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has an invalid memory-info index",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 				const auto& memory = program.memory_info[memory_index];
 				if (!ImageResourceKindMatches(memory.kind, image_info.resource_class)) {
-					return Fail(error, fmt::format("{} has invalid image-memory metadata",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(fmt::format("{} has invalid image-memory metadata",
+					                        ValueOpcodeName(inst.GetOpcode())));
 				}
 			}
 			if (inst.GetOpcode() == ValueOpcode::SetAttribute &&
 			    inst.Flags<ExportFlags>().index >= program.export_info.size()) {
-				return Fail(error, "SetAttribute has an invalid export-info index");
+				return Fail("SetAttribute has an invalid export-info index");
 			}
 			uint32_t composite_components = 0u;
 			switch (inst.GetOpcode()) {
@@ -469,25 +474,24 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 			if (composite_components != 0u &&
 			    (!inst.Arg(1).IsImmediate() || inst.Arg(1).GetType() != Type::U32 ||
 			     inst.Arg(1).U32() >= composite_components)) {
-				return Fail(error, fmt::format("{} has an invalid component index",
-				                               ValueOpcodeName(inst.GetOpcode())));
+				return Fail(fmt::format("{} has an invalid component index",
+				                        ValueOpcodeName(inst.GetOpcode())));
 			}
 			for (size_t arg_index = 0; arg_index < inst.NumArgs(); arg_index++) {
 				const auto arg = inst.Arg(arg_index);
 				if (arg.IsEmpty()) {
-					return Fail(error, fmt::format("{} has an empty argument",
-					                               ValueOpcodeName(inst.GetOpcode())));
+					return Fail(
+					    fmt::format("{} has an empty argument", ValueOpcodeName(inst.GetOpcode())));
 				}
 				if (fixed_signature && arg.GetType() != ArgTypeOf(inst.GetOpcode(), arg_index)) {
-					return Fail(error,
-					            fmt::format("{} argument {} has type {}, expected {}",
+					return Fail(fmt::format("{} argument {} has type {}, expected {}",
 					                        ValueOpcodeName(inst.GetOpcode()), arg_index,
 					                        TypeName(arg.GetType()),
 					                        TypeName(ArgTypeOf(inst.GetOpcode(), arg_index))));
 				}
 				if (const auto* definition = arg.TryInstruction();
 				    definition != nullptr && !instructions.contains(definition)) {
-					return Fail(error, "value IR argument has a foreign definition");
+					return Fail("value IR argument has a foreign definition");
 				}
 				if (const auto* definition = arg.TryInstruction(); definition != nullptr) {
 					const auto& uses = definition->Uses();
@@ -495,8 +499,7 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 						return candidate.user == &inst && candidate.operand == arg_index;
 					});
 					if (use == uses.end()) {
-						return Fail(error,
-						            fmt::format("{} argument {} is absent from {} reverse uses",
+						return Fail(fmt::format("{} argument {} is absent from {} reverse uses",
 						                        ValueOpcodeName(inst.GetOpcode()), arg_index,
 						                        ValueOpcodeName(definition->GetOpcode())));
 					}
@@ -506,7 +509,7 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 	}
 
 	if (program.blocks.empty()) {
-		return true;
+		return;
 	}
 
 	std::vector<bool>   reachable(program.blocks.size(), false);
@@ -523,7 +526,7 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 		}
 	}
 	if (!std::ranges::all_of(reachable, [](bool value) { return value; })) {
-		return Fail(error, "value IR contains an unreachable block");
+		return Fail("value IR contains an unreachable block");
 	}
 
 	std::vector<std::vector<bool>> dominators(program.blocks.size(),
@@ -577,28 +580,25 @@ bool ValidateProgram(const Program& program, bool require_ssa, std::string* erro
 					const auto* predecessor = inst.PhiBlock(arg_index);
 					if (definition->Parent() != predecessor &&
 					    !dominates(definition->Parent(), predecessor)) {
-						return Fail(error,
-						            "value IR Phi incoming definition does not dominate its edge");
+						return Fail("value IR Phi incoming definition does not dominate its edge");
 					}
 				} else if (definition->Parent() == block) {
 					if (instruction_positions.at(definition) >= instruction_positions.at(&inst)) {
-						return Fail(error,
-						            "value IR instruction uses a same-block definition before it");
+						return Fail("value IR instruction uses a same-block definition before it");
 					}
 				} else if (!dominates(definition->Parent(), block)) {
-					return Fail(error, "value IR instruction definition does not dominate its use");
+					return Fail("value IR instruction definition does not dominate its use");
 				}
 			}
 		}
 		const auto& info = program.block_info[block_index];
 		if (!info.condition.IsEmpty() && !control_dominates(info.condition, block)) {
-			return Fail(error, "value IR branch condition definition does not dominate its use");
+			return Fail("value IR branch condition definition does not dominate its use");
 		}
 		if (!info.indirect_target.IsEmpty() && !control_dominates(info.indirect_target, block)) {
-			return Fail(error, "value IR indirect selector definition does not dominate its use");
+			return Fail("value IR indirect selector definition does not dominate its use");
 		}
 	}
-	return true;
 }
 
 void ResolveControlFlowIdentities(Program& program) {

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

@@ -477,7 +477,6 @@ struct DescriptorSource {
 struct SrtRead {
 	Value    value;
 	uint32_t flat_offset = 0;
-	uint32_t use_pc      = 0;
 
 	bool operator==(const SrtRead& other) const = default;
 };
@@ -523,7 +522,7 @@ struct Program {
 
 std::string ProgramToString(const Program& program);
 
-bool  ValidateProgram(const Program& program, bool require_ssa, std::string* error);
+void  ValidateProgram(const Program& program, bool require_ssa);
 void  ResolveControlFlowIdentities(Program& program);
 bool  EquivalentValue(const Program& program, Value left, Value right);
 Value ResolveInvariantPhi(const Program& program, Value value);

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

@@ -23,12 +23,16 @@ VALUE_OPCODE(GetScc, U1)
 VALUE_OPCODE(SetScc, Void, U1)
 VALUE_OPCODE(GetExec, U1)
 VALUE_OPCODE(SetExec, Void, U1)
+VALUE_OPCODE(GetExecMaskTag, U1)
+VALUE_OPCODE(SetExecMaskTag, Void, U1)
 VALUE_OPCODE(GetExecLo, U32)
 VALUE_OPCODE(SetExecLo, Void, U32)
 VALUE_OPCODE(GetExecHi, U32)
 VALUE_OPCODE(SetExecHi, Void, U32)
 VALUE_OPCODE(GetVcc, U1)
 VALUE_OPCODE(SetVcc, Void, U1)
+VALUE_OPCODE(GetVccMaskValidTag, U1)
+VALUE_OPCODE(SetVccMaskValidTag, Void, U1)
 VALUE_OPCODE(GetVccLo, U32)
 VALUE_OPCODE(SetVccLo, Void, U32)
 VALUE_OPCODE(GetVccHi, U32)

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

@@ -1,5 +1,6 @@
 #include "graphics/shader/recompiler/ir/passes/BindingLayout.h"
 
+#include "common/assert.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 
 #include <algorithm>
@@ -9,12 +10,17 @@
 namespace Libs::Graphics::ShaderRecompiler::IR {
 namespace {
 
-constexpr auto     FirstImageBinding     = DescriptorBindingKind::Sampled1D;
+constexpr auto     FirstImageBinding = DescriptorBindingKind::Sampled1D;
 constexpr uint32_t ImageBindingCount =
     static_cast<uint32_t>(DescriptorBindingKind::StorageAtomic3D) -
     static_cast<uint32_t>(FirstImageBinding) + 1u;
 
-bool CollectUserData(const Program& program, std::vector<uint32_t>& result) {
+[[noreturn]] void BindingFail(const char* message) {
+	EXIT("shader binding layout failed: %s", message);
+	std::abort();
+}
+
+std::vector<uint32_t> CollectUserData(const Program& program) {
 	std::array<bool, NumScalarRegs> registers {};
 	for (const auto* block: program.blocks) {
 		for (const auto& inst: *block) {
@@ -22,22 +28,22 @@ bool CollectUserData(const Program& program, std::vector<uint32_t>& result) {
 				continue;
 			}
 			if (inst.Arg(0).GetType() != Type::ScalarReg) {
-				return false;
+				BindingFail("typed shader contains an invalid user-data register");
 			}
 			const auto index = RegIndex(inst.Arg(0).ScalarRegister());
 			if (index >= NumScalarRegs) {
-				return false;
+				BindingFail("typed shader contains an invalid user-data register");
 			}
 			registers[index] = true;
 		}
 	}
-	result.clear();
+	std::vector<uint32_t> result;
 	for (uint32_t index = 0; index < registers.size(); index++) {
 		if (registers[index]) {
 			result.push_back(index);
 		}
 	}
-	return true;
+	return result;
 }
 
 void AddBinding(BindingLayout& layout, DescriptorBindingKind kind,
@@ -45,8 +51,8 @@ void AddBinding(BindingLayout& layout, DescriptorBindingKind kind,
 	layout.descriptors.push_back({kind, std::move(resources)});
 }
 
-bool UsesGds(const Program& program, bool& uses_gds) {
-	uses_gds = false;
+bool UsesGds(const Program& program) {
+	bool uses_gds = false;
 	for (const auto* block: program.blocks) {
 		for (const auto& inst: *block) {
 			if (SharedAccessOf(inst.GetOpcode()) == SharedAccess::None) {
@@ -54,44 +60,36 @@ bool UsesGds(const Program& program, bool& uses_gds) {
 			}
 			const auto index = inst.Flags<MemoryFlags>().index;
 			if (index >= program.memory_info.size()) {
-				return false;
+				BindingFail("typed shader contains invalid shared-memory metadata");
 			}
 			const auto kind = program.memory_info[index].kind;
 			if (kind != ResourceKind::Lds && kind != ResourceKind::Gds) {
-				return false;
+				BindingFail("typed shader contains invalid shared-memory metadata");
 			}
 			uses_gds |= kind == ResourceKind::Gds;
 		}
 	}
-	return true;
+	return uses_gds;
 }
 
 } // namespace
 
-bool AllocateBindings(Program& program, uint32_t push_constant_offset, std::string* error) {
+void AllocateBindings(Program& program, uint32_t push_constant_offset) {
 	if (!program.shader_info_complete || program.binding_layout_complete) {
-		if (error != nullptr) {
-			*error = !program.shader_info_complete ? "shader info is not ready"
-			                                       : "binding layout already allocated";
-		}
-		return false;
+		EXIT("shader binding layout failed: %s", !program.shader_info_complete
+		                                             ? "shader info is not ready"
+		                                             : "binding layout already allocated");
 	}
 	if (push_constant_offset > NativePushConstantSize ||
 	    push_constant_offset % sizeof(uint32_t) != 0) {
-		if (error != nullptr) {
-			*error = "push-constant offset exceeds the Vulkan minimum guarantee or is unaligned";
-		}
-		return false;
+		EXIT("shader binding layout failed: push-constant offset %u exceeds the Vulkan minimum "
+		     "guarantee or is unaligned",
+		     push_constant_offset);
 	}
 
 	BindingLayout next;
 	next.push_constant_offset = push_constant_offset;
-	if (!CollectUserData(program, next.user_data_registers)) {
-		if (error != nullptr) {
-			*error = "typed shader contains an invalid user-data register";
-		}
-		return false;
-	}
+	next.user_data_registers = CollectUserData(program);
 	next.memory_offset_dword = static_cast<uint32_t>(next.user_data_registers.size());
 	next.memory_offset_count = static_cast<uint32_t>(program.info.buffers.size());
 
@@ -107,26 +105,18 @@ bool AllocateBindings(Program& program, uint32_t push_constant_offset, std::stri
 	for (uint32_t i = 0; i < program.info.images.size(); i++) {
 		const auto kind = DescriptorBindingForImage(program.info.images[i]);
 		if (!kind.has_value()) {
-			if (error != nullptr) {
-				*error = "shader info contains an invalid image binding class";
-			}
-			return false;
+			EXIT("shader binding layout failed: image %u has an invalid binding class", i);
 		}
 		const auto group = static_cast<uint32_t>(*kind) - static_cast<uint32_t>(FirstImageBinding);
 		if (group >= image_groups.size()) {
-			if (error != nullptr) {
-				*error = "shader info contains an unmapped image binding class";
-			}
-			return false;
+			EXIT("shader binding layout failed: image %u has an unmapped binding class", i);
 		}
 		auto&      resources = image_groups[group];
 		const auto dynamic   = program.info.images[i].mip_mode == ImageMipMode::DynamicStorage;
 		const auto count     = dynamic ? program.info.images[i].mip_count : 1u;
 		if (count == 0u || (!dynamic && program.info.images[i].mip_count != 1u)) {
-			if (error != nullptr) {
-				*error = "image has an invalid specialized mip descriptor count";
-			}
-			return false;
+			EXIT("shader binding layout failed: image %u has invalid specialized mip count %u", i,
+			     program.info.images[i].mip_count);
 		}
 		resources.insert(resources.end(), count, i);
 	}
@@ -146,14 +136,7 @@ bool AllocateBindings(Program& program, uint32_t push_constant_offset, std::stri
 		}
 		AddBinding(next, DescriptorBindingKind::Samplers, std::move(resources));
 	}
-	bool uses_gds = false;
-	if (!UsesGds(program, uses_gds)) {
-		if (error != nullptr) {
-			*error = "typed shader contains invalid shared-memory metadata";
-		}
-		return false;
-	}
-	if (uses_gds) {
+	if (UsesGds(program)) {
 		AddBinding(next, DescriptorBindingKind::Gds);
 	}
 	if (program.info.uses_dma) {
@@ -178,7 +161,6 @@ bool AllocateBindings(Program& program, uint32_t push_constant_offset, std::stri
 
 	program.bindings                = std::move(next);
 	program.binding_layout_complete = true;
-	return true;
 }
 
 const DescriptorBinding* FindBinding(const BindingLayout& layout, DescriptorBindingKind kind) {

src/graphics/shader/recompiler/ir/passes/BindingLayout.h

@@ -5,7 +5,7 @@
 
 namespace Libs::Graphics::ShaderRecompiler::IR {
 
-bool AllocateBindings(Program& program, uint32_t push_constant_offset, std::string* error);
+void AllocateBindings(Program& program, uint32_t push_constant_offset);
 
 const DescriptorBinding* FindBinding(const BindingLayout& layout, DescriptorBindingKind kind);
 

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

@@ -1,5 +1,6 @@
 #include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 
+#include "common/assert.h"
 #include "graphics/guest_gpu/gpu_format.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 #include "graphics/shader/shaderBindings.h"
@@ -18,6 +19,11 @@ namespace {
 constexpr uint64_t AddressMask            = 0x0000ffffffffffffull;
 constexpr uint64_t MaxIndirectImageProbes = 65536u;
 
+[[noreturn]] void SpecializationFail(std::string_view message) {
+	EXIT("shader resource specialization failed: %s", std::string(message).c_str());
+	std::abort();
+}
+
 Decoder::ImageDimension DescriptorDimension(const DescriptorValue&  descriptor,
                                             Decoder::ImageDimension requested) {
 	const bool is_array = requested == Decoder::ImageDimension::Dim1DArray ||
@@ -79,7 +85,7 @@ uint32_t DescriptorImageSwizzle(const DescriptorValue& descriptor) {
 
 Prospero::BufferFormat ImageConversionFormat(Prospero::BufferFormat format) {
 	return Prospero::RemapTextureFormat(format) != format ? format
-	                                                     : Prospero::BufferFormat::kInvalid;
+	                                                      : Prospero::BufferFormat::kInvalid;
 }
 
 bool DescriptorIsCube(const DescriptorValue& descriptor) {
@@ -153,8 +159,7 @@ bool ReadSpecializationWord(const SrtRuntime& runtime, uint64_t address, uint32_
 }
 
 bool ReadScalarBufferWord(const ShaderBufferResource& descriptor, uint32_t dynamic_offset,
-                          uint32_t immediate_offset, const SrtRuntime& runtime, uint32_t& word,
-                          std::string* error, uint32_t use_pc) {
+                          uint32_t immediate_offset, const SrtRuntime& runtime, uint32_t& word) {
 	const auto byte_offset = static_cast<uint64_t>(dynamic_offset) + immediate_offset;
 	const auto aligned     = byte_offset & ~uint64_t {3};
 	const auto size        = ScalarBufferSize(descriptor);
@@ -164,20 +169,10 @@ bool ReadScalarBufferWord(const ShaderBufferResource& descriptor, uint32_t dynam
 	}
 	const auto base = descriptor.Base48() & ~uint64_t {3};
 	if (aligned > AddressMask - base) {
-		if (error != nullptr) {
-			*error = fmt::format("indirect image scalar read at pc 0x{:08x} exceeds the "
-			                     "48-bit address space",
-			                     use_pc);
-		}
 		return false;
 	}
 	const auto address = base + aligned;
 	if (!ReadSpecializationWord(runtime, address, word)) {
-		if (error != nullptr) {
-			*error = fmt::format("indirect image scalar read at pc 0x{:08x} failed at "
-			                     "0x{:016x}",
-			                     use_pc, address);
-		}
 		return false;
 	}
 	return true;
@@ -186,23 +181,14 @@ bool ReadScalarBufferWord(const ShaderBufferResource& descriptor, uint32_t dynam
 bool MaterializeIndirectImage(const DescriptorSource::IndirectImage& indirect,
                               const DescriptorValue&                 material_value,
                               const DescriptorValue& heap_value, const SrtRuntime& runtime,
-                              ResourceSnapshot::IndirectImage& result, std::string* error,
-                              uint32_t use_pc) {
+                              ResourceSnapshot::IndirectImage& result) {
 	ShaderBufferResource material;
 	ShaderBufferResource heap;
 	if (!DecodeBufferDescriptor(material_value, material) ||
 	    !DecodeBufferDescriptor(heap_value, heap)) {
-		if (error != nullptr) {
-			*error = fmt::format("indirect image tables at pc 0x{:08x} have invalid descriptors",
-			                     use_pc);
-		}
 		return false;
 	}
 	if (material.Stride() != indirect.selector_stride) {
-		if (error != nullptr) {
-			*error =
-			    fmt::format("indirect image material table at pc 0x{:08x} changed stride", use_pc);
-		}
 		return false;
 	}
 
@@ -215,11 +201,6 @@ bool MaterializeIndirectImage(const DescriptorSource::IndirectImage& indirect,
 	const auto limit       = std::min<uint64_t>(UINT32_MAX, size + 3u);
 	const auto probe_count = residue <= limit ? (limit - residue) / step + 1u : 0u;
 	if (probe_count > MaxIndirectImageProbes) {
-		if (error != nullptr) {
-			*error = fmt::format("indirect image material table at pc 0x{:08x} requires {} "
-			                     "bounded probes",
-			                     use_pc, probe_count);
-		}
 		return false;
 	}
 
@@ -227,8 +208,7 @@ bool MaterializeIndirectImage(const DescriptorSource::IndirectImage& indirect,
 	std::unordered_set<uint32_t> seen {0u};
 	for (uint64_t offset = residue; offset <= limit && probe_count != 0u; offset += step) {
 		uint32_t key = 0;
-		if (!ReadScalarBufferWord(material, static_cast<uint32_t>(offset), 0u, runtime, key, error,
-		                          use_pc)) {
+		if (!ReadScalarBufferWord(material, static_cast<uint32_t>(offset), 0u, runtime, key)) {
 			return false;
 		}
 		if (seen.insert(key).second) {
@@ -248,7 +228,7 @@ bool MaterializeIndirectImage(const DescriptorSource::IndirectImage& indirect,
 		const auto heap_offset = key << 5u;
 		for (uint32_t dword = 0; dword < candidate.dword_count; dword++) {
 			if (!ReadScalarBufferWord(heap, heap_offset, dword * sizeof(uint32_t), runtime,
-			                          candidate.dwords[dword], error, use_pc)) {
+			                          candidate.dwords[dword])) {
 				return false;
 			}
 		}
@@ -281,35 +261,22 @@ size_t FlattenedRuntimeDwords(const Program& program) {
 
 } // namespace
 
-bool ValidateResourceSnapshot(const Program& program, const ResourceSnapshot& snapshot,
-                              std::string* error) {
+bool ValidateResourceSnapshot(const Program& program, const ResourceSnapshot& snapshot) {
 	if (!program.resource_tracking_complete) {
-		if (error != nullptr) {
-			*error = "shader resources were not tracked";
-		}
 		return false;
 	}
 	if (snapshot.buffers.size() != program.info.buffers.size() ||
 	    snapshot.images.size() != program.info.images.size() ||
 	    snapshot.samplers.size() != program.info.samplers.size()) {
-		if (error != nullptr) {
-			*error = "resource snapshot does not match dense shader topology";
-		}
 		return false;
 	}
 	if (snapshot.flattened_srt.size() != FlattenedRuntimeDwords(program)) {
-		if (error != nullptr) {
-			*error = "flattened SRT snapshot does not match the shader plan";
-		}
 		return false;
 	}
 	if (program.binding_layout_complete) {
 		for (const auto reg: program.bindings.user_data_registers) {
 			if (reg < program.user_data_base ||
 			    reg - program.user_data_base >= snapshot.user_data.size()) {
-				if (error != nullptr) {
-					*error = fmt::format("runtime snapshot is missing user SGPR {}", reg);
-				}
 				return false;
 			}
 		}
@@ -317,19 +284,12 @@ bool ValidateResourceSnapshot(const Program& program, const ResourceSnapshot& sn
 	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {
 		const auto alias = program.info.buffers[i].image_alias;
 		if (alias != BufferResource::NoImageAlias && alias >= program.info.images.size()) {
-			if (error != nullptr) {
-				*error = fmt::format("buffer resource {} has invalid image alias {}", i, alias);
-			}
 			return false;
 		}
 	}
-	const auto CheckWidth = [&](const auto& values, uint32_t width, const char* kind) {
+	const auto CheckWidth = [&](const auto& values, uint32_t width) {
 		for (uint32_t i = 0; i < values.size(); i++) {
 			if (values[i].dword_count != width) {
-				if (error != nullptr) {
-					*error = fmt::format("{} descriptor {} has {} dwords", kind, i,
-					                     values[i].dword_count);
-				}
 				return false;
 			}
 		}
@@ -345,56 +305,37 @@ bool ValidateResourceSnapshot(const Program& program, const ResourceSnapshot& sn
 		    table.keys.empty() || table.keys.size() != table.candidates.size() ||
 		    table.capacity < table.keys.size() || table.descriptors.empty() ||
 		    !indirect_resources.insert(table.resource).second) {
-			if (error != nullptr) {
-				*error = "indirect image snapshot does not match the shader plan";
-			}
 			return false;
 		}
 		for (const auto candidate: table.candidates) {
 			if (candidate >= table.descriptors.size()) {
-				if (error != nullptr) {
-					*error = "indirect image snapshot has an invalid candidate";
-				}
 				return false;
 			}
 		}
 		if (snapshot.images[table.resource] != table.descriptors[table.candidates[0]]) {
-			if (error != nullptr) {
-				*error = "indirect image snapshot has an inconsistent root descriptor";
-			}
 			return false;
 		}
 	}
-	return CheckWidth(snapshot.buffers, 4, "buffer") && CheckWidth(snapshot.images, 8, "image") &&
-	       CheckWidth(snapshot.samplers, 4, "sampler");
+	return CheckWidth(snapshot.buffers, 4) && CheckWidth(snapshot.images, 8) &&
+	       CheckWidth(snapshot.samplers, 4);
 }
 
-bool ValidateResourceSpecialization(const Program& program, const ResourceSnapshot& snapshot,
-                                    std::string* error) {
-	if (!ValidateResourceSnapshot(program, snapshot, error)) {
+bool ValidateResourceSpecialization(const Program& program, const ResourceSnapshot& snapshot) {
+	if (!ValidateResourceSnapshot(program, snapshot)) {
 		return false;
 	}
 	if (!snapshot.indirect_images.empty()) {
-		if (error != nullptr) {
-			*error = "indirect image snapshot was not specialized";
-		}
 		return false;
 	}
 	for (uint32_t i = 0; i < program.info.buffers.size(); i++) {
 		const auto&          buffer = program.info.buffers[i];
 		ShaderBufferResource descriptor;
 		if (!DecodeBufferDescriptor(snapshot.buffers[i], descriptor)) {
-			if (error != nullptr) {
-				*error = fmt::format("buffer descriptor {} has invalid width", i);
-			}
 			return false;
 		}
 		if (buffer.packed_stride != descriptor.PackedStride() ||
 		    buffer.descriptor_format != descriptor.Format() ||
 		    buffer.descriptor_swizzle != descriptor.DstSelXYZW()) {
-			if (error != nullptr) {
-				*error = fmt::format("buffer descriptor {} no longer matches specialization", i);
-			}
 			return false;
 		}
 	}
@@ -403,9 +344,6 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 		const auto& descriptor = snapshot.images[i];
 		const auto  mip_count  = StorageMipCount(image, descriptor);
 		if (mip_count == 0u || mip_count != image.mip_count) {
-			if (error != nullptr) {
-				*error = fmt::format("storage image descriptor {} changed mip range", i);
-			}
 			return false;
 		}
 		if (NullImageDescriptor(descriptor)) {
@@ -419,10 +357,6 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 				    program.info.images[root].conversion_format != image.conversion_format ||
 				    program.info.images[root].shader_swizzle != image.shader_swizzle ||
 				    program.info.images[root].cube != image.cube) {
-					if (error != nullptr) {
-						*error = fmt::format(
-						    "null indirect image descriptor {} changed binding class", i);
-					}
 					return false;
 				}
 				continue;
@@ -434,10 +368,6 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 			}
 			if (image.dimension != Decoder::ImageDimension::Dim2D || image.cube ||
 			    !canonical_kind) {
-				if (error != nullptr) {
-					*error = fmt::format(
-					    "image descriptor {} no longer matches canonical null specialization", i);
-				}
 				return false;
 			}
 			continue;
@@ -445,14 +375,6 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 		const auto dimension = DescriptorDimension(descriptor, image.dimension);
 		if (dimension == Decoder::ImageDimension::Unknown || dimension != image.dimension ||
 		    DescriptorIsCube(descriptor) != image.cube) {
-			if (error != nullptr) {
-				*error =
-				    fmt::format("image descriptor {} no longer matches specialized dimension: "
-				                "{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x}",
-				                i, descriptor.dwords[0], descriptor.dwords[1], descriptor.dwords[2],
-				                descriptor.dwords[3], descriptor.dwords[4], descriptor.dwords[5],
-				                descriptor.dwords[6], descriptor.dwords[7]);
-			}
 			return false;
 		}
 		if (image.kind == ResourceKind::Image || image.kind == ResourceKind::ImageUint ||
@@ -463,49 +385,25 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 			const auto format =
 			    static_cast<Prospero::BufferFormat>((descriptor.dwords[1] >> 20u) & 0x1ffu);
 			if (image.atomic && format != Prospero::BufferFormat::k32UInt) {
-				if (error != nullptr) {
-					*error = fmt::format(
-					    "atomic image descriptor {} changed to unsupported format {}", i,
-					    static_cast<uint32_t>(format));
-				}
 				return false;
 			}
 			const auto conversion_format = ImageConversionFormat(format);
 			if (image.conversion_format != conversion_format) {
-				if (error != nullptr) {
-					*error = fmt::format("image descriptor {} changed format conversion", i);
-				}
 				return false;
 			}
 			if ((storage || conversion_format != Prospero::BufferFormat::kInvalid) &&
 			    image.shader_swizzle != DescriptorImageSwizzle(descriptor)) {
-				if (error != nullptr) {
-					*error = fmt::format("image descriptor {} changed swizzle", i);
-				}
 				return false;
 			}
-			const bool raw_sint_storage =
-			    storage && format == Prospero::BufferFormat::k32SInt && image.written &&
-			    !image.read && !image.atomic;
-			const bool uint_descriptor =
-			    Prospero::IsUintTextureFormat(format) || raw_sint_storage;
-			const auto uint_program = image.kind == ResourceKind::ImageUint ||
-			                          image.kind == ResourceKind::StorageImageUint;
+			const bool raw_sint_storage = storage && format == Prospero::BufferFormat::k32SInt &&
+			                              image.written && !image.read && !image.atomic;
+			const bool uint_descriptor  = Prospero::IsUintTextureFormat(format) || raw_sint_storage;
+			const auto uint_program     = image.kind == ResourceKind::ImageUint ||
+			                              image.kind == ResourceKind::StorageImageUint;
 			if (uint_descriptor != uint_program && !(image.atomic && uint_program)) {
-				if (error != nullptr) {
-					*error = fmt::format(
-					    "image descriptor {} no longer matches specialized format: "
-					    "{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x},{:08x}",
-					    i, descriptor.dwords[0], descriptor.dwords[1], descriptor.dwords[2],
-					    descriptor.dwords[3], descriptor.dwords[4], descriptor.dwords[5],
-					    descriptor.dwords[6], descriptor.dwords[7]);
-				}
 				return false;
 			}
 			if (image.depth_compare && uint_program) {
-				if (error != nullptr) {
-					*error = fmt::format("integer image descriptor {} uses depth comparison", i);
-				}
 				return false;
 			}
 		}
@@ -514,11 +412,8 @@ bool ValidateResourceSpecialization(const Program& program, const ResourceSnapsh
 }
 
 bool MaterializeResources(const Program& program, const SrtRuntime& runtime,
-                          ResourceSnapshot& snapshot, std::string* error) {
+                          ResourceSnapshot& snapshot) {
 	if (!program.resource_tracking_complete) {
-		if (error != nullptr) {
-			*error = "shader resources were not tracked";
-		}
 		return false;
 	}
 
@@ -527,17 +422,12 @@ bool MaterializeResources(const Program& program, const SrtRuntime& runtime,
 	requests.reserve(program.info.buffers.size() + program.info.images.size() * 2u +
 	                 program.info.samplers.size());
 	for (const auto& buffer: program.info.buffers) {
-		requests.push_back({buffer.source, buffer.first_use_pc});
+		requests.push_back({buffer.source});
 	}
 	for (const auto& image: program.info.images) {
 		const auto* source = Source(program, image.source);
 		if (source != nullptr && source->indirect_image.has_value()) {
 			if (runtime.read_specialization_memory == nullptr) {
-				if (error != nullptr) {
-					*error = fmt::format("indirect image table at pc 0x{:08x} has no clean "
-					                     "memory reader",
-					                     image.first_use_pc);
-				}
 				return false;
 			}
 			MarkCleanFlatSlots(program, Source(program, source->indirect_image->material_source),
@@ -545,16 +435,16 @@ bool MaterializeResources(const Program& program, const SrtRuntime& runtime,
 			MarkCleanFlatSlots(program, Source(program, source->indirect_image->heap_source),
 			                   clean_flat_slots);
 		} else {
-			requests.push_back({image.source, image.first_use_pc});
+			requests.push_back({image.source});
 		}
 	}
 	for (const auto& sampler: program.info.samplers) {
-		requests.push_back({sampler.source, sampler.first_use_pc});
+		requests.push_back({sampler.source});
 	}
 	std::vector<DescriptorValue> values;
 	std::vector<uint32_t>        flattened_srt;
-	if (!EvaluateRuntimeSources(program, requests, runtime, values, flattened_srt, clean_flat_slots,
-	                            error)) {
+	if (!EvaluateRuntimeSources(program, requests, runtime, values, flattened_srt,
+	                            clean_flat_slots)) {
 		return false;
 	}
 
@@ -581,20 +471,19 @@ bool MaterializeResources(const Program& program, const SrtRuntime& runtime,
 				continue;
 			}
 			const std::array requests {
-			    DescriptorSourceRequest {source->indirect_image->material_source,
-			                             image.first_use_pc},
-			    DescriptorSourceRequest {source->indirect_image->heap_source, image.first_use_pc}};
+			    DescriptorSourceRequest {source->indirect_image->material_source},
+			    DescriptorSourceRequest {source->indirect_image->heap_source}};
 			SrtRuntime clean_runtime  = runtime;
 			clean_runtime.read_memory = runtime.read_specialization_memory;
 			std::vector<DescriptorValue> tables;
-			if (!EvaluateDescriptorSources(program, requests, clean_runtime, tables, error)) {
+			if (!EvaluateDescriptorSources(program, requests, clean_runtime, tables)) {
 				return false;
 			}
 			const auto&                     material = tables[0];
 			const auto&                     heap     = tables[1];
 			ResourceSnapshot::IndirectImage table;
-			if (!MaterializeIndirectImage(*source->indirect_image, material, heap, runtime, table,
-			                              error, image.first_use_pc)) {
+			if (!MaterializeIndirectImage(*source->indirect_image, material, heap, runtime,
+			                              table)) {
 				return false;
 			}
 			if (image.indirect_root == ImageResource::NoIndirectImage) {
@@ -608,20 +497,12 @@ bool MaterializeResources(const Program& program, const SrtRuntime& runtime,
 			}
 			if (table.keys.size() > image.indirect_mapping_capacity ||
 			    table.descriptors.size() > image.indirect_resources.size()) {
-				if (error != nullptr) {
-					*error = fmt::format(
-					    "indirect image table at pc 0x{:08x} changed candidate topology",
-					    image.first_use_pc);
-				}
 				return false;
 			}
 			for (uint32_t candidate = 0; candidate < image.indirect_resources.size(); candidate++) {
 				const auto resource = image.indirect_resources[candidate];
 				if (resource >= program.info.images.size() ||
 				    program.info.images[resource].indirect_root != image_index) {
-					if (error != nullptr) {
-						*error = "indirect image specialization has an invalid candidate resource";
-					}
 					return false;
 				}
 				next.images[resource] =
@@ -649,31 +530,26 @@ bool MaterializeResources(const Program& program, const SrtRuntime& runtime,
 		}
 	}
 	if (std::ranges::find(image_written, uint8_t {0}) != image_written.end()) {
-		if (error != nullptr) {
-			*error = "indirect image specialization left an unmaterialized candidate";
-		}
 		return false;
 	}
 	next.samplers.assign(cursor, cursor + program.info.samplers.size());
 	next.user_data.assign(runtime.user_data.begin(), runtime.user_data.end());
-	if (!ValidateResourceSnapshot(program, next, error)) {
+	if (!ValidateResourceSnapshot(program, next)) {
 		return false;
 	}
 	snapshot = std::move(next);
 	return true;
 }
 
-bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::string* error) {
+void SpecializeResources(Program& program, ResourceSnapshot& snapshot) {
 	if (!program.resource_tracking_complete || program.shader_info_complete ||
 	    program.binding_layout_complete) {
-		if (error != nullptr) {
-			*error = !program.resource_tracking_complete ? "shader resources were not tracked"
-			                                             : "resource specialization is too late";
-		}
-		return false;
+		return SpecializationFail(!program.resource_tracking_complete
+		                              ? "shader resources were not tracked"
+		                              : "resource specialization is too late");
 	}
-	if (!ValidateResourceSnapshot(program, snapshot, error)) {
-		return false;
+	if (!ValidateResourceSnapshot(program, snapshot)) {
+		return SpecializationFail("invalid resource snapshot");
 	}
 
 	auto next          = program.info;
@@ -681,10 +557,8 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 	for (const auto& table: snapshot.indirect_images) {
 		if (table.resource >= next.images.size() || table.descriptors.size() < 2u ||
 		    next.images.size() + table.descriptors.size() - 1u > ShaderInfo::MaxImages) {
-			if (error != nullptr) {
-				*error = "indirect image candidates exceed the dense image resource limit";
-			}
-			return false;
+			return SpecializationFail(
+			    "indirect image candidates exceed the dense image resource limit");
 		}
 		const auto            root_image = next.images[table.resource];
 		std::vector<uint32_t> resources(table.descriptors.size());
@@ -720,10 +594,7 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 	for (uint32_t i = 0; i < next.buffers.size(); i++) {
 		ShaderBufferResource descriptor;
 		if (!DecodeBufferDescriptor(next_snapshot.buffers[i], descriptor)) {
-			if (error != nullptr) {
-				*error = fmt::format("buffer descriptor {} has invalid width", i);
-			}
-			return false;
+			return SpecializationFail(fmt::format("buffer descriptor {} has invalid width", i));
 		}
 		next.buffers[i].packed_stride      = descriptor.PackedStride();
 		next.buffers[i].descriptor_format  = descriptor.Format();
@@ -734,10 +605,8 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 		auto&       image      = next.images[i];
 		image.mip_count        = StorageMipCount(image, descriptor);
 		if (image.mip_count == 0u) {
-			if (error != nullptr) {
-				*error = fmt::format("storage image descriptor {} has an invalid mip range", i);
-			}
-			return false;
+			return SpecializationFail(
+			    fmt::format("storage image descriptor {} has an invalid mip range", i));
 		}
 		if (NullImageDescriptor(descriptor)) {
 			image.dimension = Decoder::ImageDimension::Dim2D;
@@ -755,37 +624,31 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 		}
 		const auto descriptor_dimension = DescriptorDimension(descriptor, image.dimension);
 		if (descriptor_dimension == Decoder::ImageDimension::Unknown) {
-			if (error != nullptr) {
-				*error = fmt::format(
-				    "image descriptor {} has unsupported type {}: {:08x},{:08x},{:08x},{:08x},"
-				    "{:08x},{:08x},{:08x},{:08x}",
-				    i, (descriptor.dwords[3] >> 28u) & 0xfu, descriptor.dwords[0],
-				    descriptor.dwords[1], descriptor.dwords[2], descriptor.dwords[3],
-				    descriptor.dwords[4], descriptor.dwords[5], descriptor.dwords[6],
-				    descriptor.dwords[7]);
-			}
-			return false;
+			return SpecializationFail(fmt::format(
+			    "image descriptor {} has unsupported type {}: {:08x},{:08x},{:08x},{:08x},"
+			    "{:08x},{:08x},{:08x},{:08x}",
+			    i, (descriptor.dwords[3] >> 28u) & 0xfu, descriptor.dwords[0], descriptor.dwords[1],
+			    descriptor.dwords[2], descriptor.dwords[3], descriptor.dwords[4],
+			    descriptor.dwords[5], descriptor.dwords[6], descriptor.dwords[7]));
 		}
 		image.dimension = descriptor_dimension;
 		image.cube      = DescriptorIsCube(descriptor);
 		const auto format =
 		    static_cast<Prospero::BufferFormat>((descriptor.dwords[1] >> 20u) & 0x1ffu);
 		if (image.atomic && format != Prospero::BufferFormat::k32UInt) {
-			if (error != nullptr) {
-				*error = fmt::format("atomic image descriptor {} uses unsupported format {}", i,
-				                     static_cast<uint32_t>(format));
-			}
-			return false;
+			return SpecializationFail(
+			    fmt::format("atomic image descriptor {} uses unsupported format {}", i,
+			                static_cast<uint32_t>(format)));
 		}
-		const bool storage = image.kind == ResourceKind::StorageImage ||
-		                     image.kind == ResourceKind::StorageImageUint;
+		const bool storage      = image.kind == ResourceKind::StorageImage ||
+		                          image.kind == ResourceKind::StorageImageUint;
 		image.conversion_format = ImageConversionFormat(format);
 		if (storage || image.conversion_format != Prospero::BufferFormat::kInvalid) {
 			image.shader_swizzle = DescriptorImageSwizzle(descriptor);
 		}
 		const bool raw_sint_storage = storage && format == Prospero::BufferFormat::k32SInt &&
 		                              image.written && !image.read && !image.atomic;
-		const bool uint_image = Prospero::IsUintTextureFormat(format) || raw_sint_storage;
+		const bool uint_image       = Prospero::IsUintTextureFormat(format) || raw_sint_storage;
 		if (uint_image) {
 			switch (image.kind) {
 				case ResourceKind::Image: image.kind = ResourceKind::ImageUint; break;
@@ -803,19 +666,13 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 		    static_cast<size_t>(root.indirect_mapping_offset) + 1u +
 		            static_cast<size_t>(root.indirect_mapping_capacity) * 2u >
 		        next_snapshot.flattened_srt.size()) {
-			if (error != nullptr) {
-				*error = "indirect image specialization has an invalid key mapping";
-			}
-			return false;
+			return SpecializationFail("indirect image specialization has an invalid key mapping");
 		}
 		uint32_t exemplar = ImageResource::NoIndirectImage;
 		for (const auto resource: root.indirect_resources) {
 			if (resource >= next.images.size() ||
 			    next.images[resource].indirect_root != root_index) {
-				if (error != nullptr) {
-					*error = "indirect image specialization has an invalid candidate";
-				}
-				return false;
+				return SpecializationFail("indirect image specialization has an invalid candidate");
 			}
 			if (!NullImageDescriptor(next_snapshot.images[resource])) {
 				exemplar = resource;
@@ -823,10 +680,7 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 			}
 		}
 		if (exemplar == ImageResource::NoIndirectImage) {
-			if (error != nullptr) {
-				*error = "indirect image specialization has no typed candidate";
-			}
-			return false;
+			return SpecializationFail("indirect image specialization has no typed candidate");
 		}
 		const auto& image_class = next.images[exemplar];
 		for (uint32_t candidate = 0; candidate < next.images.size(); candidate++) {
@@ -849,12 +703,9 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 			    image.shader_swizzle != image_class.shader_swizzle ||
 			    image.depth_compare != image_class.depth_compare ||
 			    image.cube != image_class.cube) {
-				if (error != nullptr) {
-					*error = fmt::format(
-					    "indirect image table at pc 0x{:08x} has incompatible candidates",
-					    root.first_use_pc);
-				}
-				return false;
+				return SpecializationFail(
+				    fmt::format("indirect image table at pc 0x{:08x} has incompatible candidates",
+				                root.first_use_pc));
 			}
 		}
 	}
@@ -867,11 +718,8 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 	std::vector<SamplerUsage> sampler_usage(next.samplers.size());
 	for (const auto& pair: next.sampled_pairs) {
 		if (pair.image >= next.images.size() || pair.sampler >= next.samplers.size()) {
-			if (error != nullptr) {
-				*error = fmt::format("sampled resource pair at pc 0x{:08x} is invalid",
-				                     pair.first_use_pc);
-			}
-			return false;
+			return SpecializationFail(
+			    fmt::format("sampled resource pair at pc 0x{:08x} is invalid", pair.first_use_pc));
 		}
 		auto& usage = sampler_usage[pair.sampler];
 		if (next.images[pair.image].conversion_format != Prospero::BufferFormat::kInvalid) {
@@ -891,10 +739,8 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 			continue;
 		}
 		if (next.samplers.size() >= ShaderInfo::MaxSamplers) {
-			if (error != nullptr) {
-				*error = "point-filter sampler variants exceed the dense sampler resource limit";
-			}
-			return false;
+			return SpecializationFail(
+			    "point-filter sampler variants exceed the dense sampler resource limit");
 		}
 		point_sampler[i]              = static_cast<uint32_t>(next.samplers.size());
 		auto sampler                  = next.samplers[i];
@@ -915,19 +761,13 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 			}
 			const auto index = inst.Flags<MemoryFlags>().index;
 			if (index >= memory_info.size()) {
-				if (error != nullptr) {
-					*error = fmt::format("typed image instruction has invalid memory metadata {}",
-					                     index);
-				}
-				return false;
+				return SpecializationFail(
+				    fmt::format("typed image instruction has invalid memory metadata {}", index));
 			}
 			auto& memory = memory_info[index];
 			if (memory.resource >= next.images.size()) {
-				if (error != nullptr) {
-					*error = fmt::format("typed image instruction has invalid resource {}",
-					                     memory.resource);
-				}
-				return false;
+				return SpecializationFail(fmt::format(
+				    "typed image instruction has invalid resource {}", memory.resource));
 			}
 			const auto& image = next.images[memory.resource];
 			if (image_opcode.needs_sampler &&
@@ -937,22 +777,18 @@ bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::stri
 			}
 			if (image.indirect_root == memory.resource &&
 			    inst.GetOpcode() != ValueOpcode::ImageSampleRaw) {
-				if (error != nullptr) {
-					*error = fmt::format(
-					    "indirect image table at pc 0x{:08x} is used by unsupported {}",
-					    inst.Flags<MemoryFlags>().pc, ValueOpcodeName(inst.GetOpcode()));
-				}
-				return false;
+				return SpecializationFail(
+				    fmt::format("indirect image table at pc 0x{:08x} is used by unsupported {}",
+				                inst.Flags<MemoryFlags>().pc, ValueOpcodeName(inst.GetOpcode())));
 			}
 			memory.kind            = image.kind;
 			memory.image_dimension = image.dimension;
 			memory.image_cube      = image.cube;
 		}
 	}
-	program.info                = std::move(next);
+	program.info        = std::move(next);
 	program.memory_info = std::move(memory_info);
-	snapshot                    = std::move(next_snapshot);
-	return true;
+	snapshot            = std::move(next_snapshot);
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR

src/graphics/shader/recompiler/ir/passes/ResourceMaterialization.h

@@ -22,19 +22,17 @@ struct ResourceSnapshot {
 	std::vector<IndirectImage>   indirect_images;
 };
 
-bool ValidateResourceSnapshot(const Program& program, const ResourceSnapshot& snapshot,
-                              std::string* error);
-bool ValidateResourceSpecialization(const Program& program, const ResourceSnapshot& snapshot,
-                                    std::string* error);
+bool ValidateResourceSnapshot(const Program& program, const ResourceSnapshot& snapshot);
+bool ValidateResourceSpecialization(const Program& program, const ResourceSnapshot& snapshot);
 
 // Resolves the immutable dense resource topology against one runtime user-data/SRT snapshot.
 // On failure the destination is unchanged.
 bool MaterializeResources(const Program& program, const SrtRuntime& runtime,
-                          ResourceSnapshot& snapshot, std::string* error);
+                          ResourceSnapshot& snapshot);
 
 // Applies runtime descriptor shape/format facts to a copied dense topology before layout and
-// emission. On failure the program is unchanged.
-bool SpecializeResources(Program& program, ResourceSnapshot& snapshot, std::string* error);
+// emission. Invalid specializations are fatal.
+void SpecializeResources(Program& program, ResourceSnapshot& snapshot);
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR
 

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

@@ -1,8 +1,8 @@
 #include "graphics/shader/recompiler/ir/passes/ResourceTracking.h"
 
+#include "common/assert.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 #include "graphics/shader/recompiler/ir/passes/SrtWalker.h"
-#include "graphics/shader/shader.h"
 
 #include <algorithm>
 #include <fmt/format.h>
@@ -12,7 +12,7 @@
 namespace Libs::Graphics::ShaderRecompiler::IR {
 namespace {
 
-constexpr uint32_t SamplerBorderClampMask = (1u << 2u) | (1u << 5u) | (1u << 8u);
+constexpr uint32_t SamplerBorderClampMask    = (1u << 2u) | (1u << 5u) | (1u << 8u);
 constexpr uint32_t SamplerDword3ReservedMask = 0x3ffff000u;
 
 uint32_t PossibleU32Bits(Value value) {
@@ -101,19 +101,17 @@ public:
 		m_info.uses_dma = false;
 	}
 
-	bool Run(std::string* error) {
+	void Run() {
 		if (m_program.resource_tracking_complete) {
-			return Fail(0, error, "resources already tracked");
+			Fail(0, "resources already tracked");
 		}
 		if (!m_program.srt_plan_complete) {
-			return Fail(0, error, "SRT plan is not ready");
+			Fail(0, "SRT plan is not ready");
 		}
 		PlanIndirectImages();
 		for (auto* block: m_program.blocks) {
 			for (auto& inst: *block) {
-				if (!Collect(inst, error)) {
-					return false;
-				}
+				Collect(inst);
 			}
 		}
 		LinkImageAliases();
@@ -145,10 +143,9 @@ public:
 				return std::ranges::find(plan.reads, inst) != plan.reads.end();
 			});
 		});
-		m_program.descriptor_sources          = std::move(m_sources);
+		m_program.descriptor_sources         = std::move(m_sources);
 		m_program.info                       = std::move(m_info);
 		m_program.resource_tracking_complete = true;
-		return true;
 	}
 
 private:
@@ -173,18 +170,19 @@ private:
 		std::array<const Inst*, 8> reads {};
 	};
 
-	bool Fail(uint32_t pc, std::string* error, const std::string& reason) const {
-		return ShaderError::Fail(
-		    error, fmt::format("shader resource tracking: hash=0x{:016x} stage={} pc=0x{:08x} {}",
-		                       m_program.shader_hash, StageName(m_program.stage), pc, reason));
+	[[noreturn]] void Fail(uint32_t pc, const std::string& reason) const {
+		const auto message =
+		    fmt::format("shader resource tracking: hash=0x{:016x} stage={} pc=0x{:08x} {}",
+		                m_program.shader_hash, StageName(m_program.stage), pc, reason);
+		EXIT("%s", message.c_str());
+		std::abort();
 	}
 
-	bool MakeSource(const Inst& handle, uint32_t width, bool sampler, bool sample_adjust,
-	                DescriptorSource& descriptor, uint32_t pc, std::string* error) const {
+	void MakeSource(const Inst& handle, uint32_t width, bool sampler, bool sample_adjust,
+	                DescriptorSource& descriptor, uint32_t pc) const {
 		if (handle.NumArgs() != width) {
-			return Fail(pc, error,
-			            fmt::format("{} has {} descriptor dwords, expected {}",
-			                        ValueOpcodeName(handle.GetOpcode()), handle.NumArgs(), width));
+			Fail(pc, fmt::format("{} has {} descriptor dwords, expected {}",
+			                     ValueOpcodeName(handle.GetOpcode()), handle.NumArgs(), width));
 		}
 		descriptor.dword_count = width;
 		for (uint32_t i = 0; i < width; i++) {
@@ -199,18 +197,15 @@ private:
 			// Border color and its table index are unused unless a clamp axis selects border mode.
 			descriptor.dwords[3] = Value(0u);
 		}
-		return true;
 	}
 
-	bool ValidateSource(const DescriptorSource& descriptor, std::string& reason,
-	                    uint32_t& bad_dword) const {
+	bool ValidateSource(const DescriptorSource& descriptor, uint32_t& bad_dword) const {
 		for (uint32_t i = 0; i < descriptor.dword_count; i++) {
 			bad_dword = i;
 			if (descriptor.dwords[i].Resolve().GetType() != Type::U32) {
-				reason = "is not U32";
 				return false;
 			}
-			if (!ValidateRuntimeValue(m_program, descriptor.dwords[i], reason)) {
+			if (!ValidateRuntimeValue(m_program, descriptor.dwords[i])) {
 				return false;
 			}
 		}
@@ -286,13 +281,12 @@ private:
 
 	bool MakeRuntimeBufferSource(const Inst& handle, uint32_t pc, uint32_t& source,
 	                             DescriptorSource& descriptor) {
-		if (handle.GetOpcode() != ValueOpcode::GetBufferResource ||
-		    !MakeSource(handle, 4u, false, false, descriptor, pc, nullptr)) {
+		if (handle.GetOpcode() != ValueOpcode::GetBufferResource) {
 			return false;
 		}
-		std::string reason;
-		uint32_t    bad_dword = 0;
-		if (!ValidateSource(descriptor, reason, bad_dword)) {
+		MakeSource(handle, 4u, false, false, descriptor, pc);
+		uint32_t bad_dword = 0;
+		if (!ValidateSource(descriptor, bad_dword)) {
 			return false;
 		}
 		source = InternSource(descriptor);
@@ -469,38 +463,30 @@ private:
 		}
 	}
 
-	bool GetHandle(Value value, ValueOpcode expected, uint32_t width, uint32_t pc, Inst*& handle,
-	               uint32_t& source, std::string* error, bool sampler = false,
-	               bool sample_adjust = false) {
+	void GetHandle(Value value, ValueOpcode expected, uint32_t width, uint32_t pc, Inst*& handle,
+	               uint32_t& source, bool sampler = false, bool sample_adjust = false) {
 		handle = value.Resolve().TryInstruction();
 		if (handle == nullptr || handle->GetOpcode() != expected) {
-			return Fail(pc, error,
-			            fmt::format("memory operation requires {}", ValueOpcodeName(expected)));
+			Fail(pc, fmt::format("memory operation requires {}", ValueOpcodeName(expected)));
 		}
 		DescriptorSource descriptor;
-		if (!MakeSource(*handle, width, sampler, sample_adjust, descriptor, pc, error)) {
-			return false;
-		}
-		std::string reason;
-		uint32_t    bad_dword = 0;
-		if (!ValidateSource(descriptor, reason, bad_dword)) {
-			return Fail(
-			    pc, error,
-			    fmt::format("{} dword {} {}", ValueOpcodeName(expected), bad_dword, reason));
+		MakeSource(*handle, width, sampler, sample_adjust, descriptor, pc);
+		uint32_t bad_dword = 0;
+		if (!ValidateSource(descriptor, bad_dword)) {
+			Fail(pc, fmt::format("{} dword {} is not a valid runtime value",
+			                     ValueOpcodeName(expected), bad_dword));
 		}
 		source = InternSource(descriptor);
-		return true;
 	}
 
-	bool ValidateAddressHandle(Value value, uint32_t pc, std::string* error) const {
+	void ValidateAddressHandle(Value value, uint32_t pc) const {
 		const auto* handle = value.Resolve().TryInstruction();
 		if (handle == nullptr || handle->GetOpcode() != ValueOpcode::GetAddressResource) {
-			return Fail(pc, error, "address operation requires GetAddressResource");
+			Fail(pc, "address operation requires GetAddressResource");
 		}
 		if (handle->NumArgs() != 2) {
-			return Fail(pc, error, "GetAddressResource must have two address dwords");
+			Fail(pc, "GetAddressResource must have two address dwords");
 		}
-		return true;
 	}
 
 	uint32_t AddBuffer(uint32_t source, const MemoryInfo& memory, ValueOpcode op, uint32_t pc) {
@@ -540,7 +526,7 @@ private:
 		const auto mip   = MipMode(memory);
 		const bool depth = (memory.image_sample_flags & Decoder::ImageSampleFlagCompare) != 0;
 		for (uint32_t i = 0; i < m_info.images.size(); i++) {
-			auto& image = m_info.images[i];
+			auto&      image           = m_info.images[i];
 			const bool compatible_kind = image.kind == memory.kind || (IsStorageImage(image.kind) &&
 			                                                           IsStorageImage(memory.kind));
 			if (image.source == source && compatible_kind &&
@@ -593,155 +579,139 @@ private:
 		return static_cast<uint32_t>(m_info.samplers.size() - 1);
 	}
 
-	bool AddSampledPair(uint32_t image, uint32_t sampler, uint32_t pc, std::string* error) {
+	void AddSampledPair(uint32_t image, uint32_t sampler, uint32_t pc) {
 		for (auto& pair: m_info.sampled_pairs) {
 			if (pair.image == image && pair.sampler == sampler) {
 				pair.first_use_pc = std::min(pair.first_use_pc, pc);
-				return true;
+				return;
 			}
 		}
 		if (m_info.sampled_pairs.size() >= ShaderInfo::MaxSampledPairs) {
-			return Fail(pc, error, "sampled image/sampler pair limit exceeded");
+			Fail(pc, "sampled image/sampler pair limit exceeded");
 		}
 		m_info.sampled_pairs.push_back({image, sampler, pc});
-		return true;
 	}
 
-	bool AddHandlePatch(Inst* handle, uint32_t resource, uint32_t pc, std::string* error) {
+	void AddHandlePatch(Inst* handle, uint32_t resource, uint32_t pc) {
 		for (const auto& patch: m_handle_patches) {
 			if (patch.handle == handle) {
-				return patch.resource == resource ||
-				       Fail(pc, error,
-				            fmt::format("{} is reused with incompatible resource classes",
-				                        ValueOpcodeName(handle->GetOpcode())));
+				if (patch.resource != resource) {
+					Fail(pc, fmt::format("{} is reused with incompatible resource classes",
+					                     ValueOpcodeName(handle->GetOpcode())));
+				}
+				return;
 			}
 		}
 		m_handle_patches.push_back({handle, resource});
-		return true;
 	}
 
-	bool AddMemoryPatch(uint32_t index, uint32_t resource, uint32_t sampler, bool has_sampler,
-	                    uint32_t pc, std::string* error) {
+	void AddMemoryPatch(uint32_t index, uint32_t resource, uint32_t sampler, bool has_sampler,
+	                    uint32_t pc) {
 		for (auto& patch: m_memory_patches) {
 			if (patch.index != index) {
 				continue;
 			}
 			if (patch.resource != resource ||
 			    (has_sampler && patch.has_sampler && patch.sampler != sampler)) {
-				return Fail(pc, error, "memory metadata is reused with incompatible resources");
+				Fail(pc, "memory metadata is reused with incompatible resources");
 			}
 			if (has_sampler) {
 				patch.sampler     = sampler;
 				patch.has_sampler = true;
 			}
-			return true;
+			return;
 		}
 		m_memory_patches.push_back({index, resource, sampler, has_sampler});
-		return true;
 	}
 
-	bool Collect(Inst& inst, std::string* error) {
+	void Collect(Inst& inst) {
 		const auto op           = inst.GetOpcode();
 		const auto buffer       = BufferAccessOf(op);
 		const auto address_info = AddressOpcodeInfoOf(op);
 		const auto image_info   = ImageOpcodeInfoOf(op);
 		if (buffer == BufferAccess::None && address_info.access == AddressAccess::None &&
 		    image_info.access == ImageAccess::None) {
-			return true;
+			return;
 		}
 		const auto flags = inst.Flags<MemoryFlags>();
 		if (flags.index >= m_program.memory_info.size()) {
-			return Fail(flags.pc, error,
-			            fmt::format("memory metadata index {} is out of range", flags.index));
+			Fail(flags.pc, fmt::format("memory metadata index {} is out of range", flags.index));
 		}
 		if (inst.NumArgs() == 0) {
-			return Fail(flags.pc, error, "memory operation has no resource handle");
+			Fail(flags.pc, "memory operation has no resource handle");
 		}
 		const auto& memory = m_program.memory_info[flags.index];
 		if (memory.planning_only || IsIndirectPlanningMemory(flags.index)) {
-			return true;
+			return;
 		}
 		Inst*    handle   = nullptr;
 		uint32_t source   = 0;
 		uint32_t resource = 0;
 
 		if (buffer != BufferAccess::None) {
-			if (!GetHandle(inst.Arg(0), ValueOpcode::GetBufferResource, 4, flags.pc, handle, source,
-			               error)) {
-				return false;
-			}
+			GetHandle(inst.Arg(0), ValueOpcode::GetBufferResource, 4, flags.pc, handle, source);
 			resource = AddBuffer(source, memory, op, flags.pc);
 			if (resource == UINT32_MAX) {
-				return Fail(flags.pc, error, "buffer resource limit exceeded");
+				Fail(flags.pc, "buffer resource limit exceeded");
 			}
-			return AddHandlePatch(handle, resource, flags.pc, error) &&
-			       AddMemoryPatch(flags.index, resource, 0, false, flags.pc, error);
+			AddHandlePatch(handle, resource, flags.pc);
+			AddMemoryPatch(flags.index, resource, 0, false, flags.pc);
+			return;
 		}
 		if (address_info.access != AddressAccess::None) {
 			if (!IsAddressResourceKind(memory.kind)) {
-				return Fail(flags.pc, error, "address operation has invalid resource kind");
+				Fail(flags.pc, "address operation has invalid resource kind");
 			}
 			if (memory.kind == ResourceKind::Scratch) {
 				handle = inst.Arg(0).Resolve().TryInstruction();
 				if (handle == nullptr || handle->GetOpcode() != ValueOpcode::GetScratchResource ||
 				    handle->NumArgs() != 0) {
-					return Fail(flags.pc, error, "scratch operation requires GetScratchResource");
+					Fail(flags.pc, "scratch operation requires GetScratchResource");
 				}
 				if (m_program.scratch_dwords == 0) {
-					return Fail(flags.pc, error,
-					            "scratch operation requires a nonzero AGC per-thread size");
+					Fail(flags.pc, "scratch operation requires a nonzero AGC per-thread size");
 				}
-				return true;
-			}
-			if (!ValidateAddressHandle(inst.Arg(0), flags.pc, error)) {
-				return false;
+				return;
 			}
+			ValidateAddressHandle(inst.Arg(0), flags.pc);
 			m_info.uses_dma = true;
-			return true;
+			return;
 		}
 
 		if (!ImageResourceKindMatches(memory.kind, image_info.resource_class)) {
-			return Fail(flags.pc, error, "image operation has invalid resource kind");
+			Fail(flags.pc, "image operation has invalid resource kind");
 		}
 		handle               = inst.Arg(0).Resolve().TryInstruction();
 		const auto* indirect = handle != nullptr ? FindIndirectImage(*handle) : nullptr;
 		if (indirect != nullptr) {
 			source = indirect->source;
-		} else if (!GetHandle(inst.Arg(0), ValueOpcode::GetImageResource, 8, flags.pc, handle,
-		                      source, error)) {
-			return false;
+		} else {
+			GetHandle(inst.Arg(0), ValueOpcode::GetImageResource, 8, flags.pc, handle, source);
 		}
 		resource = AddImage(source, memory, op, flags.pc);
 		if (resource == UINT32_MAX) {
-			return Fail(flags.pc, error, "image resource limit exceeded");
-		}
-		if (!AddHandlePatch(handle, resource, flags.pc, error)) {
-			return false;
+			Fail(flags.pc, "image resource limit exceeded");
 		}
+		AddHandlePatch(handle, resource, flags.pc);
 		uint32_t sampler = 0;
 		if (image_info.needs_sampler) {
 			if (inst.NumArgs() < 2) {
-				return Fail(flags.pc, error, "sampled image operation has no sampler handle");
+				Fail(flags.pc, "sampled image operation has no sampler handle");
 			}
-			Inst*    sampler_handle = nullptr;
-			uint32_t sampler_source = 0;
+			Inst*      sampler_handle = nullptr;
+			uint32_t   sampler_source = 0;
 			const bool sample_adjust =
 			    (memory.image_sample_flags & Decoder::ImageSampleFlagAdjust) != 0;
-			if (!GetHandle(inst.Arg(1), ValueOpcode::GetSamplerResource, 4, flags.pc,
-			               sampler_handle, sampler_source, error, true, sample_adjust)) {
-				return false;
-			}
+			GetHandle(inst.Arg(1), ValueOpcode::GetSamplerResource, 4, flags.pc, sampler_handle,
+			          sampler_source, true, sample_adjust);
 			sampler = AddSampler(sampler_source, flags.pc);
 			if (sampler == UINT32_MAX) {
-				return Fail(flags.pc, error, "sampler resource limit exceeded");
-			}
-			if (!AddHandlePatch(sampler_handle, sampler, flags.pc, error) ||
-			    !AddSampledPair(resource, sampler, flags.pc, error)) {
-				return false;
+				Fail(flags.pc, "sampler resource limit exceeded");
 			}
+			AddHandlePatch(sampler_handle, sampler, flags.pc);
+			AddSampledPair(resource, sampler, flags.pc);
 		}
-		return AddMemoryPatch(flags.index, resource, sampler, image_info.needs_sampler, flags.pc,
-		                      error);
+		AddMemoryPatch(flags.index, resource, sampler, image_info.needs_sampler, flags.pc);
 	}
 
 	const DescriptorSource* Source(uint32_t source) const {
@@ -783,8 +753,8 @@ private:
 
 } // namespace
 
-bool TrackResources(Program& program, std::string* error) {
-	return Tracker(program).Run(error);
+void TrackResources(Program& program) {
+	Tracker(program).Run();
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR

src/graphics/shader/recompiler/ir/passes/ResourceTracking.h

@@ -7,7 +7,7 @@ namespace Libs::Graphics::ShaderRecompiler::IR {
 
 // Collects immutable resource topology from typed SSA handles, interns their resolved dwords in
 // descriptor_sources, then writes dense indices to handle flags and MemoryInfo.
-bool TrackResources(Program& program, std::string* error);
+void TrackResources(Program& program);
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR
 

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

@@ -1,5 +1,6 @@
 #include "graphics/shader/recompiler/ir/passes/ShaderInfoCollection.h"
 
+#include "common/assert.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 
 #include <algorithm>
@@ -8,6 +9,11 @@
 namespace Libs::Graphics::ShaderRecompiler::IR {
 namespace {
 
+[[noreturn]] void Fail(std::string_view message) {
+	EXIT("shader info collection failed: %s", std::string(message).c_str());
+	std::abort();
+}
+
 void AddInput(ShaderInfo& info, StageInputKind kind, uint32_t location, uint32_t components,
               std::string name, bool per_vertex = false) {
 	const auto input = std::find_if(info.inputs.begin(), info.inputs.end(), [=](const auto& value) {
@@ -34,38 +40,38 @@ void AddOutput(ShaderInfo& info, StageOutputKind kind, uint32_t index, uint32_t
 	}
 }
 
-bool ValidateOptions(const Program& program, const ShaderInfoOptions& options, std::string* error) {
-	auto Fail = [&](const char* message) {
-		if (error != nullptr) {
-			*error = message;
-		}
-		return false;
-	};
+void ValidateOptions(const Program& program, const ShaderInfoOptions& options) {
 	switch (program.stage) {
 		case ShaderType::Vertex:
+			if (options.vertex == nullptr) {
+				return Fail("vertex shader has no input metadata");
+			}
 			if (options.vertex->resources_num < 0 ||
 			    options.vertex->resources_num > ShaderVertexInputInfo::RES_MAX) {
 				return Fail("vertex resource count is out of range");
 			}
-			return true;
+			return;
 		case ShaderType::Pixel:
-			return options.pixel->input_num <= std::size(options.pixel->interpolator_settings) ||
-			       Fail("pixel input count is out of range");
+			if (options.pixel == nullptr) {
+				return Fail("pixel shader has no input metadata");
+			}
+			if (options.pixel->input_num > std::size(options.pixel->interpolator_settings)) {
+				return Fail("pixel input count is out of range");
+			}
+			return;
 		case ShaderType::Compute:
-			return (options.compute->thread_ids_num >= 0 && options.compute->thread_ids_num <= 3) ||
-			       Fail("compute thread ID count is out of range");
+			if (options.compute == nullptr) {
+				return Fail("compute shader has no input metadata");
+			}
+			if (options.compute->thread_ids_num < 0 || options.compute->thread_ids_num > 3) {
+				return Fail("compute thread ID count is out of range");
+			}
+			return;
 		default: return Fail("unsupported shader stage for info collection");
 	}
 }
 
-bool ValidateValueReferences(const Program& program, const ShaderInfoOptions& options,
-                             std::string* error) {
-	const auto Fail = [&](const char* message) {
-		if (error != nullptr) {
-			*error = message;
-		}
-		return false;
-	};
+void ValidateValueReferences(const Program& program, const ShaderInfoOptions& options) {
 	for (const auto* block: program.blocks) {
 		for (const auto& inst: *block) {
 			switch (inst.GetOpcode()) {
@@ -148,7 +154,6 @@ bool ValidateValueReferences(const Program& program, const ShaderInfoOptions& op
 			}
 		}
 	}
-	return true;
 }
 
 void CollectVertexInputs(const Program& program, const ShaderVertexInputInfo* vertex,
@@ -269,14 +274,8 @@ void CollectBuiltinInputs(const Program& program, ShaderInfo& info) {
 	}
 }
 
-bool CollectOutputs(const Program& program, const ShaderVertexInputInfo* vertex,
-	                const ShaderPixelInputInfo* pixel, ShaderInfo& info, std::string* error) {
-	auto Fail = [&](std::string message) {
-		if (error != nullptr) {
-			*error = std::move(message);
-		}
-		return false;
-	};
+void CollectOutputs(const Program& program, const ShaderVertexInputInfo* vertex,
+                    const ShaderPixelInputInfo* pixel, ShaderInfo& info) {
 	for (const auto* block: program.blocks) {
 		for (const auto& inst: *block) {
 			if (inst.GetOpcode() != ValueOpcode::SetAttribute) {
@@ -347,31 +346,23 @@ bool CollectOutputs(const Program& program, const ShaderVertexInputInfo* vertex,
 			}
 		}
 	}
-	return true;
 }
 
 } // namespace
 
-bool CollectShaderInfo(Program& program, const ShaderInfoOptions& options, std::string* error) {
+void CollectShaderInfo(Program& program, const ShaderInfoOptions& options) {
 	if (!program.resource_tracking_complete || program.shader_info_complete) {
-		if (error != nullptr) {
-			*error = !program.resource_tracking_complete ? "shader resources were not tracked"
-			                                             : "shader info already collected";
-		}
-		return false;
-	}
-	if (!ValidateOptions(program, options, error)) {
-		return false;
-	}
-	if (!ValidateValueReferences(program, options, error)) {
-		return false;
+		return Fail(!program.resource_tracking_complete ? "shader resources were not tracked"
+		                                                : "shader info already collected");
 	}
+	ValidateOptions(program, options);
+	ValidateValueReferences(program, options);
 
 	auto next = program.info;
 	next.inputs.clear();
 	next.outputs.clear();
-	next.has_bitwise_xor = std::any_of(
-	    program.blocks.begin(), program.blocks.end(), [](const auto* block) {
+	next.has_bitwise_xor =
+	    std::any_of(program.blocks.begin(), program.blocks.end(), [](const auto* block) {
 		    return std::any_of(block->begin(), block->end(), [](const auto& inst) {
 			    return inst.GetOpcode() == ValueOpcode::BitwiseXor32;
 		    });
@@ -380,15 +371,12 @@ bool CollectShaderInfo(Program& program, const ShaderInfoOptions& options, std::
 		case ShaderType::Vertex: CollectVertexInputs(program, options.vertex, next); break;
 		case ShaderType::Pixel: CollectPixelInputs(program, options.pixel, next); break;
 		case ShaderType::Compute: CollectComputeInputs(options.compute, next); break;
-		default: return false;
+		default: return Fail("unsupported shader stage for info collection");
 	}
 	CollectBuiltinInputs(program, next);
-	if (!CollectOutputs(program, options.vertex, options.pixel, next, error)) {
-		return false;
-	}
+	CollectOutputs(program, options.vertex, options.pixel, next);
 	program.info                 = std::move(next);
 	program.shader_info_complete = true;
-	return true;
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR

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

@@ -13,7 +13,7 @@ struct ShaderInfoOptions {
 
 // Completes the immutable shader interface after resource tracking. On failure Program::info and
 // all completion state remain unchanged.
-bool CollectShaderInfo(Program& program, const ShaderInfoOptions& options, std::string* error);
+void CollectShaderInfo(Program& program, const ShaderInfoOptions& options);
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR
 

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

@@ -1,7 +1,7 @@
 #include "graphics/shader/recompiler/ir/passes/SrtWalker.h"
 
+#include "common/assert.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
-#include "graphics/shader/shader.h"
 
 #include <algorithm>
 #include <bit>
@@ -120,10 +120,10 @@ class RuntimeValidator {
 public:
 	explicit RuntimeValidator(const Program& program): m_program(program) {}
 
-	bool Run(Value value, std::string& reason) { return Validate(value, reason); }
+	bool Run(Value value) { return Validate(value); }
 
 private:
-	bool Validate(Value value, std::string& reason) {
+	bool Validate(Value value) {
 		value            = value.Resolve();
 		const auto* inst = value.TryInstruction();
 		if (inst == nullptr) {
@@ -133,14 +133,10 @@ private:
 				case Type::U16:
 				case Type::U32:
 				case Type::U64: return true;
-				default:
-					reason = fmt::format("contains a non-integer immediate of type {}",
-					                     TypeName(value.GetType()));
-					return false;
+				default: return false;
 			}
 		}
 		if (!m_visiting.insert(inst).second) {
-			reason = fmt::format("contains a cyclic {} value", ValueOpcodeName(inst->GetOpcode()));
 			return false;
 		}
 		const auto finish = [&](bool valid) {
@@ -151,25 +147,21 @@ private:
 		if (op == ValueOpcode::UndefU1 || op == ValueOpcode::UndefU8 ||
 		    op == ValueOpcode::UndefU16 || op == ValueOpcode::UndefU32 ||
 		    op == ValueOpcode::UndefU64 || op == ValueOpcode::Void) {
-			reason = fmt::format("contains {}", ValueOpcodeName(op));
 			return finish(false);
 		}
 		if (op == ValueOpcode::GetUserData) {
 			if (inst->NumArgs() != 1 || inst->Arg(0).GetType() != Type::ScalarReg) {
-				reason = "contains a malformed GetUserData";
 				return finish(false);
 			}
 			const auto reg = RegIndex(inst->Arg(0).ScalarRegister());
 			if (reg < m_program.user_data_base ||
 			    reg - m_program.user_data_base >= m_program.user_data_count) {
-				reason = fmt::format("references unavailable user SGPR {}", reg);
 				return finish(false);
 			}
 			return finish(true);
 		}
 		if (op == ValueOpcode::GetShaderBase) {
 			if (inst->NumArgs() != 0) {
-				reason = "contains a malformed GetShaderBase";
 				return finish(false);
 			}
 			return finish(true);
@@ -177,14 +169,12 @@ private:
 		if (op == ValueOpcode::Phi) {
 			const auto invariant = ResolveInvariantPhi(m_program, value);
 			if (invariant.IsEmpty()) {
-				reason = "contains a control-dependent phi";
 				return finish(false);
 			}
-			return finish(Validate(invariant, reason));
+			return finish(Validate(invariant));
 		}
 		if (op == ValueOpcode::GetSrtResource) {
 			if (inst->NumArgs() != 0) {
-				reason = "contains a malformed GetSrtResource";
 				return finish(false);
 			}
 			return finish(true);
@@ -196,7 +186,6 @@ private:
 			        ValueOpcode::GetSrtResource ||
 			    !slot.IsImmediate() || slot.GetType() != Type::U32 ||
 			    slot.U32() >= m_program.srt_reads.size()) {
-				reason = "contains a malformed flattened SRT read";
 				return finish(false);
 			}
 		} else if (op == ValueOpcode::LoadAddressU32 || op == ValueOpcode::ReadConstBuffer) {
@@ -206,13 +195,11 @@ private:
 			const auto* handle = inst->NumArgs() != 0 ? inst->Arg(0).ResolveInstruction() : nullptr;
 			if (!IsRawRead(m_program, *inst) || handle == nullptr ||
 			    handle->GetOpcode() != expected) {
-				reason = fmt::format("contains a non-scalar {}", ValueOpcodeName(op));
 				return finish(false);
 			}
 		} else if (op == ValueOpcode::CompositeExtractU64) {
 			const auto index = inst->NumArgs() == 2 ? inst->Arg(1).Resolve() : Value {};
 			if (!index.IsImmediate() || index.GetType() != Type::U32 || index.U32() >= 2u) {
-				reason = "contains an invalid U64 component index";
 				return finish(false);
 			}
 		} else if (op == ValueOpcode::CompositeExtractU32x2) {
@@ -222,7 +209,6 @@ private:
 			    index.U32() >= 2u ||
 			    (source->GetOpcode() != ValueOpcode::CompositeConstructU32x2 &&
 			     source->GetOpcode() != ValueOpcode::IAddCarry32)) {
-				reason = "contains an unsupported composite runtime source";
 				return finish(false);
 			}
 		}
@@ -233,17 +219,14 @@ private:
 			                        : op == ValueOpcode::GetSamplerResource ? 4u
 			                                                                : 2u;
 			if (inst->NumArgs() != expected) {
-				reason = fmt::format("contains a malformed {}", ValueOpcodeName(op));
 				return finish(false);
 			}
 		} else if (op != ValueOpcode::ReadConst && op != ValueOpcode::ReadConstBuffer &&
 		           op != ValueOpcode::LoadAddressU32 && !IsRuntimeIntegerOp(op)) {
-			reason =
-			    fmt::format("contains unsupported or control-dependent {}", ValueOpcodeName(op));
 			return finish(false);
 		}
 		for (size_t index = 0; index < inst->NumArgs(); index++) {
-			if (!Validate(inst->Arg(index), reason)) {
+			if (!Validate(inst->Arg(index))) {
 				return finish(false);
 			}
 		}
@@ -258,7 +241,7 @@ class PlanBuilder {
 public:
 	explicit PlanBuilder(Program& program): m_program(program) {}
 
-	bool Run(std::string* error) {
+	void Run() {
 		m_program.srt_reads.clear();
 		m_program.dynamic_reads.clear();
 		for (auto* block: m_program.blocks) {
@@ -273,34 +256,29 @@ public:
 						                        (op == ValueOpcode::ReadConstBuffer &&
 						                         kind == ResourceKind::ScalarAddress);
 						if (crosswired) {
-							return Fail(flags.pc, error,
-							            fmt::format("{} has incompatible scalar memory metadata",
-							                        ValueOpcodeName(op)));
+							Fail(flags.pc,
+							     fmt::format("{} has incompatible scalar memory metadata",
+							                 ValueOpcodeName(op)));
 						}
 					}
 				}
 				if (IsDescriptorHandle(inst.GetOpcode())) {
 					for (size_t index = 0; index < inst.NumArgs(); index++) {
-						if (!Collect(inst.Arg(index), 0, error)) {
-							return false;
-						}
+						Collect(inst.Arg(index), 0);
 					}
 				}
 			}
 		}
 		for (auto* block: m_program.blocks) {
 			for (auto& inst: *block) {
-				std::string reason;
 				if (inst.GetOpcode() == ValueOpcode::LoadAddressU32 && IsRawRead(m_program, inst) &&
 				    inst.Arg(1).Resolve().IsImmediate() &&
-				    ValidateRuntimeValue(m_program, Value(&inst), reason) &&
-				    !Collect(Value(&inst), inst.Flags<MemoryFlags>().pc, error)) {
-					return false;
+				    ValidateRuntimeValue(m_program, Value(&inst))) {
+					Collect(Value(&inst), inst.Flags<MemoryFlags>().pc);
 				}
 			}
 		}
 		PatchReads();
-		return true;
 	}
 
 private:
@@ -310,18 +288,20 @@ private:
 		bool     keep = false;
 	};
 
-	bool Fail(uint32_t pc, std::string* error, const std::string& message) const {
-		return ShaderError::Fail(error, Diagnostic(m_program, pc, message));
+	[[noreturn]] void Fail(uint32_t pc, const std::string& message) const {
+		const auto diagnostic = Diagnostic(m_program, pc, message);
+		EXIT("shader SRT planning failed: %s", diagnostic.c_str());
+		std::abort();
 	}
 
-	bool Collect(Value value, uint32_t use_pc, std::string* error) {
+	void Collect(Value value, uint32_t use_pc) {
 		value = value.Resolve();
 		if (value.IsImmediate()) {
-			return true;
+			return;
 		}
 		auto* inst = value.TryInstruction();
 		if (inst == nullptr) {
-			return Fail(use_pc, error, "invalid typed planning value");
+			Fail(use_pc, "invalid typed planning value");
 		}
 		const auto cycle = std::ranges::find(m_visiting, inst);
 		if (cycle != m_visiting.end()) {
@@ -329,44 +309,40 @@ private:
 				return value->GetOpcode() == ValueOpcode::Phi;
 			});
 			if (contains_phi) {
-				return true;
+				return;
 			}
-			return Fail(use_pc, error,
-			            fmt::format("cyclic typed planning value {} without a phi",
-			                        ValueOpcodeName(inst->GetOpcode())));
+			Fail(use_pc, fmt::format("cyclic typed planning value {} without a phi",
+			                         ValueOpcodeName(inst->GetOpcode())));
 		}
 		if (std::ranges::find(m_visited, inst) != m_visited.end()) {
-			return true;
+			return;
 		}
 		m_visiting.push_back(inst);
 		for (size_t index = 0; index < inst->NumArgs(); index++) {
-			if (!Collect(inst->Arg(index), use_pc, error)) {
-				return false;
-			}
+			Collect(inst->Arg(index), use_pc);
 		}
 		m_visiting.pop_back();
 		m_visited.push_back(inst);
 		if (!IsRawRead(m_program, *inst)) {
-			return true;
+			return;
 		}
 		const auto offset = inst->Arg(1).Resolve();
 		if (!offset.IsImmediate() || offset.GetType() != Type::U32) {
-			if (std::ranges::find(m_program.dynamic_reads, value) == m_program.dynamic_reads.end()) {
+			if (std::ranges::find(m_program.dynamic_reads, value) ==
+			    m_program.dynamic_reads.end()) {
 				m_program.dynamic_reads.push_back(value);
 			}
-			return true;
+			return;
 		}
-		use_pc = inst->Flags<MemoryFlags>().pc;
 		for (uint32_t slot = 0; slot < m_program.srt_reads.size(); slot++) {
 			if (EquivalentValue(m_program, value, m_program.srt_reads[slot].value)) {
 				m_patches.push_back({inst, slot, false});
-				return true;
+				return;
 			}
 		}
 		const auto slot = static_cast<uint32_t>(m_program.srt_reads.size());
-		m_program.srt_reads.push_back({value, slot, use_pc});
+		m_program.srt_reads.push_back({value, slot});
 		m_patches.push_back({inst, slot, true});
-		return true;
 	}
 
 	void PatchReads() {
@@ -411,14 +387,12 @@ class Evaluator {
 public:
 	Evaluator(const Program& program, const SrtRuntime& runtime,
 	          std::span<const uint8_t> clean_flat_slots = {}, Evaluator* clean_evaluator = nullptr)
-	    : m_program(program), m_runtime(runtime),
-	      m_clean_flat_slots(clean_flat_slots), m_clean_evaluator(clean_evaluator) {}
+	    : m_program(program), m_runtime(runtime), m_clean_flat_slots(clean_flat_slots),
+	      m_clean_evaluator(clean_evaluator) {}
 
-	void SetUsePc(uint32_t pc) { m_use_pc = pc; }
-
-	bool Evaluate(Value value, uint32_t& result, std::string* error) {
+	bool Evaluate(Value value, uint32_t& result) {
 		uint64_t wide = 0;
-		if (!EvaluateWide(value, wide, error)) {
+		if (!EvaluateWide(value, wide)) {
 			return false;
 		}
 		result = static_cast<uint32_t>(wide);
@@ -426,11 +400,7 @@ public:
 	}
 
 private:
-	bool Fail(std::string* error, const std::string& message) const {
-		return ShaderError::Fail(error, Diagnostic(m_program, m_use_pc, message));
-	}
-
-	bool EvaluateWide(Value value, uint64_t& result, std::string* error) {
+	bool EvaluateWide(Value value, uint64_t& result) {
 		value = value.Resolve();
 		if (value.IsImmediate()) {
 			switch (value.GetType()) {
@@ -439,23 +409,23 @@ private:
 				case Type::U16: result = value.U16(); return true;
 				case Type::U32: result = value.U32(); return true;
 				case Type::U64: result = value.U64(); return true;
-				default: return Fail(error, "non-integer immediate in runtime expression");
+				default: return false;
 			}
 		}
 		auto* inst = value.TryInstruction();
 		if (inst == nullptr) {
-			return Fail(error, "invalid typed runtime value");
+			return false;
 		}
 		if (const auto found = m_cache.find(inst); found != m_cache.end()) {
 			result = found->second;
 			return true;
 		}
 		if (std::ranges::find(m_visiting, inst) != m_visiting.end()) {
-			return Fail(error, "cyclic typed runtime value");
+			return false;
 		}
 		m_visiting.push_back(inst);
 		uint64_t out = 0;
-		if (!EvaluateInst(*inst, out, error)) {
+		if (!EvaluateInst(*inst, out)) {
 			return false;
 		}
 		m_visiting.pop_back();
@@ -464,28 +434,27 @@ private:
 		return true;
 	}
 
-	bool Arg(const Inst& inst, size_t index, uint64_t& result, std::string* error) {
-		return EvaluateWide(inst.Arg(index), result, error);
+	bool Arg(const Inst& inst, size_t index, uint64_t& result) {
+		return EvaluateWide(inst.Arg(index), result);
 	}
 
-	bool EvaluatePhi(const Inst& inst, uint64_t& result, std::string* error) {
+	bool EvaluatePhi(const Inst& inst, uint64_t& result) {
 		const auto value = ResolveInvariantPhi(m_program, Value(const_cast<Inst*>(&inst)));
-		return !value.IsEmpty() ? EvaluateWide(value, result, error)
-		                        : Fail(error, "typed phi has runtime-dependent values");
+		return !value.IsEmpty() && EvaluateWide(value, result);
 	}
 
-	bool EvaluateExtract(const Inst& inst, uint64_t& result, std::string* error) {
+	bool EvaluateExtract(const Inst& inst, uint64_t& result) {
 		const auto index = inst.Arg(1).Resolve();
 		if (!index.IsImmediate() || index.GetType() != Type::U32) {
-			return Fail(error, "dynamic composite extract in runtime expression");
+			return false;
 		}
 		const auto component = index.U32();
 		if (component >= 2u) {
-			return Fail(error, "unsupported composite runtime source");
+			return false;
 		}
 		if (inst.GetOpcode() == ValueOpcode::CompositeExtractU64) {
 			uint64_t packed = 0;
-			if (!Arg(inst, 0, packed, error)) {
+			if (!Arg(inst, 0, packed)) {
 				return false;
 			}
 			result = static_cast<uint32_t>(packed >> (component * 32u));
@@ -493,15 +462,15 @@ private:
 		}
 		const auto* source = inst.Arg(0).ResolveInstruction();
 		if (source == nullptr) {
-			return Fail(error, "unsupported composite runtime source");
+			return false;
 		}
 		if (source->GetOpcode() == ValueOpcode::CompositeConstructU32x2) {
-			return EvaluateWide(source->Arg(component), result, error);
+			return EvaluateWide(source->Arg(component), result);
 		}
 		if (source->GetOpcode() == ValueOpcode::IAddCarry32) {
 			uint64_t lhs = 0;
 			uint64_t rhs = 0;
-			if (!Arg(*source, 0, lhs, error) || !Arg(*source, 1, rhs, error)) {
+			if (!Arg(*source, 0, lhs) || !Arg(*source, 1, rhs)) {
 				return false;
 			}
 			const auto sum =
@@ -510,24 +479,23 @@ private:
 			    component == 0u ? static_cast<uint32_t>(sum) : static_cast<uint32_t>(sum >> 32u);
 			return true;
 		}
-		return Fail(error, "unsupported composite runtime source");
+		return false;
 	}
 
-	bool EvaluateRawRead(const Inst& inst, uint64_t& result, std::string* error) {
+	bool EvaluateRawRead(const Inst& inst, uint64_t& result) {
 		const auto flags = inst.Flags<MemoryFlags>();
 		if (flags.index >= m_program.memory_info.size()) {
-			return Fail(error, "raw scalar read has invalid metadata");
+			return false;
 		}
 		const auto& mem    = m_program.memory_info[flags.index];
 		const auto* handle = inst.Arg(0).ResolveInstruction();
 		if (handle == nullptr) {
-			return Fail(error, "raw scalar read has no descriptor handle");
+			return false;
 		}
 		uint64_t low    = 0;
 		uint64_t high   = 0;
 		uint64_t offset = 0;
-		if (!Arg(*handle, 0, low, error) || !Arg(*handle, 1, high, error) ||
-		    !Arg(inst, 1, offset, error)) {
+		if (!Arg(*handle, 0, low) || !Arg(*handle, 1, high) || !Arg(inst, 1, offset)) {
 			return false;
 		}
 		const auto base      = ((high << 32u) | static_cast<uint32_t>(low)) & AddressMask;
@@ -536,12 +504,11 @@ private:
 		if (inst.GetOpcode() == ValueOpcode::ReadConstBuffer) {
 			uint64_t records = 0;
 			uint64_t word3   = 0;
-			if (handle->NumArgs() != 4u || !Arg(*handle, 2, records, error) ||
-			    !Arg(*handle, 3, word3, error)) {
-				return Fail(error, "constant-buffer descriptor has invalid width");
+			if (handle->NumArgs() != 4u || !Arg(*handle, 2, records) || !Arg(*handle, 3, word3)) {
+				return false;
 			}
 			if (immediate < 0) {
-				return Fail(error, "constant-buffer read has a negative immediate offset");
+				return false;
 			}
 			const auto byte_offset =
 			    static_cast<uint64_t>(immediate) + static_cast<uint32_t>(offset);
@@ -551,21 +518,20 @@ private:
 			                      ? static_cast<uint64_t>(static_cast<uint32_t>(records))
 			                      : static_cast<uint64_t>(stride) * static_cast<uint32_t>(records);
 			if (aligned > size || size - aligned < sizeof(uint32_t)) {
-				return Fail(
-				    error, fmt::format("constant-buffer offset {} exceeds size {}", aligned, size));
+				return false;
 			}
 			address = ((base & ~uint64_t {3}) + byte_offset) & ~uint64_t {3};
 		} else {
 			const auto relative = (immediate & ~int64_t {3}) +
 			                      static_cast<int64_t>(static_cast<uint32_t>(offset) & ~3u);
 			if (!AddSignedAddress(base & ~uint64_t {3}, relative, address)) {
-				return Fail(error, "raw scalar read is outside the 48-bit address space");
+				return false;
 			}
 		}
 		uint32_t word = 0;
 		if (m_runtime.read_memory != nullptr) {
 			if (!m_runtime.read_memory(m_runtime.userdata, address, &word)) {
-				return Fail(error, fmt::format("constant read failed at 0x{:016x}", address));
+				return false;
 			}
 		} else {
 			std::memcpy(&word, reinterpret_cast<const void*>(address), sizeof(word));
@@ -574,28 +540,28 @@ private:
 		return true;
 	}
 
-	bool EvaluateInst(const Inst& inst, uint64_t& result, std::string* error) {
+	bool EvaluateInst(const Inst& inst, uint64_t& result) {
 		uint64_t   a       = 0;
 		uint64_t   b       = 0;
 		uint64_t   c       = 0;
-		const auto binary  = [&]() { return Arg(inst, 0, a, error) && Arg(inst, 1, b, error); };
+		const auto binary  = [&]() { return Arg(inst, 0, a) && Arg(inst, 1, b); };
 		const auto ternary = [&]() {
-			return Arg(inst, 0, a, error) && Arg(inst, 1, b, error) && Arg(inst, 2, c, error);
+			return Arg(inst, 0, a) && Arg(inst, 1, b) && Arg(inst, 2, c);
 		};
 		switch (inst.GetOpcode()) {
 			case ValueOpcode::GetUserData: {
 				const auto reg = RegIndex(inst.Arg(0).ScalarRegister());
 				if (reg < m_program.user_data_base ||
 				    reg - m_program.user_data_base >= m_runtime.user_data.size()) {
-					return Fail(error, fmt::format("user SGPR {} is unavailable", reg));
+					return false;
 				}
 				result = m_runtime.user_data[reg - m_program.user_data_base];
 				return true;
 			}
 			case ValueOpcode::GetShaderBase: result = m_runtime.shader_base; return true;
-			case ValueOpcode::Phi: return EvaluatePhi(inst, result, error);
+			case ValueOpcode::Phi: return EvaluatePhi(inst, result);
 			case ValueOpcode::CompositeExtractU64:
-			case ValueOpcode::CompositeExtractU32x2: return EvaluateExtract(inst, result, error);
+			case ValueOpcode::CompositeExtractU32x2: return EvaluateExtract(inst, result);
 			case ValueOpcode::CompositeConstructU64:
 				if (!binary()) {
 					return false;
@@ -607,20 +573,19 @@ private:
 				const auto slot = inst.Arg(1).Resolve();
 				if (!slot.IsImmediate() || slot.GetType() != Type::U32 ||
 				    slot.U32() >= m_program.srt_reads.size()) {
-					return Fail(error, "invalid flattened SRT index");
+					return false;
 				}
 				if (slot.U32() < m_clean_flat_slots.size() &&
 				    m_clean_flat_slots[slot.U32()] != 0u && m_clean_evaluator != nullptr) {
-					m_clean_evaluator->SetUsePc(m_use_pc);
 					return m_clean_evaluator->EvaluateWide(m_program.srt_reads[slot.U32()].value,
-					                                       result, error);
+					                                       result);
 				}
-				return EvaluateWide(m_program.srt_reads[slot.U32()].value, result, error);
+				return EvaluateWide(m_program.srt_reads[slot.U32()].value, result);
 			}
 			case ValueOpcode::LoadAddressU32:
 			case ValueOpcode::ReadConstBuffer:
 				if (IsRawRead(m_program, inst)) {
-					return EvaluateRawRead(inst, result, error);
+					return EvaluateRawRead(inst, result);
 				}
 				break;
 			case ValueOpcode::IAdd32:
@@ -690,7 +655,7 @@ private:
 				}
 				return false;
 			case ValueOpcode::BitwiseNot32:
-				if (Arg(inst, 0, a, error)) {
+				if (Arg(inst, 0, a)) {
 					result = ~static_cast<uint32_t>(a);
 					return true;
 				}
@@ -737,7 +702,7 @@ private:
 					const auto offset = static_cast<uint32_t>(b);
 					const auto width  = static_cast<uint32_t>(c);
 					if (offset > 32u || width > 32u - offset) {
-						return Fail(error, "invalid unsigned bit-field range");
+						return false;
 					}
 					const auto mask = width == 32u  ? UINT32_MAX
 					                  : width == 0u ? 0u
@@ -751,7 +716,7 @@ private:
 					const auto offset = static_cast<uint32_t>(b);
 					const auto width  = static_cast<uint32_t>(c);
 					if (offset > 32u || width > 32u - offset) {
-						return Fail(error, "invalid signed bit-field range");
+						return false;
 					}
 					if (width == 0u) {
 						result = 0;
@@ -768,13 +733,13 @@ private:
 				return false;
 			case ValueOpcode::BitFieldInsert: {
 				uint64_t d = 0;
-				if (!ternary() || !Arg(inst, 3, d, error)) {
+				if (!ternary() || !Arg(inst, 3, d)) {
 					return false;
 				}
 				const auto offset = static_cast<uint32_t>(c);
 				const auto width  = static_cast<uint32_t>(d);
 				if (offset > 32u || width > 32u - offset) {
-					return Fail(error, "invalid inserted bit-field range");
+					return false;
 				}
 				if (width == 0u) {
 					result = static_cast<uint32_t>(a);
@@ -836,7 +801,7 @@ private:
 				}
 				return false;
 			case ValueOpcode::LogicalNot:
-				if (Arg(inst, 0, a, error)) {
+				if (Arg(inst, 0, a)) {
 					result = a == 0u;
 					return true;
 				}
@@ -845,18 +810,16 @@ private:
 			case ValueOpcode::UndefU8:
 			case ValueOpcode::UndefU16:
 			case ValueOpcode::UndefU32:
-			case ValueOpcode::UndefU64: return Fail(error, "undefined typed runtime value");
+			case ValueOpcode::UndefU64: return false;
 			default: break;
 		}
-		return Fail(error, fmt::format("unsupported typed runtime opcode {}",
-		                               ValueOpcodeName(inst.GetOpcode())));
+		return false;
 	}
 
 	const Program&                            m_program;
 	const SrtRuntime&                         m_runtime;
 	std::span<const uint8_t>                  m_clean_flat_slots;
 	Evaluator*                                m_clean_evaluator = nullptr;
-	uint32_t                                  m_use_pc          = 0;
 	std::unordered_map<const Inst*, uint64_t> m_cache;
 	std::vector<const Inst*>                  m_visiting;
 };
@@ -872,18 +835,12 @@ bool EvaluateRuntimeSourcesImpl(const Program&                           program
                                 std::span<const DescriptorSourceRequest> requests,
                                 const SrtRuntime& runtime, std::vector<DescriptorValue>& results,
                                 std::vector<uint32_t>& flat, bool evaluate_flat,
-                                std::span<const uint8_t> clean_flat_slots, std::string* error) {
+                                std::span<const uint8_t> clean_flat_slots) {
 	if (!program.srt_plan_complete) {
-		if (error != nullptr) {
-			*error = Diagnostic(program, 0, "typed SRT plan is not ready");
-		}
 		return false;
 	}
 	if (std::ranges::any_of(clean_flat_slots, [](uint8_t clean) { return clean != 0u; }) &&
 	    runtime.read_specialization_memory == nullptr) {
-		if (error != nullptr) {
-			*error = Diagnostic(program, 0, "clean flattened SRT read has no memory reader");
-		}
 		return false;
 	}
 	SrtRuntime clean_runtime  = runtime;
@@ -895,17 +852,12 @@ bool EvaluateRuntimeSourcesImpl(const Program&                           program
 	for (const auto& request: requests) {
 		const auto* source = Source(program, request.source);
 		if (source == nullptr) {
-			if (error != nullptr) {
-				*error = Diagnostic(program, request.use_pc,
-				                    fmt::format("invalid descriptor source {}", request.source));
-			}
 			return false;
 		}
-		evaluator.SetUsePc(request.use_pc);
 		DescriptorValue value;
 		value.dword_count = source->dword_count;
 		for (uint32_t index = 0; index < source->dword_count; index++) {
-			if (!evaluator.Evaluate(source->dwords[index], value.dwords[index], error)) {
+			if (!evaluator.Evaluate(source->dwords[index], value.dwords[index])) {
 				return false;
 			}
 		}
@@ -918,9 +870,8 @@ bool EvaluateRuntimeSourcesImpl(const Program&                           program
 			const bool clean    = read.flat_offset < clean_flat_slots.size() &&
 			                      clean_flat_slots[read.flat_offset] != 0u;
 			auto&      selected = clean ? clean_evaluator : evaluator;
-			selected.SetUsePc(read.use_pc);
 			if (read.flat_offset >= flattened.size() ||
-			    !selected.Evaluate(read.value, flattened[read.flat_offset], error)) {
+			    !selected.Evaluate(read.value, flattened[read.flat_offset])) {
 				return false;
 			}
 		}
@@ -934,31 +885,24 @@ bool EvaluateRuntimeSourcesImpl(const Program&                           program
 
 } // namespace
 
-bool ValidateRuntimeValue(const Program& program, Value value, std::string& reason) {
-	return RuntimeValidator(program).Run(value, reason);
+bool ValidateRuntimeValue(const Program& program, Value value) {
+	return RuntimeValidator(program).Run(value);
 }
 
-bool BuildSrtPlan(Program& program, std::string* error) {
+void BuildSrtPlan(Program& program) {
 	if (program.resource_tracking_complete) {
-		if (error != nullptr) {
-			*error = "cannot rebuild SRT after resource tracking";
-		}
-		return false;
+		EXIT("shader SRT planning failed: cannot rebuild SRT after resource tracking");
 	}
 	program.srt_plan_complete = false;
-	if (!PlanBuilder(program).Run(error)) {
-		return false;
-	}
+	PlanBuilder(program).Run();
 	program.srt_plan_complete = true;
-	return true;
 }
 
-bool EvaluateDescriptorSource(const Program& program, uint32_t source, uint32_t use_pc,
-                              const SrtRuntime& runtime, DescriptorValue& result,
-                              std::string* error) {
-	const DescriptorSourceRequest request {source, use_pc};
+bool EvaluateDescriptorSource(const Program& program, uint32_t source, const SrtRuntime& runtime,
+                              DescriptorValue& result) {
+	const DescriptorSourceRequest request {source};
 	std::vector<DescriptorValue>  results;
-	if (!EvaluateDescriptorSources(program, std::span {&request, 1}, runtime, results, error)) {
+	if (!EvaluateDescriptorSources(program, std::span {&request, 1}, runtime, results)) {
 		return false;
 	}
 	result = results.front();
@@ -967,26 +911,23 @@ bool EvaluateDescriptorSource(const Program& program, uint32_t source, uint32_t
 
 bool EvaluateDescriptorSources(const Program&                           program,
                                std::span<const DescriptorSourceRequest> requests,
-                               const SrtRuntime& runtime, std::vector<DescriptorValue>& results,
-                               std::string* error) {
+                               const SrtRuntime& runtime, std::vector<DescriptorValue>& results) {
 	std::vector<uint32_t> ignored;
-	return EvaluateRuntimeSourcesImpl(program, requests, runtime, results, ignored, false, {},
-	                                  error);
+	return EvaluateRuntimeSourcesImpl(program, requests, runtime, results, ignored, false, {});
 }
 
 bool EvaluateRuntimeSources(const Program&                           program,
                             std::span<const DescriptorSourceRequest> requests,
                             const SrtRuntime& runtime, std::vector<DescriptorValue>& results,
-                            std::vector<uint32_t>& flat, std::span<const uint8_t> clean_flat_slots,
-                            std::string* error) {
+                            std::vector<uint32_t>&   flat,
+                            std::span<const uint8_t> clean_flat_slots) {
 	return EvaluateRuntimeSourcesImpl(program, requests, runtime, results, flat, true,
-	                                  clean_flat_slots, error);
+	                                  clean_flat_slots);
 }
 
-bool WalkSrt(const Program& program, const SrtRuntime& runtime, std::vector<uint32_t>& flat,
-             std::string* error) {
+bool WalkSrt(const Program& program, const SrtRuntime& runtime, std::vector<uint32_t>& flat) {
 	std::vector<DescriptorValue> ignored;
-	return EvaluateRuntimeSources(program, {}, runtime, ignored, flat, {}, error);
+	return EvaluateRuntimeSources(program, {}, runtime, ignored, flat, {});
 }
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR

src/graphics/shader/recompiler/ir/passes/SrtWalker.h

@@ -13,43 +13,38 @@ using SrtMemoryReader = bool (*)(void* userdata, uint64_t address, uint32_t* val
 
 struct SrtRuntime {
 	std::span<const uint32_t> user_data;
-	uint64_t                  shader_base = 0;
-	SrtMemoryReader           read_memory = nullptr;
-	void*                     userdata    = nullptr;
+	uint64_t                  shader_base                = 0;
+	SrtMemoryReader           read_memory                = nullptr;
+	void*                     userdata                   = nullptr;
 	SrtMemoryReader           read_specialization_memory = nullptr;
 };
 
 struct DescriptorSourceRequest {
 	uint32_t source = 0;
-	uint32_t use_pc = 0;
 };
 
 // Collects reachable ReadConst values. Immediate offsets receive compact flat-buffer slots;
 // dynamic offsets remain explicit and are never assigned a fake slot.
-bool BuildSrtPlan(Program& program, std::string* error);
-bool ValidateRuntimeValue(const Program& program, Value value, std::string& reason);
+void BuildSrtPlan(Program& program);
+bool ValidateRuntimeValue(const Program& program, Value value);
 
-bool EvaluateDescriptorSource(const Program& program, uint32_t source, uint32_t use_pc,
-                              const SrtRuntime& runtime, DescriptorValue& result,
-                              std::string* error);
+bool EvaluateDescriptorSource(const Program& program, uint32_t source, const SrtRuntime& runtime,
+                              DescriptorValue& result);
 
 // Evaluates one runtime snapshot transactionally. Scalar values and ReadConst results shared by
 // several descriptors are memoized once across the batch.
 bool EvaluateDescriptorSources(const Program&                           program,
                                std::span<const DescriptorSourceRequest> requests,
-                               const SrtRuntime& runtime, std::vector<DescriptorValue>& results,
-                               std::string* error);
+                               const SrtRuntime& runtime, std::vector<DescriptorValue>& results);
 
 // Evaluates descriptor sources and the flattened immediate SRT with one memoized scalar walk.
 // On failure neither destination is changed.
 bool EvaluateRuntimeSources(const Program&                           program,
                             std::span<const DescriptorSourceRequest> requests,
                             const SrtRuntime& runtime, std::vector<DescriptorValue>& results,
-                            std::vector<uint32_t>& flat, std::span<const uint8_t> clean_flat_slots,
-                            std::string* error);
+                            std::vector<uint32_t>& flat, std::span<const uint8_t> clean_flat_slots);
 
-bool WalkSrt(const Program& program, const SrtRuntime& runtime, std::vector<uint32_t>& flat,
-             std::string* error);
+bool WalkSrt(const Program& program, const SrtRuntime& runtime, std::vector<uint32_t>& flat);
 
 } // namespace Libs::Graphics::ShaderRecompiler::IR
 

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

@@ -22,6 +22,9 @@ struct ScalarMaskTag {
 struct ExecTag {
 	auto operator<=>(const ExecTag&) const = default;
 };
+struct ExecMaskTag {
+	auto operator<=>(const ExecMaskTag&) const = default;
+};
 struct ExecLoTag {
 	auto operator<=>(const ExecLoTag&) const = default;
 };
@@ -31,6 +34,9 @@ struct ExecHiTag {
 struct VccTag {
 	auto operator<=>(const VccTag&) const = default;
 };
+struct VccMaskValidTag {
+	auto operator<=>(const VccMaskValidTag&) const = default;
+};
 struct VccLoTag {
 	auto operator<=>(const VccLoTag&) const = default;
 };
@@ -45,9 +51,9 @@ struct GotoVariable {
 	auto     operator<=>(const GotoVariable&) const = default;
 };
 
-using Variable =
-    std::variant<ScalarReg, ThreadBitScalarReg, ScalarMaskTag, VectorReg, GotoVariable, SccTag,
-                 ExecTag, ExecLoTag, ExecHiTag, VccTag, VccLoTag, VccHiTag, M0Tag>;
+using Variable = std::variant<ScalarReg, ThreadBitScalarReg, ScalarMaskTag, VectorReg, GotoVariable,
+                              SccTag, ExecTag, ExecMaskTag, ExecLoTag, ExecHiTag, VccTag,
+                              VccMaskValidTag, VccLoTag, VccHiTag, M0Tag>;
 using ValueMap = std::unordered_map<Block*, Value>;
 
 struct DefTable {
@@ -99,12 +105,16 @@ struct DefTable {
 	void         Set(Block* block, SccTag, Value value) { scc[block] = value; }
 	const Value& Get(Block* block, ExecTag) { return exec[block]; }
 	void         Set(Block* block, ExecTag, Value value) { exec[block] = value; }
+	const Value& Get(Block* block, ExecMaskTag) { return exec_mask_tag[block]; }
+	void         Set(Block* block, ExecMaskTag, Value value) { exec_mask_tag[block] = value; }
 	const Value& Get(Block* block, ExecLoTag) { return exec_lo[block]; }
 	void         Set(Block* block, ExecLoTag, Value value) { exec_lo[block] = value; }
 	const Value& Get(Block* block, ExecHiTag) { return exec_hi[block]; }
 	void         Set(Block* block, ExecHiTag, Value value) { exec_hi[block] = value; }
 	const Value& Get(Block* block, VccTag) { return vcc[block]; }
 	void         Set(Block* block, VccTag, Value value) { vcc[block] = value; }
+	const Value& Get(Block* block, VccMaskValidTag) { return vcc_mask_valid_tag[block]; }
+	void Set(Block* block, VccMaskValidTag, Value value) { vcc_mask_valid_tag[block] = value; }
 	const Value& Get(Block* block, VccLoTag) { return vcc_lo[block]; }
 	void         Set(Block* block, VccLoTag, Value value) { vcc_lo[block] = value; }
 	const Value& Get(Block* block, VccHiTag) { return vcc_hi[block]; }
@@ -114,9 +124,11 @@ struct DefTable {
 
 	ValueMap                               scc;
 	ValueMap                               exec;
+	ValueMap                               exec_mask_tag;
 	ValueMap                               exec_lo;
 	ValueMap                               exec_hi;
 	ValueMap                               vcc;
+	ValueMap                               vcc_mask_valid_tag;
 	ValueMap                               vcc_lo;
 	ValueMap                               vcc_hi;
 	ValueMap                               m0;
@@ -144,6 +156,9 @@ ValueOpcode UndefOpcode(SccTag) {
 ValueOpcode UndefOpcode(ExecTag) {
 	return ValueOpcode::UndefU1;
 }
+ValueOpcode UndefOpcode(ExecMaskTag) {
+	return ValueOpcode::UndefU1;
+}
 ValueOpcode UndefOpcode(ExecLoTag) {
 	return ValueOpcode::UndefU32;
 }
@@ -153,6 +168,9 @@ ValueOpcode UndefOpcode(ExecHiTag) {
 ValueOpcode UndefOpcode(VccTag) {
 	return ValueOpcode::UndefU1;
 }
+ValueOpcode UndefOpcode(VccMaskValidTag) {
+	return ValueOpcode::UndefU1;
+}
 ValueOpcode UndefOpcode(VccLoTag) {
 	return ValueOpcode::UndefU32;
 }
@@ -184,6 +202,9 @@ Value InitialValue(SccTag) {
 Value InitialValue(ExecTag) {
 	return Value(false);
 }
+Value InitialValue(ExecMaskTag) {
+	return Value(false);
+}
 Value InitialValue(ExecLoTag) {
 	return Value(0u);
 }
@@ -193,6 +214,9 @@ Value InitialValue(ExecHiTag) {
 Value InitialValue(VccTag) {
 	return Value(false);
 }
+Value InitialValue(VccMaskValidTag) {
+	return Value(false);
+}
 Value InitialValue(VccLoTag) {
 	return Value(0u);
 }
@@ -351,9 +375,15 @@ void VisitInstruction(Pass& pass, Block* block, Inst& inst) {
 			break;
 		case ValueOpcode::SetScc: pass.Write(SccTag {}, block, inst.Arg(0)); break;
 		case ValueOpcode::SetExec: pass.Write(ExecTag {}, block, inst.Arg(0)); break;
+		case ValueOpcode::SetExecMaskTag:
+			pass.Write(ExecMaskTag {}, block, inst.Arg(0));
+			break;
 		case ValueOpcode::SetExecLo: pass.Write(ExecLoTag {}, block, inst.Arg(0)); break;
 		case ValueOpcode::SetExecHi: pass.Write(ExecHiTag {}, block, inst.Arg(0)); break;
 		case ValueOpcode::SetVcc: pass.Write(VccTag {}, block, inst.Arg(0)); break;
+		case ValueOpcode::SetVccMaskValidTag:
+			pass.Write(VccMaskValidTag {}, block, inst.Arg(0));
+			break;
 		case ValueOpcode::SetVccLo: pass.Write(VccLoTag {}, block, inst.Arg(0)); break;
 		case ValueOpcode::SetVccHi: pass.Write(VccHiTag {}, block, inst.Arg(0)); break;
 		case ValueOpcode::SetM0: pass.Write(M0Tag {}, block, inst.Arg(0)); break;
@@ -375,9 +405,15 @@ void VisitInstruction(Pass& pass, Block* block, Inst& inst) {
 			break;
 		case ValueOpcode::GetScc: inst.ReplaceUsesWith(pass.Read(SccTag {}, block)); break;
 		case ValueOpcode::GetExec: inst.ReplaceUsesWith(pass.Read(ExecTag {}, block)); break;
+		case ValueOpcode::GetExecMaskTag:
+			inst.ReplaceUsesWith(pass.Read(ExecMaskTag {}, block));
+			break;
 		case ValueOpcode::GetExecLo: inst.ReplaceUsesWith(pass.Read(ExecLoTag {}, block)); break;
 		case ValueOpcode::GetExecHi: inst.ReplaceUsesWith(pass.Read(ExecHiTag {}, block)); break;
 		case ValueOpcode::GetVcc: inst.ReplaceUsesWith(pass.Read(VccTag {}, block)); break;
+		case ValueOpcode::GetVccMaskValidTag:
+			inst.ReplaceUsesWith(pass.Read(VccMaskValidTag {}, block));
+			break;
 		case ValueOpcode::GetVccLo: inst.ReplaceUsesWith(pass.Read(VccLoTag {}, block)); break;
 		case ValueOpcode::GetVccHi: inst.ReplaceUsesWith(pass.Read(VccHiTag {}, block)); break;
 		case ValueOpcode::GetM0: inst.ReplaceUsesWith(pass.Read(M0Tag {}, block)); break;

src/graphics/shader/shader.cpp

@@ -841,23 +841,22 @@ static void ApplyPixelOutputs(ShaderPixelInputInfo&                info,
 	}
 }
 
-static bool LogPermutationMismatch(const ShaderRecompiler::IR::Program& program,
-                                   const char* stage, const std::string& error) {
+static bool LogPermutationMismatch(const ShaderRecompiler::IR::Program* program,
+                                   const char* stage) {
 	static std::atomic<uint32_t> log_count {0};
 	if (Config::GraphicsDebugDumpEnabled() &&
 	    log_count.fetch_add(1, std::memory_order_relaxed) < 64) {
-		LOGF("ShaderProgramCache native runtime mismatch %s shader=0x%016" PRIx64 ": %s\n",
-		     stage, program.shader_hash, error.c_str());
+		LOGF("ShaderProgramCache native runtime mismatch %s shader=0x%016" PRIx64 "\n",
+		     stage, program != nullptr ? program->shader_hash : 0);
 	}
 	return false;
 }
 
 bool MaterializeProgram(const std::shared_ptr<const ShaderRecompiler::IR::Program>& program,
                         const ShaderParams& params, ShaderVertexInputInfo& info) {
-	std::string error;
-	if (!ShaderMaterializeStageRuntime(program, params.user_data, params.Base(), info.stage, &error,
+	if (!ShaderMaterializeStageRuntime(program, params.user_data, params.Base(), info.stage,
 	                                   ReadShaderGuestMemory)) {
-		return LogPermutationMismatch(*program, "VS", error);
+		return LogPermutationMismatch(program.get(), "VS");
 	}
 	ApplyVertexOutputs(info, *program);
 	return true;
@@ -865,10 +864,9 @@ bool MaterializeProgram(const std::shared_ptr<const ShaderRecompiler::IR::Progra
 
 bool MaterializeProgram(const std::shared_ptr<const ShaderRecompiler::IR::Program>& program,
                         const ShaderParams& params, ShaderPixelInputInfo& info) {
-	std::string error;
-	if (!ShaderMaterializeStageRuntime(program, params.user_data, params.Base(), info.stage, &error,
+	if (!ShaderMaterializeStageRuntime(program, params.user_data, params.Base(), info.stage,
 	                                   ReadShaderGuestMemory)) {
-		return LogPermutationMismatch(*program, "PS", error);
+		return LogPermutationMismatch(program.get(), "PS");
 	}
 	ApplyPixelOutputs(info, *program);
 	return true;
@@ -876,10 +874,9 @@ bool MaterializeProgram(const std::shared_ptr<const ShaderRecompiler::IR::Progra
 
 bool MaterializeProgram(const std::shared_ptr<const ShaderRecompiler::IR::Program>& program,
                         const ShaderParams& params, ShaderComputeInputInfo& info) {
-	std::string error;
-	if (!ShaderMaterializeStageRuntime(program, params.user_data, params.Base(), info.stage, &error,
+	if (!ShaderMaterializeStageRuntime(program, params.user_data, params.Base(), info.stage,
 	                                   ReadShaderGuestMemory)) {
-		return LogPermutationMismatch(*program, "CS", error);
+		return LogPermutationMismatch(program.get(), "CS");
 	}
 	return true;
 }
@@ -1252,11 +1249,7 @@ static vk::ShaderModule CompileModule(vk::Device device, const ShaderParams& par
                                       ShaderStageRuntime& stage) {
 	const auto* stage_name = ShaderStageName(options.stage);
 	const auto* label      = ShaderStageLabel(options.stage);
-	ShaderRecompiler::CompileResult result;
-	std::string                     error;
-	if (!ShaderRecompiler::TryRecompile(params.code, options, result, &error)) {
-		ExitShaderRecompilerFailure(label, options.shader_hash, error.c_str());
-	}
+	auto result = ShaderRecompiler::Recompile(params.code, options);
 
 	DumpShaderRecompilerOriginal(stage_name, options.shader_hash, params.code, result.decoded_dump);
 	if (!SpirvValidateBinary(label, options.shader_hash, result.spirv)) {

src/graphics/shader/shader.h

@@ -57,7 +57,7 @@ using ShaderSpecializationMemoryReader = bool (*)(void* userdata, uint64_t addre
 bool ShaderMaterializeStageRuntime(
     std::shared_ptr<const ShaderRecompiler::IR::Program> program,
     std::span<const uint32_t> user_data, uint64_t shader_base, ShaderStageRuntime& stage,
-    std::string* error, ShaderSpecializationMemoryReader read_specialization_memory = nullptr,
+    ShaderSpecializationMemoryReader read_specialization_memory = nullptr,
     void* read_memory_data = nullptr);
 
 constexpr uint32_t DstSel(uint32_t x, uint32_t y = 0, uint32_t z = 0, uint32_t w = 0) {

src/graphics/shader/shaderStageRuntime.cpp

@@ -7,13 +7,10 @@ namespace Libs::Graphics {
 
 bool ShaderMaterializeStageRuntime(std::shared_ptr<const ShaderRecompiler::IR::Program> program,
                                    std::span<const uint32_t> user_data, uint64_t shader_base,
-                                   ShaderStageRuntime& stage, std::string* error,
+                                   ShaderStageRuntime& stage,
                                    ShaderSpecializationMemoryReader read_specialization_memory,
                                    void*                            read_memory_data) {
 	if (program == nullptr) {
-		if (error != nullptr) {
-			*error = "missing native shader plan";
-		}
 		return false;
 	}
 	ShaderRecompiler::IR::SrtRuntime runtime;
@@ -22,8 +19,8 @@ bool ShaderMaterializeStageRuntime(std::shared_ptr<const ShaderRecompiler::IR::P
 	runtime.read_specialization_memory = read_specialization_memory;
 	runtime.userdata                   = read_memory_data;
 	ShaderRecompiler::IR::ResourceSnapshot snapshot;
-	if (!ShaderRecompiler::IR::MaterializeResources(*program, runtime, snapshot, error) ||
-	    !ShaderRecompiler::IR::ValidateResourceSpecialization(*program, snapshot, error)) {
+	if (!ShaderRecompiler::IR::MaterializeResources(*program, runtime, snapshot) ||
+	    !ShaderRecompiler::IR::ValidateResourceSpecialization(*program, snapshot)) {
 		return false;
 	}
 	auto resources =

src/libs/agc.cpp

@@ -3240,6 +3240,27 @@ uint32_t KYTY_SYSV_ABI AgcAcbCondExecGetSize() {
 	return AgcDcbCondExecGetSize();
 }
 
+uint32_t* KYTY_SYSV_ABI AgcAcbJump(CommandBuffer* buf, uint8_t cache_policy,
+                                   const uint32_t* target, uint32_t size_in_dwords) {
+	if (buf == nullptr) {
+		return nullptr;
+	}
+
+	auto* cmd = buf->AllocateDW(4);
+	if (cmd == nullptr) {
+		return nullptr;
+	}
+
+	auto target_address = reinterpret_cast<uint64_t>(target);
+	cmd[0]              = KYTY_PM4(4, Pm4::IT_INDIRECT_BUFFER, 0u);
+	cmd[1]              = static_cast<uint32_t>(target_address) & ~0x3u;
+	cmd[2]              = static_cast<uint32_t>(target_address >> 32u);
+	cmd[3] = 0x0f900000u | ((static_cast<uint32_t>(cache_policy) & 0x3u) << 28u) |
+	         (size_in_dwords & 0xfffffu);
+
+	return cmd;
+}
+
 uint32_t KYTY_SYSV_ABI AgcAcbJumpGetSize() {
 	return 0x10u;
 }
@@ -3251,6 +3272,12 @@ uint32_t* KYTY_SYSV_ABI AgcAcbWaitRegMem(CommandBuffer* buf, uint8_t size, uint8
 	                        poll_cycles);
 }
 
+uint64_t KYTY_SYSV_ABI AgcAcbWaitOnAddressGetSize(uint8_t size) {
+	PRINT_NAME();
+
+	return AgcDcbWaitOnAddressGetSize(size);
+}
+
 uint32_t* KYTY_SYSV_ABI AgcAcbDmaData(CommandBuffer* buf, uint8_t dst, uint8_t dst_cache_policy,
                                       uint64_t dst_address_or_offset, uint8_t src,
                                       uint8_t  src_cache_policy,
@@ -3322,6 +3349,12 @@ uint32_t* KYTY_SYSV_ABI AgcAcbCopyData(CommandBuffer* buf, uint8_t dst, uint8_t
 	return cmd;
 }
 
+uint64_t KYTY_SYSV_ABI AgcAcbCopyDataGetSize() {
+	PRINT_NAME();
+
+	return AgcDcbCopyDataGetSize();
+}
+
 uint32_t* KYTY_SYSV_ABI AgcAcbDispatchIndirect(CommandBuffer*       buf,
                                                const volatile void* indirect_args,
                                                uint32_t             modifier) {

src/libs/agc.h

@@ -204,10 +204,13 @@ uint32_t KYTY_SYSV_ABI  AgcAcbAcquireMemGetSize();
 uint32_t* KYTY_SYSV_ABI AgcAcbCondExec(CommandBuffer* buf, const volatile uint32_t* address,
                                        uint32_t num_dwords);
 uint32_t KYTY_SYSV_ABI  AgcAcbCondExecGetSize();
+uint32_t* KYTY_SYSV_ABI AgcAcbJump(CommandBuffer* buf, uint8_t cache_policy,
+                                   const uint32_t* target, uint32_t size_in_dwords);
 uint32_t KYTY_SYSV_ABI  AgcAcbJumpGetSize();
 uint32_t* KYTY_SYSV_ABI AgcAcbWaitRegMem(CommandBuffer* buf, uint8_t size, uint8_t compare_function,
                                          uint8_t cache_policy, const volatile void* address,
                                          uint64_t reference, uint64_t mask, uint32_t poll_cycles);
+uint64_t KYTY_SYSV_ABI  AgcAcbWaitOnAddressGetSize(uint8_t size);
 uint32_t* KYTY_SYSV_ABI AgcAcbDmaData(CommandBuffer* buf, uint8_t dst, uint8_t dst_cache_policy,
                                       uint64_t dst_address_or_offset, uint8_t src,
                                       uint8_t  src_cache_policy,
@@ -218,6 +221,7 @@ uint32_t* KYTY_SYSV_ABI AgcAcbCopyData(CommandBuffer* buf, uint8_t dst, uint8_t
                                        uint64_t dst_address, uint8_t src, uint8_t src_cache_policy,
                                        uint64_t src_address_or_immediate, uint8_t item_size,
                                        uint8_t write_confirm);
+uint64_t KYTY_SYSV_ABI  AgcAcbCopyDataGetSize();
 uint32_t* KYTY_SYSV_ABI AgcAcbDispatchIndirect(CommandBuffer*       buf,
                                                const volatile void* indirect_args,
                                                uint32_t             modifier);

src/libs/libAgcDriver.cpp

@@ -73,10 +73,13 @@ LIB_DEFINE(InitAgcDriver_1) {
 	LIB_FUNC("ewobAQeMo5k", Gen5::AgcAcbAcquireMemGetSize);
 	LIB_FUNC("qyM2bxYFPAk", Gen5::AgcAcbCondExec);
 	LIB_FUNC("ozKzBP4aki4", Gen5::AgcAcbCondExecGetSize);
+	LIB_FUNC("e1DFTg+Sd8U", Gen5::AgcAcbJump);
 	LIB_FUNC("b-oySn+G2tE", Gen5::AgcAcbJumpGetSize);
 	LIB_FUNC("htn36gPnBk4", Gen5::AgcAcbWaitRegMem);
+	LIB_FUNC("idlaArvdXEs", Gen5::AgcAcbWaitOnAddressGetSize);
 	LIB_FUNC("-RnpfpxIhec", Gen5::AgcAcbDmaData);
 	LIB_FUNC("qzMN2XKGA4k", Gen5::AgcAcbCopyData);
+	LIB_FUNC("CbQh3DKMSno", Gen5::AgcAcbCopyDataGetSize);
 	LIB_FUNC("j3EtxFkSIhQ", Gen5::AgcAcbDispatchIndirect);
 	LIB_FUNC("eZ4+17OQz4Q", Gen5::AgcAcbWriteData);
 	LIB_FUNC("xAeBOa0A3kk", Gen5::AgcAcbSetMarker);

src/main.cpp

@@ -52,6 +52,8 @@ static void PrintUsage() {
 	         Config::DEFAULT_USER_ID);
 	::printf(
 	    "  --present-mode <value>               Fifo, Mailbox, or Immediate. Default: Fifo.\n");
+	::printf(
+	    "  --gpu <index>                        Vulkan physical device index. Default: auto.\n");
 	::printf("  --fullscreen                         Run in borderless desktop fullscreen.\n");
 	::printf("  --vblank-frequency <num>             Virtual vblank frequency. Default: 60.\n");
 	::printf("  --console-language <0-29>            Console language. Default: 1 (English US).\n");
@@ -234,6 +236,8 @@ static bool ParseArgs(int argc, char* argv[], RunOptions& options, bool& show_he
 				::printf("invalid present mode: %s\n", value.c_str());
 				return false;
 			}
+		} else if (arg == "--gpu") {
+			options.config.gpu_index = Common::ToInt32(value);
 		} else if (arg == "--vblank-frequency") {
 			const int32_t vblank_frequency = Common::ToInt32(value);
 			options.config.vblank_frequency =

tests/ResourceTrackingTests.cpp

@@ -29,6 +29,19 @@ void Check(bool condition, const char *message) {
   }
 }
 
+template <typename F>
+void CheckFatal(F &&function, std::string_view expected, const char *message) {
+  try {
+    function();
+  } catch (const std::runtime_error &error) {
+    Check(std::string_view(error.what()).find(expected) !=
+              std::string_view::npos,
+          message);
+    return;
+  }
+  Check(false, message);
+}
+
 struct Fixture {
   Program program;
   Block *block = nullptr;
@@ -75,8 +88,7 @@ struct Fixture {
   }
 
   MemoryFlags AddMemory(MemoryInfo memory, uint32_t pc) {
-    const auto index =
-        static_cast<uint32_t>(program.memory_info.size());
+    const auto index = static_cast<uint32_t>(program.memory_info.size());
     program.memory_info.push_back(memory);
     return {index, pc};
   }
@@ -113,10 +125,8 @@ struct Fixture {
   }
 
   void PlanAndTrack() {
-    std::string error;
-    if (!BuildSrtPlan(program, &error) || !TrackResources(program, &error)) {
-      throw std::runtime_error(error);
-    }
+    BuildSrtPlan(program);
+    TrackResources(program);
   }
 };
 
@@ -241,10 +251,7 @@ void TestInvariantIndirectImageMaterialization() {
   auto fixture = MakeIndirectImageFixture(false);
   fixture->PlanAndTrack();
   EliminateDeadCode(fixture->program.blocks);
-  std::string validation_error;
-  Check(
-      ValidateProgram(fixture->program, true, &validation_error),
-      "post-tracking dead-code elimination invalidated descriptor provenance");
+  ValidateProgram(fixture->program, true);
 
   Check(fixture->program.info.buffers.size() == 1 &&
             fixture->program.info.images.size() == 1 &&
@@ -291,7 +298,6 @@ void TestInvariantIndirectImageMaterialization() {
                      .userdata = &memory,
                      .read_specialization_memory = ReadLinearTestMemory};
   ResourceSnapshot snapshot;
-  std::string error;
   const auto same_snapshot = [](const ResourceSnapshot &lhs,
                                 const ResourceSnapshot &rhs) {
     return lhs.buffers == rhs.buffers && lhs.images == rhs.images &&
@@ -300,7 +306,7 @@ void TestInvariantIndirectImageMaterialization() {
            lhs.user_data == rhs.user_data &&
            lhs.indirect_images.empty() == rhs.indirect_images.empty();
   };
-  Check(MaterializeResources(fixture->program, runtime, snapshot, &error) &&
+  Check(MaterializeResources(fixture->program, runtime, snapshot) &&
             snapshot.images.size() == 1 &&
             std::equal(image_descriptor.begin(), image_descriptor.end(),
                        snapshot.images[0].dwords.begin()),
@@ -308,8 +314,7 @@ void TestInvariantIndirectImageMaterialization() {
 
   const auto prior_snapshot = snapshot;
   memory.fail_address = 0x1004u;
-  Check(!MaterializeResources(fixture->program, runtime, snapshot, &error) &&
-            error.find("scalar read") != std::string::npos &&
+  Check(!MaterializeResources(fixture->program, runtime, snapshot) &&
             same_snapshot(snapshot, prior_snapshot),
         "rejected planning memory read mutated the snapshot");
   memory.fail_address = UINT64_MAX;
@@ -325,12 +330,11 @@ void TestInvariantIndirectImageMaterialization() {
   memory.words[(0x2020u - memory.base) / 4u + 3u] =
       image_descriptor[3] ^ (1u << 28u);
   ResourceSnapshot null_snapshot;
-  Check(
-      MaterializeResources(fixture->program, runtime, null_snapshot, &error) &&
-          null_snapshot.indirect_images.empty() &&
-          std::ranges::all_of(null_snapshot.images[0].dwords,
-                              [](uint32_t dword) { return dword == 0u; }),
-      "stale typed null image descriptors were not canonicalized");
+  Check(MaterializeResources(fixture->program, runtime, null_snapshot) &&
+            null_snapshot.indirect_images.empty() &&
+            std::ranges::all_of(null_snapshot.images[0].dwords,
+                                [](uint32_t dword) { return dword == 0u; }),
+        "stale typed null image descriptors were not canonicalized");
 
   for (uint32_t dword = 0; dword < image_descriptor.size(); dword++) {
     memory.words[(0x2000u - memory.base) / 4u + dword] =
@@ -341,13 +345,13 @@ void TestInvariantIndirectImageMaterialization() {
   memory.words[(0x2020u - memory.base) / 4u] ^= 1u;
   memory.words[(0x1000u - memory.base + 36u) / 4u] = 1u;
   ResourceSnapshot dynamic_snapshot;
-  Check(MaterializeResources(fixture->program, runtime, dynamic_snapshot,
-                             &error) &&
+  Check(MaterializeResources(fixture->program, runtime, dynamic_snapshot) &&
             dynamic_snapshot.images.size() == 1 &&
             dynamic_snapshot.indirect_images.size() == 1 &&
-            dynamic_snapshot.indirect_images[0].descriptors.size() == 2 &&
-            SpecializeResources(fixture->program, dynamic_snapshot, &error) &&
-            fixture->program.info.images.size() == 2 &&
+            dynamic_snapshot.indirect_images[0].descriptors.size() == 2,
+        "dynamic indirect image table did not materialize");
+  SpecializeResources(fixture->program, dynamic_snapshot);
+  Check(fixture->program.info.images.size() == 2 &&
             fixture->program.info.images[0].indirect_root == 0 &&
             fixture->program.info.images[0].indirect_mapping_capacity != 0 &&
             fixture->program.info.images[0].indirect_resources.size() == 2 &&
@@ -364,27 +368,21 @@ void TestInvariantIndirectImageMaterialization() {
   memory.words[(0x2000u - memory.base) / 4u] += 0x100u;
   memory.words[(0x2020u - memory.base) / 4u] += 0x101u;
   ResourceSnapshot rebound_snapshot;
-  Check(MaterializeResources(fixture->program, runtime, rebound_snapshot,
-                             &error) &&
-            ValidateResourceSpecialization(fixture->program, rebound_snapshot,
-                                           &error),
+  Check(MaterializeResources(fixture->program, runtime, rebound_snapshot) &&
+            ValidateResourceSpecialization(fixture->program, rebound_snapshot),
         "stable indirect key mapping did not accept changed image addresses");
   memory.words[(0x2020u - memory.base) / 4u] =
       memory.words[(0x2000u - memory.base) / 4u];
-  Check(MaterializeResources(fixture->program, runtime, rebound_snapshot,
-                             &error) &&
-            ValidateResourceSpecialization(fixture->program, rebound_snapshot,
-                                           &error),
+  Check(MaterializeResources(fixture->program, runtime, rebound_snapshot) &&
+            ValidateResourceSpecialization(fixture->program, rebound_snapshot),
         "runtime indirect key mapping did not accept collapsed candidates");
   const auto collapsed_snapshot = rebound_snapshot;
   ResourceSnapshot capacity_snapshot;
   for (const uint32_t records : {1u, 3u}) {
     user_data[2] = records;
     Check(
-        MaterializeResources(fixture->program, runtime, capacity_snapshot,
-                             &error) &&
-            ValidateResourceSpecialization(fixture->program, capacity_snapshot,
-                                           &error),
+        MaterializeResources(fixture->program, runtime, capacity_snapshot) &&
+            ValidateResourceSpecialization(fixture->program, capacity_snapshot),
         "runtime indirect key mapping rejected a fitting material-table size");
   }
   user_data[2] = 2u;
@@ -398,9 +396,7 @@ void TestInvariantIndirectImageMaterialization() {
   }
   memory.words[(0x1000u - memory.base + 68u) / 4u] = 2u;
   Check(
-      !MaterializeResources(fixture->program, runtime, rebound_snapshot,
-                            &error) &&
-          error.find("candidate topology") != std::string::npos &&
+      !MaterializeResources(fixture->program, runtime, rebound_snapshot) &&
           same_snapshot(rebound_snapshot, collapsed_snapshot),
       "larger indirect candidate topology reused or mutated a cached snapshot");
 
@@ -418,26 +414,26 @@ void TestInvariantIndirectImageMaterialization() {
                                    .read_specialization_memory =
                                        ReadLinearTestMemory};
   Check(!MaterializeResources(memory_backed->program, memory_backed_runtime,
-                              snapshot, &error) &&
-            error.find("constant read failed") != std::string::npos &&
+                              snapshot) &&
             same_snapshot(snapshot, prior_snapshot),
         "rejected indirect table descriptor read mutated the snapshot");
   memory.fail_address = UINT64_MAX;
 
   auto malformed = MakeIndirectImageFixture(true);
-  Check(BuildSrtPlan(malformed->program, &error) &&
-            !TrackResources(malformed->program, &error) &&
-            error.find("ReadFirstLane") != std::string::npos &&
-            !malformed->program.resource_tracking_complete &&
+  BuildSrtPlan(malformed->program);
+  CheckFatal([&] { TrackResources(malformed->program); }, "not a valid runtime value",
+             "malformed indirect image pattern was accepted");
+  Check(!malformed->program.resource_tracking_complete &&
             malformed->program.info.images.empty() &&
             malformed->program.descriptor_sources.empty(),
         "malformed indirect image pattern was partially accepted");
 
   auto wrapped_immediate = MakeIndirectImageFixture(false, 4u);
-  Check(BuildSrtPlan(wrapped_immediate->program, &error) &&
-            !TrackResources(wrapped_immediate->program, &error) &&
-            error.find("ReadFirstLane") != std::string::npos &&
-            !wrapped_immediate->program.resource_tracking_complete,
+  BuildSrtPlan(wrapped_immediate->program);
+  CheckFatal([&] { TrackResources(wrapped_immediate->program); },
+             "not a valid runtime value",
+             "wrapped scalar immediate entered the invariant image proof");
+  Check(!wrapped_immediate->program.resource_tracking_complete,
         "wrapped scalar immediate entered the invariant image proof");
 }
 
@@ -494,16 +490,12 @@ void TestDenseBufferTracking() {
             second.Instruction()->Flags<uint32_t>() == 1,
         "typed handles were not assigned dense indices");
   Check(fixture.program.memory_info[load_flags.index].resource == 0 &&
-            fixture.program.memory_info[store_flags.index].resource ==
-                0 &&
-            fixture.program.memory_info[other_flags.index].resource ==
-                1,
+            fixture.program.memory_info[store_flags.index].resource == 0 &&
+            fixture.program.memory_info[other_flags.index].resource == 1,
         "typed memory metadata was not patched to dense indices");
 
-  std::string error;
-  Check(!TrackResources(fixture.program, &error) &&
-            error.find("already tracked") != std::string::npos,
-        "resource tracking allowed a second mutation pass");
+  CheckFatal([&] { TrackResources(fixture.program); }, "already tracked",
+             "resource tracking allowed a second mutation pass");
 }
 
 void TestScalarAndVectorBufferAlias() {
@@ -531,8 +523,7 @@ void TestScalarAndVectorBufferAlias() {
             fixture.program.info.buffers[0].scalar,
         "typed scalar and vector uses of one descriptor were split");
   Check(fixture.program.memory_info[scalar_flags.index].resource == 0 &&
-            fixture.program.memory_info[vector_flags.index].resource ==
-                0,
+            fixture.program.memory_info[vector_flags.index].resource == 0,
         "scalar/vector alias did not share a dense index");
 }
 
@@ -552,16 +543,13 @@ void TestRuntimeUnsignedMinDescriptor() {
   std::array<uint32_t, 1> user_data{0xffffffffu};
   SrtRuntime runtime{.user_data = user_data};
   DescriptorValue value;
-  std::string error;
   const auto source = fixture.program.info.buffers[0].source;
-  Check(EvaluateDescriptorSource(fixture.program, source, 0x330, runtime, value,
-                                 &error) &&
+  Check(EvaluateDescriptorSource(fixture.program, source, runtime, value) &&
             value.dwords[3] == 0x100u,
         "runtime descriptor unsigned minimum did not clamp its first operand");
   user_data[0] = 0x80u;
   Check(
-      EvaluateDescriptorSource(fixture.program, source, 0x330, runtime, value,
-                               &error) &&
+      EvaluateDescriptorSource(fixture.program, source, runtime, value) &&
           value.dwords[3] == 0x80u,
       "runtime descriptor unsigned minimum did not preserve its first operand");
 }
@@ -624,9 +612,7 @@ void TestImagesSamplersAndAliases() {
             repeated.second.Instruction()->Flags<uint32_t>() == 0,
         "unused sampler border colors prevented source interning");
   const auto sampler_source = fixture.program.info.samplers[0].source;
-  Check(fixture.program.descriptor_sources[sampler_source]
-                .dwords[3]
-                .U32() == 0,
+  Check(fixture.program.descriptor_sources[sampler_source].dwords[3].U32() == 0,
         "unused sampler border color was not canonicalized");
   Check(fixture.program.info.buffers[0].image_alias == 0,
         "buffer/image descriptor alias was not linked");
@@ -673,9 +659,7 @@ void TestSampleAdjustSamplerScratch() {
   std::array<uint32_t, 4> user_data{4u, 1u, 2u, 0x80000abcu};
   SrtRuntime runtime{.user_data = user_data};
   DescriptorValue descriptor;
-  std::string error;
-  Check(EvaluateDescriptorSource(fixture.program, source, 0x1ec, runtime,
-                                 descriptor, &error) &&
+  Check(EvaluateDescriptorSource(fixture.program, source, runtime, descriptor) &&
             descriptor.dwords[3] == 0x80000abcu,
         "SampleAdjust canonicalization lost sampler border fields");
 
@@ -704,11 +688,9 @@ void TestSampleAdjustSamplerScratch() {
     rejected.Emit(ValueOpcode::ImageSampleRaw,
                   {rejected_image, rejected_sampler, rejected.ImageAddress()},
                   rejected.AddMemory(rejected_memory, 0x200));
-    std::string rejected_error;
-    Check(BuildSrtPlan(rejected.program, &rejected_error) &&
-              !TrackResources(rejected.program, &rejected_error) &&
-              rejected_error.find("WqmMask") != std::string::npos,
-          message);
+    BuildSrtPlan(rejected.program);
+    CheckFatal([&] { TrackResources(rejected.program); },
+               "not a valid runtime value", message);
   };
   CheckRejected(0u, 12u,
                 "ordinary sampling accepted SampleAdjust reserved scratch");
@@ -778,15 +760,13 @@ void TestDynamicStorageMipTracking() {
   descriptor.dword_count = 8;
   ResourceSnapshot snapshot;
   snapshot.images.assign(images.size(), descriptor);
-  std::string error;
-  Check(SpecializeResources(fixture.program, snapshot, &error) &&
-            fixture.program.info.images[1].mip_count == 3 &&
-            ValidateResourceSpecialization(fixture.program, snapshot, &error),
+  SpecializeResources(fixture.program, snapshot);
+  Check(fixture.program.info.images[1].mip_count == 3 &&
+            ValidateResourceSpecialization(fixture.program, snapshot),
         "base-1 through last-3 dynamic storage range was not specialized");
   ShaderComputeInputInfo compute{};
-  Check(CollectShaderInfo(fixture.program, {.compute = &compute}, &error) &&
-            AllocateBindings(fixture.program, 0, &error),
-        "dynamic storage mip bindings were not allocated");
+  CollectShaderInfo(fixture.program, {.compute = &compute});
+  AllocateBindings(fixture.program, 0);
   const auto *storage_binding =
       FindBinding(fixture.program.bindings, DescriptorBindingKind::Storage2D);
   Check(storage_binding != nullptr &&
@@ -805,19 +785,19 @@ void TestDynamicStorageMipTracking() {
   DescriptorValue null_descriptor{};
   null_descriptor.dword_count = 8;
   null_snapshot.images.push_back(null_descriptor);
-  Check(SpecializeResources(null_program, null_snapshot, &error) &&
-            null_program.info.images[0].mip_count == 1 &&
-            ValidateResourceSpecialization(null_program, null_snapshot, &error),
+  SpecializeResources(null_program, null_snapshot);
+  Check(null_program.info.images[0].mip_count == 1 &&
+            ValidateResourceSpecialization(null_program, null_snapshot),
         "canonical null dynamic storage image did not retain one descriptor");
 
   snapshot.images[1].dwords[3] =
       (snapshot.images[1].dwords[3] & ~(0xfu << 16u)) | (2u << 16u);
-  Check(!ValidateResourceSpecialization(fixture.program, snapshot, &error),
+  Check(!ValidateResourceSpecialization(fixture.program, snapshot),
         "a changed dynamic storage mip count reused the specialization");
   snapshot.images[1].dwords[3] =
       (snapshot.images[1].dwords[3] & ~((0xfu << 12u) | (0xfu << 16u))) |
       (4u << 12u) | (3u << 16u);
-  Check(!ValidateResourceSpecialization(fixture.program, snapshot, &error),
+  Check(!ValidateResourceSpecialization(fixture.program, snapshot),
         "an inverted dynamic storage mip range was accepted");
 }
 
@@ -861,11 +841,9 @@ void TestSrtFlatteningAndRuntimeMemoization() {
                      .userdata = &memory};
   std::vector<DescriptorValue> descriptors;
   std::vector<uint32_t> flat;
-  const DescriptorSourceRequest request{fixture.program.info.buffers[0].source,
-                                        12};
-  std::string error;
+  const DescriptorSourceRequest request{fixture.program.info.buffers[0].source};
   Check(EvaluateRuntimeSources(fixture.program, std::span{&request, 1}, runtime,
-                               descriptors, flat, {}, &error),
+                               descriptors, flat, {}),
         "typed runtime source evaluation failed");
   Check(descriptors.size() == 1 && descriptors[0].dwords[0] == 0xdeadbeefu &&
             flat == std::vector<uint32_t>{0xdeadbeefu} && memory.reads == 1,
@@ -876,16 +854,16 @@ void TestSrtFlatteningAndRuntimeMemoization() {
   descriptors = {{{1u}, 1u}};
   flat = {2u};
   Check(!EvaluateRuntimeSources(fixture.program, std::span{&request, 1},
-                                runtime, descriptors, flat, {}, &error) &&
+                                runtime, descriptors, flat, {}) &&
             descriptors == std::vector<DescriptorValue>{{{1u}, 1u}} &&
             flat == std::vector<uint32_t>{2u},
         "runtime evaluation failure was not transactional");
 
   ShaderComputeInputInfo compute{};
-  Check(CollectShaderInfo(fixture.program, {.compute = &compute}, &error) &&
-            AllocateBindings(fixture.program, 0, &error) &&
-            FindBinding(fixture.program.bindings,
-                        DescriptorBindingKind::FlattenedSrt) != nullptr,
+  CollectShaderInfo(fixture.program, {.compute = &compute});
+  AllocateBindings(fixture.program, 0);
+  Check(FindBinding(fixture.program.bindings,
+                    DescriptorBindingKind::FlattenedSrt) != nullptr,
         "flattened typed SRT reads did not receive a binding");
 }
 
@@ -919,18 +897,16 @@ void TestDynamicSrtReadRemainsExplicit() {
                      .read_memory = ReadTestMemory,
                      .userdata = &memory};
   DescriptorValue value;
-  std::string error;
   Check(EvaluateDescriptorSource(fixture.program,
-                                 fixture.program.info.buffers[0].source, 8,
-                                 runtime, value, &error) &&
+                                 fixture.program.info.buffers[0].source, runtime, value) &&
             value.dwords[0] == 0xabcdef01u && memory.reads == 1,
         "dynamic typed scalar descriptor source was not evaluated");
 
   ShaderComputeInputInfo compute{};
-  Check(CollectShaderInfo(fixture.program, {.compute = &compute}, &error) &&
-            AllocateBindings(fixture.program, 0, &error) &&
-            FindBinding(fixture.program.bindings,
-                        DescriptorBindingKind::FlattenedSrt) == nullptr &&
+  CollectShaderInfo(fixture.program, {.compute = &compute});
+  AllocateBindings(fixture.program, 0);
+  Check(FindBinding(fixture.program.bindings,
+                    DescriptorBindingKind::FlattenedSrt) == nullptr &&
             FindBinding(fixture.program.bindings,
                         DescriptorBindingKind::BdaPagetable) != nullptr &&
             FindBinding(fixture.program.bindings,
@@ -966,12 +942,10 @@ void TestPhiValidation() {
                {handle, Value(0u), Value(0u), Value(0u), Value(true)},
                fixture.AddMemory(memory, 20), merge);
 
-  std::string error;
-  Check(BuildSrtPlan(fixture.program, &error),
-        "SRT planning rejected a well-formed typed phi");
-  Check(!TrackResources(fixture.program, &error) &&
-            error.find("control-dependent phi") != std::string::npos &&
-            !fixture.program.resource_tracking_complete &&
+  BuildSrtPlan(fixture.program);
+  CheckFatal([&] { TrackResources(fixture.program); }, "not a valid runtime value",
+             "control-dependent descriptor phi was accepted");
+  Check(!fixture.program.resource_tracking_complete &&
             fixture.program.info.buffers.empty() &&
             fixture.program.descriptor_sources.empty(),
         "control-dependent descriptor phi was not rejected transactionally");
@@ -994,9 +968,7 @@ void TestLoopCycleEnteredThroughRuntimeValue() {
                {carried, Value(0u), Value(0u), Value(0u)}, MemoryFlags{0, 12},
                loop);
 
-  std::string error;
-  Check(BuildSrtPlan(fixture.program, &error),
-        "SRT planning rejected a valid loop entered through a runtime value");
+  BuildSrtPlan(fixture.program);
 }
 
 void TestInvariantLoopPhi() {
@@ -1023,10 +995,8 @@ void TestInvariantLoopPhi() {
   std::array<uint32_t, 1> user_data{0x12345678u};
   SrtRuntime runtime{.user_data = user_data};
   DescriptorValue descriptor;
-  std::string error;
   Check(EvaluateDescriptorSource(fixture.program,
-                                 fixture.program.info.buffers[0].source, 4,
-                                 runtime, descriptor, &error) &&
+                                 fixture.program.info.buffers[0].source, runtime, descriptor) &&
             descriptor.dwords[0] == user_data[0],
         "loop-invariant descriptor phi was not evaluated through typed SSA");
 }
@@ -1057,11 +1027,9 @@ void TestDmaAddressMaterialization() {
   std::array<uint32_t, 2> user_data{0x2008u, 0u};
   SrtRuntime runtime{.user_data = user_data};
   ResourceSnapshot snapshot;
-  std::string error;
-  Check(MaterializeResources(fixture.program, runtime, snapshot, &error),
+  Check(MaterializeResources(fixture.program, runtime, snapshot),
         "DMA shader resources did not materialize");
-  Check(SpecializeResources(fixture.program, snapshot, &error),
-        "DMA shader resources were not specialized");
+  SpecializeResources(fixture.program, snapshot);
 }
 
 void TestDynamicFlatAddressesUseDma() {
@@ -1089,8 +1057,7 @@ void TestDynamicFlatAddressesUseDma() {
   std::array<uint32_t, 3> user_data{0x23456780u, 1u, 1u};
   SrtRuntime runtime{.user_data = user_data};
   ResourceSnapshot snapshot;
-  std::string error;
-  Check(MaterializeResources(fixture.program, runtime, snapshot, &error),
+  Check(MaterializeResources(fixture.program, runtime, snapshot),
         "exec-masked FLAT shader resources did not materialize");
 
   Fixture mismatch;
@@ -1137,15 +1104,15 @@ void TestBufferSwizzleSpecialization() {
           (1u << 24u)};
   SrtRuntime runtime{.user_data = user_data};
   ResourceSnapshot snapshot;
-  std::string error;
-  Check(MaterializeResources(fixture.program, runtime, snapshot, &error) &&
-            SpecializeResources(fixture.program, snapshot, &error) &&
-            fixture.program.info.buffers[0].descriptor_swizzle == swizzle &&
-            ValidateResourceSpecialization(fixture.program, snapshot, &error),
+  Check(MaterializeResources(fixture.program, runtime, snapshot),
+        "buffer resources did not materialize");
+  SpecializeResources(fixture.program, snapshot);
+  Check(fixture.program.info.buffers[0].descriptor_swizzle == swizzle &&
+            ValidateResourceSpecialization(fixture.program, snapshot),
         "buffer destination selectors were not specialized");
 
   snapshot.buffers[0].dwords[3] ^= 1u << 9u;
-  Check(!ValidateResourceSpecialization(fixture.program, snapshot, &error),
+  Check(!ValidateResourceSpecialization(fixture.program, snapshot),
         "buffer swizzle change did not invalidate specialization");
 }
 
@@ -1171,17 +1138,14 @@ void TestShaderInfoAndBindingLayout() {
 
   ShaderComputeInputInfo compute{};
   compute.dispatch_thread_dimensions = true;
-  std::string error;
-  Check(CollectShaderInfo(fixture.program, {.compute = &compute}, &error),
-        "typed shader info collection failed");
+  CollectShaderInfo(fixture.program, {.compute = &compute});
   Check(fixture.program.info.has_bitwise_xor &&
             !fixture.program.info.inputs.empty() &&
             fixture.program.info.inputs[0].kind ==
                 StageInputKind::GlobalInvocationId,
         "typed shader values were not reflected in shader info");
 
-  Check(AllocateBindings(fixture.program, 0, &error),
-        "typed binding allocation failed");
+  AllocateBindings(fixture.program, 0);
   Check(FindBinding(fixture.program.bindings, DescriptorBindingKind::Buffers) !=
                 nullptr &&
             FindBinding(fixture.program.bindings, DescriptorBindingKind::Gds) !=
@@ -1211,53 +1175,57 @@ void TestGraphicsPushConstantLayout() {
     }
     fixture.program.shader_info_complete = true;
   };
-
-  std::string error;
   Fixture vertex(ShaderType::Vertex);
   AddUserData(vertex, 9);
-  Check(AllocateBindings(vertex.program, 0, &error) &&
-            vertex.program.bindings.push_constant_offset == 0 &&
+  AllocateBindings(vertex.program, 0);
+  Check(vertex.program.bindings.push_constant_offset == 0 &&
             vertex.program.bindings.push_constant_size == 9 * sizeof(uint32_t),
         "vertex shader was not placed at the start of the graphics push bank");
 
   const auto pixel_offset = vertex.program.bindings.push_constant_size;
   Fixture pixel(ShaderType::Pixel);
   AddUserData(pixel, 4);
-  Check(AllocateBindings(pixel.program, pixel_offset, &error) &&
-            pixel.program.bindings.push_constant_offset == pixel_offset &&
-            pixel.program.bindings.push_constant_size == 4 * sizeof(uint32_t) &&
-            FindBinding(pixel.program.bindings, DescriptorBindingKind::UserData) ==
-                nullptr,
-        "pixel shader did not follow the vertex data in the graphics push bank");
+  AllocateBindings(pixel.program, pixel_offset);
+  Check(
+      pixel.program.bindings.push_constant_offset == pixel_offset &&
+          pixel.program.bindings.push_constant_size == 4 * sizeof(uint32_t) &&
+          FindBinding(pixel.program.bindings,
+                      DescriptorBindingKind::UserData) == nullptr,
+      "pixel shader did not follow the vertex data in the graphics push bank");
 
   Fixture edge(ShaderType::Pixel);
   AddUserData(edge, 1);
-  Check(AllocateBindings(edge.program, NativePushConstantSize - sizeof(uint32_t),
-                         &error) &&
-            edge.program.bindings.push_constant_size == sizeof(uint32_t),
+  AllocateBindings(edge.program, NativePushConstantSize - sizeof(uint32_t));
+  Check(edge.program.bindings.push_constant_size == sizeof(uint32_t),
         "last aligned push-constant dword did not fit in the graphics bank");
 
   Fixture spill(ShaderType::Pixel);
   AddUserData(spill, 32);
-  Check(AllocateBindings(spill.program, pixel_offset, &error) &&
-            spill.program.bindings.push_constant_size == 0 &&
-            FindBinding(spill.program.bindings, DescriptorBindingKind::UserData) !=
-                nullptr,
-        "pixel shader overlapping the vertex push data did not spill to storage");
+  AllocateBindings(spill.program, pixel_offset);
+  Check(
+      spill.program.bindings.push_constant_size == 0 &&
+          FindBinding(spill.program.bindings,
+                      DescriptorBindingKind::UserData) != nullptr,
+      "pixel shader overlapping the vertex push data did not spill to storage");
 
   Fixture full(ShaderType::Pixel);
   AddUserData(full, 1);
-  Check(AllocateBindings(full.program, NativePushConstantSize, &error) &&
-            full.program.bindings.push_constant_size == 0 &&
-            FindBinding(full.program.bindings, DescriptorBindingKind::UserData) !=
-                nullptr,
+  AllocateBindings(full.program, NativePushConstantSize);
+  Check(full.program.bindings.push_constant_size == 0 &&
+            FindBinding(full.program.bindings,
+                        DescriptorBindingKind::UserData) != nullptr,
         "full graphics push bank did not spill pixel user data to storage");
 
   Fixture invalid(ShaderType::Pixel);
   AddUserData(invalid, 1);
-  Check(!AllocateBindings(invalid.program,
-                          NativePushConstantSize + sizeof(uint32_t), &error) &&
-            !invalid.program.binding_layout_complete,
+  CheckFatal(
+      [&] {
+        AllocateBindings(invalid.program,
+                         NativePushConstantSize + sizeof(uint32_t));
+      },
+      "push-constant offset",
+      "push-constant placement beyond the graphics bank was accepted");
+  Check(!invalid.program.binding_layout_complete,
         "push-constant placement beyond the graphics bank was accepted");
 }
 
@@ -1273,12 +1241,11 @@ void TestResourceLimitIsTransactional() {
                  {handle, Value(0u), Value(0u), Value(0u), Value(true)},
                  fixture.AddMemory(memory, index * 4u));
   }
-  std::string error;
-  Check(BuildSrtPlan(fixture.program, &error),
-        "SRT plan failed before resource-limit test");
-  Check(!TrackResources(fixture.program, &error) &&
-            error.find("buffer resource limit exceeded") != std::string::npos &&
-            !fixture.program.resource_tracking_complete &&
+  BuildSrtPlan(fixture.program);
+  CheckFatal([&] { TrackResources(fixture.program); },
+             "buffer resource limit exceeded",
+             "resource-limit failure was not reported");
+  Check(!fixture.program.resource_tracking_complete &&
             fixture.program.info.buffers.empty() &&
             fixture.program.descriptor_sources.empty(),
         "resource-limit failure partially mutated typed resource state");
@@ -1293,12 +1260,11 @@ void TestMalformedMemoryKindsRejected() {
     fixture.Emit(ValueOpcode::StoreAddressU32,
                  {address, Value(0u), Value(0u), Value(1u), Value(true)},
                  fixture.AddMemory(memory, 4));
-    std::string error;
-    Check(BuildSrtPlan(fixture.program, &error) &&
-              !TrackResources(fixture.program, &error) &&
-              error.find("address operation has invalid resource kind") !=
-                  std::string::npos,
-          "resource tracking accepted an address opcode with buffer metadata");
+    BuildSrtPlan(fixture.program);
+    CheckFatal(
+        [&] { TrackResources(fixture.program); },
+        "address operation has invalid resource kind",
+        "resource tracking accepted an address opcode with buffer metadata");
   }
   {
     Fixture fixture;
@@ -1311,12 +1277,11 @@ void TestMalformedMemoryKindsRejected() {
     fixture.Emit(ValueOpcode::ImageRead,
                  {image, fixture.ImageAddress(), Value(true)},
                  fixture.AddMemory(memory, 8));
-    std::string error;
-    Check(BuildSrtPlan(fixture.program, &error) &&
-              !TrackResources(fixture.program, &error) &&
-              error.find("image operation has invalid resource kind") !=
-                  std::string::npos,
-          "resource tracking accepted an image opcode with address metadata");
+    BuildSrtPlan(fixture.program);
+    CheckFatal(
+        [&] { TrackResources(fixture.program); },
+        "image operation has invalid resource kind",
+        "resource tracking accepted an image opcode with address metadata");
   }
   {
     Fixture fixture;
@@ -1329,12 +1294,11 @@ void TestMalformedMemoryKindsRejected() {
     fixture.Emit(ValueOpcode::ImageRead,
                  {image, fixture.ImageAddress(), Value(true)},
                  fixture.AddMemory(memory, 12));
-    std::string error;
-    Check(BuildSrtPlan(fixture.program, &error) &&
-              !TrackResources(fixture.program, &error) &&
-              error.find("image operation has invalid resource kind") !=
-                  std::string::npos,
-          "resource tracking accepted a sampled read with storage metadata");
+    BuildSrtPlan(fixture.program);
+    CheckFatal(
+        [&] { TrackResources(fixture.program); },
+        "image operation has invalid resource kind",
+        "resource tracking accepted a sampled read with storage metadata");
   }
   {
     Fixture fixture;
@@ -1347,12 +1311,10 @@ void TestMalformedMemoryKindsRejected() {
     fixture.Emit(ValueOpcode::ImageAtomicIAdd32,
                  {image, fixture.ImageAddress(), Value(1u), Value(true)},
                  fixture.AddMemory(memory, 16));
-    std::string error;
-    Check(
-        BuildSrtPlan(fixture.program, &error) &&
-            !TrackResources(fixture.program, &error) &&
-            error.find("image operation has invalid resource kind") !=
-                std::string::npos,
+    BuildSrtPlan(fixture.program);
+    CheckFatal(
+        [&] { TrackResources(fixture.program); },
+        "image operation has invalid resource kind",
         "resource tracking accepted a uint atomic with float storage metadata");
   }
 }
@@ -1399,14 +1361,24 @@ int main() {
 // target links only fmt; keep assertion failures observable without widening
 // its focused build manifest.
 namespace Common {
+int DbgExitHandler(const char *, int, std::string_view text) {
+  throw std::runtime_error(std::string(text));
+}
+
+int DbgExitHandler(const char *, int, fmt::text_style, std::string_view text) {
+  throw std::runtime_error(std::string(text));
+}
+
 int DbgExitIfHandler(const char *expression, const char *file, int line) {
   throw std::runtime_error(std::string("typed IR assertion: ") + expression +
                            " at " + file + ':' + std::to_string(line));
 }
 
-int DbgNotImplementedHandler(const char *expression, const char *file, int line) {
-  throw std::runtime_error(std::string("typed IR not implemented: ") + expression +
-                           " at " + file + ':' + std::to_string(line));
+int DbgNotImplementedHandler(const char *expression, const char *file,
+                             int line) {
+  throw std::runtime_error(std::string("typed IR not implemented: ") +
+                           expression + " at " + file + ':' +
+                           std::to_string(line));
 }
 
 void DbgExit(int) { throw std::runtime_error("typed IR assertion failed"); }
@@ -1415,8 +1387,8 @@ void DbgExit(int) { throw std::runtime_error("typed IR assertion failed"); }
 // Keep this focused standalone target self-contained by amalgamating its small
 // typed-IR implementation set.
 #include "graphics/shader/recompiler/ir/Block.cpp"
+#include "graphics/shader/recompiler/ir/Program.cpp"
 #include "graphics/shader/recompiler/ir/Type.cpp"
 #include "graphics/shader/recompiler/ir/Value.cpp"
-#include "graphics/shader/recompiler/ir/Program.cpp"
 #include "graphics/shader/recompiler/ir/opcodes/ValueOpcodes.cpp"
 #include "graphics/shader/recompiler/ir/passes/DeadCodeElimination.cpp"

tests/ScalarProvenanceTests.cpp

@@ -20,6 +20,19 @@ void Check(bool condition, const char *message) {
   }
 }
 
+template <typename F>
+void CheckFatal(F &&function, std::string_view expected, const char *message) {
+  try {
+    function();
+  } catch (const std::runtime_error &error) {
+    Check(std::string_view(error.what()).find(expected) !=
+              std::string_view::npos,
+          message);
+    return;
+  }
+  Check(false, message);
+}
+
 struct Fixture {
   Program program;
 
@@ -58,10 +71,7 @@ struct Fixture {
     return static_cast<uint32_t>(program.memory_info.size() - 1u);
   }
 
-  void Plan() {
-    std::string error;
-    Check(BuildSrtPlan(program, &error), error.c_str());
-  }
+  void Plan() { BuildSrtPlan(program); }
 };
 
 struct TestMemory {
@@ -112,8 +122,7 @@ void TestImmediateFlatteningAndGvn() {
   TestMemory memory_image{{{0x1020u, 0xfeedbeefu}}};
   SrtRuntime runtime{.read_memory = ReadMemory, .userdata = &memory_image};
   std::vector<uint32_t> flat;
-  std::string error;
-  Check(WalkSrt(fixture.program, runtime, flat, &error), error.c_str());
+  Check(WalkSrt(fixture.program, runtime, flat), "flattened SRT walk failed");
   Check(flat == std::vector<uint32_t>{0xfeedbeefu} && memory_image.reads == 1,
         "flattened SRT did not evaluate its canonical read once");
 }
@@ -130,8 +139,7 @@ void TestRawScalarComponentAlignment() {
   TestMemory memory_image{{{0x1000u, 0x12345678u}}};
   SrtRuntime runtime{.read_memory = ReadMemory, .userdata = &memory_image};
   std::vector<uint32_t> flat;
-  std::string error;
-  Check(WalkSrt(fixture.program, runtime, flat, &error), error.c_str());
+  Check(WalkSrt(fixture.program, runtime, flat), "raw scalar SRT walk failed");
   Check(
       flat == std::vector<uint32_t>{0x12345678u} && memory_image.reads == 1,
       "raw scalar base, immediate, and offset were not aligned independently");
@@ -141,11 +149,9 @@ void TestScalarMemoryDomainMismatchFails() {
   Fixture raw;
   const auto raw_memory = raw.AddMemory(ResourceKind::ScalarBuffer);
   RawRead(raw, Address(raw, Value(0x1000u), Value(0u)), Value(0u), raw_memory);
-  std::string error;
-  Check(!BuildSrtPlan(raw.program, &error) &&
-            error.find("incompatible scalar memory metadata") !=
-                std::string::npos,
-        "raw scalar load accepted descriptor-buffer metadata");
+  CheckFatal([&] { BuildSrtPlan(raw.program); },
+             "incompatible scalar memory metadata",
+             "raw scalar load accepted descriptor-buffer metadata");
 
   Fixture buffer;
   const auto buffer_memory = buffer.AddMemory(ResourceKind::ScalarAddress);
@@ -154,11 +160,9 @@ void TestScalarMemoryDomainMismatchFails() {
                   {Value(0x1000u), Value(0u), Value(16u), Value(0u)});
   buffer.EmitMemory(ValueOpcode::ReadConstBuffer, {resource, Value(0u)},
                     buffer_memory);
-  error.clear();
-  Check(!BuildSrtPlan(buffer.program, &error) &&
-            error.find("incompatible scalar memory metadata") !=
-                std::string::npos,
-        "descriptor scalar load accepted raw-address metadata");
+  CheckFatal([&] { BuildSrtPlan(buffer.program); },
+             "incompatible scalar memory metadata",
+             "descriptor scalar load accepted raw-address metadata");
 }
 
 void TestDynamicReadRemainsTyped() {
@@ -191,8 +195,7 @@ void TestNestedSrtWalk() {
   TestMemory memory_image{{{0x1000u, 0x2000u}, {0x2000u, 0xabcdef01u}}};
   SrtRuntime runtime{.read_memory = ReadMemory, .userdata = &memory_image};
   std::vector<uint32_t> flat;
-  std::string error;
-  Check(WalkSrt(fixture.program, runtime, flat, &error), error.c_str());
+  Check(WalkSrt(fixture.program, runtime, flat), "nested SRT walk failed");
   Check(flat == std::vector<uint32_t>({0x2000u, 0xabcdef01u}),
         "nested typed SRT reads were not evaluated in dependency order");
 }
@@ -215,10 +218,8 @@ void TestShaderBaseAndUserData() {
   SrtRuntime runtime{.user_data = user_data,
                      .shader_base = 0x12345678abcdef00ull};
   DescriptorValue result;
-  std::string error;
-  Check(EvaluateDescriptorSource(fixture.program, 0, 0x90, runtime, result,
-                                 &error),
-        error.c_str());
+  Check(EvaluateDescriptorSource(fixture.program, 0, runtime, result),
+        "shader-relative descriptor evaluation failed");
   Check(result.dword_count == 3 && result.dwords[0] == 0xabcdef00u &&
             result.dwords[1] == 0x12345678u && result.dwords[2] == 0x24u,
         "shader-relative typed descriptor expression evaluated incorrectly");
@@ -242,9 +243,8 @@ void TestCarryAndBitFields() {
       {.dwords = {low, high, inserted, sign}, .dword_count = 4});
 
   DescriptorValue result;
-  std::string error;
-  Check(EvaluateDescriptorSource(fixture.program, 0, 0, {}, result, &error),
-        error.c_str());
+  Check(EvaluateDescriptorSource(fixture.program, 0, {}, result),
+        "carry and bit-field descriptor evaluation failed");
   Check(result.dwords[0] == 1u && result.dwords[1] == 1u &&
             result.dwords[2] == 0x89abcdefu && result.dwords[3] == 0xffffffffu,
         "typed carry or bit-field runtime evaluation is incorrect");
@@ -267,17 +267,13 @@ void TestInvariantAndDivergentPhi() {
       {.dwords = {Value(&divergent)}, .dword_count = 1});
 
   DescriptorValue result;
-  std::string error;
-  Check(
-      EvaluateDescriptorSource(fixture.program, 0, 0x100, {}, result, &error) &&
-          result.dwords[0] == 7u,
-      "loop-invariant typed phi was rejected");
+  Check(EvaluateDescriptorSource(fixture.program, 0, {}, result) &&
+            result.dwords[0] == 7u,
+        "loop-invariant typed phi was rejected");
   result.dword_count = 4;
   result.dwords[0] = 0xdeadbeefu;
-  Check(!EvaluateDescriptorSource(fixture.program, 1, 0x104, {}, result,
-                                  &error) &&
-            result.dword_count == 4 && result.dwords[0] == 0xdeadbeefu &&
-            error.find("runtime-dependent") != std::string::npos,
+  Check(!EvaluateDescriptorSource(fixture.program, 1, {}, result) &&
+            result.dword_count == 4 && result.dwords[0] == 0xdeadbeefu,
         "divergent phi did not fail transactionally");
 }
 
@@ -318,8 +314,7 @@ void TestRuntime64BitDescriptorOps() {
       {.dwords = {low, high}, .dword_count = 2});
 
   DescriptorValue result;
-  std::string error;
-  Check(EvaluateDescriptorSource(fixture.program, 0, 0, {}, result, &error) &&
+  Check(EvaluateDescriptorSource(fixture.program, 0, {}, result) &&
             result.dwords[0] == 0xabcdu && result.dwords[1] == 0x1235u,
         "64-bit typed descriptor arithmetic evaluation is incorrect");
 }
@@ -339,10 +334,9 @@ void TestConstantBufferBounds() {
   TestMemory memory_image{{{0x300cu, 0xa5a5a5a5u}}};
   SrtRuntime runtime{.read_memory = ReadMemory, .userdata = &memory_image};
   std::vector<uint32_t> flat;
-  std::string error;
-  Check(WalkSrt(fixture.program, runtime, flat, &error) &&
+  Check(WalkSrt(fixture.program, runtime, flat) &&
             flat == std::vector<uint32_t>{0xa5a5a5a5u},
-        error.c_str());
+        "constant-buffer SRT walk failed");
 
   Fixture overflow;
   const auto overflow_memory = overflow.AddMemory(ResourceKind::ScalarBuffer);
@@ -356,9 +350,8 @@ void TestConstantBufferBounds() {
                 {overflow_read, Value(0u), Value(16u), Value(0u)});
   overflow.Plan();
   flat = {0x55u};
-  Check(!WalkSrt(overflow.program, runtime, flat, &error) &&
-            flat == std::vector<uint32_t>{0x55u} &&
-            error.find("exceeds size") != std::string::npos,
+  Check(!WalkSrt(overflow.program, runtime, flat) &&
+            flat == std::vector<uint32_t>{0x55u},
         "out-of-bounds constant-buffer walk was not transactional");
 }
 
@@ -420,10 +413,9 @@ void TestControlFlowValueSurvivesReadLaneFolding() {
       Libs::Graphics::ShaderRecompiler::CFG::TerminatorKind::Return;
 
   const auto undef = fixture.Emit(ValueOpcode::UndefU32);
-  const auto write = fixture.Emit(
-      ValueOpcode::WriteLane, {undef, Value(42u), Value(5u)});
-  const auto read =
-      fixture.Emit(ValueOpcode::ReadLane, {write, Value(5u)});
+  const auto write =
+      fixture.Emit(ValueOpcode::WriteLane, {undef, Value(42u), Value(5u)});
+  const auto read = fixture.Emit(ValueOpcode::ReadLane, {write, Value(5u)});
   entry_info.condition =
       fixture.Emit(ValueOpcode::IEqual32, {read, Value(42u)});
   fixture.Emit(ValueOpcode::Reference, {entry_info.condition});
@@ -435,11 +427,9 @@ void TestControlFlowValueSurvivesReadLaneFolding() {
   RemoveIdentities(fixture.program.blocks);
   EliminateDeadCode(fixture.program.blocks);
 
-  std::string error;
-  Check(entry_info.condition == Value(true) &&
-            ValidateProgram(fixture.program, true, &error),
-        error.empty() ? "folded branch condition did not survive identity removal"
-                      : error.c_str());
+  Check(entry_info.condition == Value(true),
+        "folded branch condition did not survive identity removal");
+  ValidateProgram(fixture.program, true);
 }
 
 void TestUndefinedRuntimeValueFails() {
@@ -450,11 +440,8 @@ void TestUndefinedRuntimeValueFails() {
       {.dwords = {undef}, .dword_count = 1});
   DescriptorValue result;
   result.dword_count = 3;
-  std::string error;
-  Check(!EvaluateDescriptorSource(fixture.program, 0, 0x200, {}, result,
-                                  &error) &&
-            result.dword_count == 3 &&
-            error.find("undefined typed runtime value") != std::string::npos,
+  Check(!EvaluateDescriptorSource(fixture.program, 0, {}, result) &&
+            result.dword_count == 3,
         "undefined typed descriptor source did not fail transactionally");
 }
 
@@ -464,6 +451,14 @@ namespace Common {
 
 int DbgExitIfHandler(const char *, const char *, int) { return 1; }
 
+int DbgExitHandler(const char *, int, std::string_view text) {
+  throw std::runtime_error(std::string(text));
+}
+
+int DbgExitHandler(const char *, int, fmt::text_style, std::string_view text) {
+  throw std::runtime_error(std::string(text));
+}
+
 void DbgExit(int) { std::abort(); }
 
 } // namespace Common
@@ -496,9 +491,9 @@ int main() {
 // Keep this focused standalone target self-contained by amalgamating its small
 // typed-IR implementation set.
 #include "../src/graphics/shader/recompiler/ir/Block.cpp"
+#include "../src/graphics/shader/recompiler/ir/Program.cpp"
 #include "../src/graphics/shader/recompiler/ir/Type.cpp"
 #include "../src/graphics/shader/recompiler/ir/Value.cpp"
-#include "../src/graphics/shader/recompiler/ir/Program.cpp"
 #include "../src/graphics/shader/recompiler/ir/opcodes/ValueOpcodes.cpp"
 #include "../src/graphics/shader/recompiler/ir/passes/ConstantPropagation.cpp"
 #include "../src/graphics/shader/recompiler/ir/passes/DeadCodeElimination.cpp"

tests/ShaderRecompilerComputeTests.cpp

@@ -85,6 +85,12 @@
 #include <string>
 #include <vector>
 
+#if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
+#include <cerrno>
+#include <sys/wait.h>
+#include <unistd.h>
+#endif
+
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #ifndef NOMINMAX
 #define NOMINMAX
@@ -553,6 +559,12 @@ constexpr u32 EncodeVop1Sdwa(u32 src0, u32 dst_sel = 6, u32 dst_u = 0,
          ((s0 & 0x1u) << 23u);
 }
 
+constexpr u32 EncodeVop1Dpp(u32 src0, u32 dpp_ctrl = 0, u32 row_mask = 0xf,
+                            u32 bank_mask = 0xf) {
+  return (src0 & 0xffu) | ((dpp_ctrl & 0x1ffu) << 8u) |
+         ((bank_mask & 0xfu) << 24u) | ((row_mask & 0xfu) << 28u);
+}
+
 constexpr u32 EncodeVop2(u32 opcode, u32 dst, u32 src0, u32 src1) {
   return ((opcode & 0x3fu) << 25u) | ((dst & 0xffu) << 17u) |
          ((src1 & 0xffu) << 9u) | (src0 & 0x1ffu);
@@ -901,6 +913,31 @@ void Require(const char *shader_name, const char *stage, bool value,
   }
 }
 
+#if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
+template <typename Function>
+void ExpectFatal(const char *name, Function function) {
+  const pid_t pid = ::fork();
+  Require(name, "fatal subprocess", pid >= 0,
+          "fork failed while starting fatal shader test");
+  if (pid == 0) {
+    function();
+    ::_exit(0);
+  }
+
+  int status = 0;
+  pid_t waited = -1;
+  do {
+    waited = ::waitpid(pid, &status, 0);
+  } while (waited < 0 && errno == EINTR);
+  Require(name, "fatal subprocess", waited == pid,
+          "waitpid failed while collecting fatal shader test");
+  Require(name, "fatal exit",
+          WIFEXITED(status) && WEXITSTATUS(status) == (321 & 0xff),
+          "fatal shader path did not terminate with status 321");
+  std::printf("[host]    %-32s ok\n", name);
+}
+#endif
+
 void CheckLeastRecentlyUsedCacheOrdering() {
   Common::LeastRecentlyUsedCache<uint32_t, uint64_t> cache;
   const auto first = cache.Insert(1, 1);
@@ -973,7 +1010,6 @@ struct TestCase {
   std::vector<std::pair<std::string, size_t>> decoded_counts;
   std::vector<std::pair<std::string, size_t>> ir_counts;
   u32 expected_storage_mip_descriptors = 0;
-  std::string expected_compile_error;
 };
 
 struct GraphicsCase {
@@ -1181,17 +1217,7 @@ CompiledShader CompileCase(const TestCase &test) {
     options.wave_size = test.compute_info.wave_size;
   }
 
-  ShaderRecompiler::CompileResult result;
-  std::string error;
-  if (!ShaderRecompiler::TryRecompile(test.code, options, result, &error)) {
-    if (!test.expected_compile_error.empty() &&
-        error.find(test.expected_compile_error) != std::string::npos) {
-      return {};
-    }
-    Fail(test.name, "decode/IR", error.c_str());
-  }
-  Require(test.name, "decode/IR", test.expected_compile_error.empty(),
-          "shader unexpectedly compiled despite the required rejection");
+  auto result = ShaderRecompiler::Recompile(test.code, options);
   for (const auto &[text, expected] : test.decoded_counts) {
     const auto actual = CountText(result.decoded_dump, text);
     Require(test.name, "decoded RDNA2", actual == expected,
@@ -1239,9 +1265,7 @@ CompiledShader CompileCase(const TestCase &test) {
     }
     result.program.bindings = {};
     result.program.binding_layout_complete = false;
-    if (!ShaderRecompiler::IR::AllocateBindings(result.program, 0, &error)) {
-      Fail(test.name, "binding layout", error.c_str());
-    }
+    ShaderRecompiler::IR::AllocateBindings(result.program, 0);
     const auto *shader_data = ShaderRecompiler::IR::FindBinding(
         result.program.bindings,
         ShaderRecompiler::IR::DescriptorBindingKind::UserData);
@@ -1250,13 +1274,8 @@ CompiledShader CompileCase(const TestCase &test) {
                 result.program.bindings.push_constant_size == 0 &&
                 shader_data != nullptr,
             "oversized shader data did not use its storage fallback");
-    std::vector<u32> storage_spirv;
-    if (!ShaderRecompiler::Spirv::EmitProgram(result.program, result.resources,
-                                              options.input_info, storage_spirv,
-                                              &error)) {
-      Fail(test.name, "SPIR-V emit", error.c_str());
-    }
-    result.spirv = std::move(storage_spirv);
+    result.spirv = ShaderRecompiler::Spirv::EmitProgram(
+        result.program, result.resources, options.input_info);
   }
   Require(test.name, "SPIR-V emit", !result.spirv.empty(),
           "recompiler returned empty SPIR-V");
@@ -1332,12 +1351,7 @@ CompiledShader CompileFragmentCase(const GraphicsCase &test) {
   options.input_info.pixel = &pixel_info;
   options.user_data = user_data.data();
 
-  ShaderRecompiler::CompileResult result;
-  std::string error;
-  if (!ShaderRecompiler::TryRecompile(test.fragment_code, options, result,
-                                      &error)) {
-    Fail(test.name, "decode/IR", error.c_str());
-  }
+  auto result = ShaderRecompiler::Recompile(test.fragment_code, options);
   Require(test.name, "SPIR-V emit", !result.spirv.empty(),
           "recompiler returned empty SPIR-V");
   ValidateSpirv(test.name, result.spirv);
@@ -7573,10 +7587,7 @@ public:
       const auto allocate_bindings =
           [&](ShaderRecompiler::IR::Program &program) {
             program.shader_info_complete = true;
-            std::string error;
-            Require(name, "binding layout",
-                    ShaderRecompiler::IR::AllocateBindings(program, 0, &error),
-                    error.c_str());
+            ShaderRecompiler::IR::AllocateBindings(program, 0);
           };
 
       constexpr auto stencil_format = Prospero::BufferFormat::k8UInt;
@@ -7610,11 +7621,7 @@ public:
       ShaderRecompiler::IR::DescriptorValue null_descriptor{};
       null_descriptor.dword_count = 8;
       null_snapshot.images.assign(3, null_descriptor);
-      std::string null_error;
-      Require(name, "null specialization",
-              ShaderRecompiler::IR::SpecializeResources(
-                  null_program, null_snapshot, &null_error),
-              null_error.c_str());
+      ShaderRecompiler::IR::SpecializeResources(null_program, null_snapshot);
       allocate_bindings(null_program);
       ShaderStageRuntime null_runtime{
           std::make_shared<const ShaderRecompiler::IR::Program>(
@@ -11651,10 +11658,6 @@ void CompareGraphicsWords(const GraphicsCase &test,
 
 void RunCase(VulkanHarness *vulkan, const TestCase &test) {
   auto compiled = CompileCase(test);
-  if (!test.expected_compile_error.empty()) {
-    std::printf("[compute] %-32s ok\n", test.name);
-    return;
-  }
   if (test.image_descriptor_swizzle != DstSel(4, 5, 6, 7)) {
     Require(test.name, "resource specialization",
             !compiled.program.info.images.empty() &&
@@ -11931,6 +11934,7 @@ CoverageClass ClassifyOpcode(ShaderOpcode opcode,
   case Opcode::V_CMPX_NEQ_F32:
   case Opcode::V_CMPX_NLT_F32:
   case Opcode::V_CMP_CLASS_F32:
+  case Opcode::V_CMPX_CLASS_F32:
   case Opcode::V_CMP_LT_F16:
   case Opcode::V_CMP_EQ_F16:
   case Opcode::V_CMP_LE_F16:
@@ -12733,6 +12737,160 @@ TestCase ScalarNotB64UpdatesScc() {
            O::S_ENDPGM}};
 }
 
+TestCase ScalarMovB64PreservesRawSpecialData() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendSMovLiteral(&code, 0, 0x09500000u);
+  AppendSMovLiteral(&code, 1, 0x81234567u);
+  code.push_back(EncodeSop1(0x04, 106, 0)); // s_mov_b64 vcc, s[0:1]
+  code.push_back(EncodeSop1(0x04, 2, 106)); // s_mov_b64 s[2:3], vcc
+  code.push_back(EncodeSop1(0x08, 4, 2));   // s_not_b64 s[4:5], s[2:3]
+  code.push_back(EncodeSop2(0x0a, 6, InlineU32(1), InlineU32(0)));
+
+  AppendStoreSgprPair(&code, 2, 0);
+  AppendStoreSgprPair(&code, 4, 2);
+  AppendStoreSgpr(&code, 6, 4);
+
+  AppendSMovLiteral(&code, 8, 0x09500001u);
+  AppendSMovLiteral(&code, 9, 0x81234567u);
+  code.push_back(EncodeSop1(0x04, 126, 8));  // s_mov_b64 exec, s[8:9]
+  code.push_back(EncodeSop1(0x04, 10, 126)); // s_mov_b64 s[10:11], exec
+  code.push_back(EncodeSop1(0x08, 12, 10));  // s_not_b64 s[12:13], s[10:11]
+  code.push_back(EncodeSop2(0x0a, 14, InlineU32(1), InlineU32(0)));
+
+  AppendStoreSgprPair(&code, 10, 5);
+  AppendStoreSgprPair(&code, 12, 7);
+  AppendStoreSgpr(&code, 14, 9);
+  AppendEnd(&code);
+
+  return {"ScalarMovB64PreservesRawSpecialData",
+          code,
+          {},
+          {0x09500000u, 0x81234567u, 0xf6afffffu, 0x7edcba98u, 1u,
+           0x09500001u, 0x81234567u, 0xf6affffeu, 0x7edcba98u, 1u},
+          {O::S_MOV_B32, O::S_MOV_B64, O::S_NOT_B64, O::S_CSELECT_B32,
+           O::V_MOV_B32, O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
+}
+
+TestCase Scalar64RawSpecialOps() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  const auto capture_scc = [&](u32 dst) {
+    code.push_back(EncodeSop2(0x0a, dst, InlineU32(1), InlineU32(0)));
+  };
+
+  AppendSMovLiteral(&code, 0, 0x09500001u);
+  AppendSMovLiteral(&code, 1, 0x81234567u);
+  AppendSMovLiteral(&code, 2, 0xff00ff01u);
+  AppendSMovLiteral(&code, 3, 0x0f0f0f0fu);
+  code.push_back(EncodeSop1(0x04, 126, 0));
+  code.push_back(EncodeSop2(0x0f, 106, 126, 2));
+  code.push_back(EncodeSop1(0x04, 4, 106));
+  capture_scc(6);
+  code.push_back(EncodeSop1(0x08, 106, 106));
+  code.push_back(EncodeSop1(0x04, 8, 106));
+  capture_scc(10);
+
+  AppendSMovLiteral(&code, 12, 0x80000011u);
+  AppendSMovLiteral(&code, 13, 0x80000001u);
+  code.push_back(EncodeSop1(0x04, 106, 12));
+  code.push_back(EncodeSop1(0x0a, 106, 106));
+  code.push_back(EncodeSop1(0x04, 14, 106));
+  capture_scc(16);
+
+  code.push_back(EncodeSop1(0x0a, 26, 193u));
+  capture_scc(28);
+
+  code.push_back(EncodeSMovB32(18, InlineU32(1)));
+  code.push_back(EncodeSMovB32(19, InlineU32(0)));
+  code.push_back(EncodeSopc(0x06, InlineU32(1), InlineU32(1)));
+  code.push_back(EncodeSop2(0x0b, 126, 0, 18));
+  code.push_back(EncodeSop1(0x04, 20, 126));
+  code.push_back(EncodeSop1(0x08, 106, 126));
+  code.push_back(EncodeSop1(0x04, 22, 106));
+  capture_scc(24);
+  code.push_back(EncodeSop1(0x04, 126, 18));
+
+  AppendStoreSgprPair(&code, 4, 0);
+  AppendStoreSgpr(&code, 6, 2);
+  AppendStoreSgprPair(&code, 8, 3);
+  AppendStoreSgpr(&code, 10, 5);
+  AppendStoreSgprPair(&code, 14, 6);
+  AppendStoreSgpr(&code, 16, 8);
+  AppendStoreSgprPair(&code, 26, 9);
+  AppendStoreSgpr(&code, 28, 11);
+  AppendStoreSgprPair(&code, 20, 12);
+  AppendStoreSgprPair(&code, 22, 14);
+  AppendStoreSgpr(&code, 24, 16);
+  AppendEnd(&code);
+
+  return {"Scalar64RawSpecialOps",
+          code,
+          {},
+          {0x09000001u, 0x01030507u, 1, 0xf6fffffeu, 0xfefcfaf8u, 1,
+           0xf00000ffu, 0xf000000fu, 1, 0xffffffffu, 0xffffffffu, 1,
+           0x09500001u, 0x81234567u, 0xf6affffeu, 0x7edcba98u, 1},
+          {O::S_MOV_B32, O::S_MOV_B64, O::S_AND_B64, O::S_NOT_B64,
+           O::S_CSELECT_B32, O::S_WQM_B64, O::S_CMP_EQ_U32, O::S_CSELECT_B64,
+           O::V_MOV_B32, O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
+}
+
+TestCase ScalarMovB32SpecialAliases() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  const auto capture_scc = [&](u32 dst) {
+    code.push_back(EncodeSop2(0x0a, dst, InlineU32(1), InlineU32(0)));
+  };
+
+  AppendSMovLiteral(&code, 0, 0x09500001u);
+  AppendSMovLiteral(&code, 1, 0x81234567u);
+  AppendSMovLiteral(&code, 2, 0xaaaaaaaau);
+  AppendSMovLiteral(&code, 3, 0xbbbbbbbbu);
+  code.push_back(EncodeSMovB32(4, InlineU32(1)));
+  code.push_back(EncodeSMovB32(5, InlineU32(0)));
+  code.push_back(EncodeSop1(0x04, 126, 0));
+  code.push_back(EncodeSop1(0x04, 106, 2));
+  code.push_back(EncodeSMovB32(107, 126));
+  code.push_back(EncodeSop1(0x04, 6, 106));
+  code.push_back(EncodeSop1(0x08, 8, 106));
+  capture_scc(10);
+
+  AppendSMovLiteral(&code, 12, 0x13579bdfu);
+  AppendSMovLiteral(&code, 13, 0x2468ace0u);
+  code.push_back(EncodeSop1(0x04, 106, 12));
+  code.push_back(EncodeSMovB32(126, 106));
+  code.push_back(EncodeSop1(0x04, 14, 126));
+  code.push_back(EncodeSop1(0x04, 126, 4));
+
+  code.push_back(EncodeVopc(0xc2, InlineU32(0), 0));
+  code.push_back(EncodeSMovB32(106, 106));
+  code.push_back(EncodeSop1(0x08, 16, 106));
+  capture_scc(18);
+  code.push_back(EncodeSop1(0x04, 20, 106));
+  code.push_back(EncodeSMovB32(20, 20));
+  code.push_back(EncodeSop1(0x08, 22, 20));
+  capture_scc(24);
+
+  AppendStoreSgprPair(&code, 6, 0);
+  AppendStoreSgprPair(&code, 8, 2);
+  AppendStoreSgpr(&code, 10, 4);
+  AppendStoreSgprPair(&code, 14, 5);
+  AppendStoreSgpr(&code, 18, 7);
+  AppendStoreSgpr(&code, 24, 8);
+  AppendEnd(&code);
+
+  return {"ScalarMovB32SpecialAliases",
+          code,
+          {},
+          {0xaaaaaaaau, 0x09500001u, 0x55555555u, 0xf6affffeu, 1, 0x13579bdfu,
+           0x81234567u, 0, 0},
+          {O::S_MOV_B32, O::S_MOV_B64, O::S_NOT_B64, O::S_CSELECT_B32,
+           O::V_CMP_EQ_U32, O::V_MOV_B32, O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
+}
+
 TestCase ScalarQuadmaskB64() {
   using O = ShaderOpcode;
 
@@ -12831,8 +12989,7 @@ TestCase ScalarSaveExecOps() {
   return {"ScalarSaveExecOps",
           code,
           {},
-          // EXEC and saved EXEC values are invocation-local Booleans.
-          {1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 0, 0},
+          {1, 0, 1, 0, 0xffffffffu, 0xffffffffu, 1, 1, 0, 1, 0, 0, 0},
           {O::S_MOV_B32, O::S_AND_SAVEEXEC_B64, O::S_ORN2_SAVEEXEC_B64,
            O::S_ANDN1_SAVEEXEC_B64, O::S_AND_SAVEEXEC_B32,
            O::S_ANDN1_SAVEEXEC_B32, O::S_MOV_B64, O::V_MOV_B32,
@@ -12859,7 +13016,7 @@ TestCase ScalarOrn2SaveexecUsesSourceOrNotExec() {
   return {"ScalarOrn2SaveexecUsesSourceOrNotExec",
           code,
           {},
-          {1, 0, 1, 0, 1},
+          {0x0000000cu, 0x80000000u, 0xfffffff3u, 0x7fffffffu, 1},
           {O::S_MOV_B32, O::S_ORN2_SAVEEXEC_B64, O::S_MOV_B64, O::V_MOV_B32,
            O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
 }
@@ -14249,6 +14406,53 @@ TestCase VectorLaneAndPackedOps() {
            O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
 }
 
+TestCase Vop2PkFmacF16AccumulatesPackedHalvesIndependently() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovLiteral(&code, 0, 0x50004c00u); // low=16.0h, high=32.0h
+  AppendVMovLiteral(&code, 1, 0x42003c00u); // low=1.0h, high=3.0h
+  AppendVMovLiteral(&code, 2, 0x48004400u); // low=4.0h, high=8.0h
+
+  code.push_back(EncodeVop2(0x3c, 0, Vgpr(1), 2));
+  AppendStoreVgpr(&code, 0, 0);
+  code.push_back(EncodeVop2(0x3c, 0, Vgpr(1), 2));
+  AppendStoreVgpr(&code, 0, 1);
+
+  AppendVMovLiteral(&code, 3, 0x50004c00u);
+  code.push_back(EncodeVop2(0x3c, 3, 242, 2)); // Inline 1.0h is {1.0h, 0.0h}.
+  AppendStoreVgpr(&code, 3, 2);
+  AppendEnd(&code);
+
+  return {"Vop2PkFmacF16AccumulatesPackedHalvesIndependently",
+          code,
+          {},
+          {0x53004d00u, 0x55004e00u, 0x50004d00u},
+          {O::V_MOV_B32, O::V_PK_FMAC_F16, O::BUFFER_STORE_DWORD,
+           O::S_ENDPGM}};
+}
+
+TestCase Vop2PkFmacF16DppNegatesBothPackedHalves() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovLiteral(&code, 0, 0x50004c00u); // low=16.0h, high=32.0h
+  AppendVMovLiteral(&code, 1, 0x42003c00u); // low=1.0h, high=3.0h
+  AppendVMovLiteral(&code, 2, 0x48004400u); // low=4.0h, high=8.0h
+
+  code.push_back(EncodeVop2(0x3c, 0, 250, 2));
+  code.push_back(EncodeVop2Dpp(1, 0, 0xf, 0xf, 1));
+  AppendStoreVgpr(&code, 0, 0);
+  AppendEnd(&code);
+
+  return {"Vop2PkFmacF16DppNegatesBothPackedHalves",
+          code,
+          {},
+          {0x48004a00u},
+          {O::V_MOV_B32, O::V_PK_FMAC_F16, O::BUFFER_STORE_DWORD,
+           O::S_ENDPGM}};
+}
+
 TestCase Vop3pOpselHiUsesArchitecturalSourceBits() {
   using O = ShaderOpcode;
 
@@ -14320,6 +14524,54 @@ TestCase CvtPkrtzF16F32SubnormalRoundsTowardZero() {
            O::S_ENDPGM}};
 }
 
+TestCase CvtPkrtzF16F32SdwaAndOutputModifiers() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovLiteral(&code, 0, 0x3f000000u); // 0.5f
+  AppendVMovLiteral(&code, 1, 0x40000000u); // 2.0f
+  AppendVMovLiteral(&code, 2, 0x3f802000u); // exactly 0x3c01h
+  AppendVMovLiteral(&code, 3, 0x00000000u);
+  AppendVMovLiteral(&code, 4, 0xbf000000u); // -0.5f
+  for (u32 dst = 10; dst <= 18; dst++) {
+    AppendVMovLiteral(&code, dst, 0xa1b2c3d4u);
+  }
+
+  code.push_back(EncodeVop2(0x2f, 10, 249, 1));
+  code.push_back(EncodeVop2Sdwa(0, 6, 2, 6, 6, 0, 0, 0, 0, 0, 0, 0, 0,
+                                1));
+  code.push_back(EncodeVop2(0x2f, 11, 249, 3));
+  code.push_back(EncodeVop2Sdwa(2, 0, 0));
+  code.push_back(EncodeVop2(0x2f, 12, 249, 1));
+  code.push_back(EncodeVop2Sdwa(0, 6, 0, 3, 5));
+  code.push_back(EncodeVop2(0x2f, 13, 249, 1));
+  code.push_back(EncodeVop2Sdwa(4, 6, 0, 6, 6, 0, 0, 0, 1, 1));
+
+  code.push_back(EncodeVop2(0x2f, 14, 250, 1));
+  code.push_back(EncodeVop2Dpp(0, 0, 0xf, 0xf, 1));
+  AppendVop3(&code, 0x12f, 15, Vgpr(0), Vgpr(1), 0, 0, 0, true);
+  code.push_back(EncodeVop2(0x2f, 16, 249, 1));
+  code.push_back(EncodeVop2Sdwa(0, 4, 2));
+  code.push_back(EncodeVop2(0x2f, 17, 249, 1));
+  code.push_back(EncodeVop2Sdwa(0, 6, 2, 6, 6, 0, 0, 0, 0, 0, 0, 0, 0,
+                                0, 1));
+  AppendVop3(&code, 0x12f, 18, Vgpr(0), Vgpr(1), 0, 0, 0, false, 1);
+
+  for (u32 dst = 10; dst <= 18; dst++) {
+    AppendStoreVgpr(&code, dst, dst - 10);
+  }
+  AppendEnd(&code);
+
+  return {"CvtPkrtzF16F32SdwaAndOutputModifiers",
+          code,
+          {},
+          {0x3c003800u, 0x00000001u, 0x00000000u, 0xc0003800u,
+           0x4000b800u, 0x3c003800u, 0xa1b23800u, 0x44003c00u,
+           0x44003c00u},
+          {O::V_MOV_B32, O::V_CVT_PKRTZ_F16_F32, O::BUFFER_STORE_DWORD,
+           O::S_ENDPGM}};
+}
+
 TestCase PackedMinMaxF16NanAndSignedZeroEdges() {
   using O = ShaderOpcode;
 
@@ -14427,6 +14679,99 @@ TestCase VectorCvtU16F16Sdwa() {
           {O::V_MOV_B32, O::V_CVT_U16_F16, O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
 }
 
+TestCase NativeAndSdwa16BitDestinationWrites() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovLiteral(&code, 0, 0x3fc00000u); // 1.5f -> 0x3e00h
+  AppendVMovLiteral(&code, 1, 0x00004200u); // low=3.0h
+  AppendVMovLiteral(&code, 2, 0x0000c000u); // low=-2.0h
+  AppendVMovLiteral(&code, 3, 0x00003c00u); // low=1.0h
+  AppendVMovLiteral(&code, 4, 0x00004000u); // low=2.0h
+
+  const u32 sentinels[] = {
+      0x3c001234u, 0xbbbb2345u, 0xcccc3456u, 0xdddd4567u, 0xeeee5678u,
+      0xffff6789u, 0xf00d789au, 0x11118888u, 0x22229999u, 0x3333aaaau,
+      0xaaaa0000u, 0xbbbb0000u, 0xcccc0000u, 0xabcd1234u, 0xbcde2345u,
+      0xcdef3456u, 0xdef01234u, 0xef012345u, 0xf0123456u, 0x01234567u,
+      0xaaaa1234u, 0xbbbb5678u, 0x24681234u, 0x13572468u, 0x89abcdefu,
+      0xfedcba98u, 0x0bad1234u,
+  };
+  // AppendStoreVgpr uses v31 as its address scratch register.
+  const auto output_vgpr = [](u32 index) { return 10u + index + (index >= 21u ? 1u : 0u); };
+  for (u32 i = 0; i < static_cast<u32>(std::size(sentinels)); i++) {
+    AppendVMovLiteral(&code, output_vgpr(i), sentinels[i]);
+  }
+
+  code.push_back(EncodeVop1(0x0a, 10, Vgpr(0)));
+  code.push_back(EncodeVop1(0x0a, 11, 250));
+  code.push_back(EncodeVop1Dpp(0));
+  AppendVop3(&code, 0x18a, 12, Vgpr(0), 0);
+  code.push_back(EncodeVop1(0x52, 13, Vgpr(1)));
+  code.push_back(EncodeVop1(0x53, 14, Vgpr(2)));
+
+  code.push_back(EncodeVop1(0x0a, 15, 249));
+  code.push_back(EncodeVop1Sdwa(0));
+  code.push_back(EncodeVop1(0x54, 16, 249));
+  code.push_back(EncodeVop1Sdwa(10, 4, 2, 5));
+  code.push_back(EncodeVop1(0x52, 17, 249));
+  code.push_back(EncodeVop1Sdwa(1, 6, 0, 4));
+  code.push_back(EncodeVop1(0x52, 18, 249));
+  code.push_back(EncodeVop1Sdwa(1, 4, 0, 4));
+  code.push_back(EncodeVop1(0x53, 19, 249));
+  code.push_back(EncodeVop1Sdwa(2, 4, 1, 4));
+
+  code.push_back(EncodeVop2(0x32, 20, 250, 4));
+  code.push_back(EncodeVop2Dpp(3));
+  AppendVop3(&code, 0x132, 21, Vgpr(3), Vgpr(4));
+  code.push_back(EncodeVop2(0x32, 22, 249, 4));
+  code.push_back(EncodeVop2Sdwa(3, 4, 0, 4, 4));
+
+  code.push_back(EncodeVop1(0x0a, 23, 249));
+  code.push_back(EncodeVop1Sdwa(0, 4, 2, 3));
+  code.push_back(EncodeVop1(0x0a, 24, 249));
+  code.push_back(EncodeVop1Sdwa(0, 4, 2, 5));
+  code.push_back(EncodeVop1(0x0a, 25, 249));
+  code.push_back(EncodeVop1Sdwa(0, 4, 2, 6, 0, 1, 1));
+
+  code.push_back(EncodeVop1(0x0a, 26, 249));
+  code.push_back(EncodeVop1Sdwa(0, 4, 2, 6, 0, 0, 0, 0, 0, 1));
+  code.push_back(EncodeVop1(0x0a, 27, 249));
+  code.push_back(EncodeVop1Sdwa(0, 4, 2, 6, 0, 0, 0, 0, 1));
+  AppendVop3(&code, 0x18a, 28, Vgpr(0), 0, 0, 0, 0, false, 1);
+  AppendVop3(&code, 0x18a, 29, Vgpr(0), 0, 0, 0, 0, true);
+  AppendVop3(&code, 0x34b, 30, Vgpr(3), Vgpr(4), Vgpr(3));
+  AppendVop3(&code, 0x34b, 32, Vgpr(3), Vgpr(4), Vgpr(3), 0, 0x8);
+  code.push_back(EncodeVop1(0x52, 33, 249));
+  code.push_back(EncodeVop1Sdwa(1, 4, 2, 4, 0, 1, 1, 0, 1));
+  AppendVop3(&code, 0x1d3, 34, Vgpr(1), 0, 0, 1, 0, true, 0, 1);
+  code.push_back(EncodeVop1(0x07, 35, 249));
+  code.push_back(EncodeVop1Sdwa(0, 6, 0, 6, 0, 1, 1, 0, 1));
+  AppendVop3(&code, 0x188, 36, Vgpr(0), 0, 0, 1, 0, true, 0, 1);
+  code.push_back(EncodeVop1(0x54, 37, 249));
+  code.push_back(EncodeVop1Sdwa(4, 4, 2, 4, 0, 1, 1));
+
+  for (u32 i = 0; i < static_cast<u32>(std::size(sentinels)); i++) {
+    AppendStoreVgpr(&code, output_vgpr(i), i);
+  }
+  AppendEnd(&code);
+
+  return {"NativeAndSdwa16BitDestinationWrites",
+          code,
+          {},
+          {0x3c003e00u, 0xbbbb3e00u, 0xcccc3e00u, 0xdddd0003u,
+           0xeeeefffeu, 0x00003e00u, 0xf00d3c00u, 0x00000003u,
+           0x00000003u, 0xfffffffeu, 0xaaaa4200u, 0xbbbb4200u,
+           0x00004200u, 0xabcd0000u, 0xbcde0000u, 0xcdefbe00u,
+           0xdef04200u, 0xef013c00u, 0xf0124200u, 0x01233c00u,
+           0xaaaa4200u, 0x42005678u, 0x24680000u, 0x1357fffdu,
+           0x00000000u, 0xffffffffu, 0x0badb800u},
+          {O::V_MOV_B32, O::V_CVT_F16_F32, O::V_CVT_U32_F32,
+           O::V_CVT_I32_F32, O::V_CVT_U16_F16, O::V_CVT_I16_F16,
+           O::V_RCP_F16, O::V_ADD_F16, O::V_FMA_F16, O::BUFFER_STORE_DWORD,
+           O::S_ENDPGM}};
+}
+
 TestCase VectorMinMaxMed3F16Ops() {
   using O = ShaderOpcode;
 
@@ -15785,6 +16130,38 @@ TestCase VectorCompareClassF32() {
            O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
 }
 
+TestCase VectorVopcSdwaCmpxClassF32CapturedExecMask() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovLiteral(&code, 13, 0x7fc00000u); // Quiet NaN.
+  AppendVMovU32(&code, 3, 7);
+  AppendVMovU32(&code, 31, 0);
+  code.push_back(EncodeSMovB32(106, InlineU32(2))); // qNaN class mask.
+  code.insert(code.end(), {0x7d30d4f9u, 0x8606000du});
+  AppendBufferStoreDword(&code, 3, 31);
+
+  code.push_back(EncodeSMovB32(126, InlineU32(1))); // Restore lane-zero EXEC.
+  AppendVMovLiteral(&code, 13, 0x3f800000u);        // Positive normal.
+  AppendVMovU32(&code, 3, 9);
+  AppendVMovU32(&code, 31, 4);
+  code.push_back(EncodeSMovB32(106, InlineU32(2))); // qNaN class mask.
+  code.insert(code.end(), {0x7d30d4f9u, 0x8606000du});
+  AppendBufferStoreDword(&code, 3, 31);
+  AppendEnd(&code);
+
+  TestCase test{
+      "VectorVopcSdwaCmpxClassF32CapturedExecMask",
+      code,
+      {0x11111111u, 0x22222222u},
+      {7u, 0x22222222u},
+      {O::S_MOV_B32, O::V_MOV_B32, O::V_CMPX_CLASS_F32,
+       O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
+  test.decoded_counts = {
+      {"V_CMPX_CLASS_F32 exec_lo, v13, vcc_lo", 2}};
+  return test;
+}
+
 TestCase VectorCompareF16Ops() {
   using O = ShaderOpcode;
 
@@ -16269,6 +16646,30 @@ TestCase ScalarMemoryLoadVariants() {
            O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
 }
 
+TestCase ScalarLoadUsesRawExecAddress() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  code.push_back(EncodeSMovB32(0, InlineU32(0)));
+  code.push_back(EncodeSMovB32(1, InlineU32(0)));
+  code.push_back(EncodeSMovB32(2, InlineU32(4)));
+  code.push_back(EncodeSMovB32(3, InlineU32(0)));
+  code.push_back(EncodeSop1(0x04, 126, 2));
+  code.push_back(EncodeSmem0(0x00, 20, 63));
+  code.push_back(EncodeSmem1(0));
+  code.push_back(EncodeSMovB32(2, InlineU32(1)));
+  code.push_back(EncodeSop1(0x04, 126, 2));
+  AppendStoreSgpr(&code, 20, 0);
+  AppendEnd(&code);
+
+  return {"ScalarLoadUsesRawExecAddress",
+          code,
+          {0x11111111u, 0x22222222u},
+          {0x22222222u, 0x22222222u},
+          {O::S_MOV_B32, O::S_MOV_B64, O::S_LOAD_DWORD, O::V_MOV_B32,
+           O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
+}
+
 TestCase ScalarLoadSignedImmediateOffsetAddsSoffset() {
   using O = ShaderOpcode;
 
@@ -18102,8 +18503,6 @@ TestCase FlatStoreVariants() {
                 {O::V_MOV_B32, O::FLAT_STORE_BYTE, O::FLAT_STORE_SHORT,
                  O::FLAT_STORE_DWORD, O::FLAT_STORE_DWORDX2,
                  O::FLAT_STORE_DWORDX3, O::FLAT_STORE_DWORDX4, O::S_ENDPGM}};
-  test.expected_compile_error =
-      "writable FLAT/GLOBAL addresses require GPU ownership tracking";
   return test;
 }
 
@@ -19910,9 +20309,7 @@ void CheckIndirectImageKeySwitch() {
   program.info.samplers.push_back({1u, 0x10f0u});
   program.info.sampled_pairs.push_back({0u, 0u, 0x10f0u});
 
-  std::string error;
-  Require(name, "binding layout", AllocateBindings(program, 0, &error),
-          error.c_str());
+  AllocateBindings(program, 0);
   ResourceSnapshot snapshot{};
   DescriptorValue descriptor{};
   descriptor.dword_count = 8;
@@ -19936,14 +20333,9 @@ void CheckIndirectImageKeySwitch() {
   snapshot.samplers.push_back(sampler_descriptor);
 
   ShaderComputeInputInfo compute{};
-  std::vector<u32> spirv;
-  Require(name, "SPIR-V requirements",
-          ShaderRecompiler::Spirv::AnalyzeProgramRequirements(program, &error),
-          error.c_str());
-  Require(name, "SPIR-V emit",
-          ShaderRecompiler::Spirv::EmitProgram(
-              program, snapshot, {.compute = &compute}, spirv, &error),
-          error.c_str());
+  ShaderRecompiler::Spirv::AnalyzeProgramRequirements(program);
+  auto spirv = ShaderRecompiler::Spirv::EmitProgram(program, snapshot,
+                                                    {.compute = &compute});
   ValidateSpirv(name, spirv);
   spvtools::SpirvTools tools(SPV_ENV_VULKAN_1_2);
   std::string text;
@@ -20721,6 +21113,9 @@ std::vector<TestCase> MakeCases() {
   AddCase(ScalarCompareOps);
   AddCase(ScalarShiftAddAndMaskOps);
   AddCase(ScalarNotB64UpdatesScc);
+  AddCase(ScalarMovB64PreservesRawSpecialData);
+  AddCase(Scalar64RawSpecialOps);
+  AddCase(ScalarMovB32SpecialAliases);
   AddCase(ScalarQuadmaskB64);
   AddCase(ScalarFlbitI32B64Gpu);
   AddCase(ScalarSaveExecOps);
@@ -20767,12 +21162,16 @@ std::vector<TestCase> MakeCases() {
   AddCase(VectorVop3BSubCoU32UsesRdna2Opcode310);
   AddCase(VectorMadU64U32UnsignedCarryOut);
   AddCase(VectorLaneAndPackedOps);
+  AddCase(Vop2PkFmacF16AccumulatesPackedHalvesIndependently);
+  AddCase(Vop2PkFmacF16DppNegatesBothPackedHalves);
   AddCase(Vop3pOpselHiUsesArchitecturalSourceBits);
   AddCase(CvtPkU8F32PacksSelectedByte);
   AddCase(CvtPkrtzF16F32SubnormalRoundsTowardZero);
+  AddCase(CvtPkrtzF16F32SdwaAndOutputModifiers);
   AddCase(PackedMinMaxF16NanAndSignedZeroEdges);
   AddCase(VectorMinMaxF16Ops);
   AddCase(VectorCvtU16F16Sdwa);
+  AddCase(NativeAndSdwa16BitDestinationWrites);
   AddCase(VectorMinMaxMed3F16Ops);
   AddCase(VectorSpecialF16Ops);
   AddCase(VectorCosF16CapturedSdwaAndEdges);
@@ -20812,6 +21211,7 @@ std::vector<TestCase> MakeCases() {
   AddCase(VectorVopcCmpxNeU64CapturedExecMask);
   AddCase(VectorVop3CmpxNeI64CapturedExecMask);
   AddCase(VectorCompareClassF32);
+  AddCase(VectorVopcSdwaCmpxClassF32CapturedExecMask);
   AddCase(VectorCompareF16Ops);
   AddCase(Vop2SdwaCndmaskSourceModifier);
   AddCase(Vop2SdwaCndmaskCapturedByte3Source);
@@ -20829,6 +21229,7 @@ std::vector<TestCase> MakeCases() {
   AddCase(BranchVccnzUsesInvocationMask);
   AddCase(BranchVccnzUsesCarryProducedInvocationMask);
   AddCase(ScalarMemoryLoadVariants);
+  AddCase(ScalarLoadUsesRawExecAddress);
   AddCase(ScalarLoadSignedImmediateOffsetAddsSoffset);
   AddCase(ScalarLoadAlignsComponentsAndMasksAddress);
   AddCase(BufferLoadStore);
@@ -20901,7 +21302,6 @@ std::vector<TestCase> MakeCases() {
   AddCase(GlobalSignedImmediateRebasesBeforeSaddr);
   AddCase(FlatSegmentIgnoresSaddrAndMasksOffsetMsb);
   AddCase(ScratchIsPrivatePerInvocation);
-  AddCase(FlatStoreVariants);
   AddCase(DsReadWriteVariants);
   AddCase(DsWriteB16D16HiCapturedUsesHighHalf);
   AddCase(DsReadU16D16CapturedPreservesHighHalf);
@@ -21020,7 +21420,7 @@ void CheckPs5GameExampleImageClearRuntimeShape() {
   compute.thread_ids_num = 1;
   compute.workgroup_register = 8;
 
-  const auto Compile = [&](const char *stage, const std::vector<u32> &code) {
+  const auto Compile = [&](const std::vector<u32> &code) {
     ShaderRecompiler::CompileOptions options;
     options.stage = ShaderType::Compute;
     options.wave_size = 64;
@@ -21029,17 +21429,13 @@ void CheckPs5GameExampleImageClearRuntimeShape() {
     options.user_data = user_data.data();
     options.input_info.compute = &compute;
     options.dump_ir = false;
-    ShaderRecompiler::CompileResult result;
-    std::string error;
-    Require("Ps5GameExampleImageClear", stage,
-            ShaderRecompiler::TryRecompile(code, options, result, &error),
-            error);
+    auto result = ShaderRecompiler::Recompile(code, options);
     ValidateSpirv("Ps5GameExampleImageClear", result.spirv);
     return result;
   };
 
   const auto code = MakeCode();
-  auto positive = Compile("exact Prospero kernel", code);
+  auto positive = Compile(code);
 
   compute.stage.program = std::make_shared<const ShaderRecompiler::IR::Program>(
       std::move(positive.program));
@@ -21130,11 +21526,7 @@ void CheckEmbeddedFetchVertexOffset() {
     options.user_data = user_data.data();
     options.input_info.vertex = &vertex;
 
-    ShaderRecompiler::CompileResult result;
-    std::string error;
-    Require(name, "compile",
-            ShaderRecompiler::TryRecompile(code, options, result, &error),
-            error);
+    auto result = ShaderRecompiler::Recompile(code, options);
     Require(name, "fetch rewrite",
             result.program.info.vertex_fetch_components[0] == 4,
             "encoded fetch sequence was not recognized and rewritten");
@@ -23064,12 +23456,8 @@ void CheckImageSamplerSpecialization() {
   ShaderRecompiler::IR::DescriptorValue sampler_descriptor{};
   sampler_descriptor.dword_count = 4;
   mixed_sampler_snapshot.samplers.push_back(sampler_descriptor);
-  std::string specialization_error;
-  Require("ImageSamplerSpecialization", "mixed sampler specialization",
-          ShaderRecompiler::IR::SpecializeResources(mixed_sampler_program,
-                                                    mixed_sampler_snapshot,
-                                                    &specialization_error),
-          specialization_error);
+  ShaderRecompiler::IR::SpecializeResources(mixed_sampler_program,
+                                            mixed_sampler_snapshot);
   Require("ImageSamplerSpecialization", "mixed sampler variant",
           mixed_sampler_program.info.samplers.size() == 2u &&
               !mixed_sampler_program.info.samplers[0].force_point_filtering &&
@@ -23080,25 +23468,45 @@ void CheckImageSamplerSpecialization() {
           "a shared native/bit-packed sampler was not split into point and "
           "native variants");
 
-  ShaderRecompiler::IR::Program packed_atomic_program;
-  packed_atomic_program.resource_tracking_complete = true;
-  auto packed_atomic_image = sampled_image;
-  packed_atomic_image.kind =
-      ShaderRecompiler::IR::ResourceKind::StorageImageUint;
-  packed_atomic_image.written = true;
-  packed_atomic_image.atomic = true;
-  packed_atomic_program.info.images.push_back(packed_atomic_image);
-  ShaderRecompiler::IR::ResourceSnapshot packed_atomic_snapshot;
-  packed_atomic_snapshot.images.push_back(packed_image_descriptor);
-  std::string atomic_error;
-  Require("ImageSamplerSpecialization", "packed atomic rejection",
-          !ShaderRecompiler::IR::SpecializeResources(
-              packed_atomic_program, packed_atomic_snapshot, &atomic_error) &&
-              atomic_error.find("unsupported format 34") != std::string::npos,
-          "a non-atomic PS5 packed format entered the typed atomic path");
   std::printf("[host]    %-32s ok\n", "ImageSpecializationPipelineId");
 }
 
+#if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
+void CheckShaderRecompilerFatalContracts() {
+  ExpectFatal("WritableFlatStoreRejection", [] {
+    const auto test = FlatStoreVariants();
+    (void)CompileCase(test);
+  });
+
+  ExpectFatal("PackedAtomicImageRejection", [] {
+    ShaderRecompiler::IR::Program program;
+    program.resource_tracking_complete = true;
+
+    ShaderRecompiler::IR::ImageResource image;
+    image.kind = ShaderRecompiler::IR::ResourceKind::StorageImageUint;
+    image.dimension = ShaderRecompiler::Decoder::ImageDimension::Dim2D;
+    image.read = true;
+    image.written = true;
+    image.atomic = true;
+    program.info.images.push_back(image);
+
+    ShaderRecompiler::IR::DescriptorValue descriptor{};
+    descriptor.dword_count = 8;
+    descriptor.dwords[0] = 0x2000u;
+    descriptor.dwords[1] =
+        static_cast<uint32_t>(Prospero::BufferFormat::k11_11_10UInt) << 20u;
+    descriptor.dwords[2] = 3u | (3u << 14u);
+    descriptor.dwords[3] =
+        DstSel(4, 5, 6, 7) |
+        (static_cast<uint32_t>(Prospero::ImageType::kColor2D) << 28u);
+
+    ShaderRecompiler::IR::ResourceSnapshot snapshot;
+    snapshot.images.push_back(descriptor);
+    ShaderRecompiler::IR::SpecializeResources(program, snapshot);
+  });
+}
+#endif
+
 void CheckStorageTextureVolumeUploadLayout() {
   constexpr auto format = Prospero::BufferFormat::k16_16_16_16Float;
   constexpr uint32_t width = 33;
@@ -23819,10 +24227,7 @@ void CheckEmbeddedFetchLaneSpill() {
   options.user_data = user_data.data();
   options.input_info.vertex = &vertex;
 
-  ShaderRecompiler::CompileResult result;
-  std::string error;
-  Require("EmbeddedFetchLaneSpill", "compile",
-          ShaderRecompiler::TryRecompile(code, options, result, &error), error);
+  auto result = ShaderRecompiler::Recompile(code, options);
   Require("EmbeddedFetchLaneSpill", "fetch rewrite",
           result.program.info.vertex_fetch_components[0] == 4,
           "lane-spilled fetch descriptor was not recognized and rewritten");
@@ -24628,6 +25033,12 @@ int main(int argc, char **argv) {
   std::setvbuf(stdout, nullptr, _IONBF, 0);
   EnsureConfigInitialized();
   CheckLeastRecentlyUsedCacheOrdering();
+#if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
+  if (argc == 2 && std::strcmp(argv[1], "--shader-fatal-only") == 0) {
+    CheckShaderRecompilerFatalContracts();
+    return 0;
+  }
+#endif
   if (argc == 2 && std::strcmp(argv[1], "--clip-control-only") == 0) {
     CheckClipControlDepthClipState();
     return 0;
@@ -24888,6 +25299,7 @@ int main(int argc, char **argv) {
     std::fprintf(stderr, "unknown test selector: %s\n", argv[1]);
     return 2;
   }
+  CheckShaderRecompilerFatalContracts();
   VulkanHarness vulkan;
 #endif
   CheckClipControlDepthClipState();

tests/ShaderStageRuntimeTests.cpp

@@ -52,7 +52,7 @@ SrtProgram(uint64_t address) {
       ValueOpcode::LoadAddressU32,
       {Value(&handle), Value(0u), Value(0u), Value(true)});
   raw.SetFlags(MemoryFlags{.index = 0, .pc = 0x40});
-  program.srt_reads.push_back({Value(&raw), 0, 0x40});
+  program.srt_reads.push_back({Value(&raw), 0});
 
   auto &srt = value_block.AppendNewInst(ValueOpcode::GetSrtResource);
   auto &flat = value_block.AppendNewInst(ValueOpcode::ReadConst,
@@ -77,17 +77,38 @@ UnbasedFlatProgram() {
   return std::make_shared<const Program>(std::move(program));
 }
 
+std::shared_ptr<const Libs::Graphics::ShaderRecompiler::IR::Program>
+UserDataBufferProgram() {
+  using namespace Libs::Graphics::ShaderRecompiler::IR;
+  Program program;
+  program.stage = Libs::Graphics::ShaderType::Compute;
+  program.srt_plan_complete = true;
+  program.resource_tracking_complete = true;
+  auto &value_block = AddValueBlock(program);
+
+  auto &user_data = value_block.AppendNewInst(
+      ValueOpcode::GetUserData, {Value(static_cast<ScalarReg>(0))});
+  DescriptorSource source;
+  source.dwords[0] = Value(&user_data);
+  source.dwords[1] = Value(0u);
+  source.dwords[2] = Value(0u);
+  source.dwords[3] = Value(0u);
+  source.dword_count = 4;
+  program.descriptor_sources.push_back(source);
+  program.info.buffers.push_back({.source = 0});
+  return std::make_shared<const Program>(std::move(program));
+}
+
 void TestMappedSrtUsesDirectReaderByDefault() {
   using namespace Libs::Graphics;
   const uint32_t dword = 0x12345678;
   auto cached_program = SrtProgram(reinterpret_cast<uint64_t>(&dword));
   ShaderStageRuntime stage;
-  std::string error;
   uint32_t specialization_reads = 0;
-  Check(ShaderMaterializeStageRuntime(cached_program, {}, 0, stage, &error,
+  Check(ShaderMaterializeStageRuntime(cached_program, {}, 0, stage,
                                       RejectSpecializationRead,
                                       &specialization_reads),
-        error.c_str());
+        "mapped SRT stage materialization failed");
   Check(specialization_reads == 0,
         "ordinary SRT read used the specialization reader");
   Check(stage.program == cached_program && stage.resources != nullptr,
@@ -104,16 +125,36 @@ void TestUnbasedFlatCacheHitMaterializes() {
   auto prior_resources =
       std::make_shared<const ShaderRecompiler::IR::ResourceSnapshot>();
   ShaderStageRuntime stage{prior_program, prior_resources};
-  std::string error;
-  Check(ShaderMaterializeStageRuntime(cached_program, {}, 0, stage, &error), error.c_str());
+  Check(ShaderMaterializeStageRuntime(cached_program, {}, 0, stage),
+        "unbased FLAT stage materialization failed");
   Check(stage.program == cached_program && stage.resources != prior_resources,
         "unbased FLAT cache hit did not publish its BDA stage snapshot");
 }
 
+void TestFailedMaterializationPreservesPriorStage() {
+  using namespace Libs::Graphics;
+  auto cached_program = UserDataBufferProgram();
+  auto prior_program = std::make_shared<const ShaderRecompiler::IR::Program>();
+  auto prior_resources =
+      std::make_shared<const ShaderRecompiler::IR::ResourceSnapshot>();
+  ShaderStageRuntime stage{prior_program, prior_resources};
+
+  Check(!ShaderMaterializeStageRuntime(cached_program, {}, 0, stage),
+        "missing runtime user data did not reject the cached stage");
+  Check(stage.program == prior_program && stage.resources == prior_resources,
+        "failed cache materialization replaced the prior stage");
+}
+
 } // namespace
 
 namespace Common {
 
+int DbgExitHandler(const char *, int, std::string_view) { std::abort(); }
+
+int DbgExitHandler(const char *, int, fmt::text_style, std::string_view) {
+  std::abort();
+}
+
 int DbgExitIfHandler(const char *, const char *, int) { return 1; }
 
 void DbgExit(int) { std::abort(); }
@@ -123,6 +164,7 @@ void DbgExit(int) { std::abort(); }
 int main() {
   TestMappedSrtUsesDirectReaderByDefault();
   TestUnbasedFlatCacheHitMaterializes();
+  TestFailedMaterializationPreservesPriorStage();
   std::puts("ShaderStageRuntimeTests: all cases passed");
   return 0;
 }
@@ -130,7 +172,7 @@ int main() {
 // Keep this focused standalone target self-contained by amalgamating its small
 // typed-IR implementation set.
 #include "graphics/shader/recompiler/ir/Block.cpp"
+#include "graphics/shader/recompiler/ir/Program.cpp"
 #include "graphics/shader/recompiler/ir/Type.cpp"
 #include "graphics/shader/recompiler/ir/Value.cpp"
-#include "graphics/shader/recompiler/ir/Program.cpp"
 #include "graphics/shader/recompiler/ir/opcodes/ValueOpcodes.cpp"