Support reverse RGBA16F render targets

src/graphics/guest_gpu/gpu_defs.h

@@ -104,6 +104,32 @@ enum class ChannelOrder : uint32_t {
 	kAltReversed = 3,
 };
 
+// Maps a shader's logical RGBA export to the host color-attachment components. Two bits per
+// component keep this cheap to carry in shader and pipeline keys.
+struct ColorComponentMapping {
+	uint8_t packed = 0xe4u; // RGBA: 0, 1, 2, 3
+
+	[[nodiscard]] constexpr uint32_t Map(uint32_t component) const {
+		return component < 4u ? (packed >> (component * 2u)) & 0x3u : component;
+	}
+
+	[[nodiscard]] constexpr uint32_t ApplyMask(uint32_t mask) const {
+		uint32_t mapped = 0;
+		for (uint32_t component = 0; component < 4u; component++) {
+			mapped |= ((mask >> component) & 1u) << Map(component);
+		}
+		return mapped;
+	}
+
+	[[nodiscard]] constexpr bool IsIdentity() const { return packed == 0xe4u; }
+
+	bool operator==(const ColorComponentMapping&) const = default;
+};
+
+static_assert(sizeof(ColorComponentMapping) == sizeof(uint8_t));
+
+inline constexpr ColorComponentMapping ColorMappingAbgr {0x1bu}; // 3, 2, 1, 0
+
 enum class DepthFormat : uint32_t {
 	kInvalid = 0,
 	kZ16     = 1,

src/graphics/host_gpu/objects/textureCommon.cpp

@@ -92,6 +92,9 @@ RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t raw_layout, uint32_
 	if (is(Prospero::ChannelLayout::k16_16_16_16, Prospero::ChannelType::kFloat, Prospero::ChannelOrder::kStandard)) {
 		return {VK_FORMAT_R16G16B16A16_SFLOAT, 8};
 	}
+	if (is(Prospero::ChannelLayout::k16_16_16_16, Prospero::ChannelType::kFloat, Prospero::ChannelOrder::kReversed)) {
+		return {VK_FORMAT_R16G16B16A16_SFLOAT, 8, Prospero::ColorMappingAbgr};
+	}
 	if (is(Prospero::ChannelLayout::k32, Prospero::ChannelType::kFloat, Prospero::ChannelOrder::kStandard)) {
 		return {VK_FORMAT_R32_SFLOAT, 4};
 	}

src/graphics/host_gpu/objects/textureCommon.h

@@ -3,6 +3,7 @@
 
 #include "common/abi.h"
 #include "common/common.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/utils.h"
 
@@ -47,8 +48,9 @@ enum class TextureUploadDestination { MipLevels, MipAtlas };
 enum class TextureUploadSliceLayout { MipChainPerSlice, MipLevelPerSlice };
 
 struct RenderTargetFormatInfo {
-	VkFormat format            = VK_FORMAT_UNDEFINED;
-	uint32_t bytes_per_element = 0;
+	VkFormat                        format            = VK_FORMAT_UNDEFINED;
+	uint32_t                        bytes_per_element = 0;
+	Prospero::ColorComponentMapping export_mapping;
 };
 
 struct TextureUploadLayout {

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -285,7 +285,8 @@ void ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer* buffer, const H
 		mask = 0x0f;
 	}
 
-	r->target_slot = rt_slot;
+	r->target_slot    = rt_slot;
+	r->export_mapping = {};
 
 	if (rt.base.addr == 0 || mask == 0) {
 		if (graphics_debug_dump_enabled()) {
@@ -479,6 +480,7 @@ void ResolveRenderColorTarget(uint64_t submit_id, CommandBuffer* buffer, const H
 		r->extent         = view_extent;
 		r->base_mip_level = rt.view.current_mip_level;
 		r->buffer_size    = backing_size;
+		r->export_mapping = rt_format.export_mapping;
 	} else {
 		const auto layout = static_cast<Prospero::ChannelLayout>(rt.info.format);
 		const auto type   = static_cast<Prospero::ChannelType>(rt.info.channel_type);

src/graphics/host_gpu/renderer/pipelineCache.cpp

@@ -82,10 +82,10 @@ PipelineCache::GraphicsPipeline* PipelineCache::CreateGraphicsPipeline(
 	const HW::BlendColor& bclr                                     = ctx->GetBlendColor();
 	uint32_t              color_mask[RENDER_COLOR_ATTACHMENTS_MAX] = {};
 	for (uint32_t i = 0; i < color_count; i++) {
-		color_mask[i] =
-		    (colors[i].vulkan_buffer != nullptr
-		         ? render_target_mask_slot(ctx->GetRenderTargetMask(), colors[i].target_slot)
-		         : 0);
+		color_mask[i] = (colors[i].vulkan_buffer != nullptr
+		                     ? colors[i].export_mapping.ApplyMask(render_target_mask_slot(
+		                           ctx->GetRenderTargetMask(), colors[i].target_slot))
+		                     : 0);
 	}
 	const HW::ModeControl& mc = ctx->GetModeControl();
 

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -748,6 +748,12 @@ static void RefreshShaders(HW::Context* ctx, HW::Shader* sh_ctx, const DrawCallI
 	state->vs_shader     = {};
 	state->ps_shader     = {};
 	state->ps_input_info = {};
+	std::array<Prospero::ColorComponentMapping, RENDER_COLOR_ATTACHMENTS_MAX>
+	    target_export_mapping {};
+	for (uint32_t i = 0; i < state->color_count; i++) {
+		target_export_mapping[state->color_info[i].target_slot] =
+		    state->color_info[i].export_mapping;
+	}
 	EXIT_IF(g_render_ctx == nullptr || g_render_ctx->GetGraphicCtx() == nullptr);
 	const auto lane_mask_mode = SelectGraphicsLaneMaskMode(*g_render_ctx->GetGraphicCtx(), 64u);
 
@@ -766,7 +772,8 @@ static void RefreshShaders(HW::Context* ctx, HW::Shader* sh_ctx, const DrawCallI
 		LogDrawPhase(draw.name, "ShaderCompileInfoPS");
 	}
 	if (!ShaderCompileInfoPS(&pixel_shader_info, &shader_regs, lane_mask_mode,
-	                         &state->vs_input_info, &state->ps_input_info, &state->ps_shader)) {
+	                         &state->vs_input_info, target_export_mapping, &state->ps_input_info,
+	                         &state->ps_shader)) {
 		EXIT("ShaderCompileInfoPS failed for draw %s\n", draw.name);
 	}
 }

src/graphics/host_gpu/renderer/renderState.h

@@ -4,6 +4,7 @@
 #include "common/abi.h"
 #include "common/assert.h"
 #include "common/common.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 
 #include <cstdint>
 #include <string>
@@ -241,6 +242,7 @@ struct RenderColorInfo {
 	uint64_t          base_addr          = 0;
 	uint64_t          buffer_size        = 0;
 	uint32_t          target_slot        = 0;
+	Prospero::ColorComponentMapping export_mapping;
 	bool              color_clear_enable = false;
 	VkClearColorValue color_clear_value {};
 };

src/graphics/shader/recompiler/spirvEmitter/spirvEmitterExport.cpp

@@ -50,6 +50,23 @@ uint32_t EmitExportVec4F32(EmitterState* state, const IR::Instruction& inst) {
 	return vec;
 }
 
+uint32_t ApplyMrtExportMapping(EmitterState* state, const IR::Instruction& inst, uint32_t value) {
+	if (inst.export_info.kind != IR::ExportTargetKind::Mrt || state->pixel_input_info == nullptr ||
+	    inst.export_info.index >= state->pixel_input_info->target_export_mapping.size()) {
+		return value;
+	}
+
+	const auto mapping = state->pixel_input_info->target_export_mapping[inst.export_info.index];
+	if (mapping.IsIdentity()) {
+		return value;
+	}
+
+	const auto mapped = state->builder.AllocateId();
+	state->builder.AddFunction({OpVectorShuffle, state->vec4_float_type, mapped, value, value,
+	                            mapping.Map(0), mapping.Map(1), mapping.Map(2), mapping.Map(3)});
+	return mapped;
+}
+
 bool ExportWritesData(const IR::Instruction& inst) {
 	switch (inst.export_info.kind) {
 		case IR::ExportTargetKind::Null:
@@ -97,7 +114,7 @@ void EmitExport(EmitterState* state, const IR::Instruction& inst) {
 		return;
 	}
 
-	const auto value = EmitExportVec4F32(state, inst);
+	const auto value = ApplyMrtExportMapping(state, inst, EmitExportVec4F32(state, inst));
 	if (inst.export_info.kind == IR::ExportTargetKind::Position) {
 		const auto pointer = state->builder.AllocateId();
 		state->builder.AddFunction({OpAccessChain, state->ptr_output_vec4_float, pointer, variable,

src/graphics/shader/recompiler/spirvEmitter/spirvEmitterInternal.h

@@ -135,6 +135,7 @@ enum : uint32_t {
 	OpArrayLength                  = 68,
 	OpDecorate                     = 71,
 	OpMemberDecorate               = 72,
+	OpVectorShuffle                = 79,
 	OpCompositeConstruct           = 80,
 	OpCompositeExtract             = 81,
 	OpSampledImage                 = 86,

src/graphics/shader/shader.cpp

@@ -791,10 +791,11 @@ static bool ShaderGetStaticInputInfoVS(const HW::VertexShaderInfo* regs,
 	return true;
 }
 
-static void ShaderGetStaticInputInfoPS(const HW::PixelShaderInfo*   regs,
-                                       const HW::ShaderRegisters*   sh,
-                                       const ShaderVertexInputInfo* vs_info,
-                                       ShaderPixelInputInfo*        ps_info) {
+static void ShaderGetStaticInputInfoPS(
+    const HW::PixelShaderInfo* regs, const HW::ShaderRegisters* sh,
+    const ShaderVertexInputInfo* vs_info,
+    std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
+    ShaderPixelInputInfo* ps_info) {
 	KYTY_PROFILER_FUNCTION();
 
 	EXIT_IF(vs_info == nullptr);
@@ -834,6 +835,9 @@ static void ShaderGetStaticInputInfoPS(const HW::PixelShaderInfo*   regs,
 
 	for (int i = 0; i < 8; i++) {
 		ps_info->target_output_mode[i] = sh->target_output_mode[i];
+		ps_info->target_export_mapping[i] =
+		    sh->target_output_mode[i] != 0 ? target_export_mapping[i]
+		                                   : Prospero::ColorComponentMapping {};
 	}
 	ps_info->mrt_output_mask = 0;
 }
@@ -1105,11 +1109,12 @@ bool ShaderCompileInfoVS(const HW::VertexShaderInfo* regs, const HW::ShaderRegis
 
 bool ShaderCompileInfoPS(const HW::PixelShaderInfo* regs, const HW::ShaderRegisters* sh,
                          ShaderLaneMaskMode lane_mask_mode, const ShaderVertexInputInfo* vs_info,
+                         std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
                          ShaderPixelInputInfo* ps_info, std::span<const uint32_t>* spirv) {
 	EXIT_IF(spirv == nullptr);
 	*spirv = {};
 
-	ShaderGetStaticInputInfoPS(regs, sh, vs_info, ps_info);
+	ShaderGetStaticInputInfoPS(regs, sh, vs_info, target_export_mapping, ps_info);
 	const auto shader_hash =
 	    regs->ps_regs.chksum != 0 ? regs->ps_regs.chksum : regs->ps_regs.data_addr;
 	const auto program_id = ShaderGetIdPS(regs, ps_info, false);
@@ -1632,6 +1637,14 @@ ShaderId ShaderGetIdPS(const HW::PixelShaderInfo* regs, const ShaderPixelInputIn
 	for (auto mode: input_info->target_output_mode) {
 		ret.ids.push_back(mode);
 	}
+	for (uint32_t base = 0; base < input_info->target_export_mapping.size(); base += 4u) {
+		uint32_t packed = 0;
+		for (uint32_t i = 0; i < 4u; i++) {
+			packed |= static_cast<uint32_t>(input_info->target_export_mapping[base + i].packed)
+			          << (i * 8u);
+		}
+		ret.ids.push_back(packed);
+	}
 	ret.ids.push_back(input_info->mrt_output_mask);
 
 	for (uint32_t i = 0; i < input_info->input_num; i++) {

src/graphics/shader/shader.h

@@ -3,8 +3,10 @@
 
 #include "common/abi.h"
 #include "common/common.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/shader/shaderBindings.h"
 
+#include <array>
 #include <memory>
 #include <span>
 #include <string>
@@ -98,6 +100,7 @@ struct ShaderPixelInputInfo {
 	uint32_t           input_num                    = 0;
 	uint32_t           ps_system_input_base         = 0;
 	uint8_t            target_output_mode[8]        = {};
+	std::array<Prospero::ColorComponentMapping, 8> target_export_mapping = {};
 	uint32_t           mrt_output_mask              = 0;
 	uint32_t           descriptor_set               = 0;
 	bool               ps_pos_x                     = false;
@@ -227,6 +230,7 @@ bool ShaderCompileInfoVS(const HW::VertexShaderInfo* regs, const HW::ShaderRegis
                          std::span<const uint32_t>* spirv);
 bool ShaderCompileInfoPS(const HW::PixelShaderInfo* regs, const HW::ShaderRegisters* sh,
                          ShaderLaneMaskMode lane_mask_mode, const ShaderVertexInputInfo* vs_info,
+                         std::span<const Prospero::ColorComponentMapping, 8> target_export_mapping,
                          ShaderPixelInputInfo* input_info, std::span<const uint32_t>* spirv);
 bool ShaderCompileInfoCS(const HW::ComputeShaderInfo* regs, const HW::ShaderRegisters* sh,
                          ShaderComputeInputInfo* input_info, std::span<const uint32_t>* spirv);

tests/ShaderRecompilerComputeTests.cpp

@@ -8676,6 +8676,51 @@ bool CacheFault(void *opaque, PageFaultAccess access, uint64_t vaddr,
          context->texture->InvalidateMemory(access, vaddr, size, phase);
 }
 
+[[noreturn]] void RunReverseRenderTargetDeathCase() {
+  (void)TextureGetRenderTargetFormat(12u, 7u, 3u);
+  std::_Exit(0x7f);
+}
+
+void CheckReverseRenderTargetFormatContract() {
+  const auto format = TextureGetRenderTargetFormat(12u, 7u, 2u);
+  Require("ReverseRenderTarget", "exact format",
+          format.format == VK_FORMAT_R16G16B16A16_SFLOAT &&
+              format.bytes_per_element == 8u &&
+              format.export_mapping == Prospero::ColorMappingAbgr,
+          "exact reverse RGBA16F render-target tuple was rejected");
+  Require("ReverseRenderTarget", "write masks",
+          format.export_mapping.ApplyMask(0x1u) == 0x8u &&
+              format.export_mapping.ApplyMask(0x2u) == 0x4u &&
+              format.export_mapping.ApplyMask(0x4u) == 0x2u &&
+              format.export_mapping.ApplyMask(0x8u) == 0x1u &&
+              format.export_mapping.ApplyMask(0xfu) == 0xfu,
+          "reverse RGBA16F component mask was not mapped exactly once");
+
+  char path[MAX_PATH]{};
+  Require("ReverseRenderTarget", "host",
+          GetModuleFileNameA(nullptr, path, MAX_PATH) != 0,
+          "GetModuleFileName failed");
+  std::string command = std::string("\"") + path + "\" --reverse-rt-death";
+  std::vector<char> mutable_command(command.begin(), command.end());
+  mutable_command.push_back('\0');
+  STARTUPINFOA startup{sizeof(startup)};
+  PROCESS_INFORMATION process{};
+  Require("ReverseRenderTarget", "host",
+          CreateProcessA(nullptr, mutable_command.data(), nullptr, nullptr, FALSE,
+                         CREATE_NO_WINDOW, nullptr, nullptr, &startup, &process) != 0,
+          "CreateProcess failed");
+  Require("ReverseRenderTarget", "host",
+          WaitForSingleObject(process.hProcess, 10000) == WAIT_OBJECT_0,
+          "unsupported adjacent render-target tuple timed out");
+  DWORD exit_code = 0;
+  const bool exited = GetExitCodeProcess(process.hProcess, &exit_code) != 0;
+  CloseHandle(process.hThread);
+  CloseHandle(process.hProcess);
+  Require("ReverseRenderTarget", "hard failure", exited && exit_code == 321,
+          "adjacent unproven render-target tuple did not retain the fatal guard");
+  std::printf("[host]    %-32s ok\n", "ReverseRenderTargetFormat");
+}
+
 [[noreturn]] void RunImageViewDeathCase(const char *kind) {
   if (std::strcmp(kind, "sampled") == 0) {
     (void)SelectSampledColorView(VK_FORMAT_R8G8B8A8_UNORM,
@@ -11729,6 +11774,13 @@ int main(int argc, char **argv) {
     return 0;
   }
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+  if (argc == 2 && std::strcmp(argv[1], "--reverse-rt-death") == 0) {
+    RunReverseRenderTargetDeathCase();
+  }
+  if (argc == 2 && std::strcmp(argv[1], "--reverse-rt-only") == 0) {
+    CheckReverseRenderTargetFormatContract();
+    return 0;
+  }
   if (argc == 2 && std::strcmp(argv[1], "--image-overlap-only") == 0) {
     CheckImageOverlapResolution();
     return 0;
@@ -11802,6 +11854,7 @@ int main(int argc, char **argv) {
   if (argc == 3 && std::strcmp(argv[1], "--metadata-descriptor-death") == 0) {
     RunMetadataDescriptorDeathCase(argv[2]);
   }
+  CheckReverseRenderTargetFormatContract();
   CheckSampledColorViews();
   CheckSampledDepthResource();
   CheckSampledDepthDescriptor();

tests/shaderCfgTests.cpp

@@ -4,6 +4,7 @@
 #include "common/threads.h"
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/guest_gpu/pm4.h"
+#include "graphics/host_gpu/objects/textureCommon.h"
 #include "graphics/host_gpu/renderer/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/shaderSubgroup.h"
 #include "graphics/shader/recompiler/ExecMask.h"
@@ -135,6 +136,24 @@ uint32_t SpirvInstructionOpcodeCount(const std::vector<uint32_t>& binary, uint32
 	return count;
 }
 
+bool SpirvContainsVectorShuffle(const std::vector<uint32_t>&   binary,
+                                const std::array<uint32_t, 4>& selectors) {
+	for (size_t i = 5; i < binary.size();) {
+		const uint32_t word       = binary[i];
+		const uint32_t op         = word & 0xffffu;
+		const uint32_t word_count = word >> 16u;
+		if (word_count == 0 || i + word_count > binary.size()) {
+			return false;
+		}
+		if (op == 79u && word_count == 9u &&
+		    std::equal(selectors.begin(), selectors.end(), binary.begin() + i + 5u)) {
+			return true;
+		}
+		i += word_count;
+	}
+	return false;
+}
+
 bool SpirvContainsCapability(const std::vector<uint32_t>& binary, uint32_t capability) {
 	for (size_t i = 5; i < binary.size();) {
 		const uint32_t word       = binary[i];
@@ -5696,6 +5715,82 @@ void TestNewShaderRecompilerExpPixelOutputs() {
 	CheckSpirvBinaryValidates(result.spirv);
 }
 
+void TestRenderTargetReverseFloat16ExportMapping() {
+	const auto format =
+	    TextureGetRenderTargetFormat(Prospero::GpuEnumValue(Prospero::ChannelLayout::k16_16_16_16),
+	                                 Prospero::GpuEnumValue(Prospero::ChannelType::kFloat),
+	                                 Prospero::GpuEnumValue(Prospero::ChannelOrder::kReversed));
+	Check(format.format == VK_FORMAT_R16G16B16A16_SFLOAT && format.bytes_per_element == 8u,
+	      "reverse RGBA16F render target did not retain its native host format "
+	      "and size");
+	Check(format.export_mapping == Prospero::ColorMappingAbgr &&
+	          format.export_mapping.ApplyMask(0x1u) == 0x8u &&
+	          format.export_mapping.ApplyMask(0x2u) == 0x4u &&
+	          format.export_mapping.ApplyMask(0x4u) == 0x2u &&
+	          format.export_mapping.ApplyMask(0x8u) == 0x1u &&
+	          format.export_mapping.ApplyMask(0xfu) == 0xfu,
+	      "reverse RGBA16F render-target export or write-mask mapping is "
+	      "incorrect");
+	const auto legacy_alt = TextureGetRenderTargetFormat(
+	    Prospero::GpuEnumValue(Prospero::ChannelLayout::k8_8_8_8),
+	    Prospero::GpuEnumValue(Prospero::ChannelType::kUNorm),
+	    Prospero::GpuEnumValue(Prospero::ChannelOrder::kAlt));
+	Check(legacy_alt.format == VK_FORMAT_B8G8R8A8_UNORM && legacy_alt.export_mapping.IsIdentity(),
+	      "legacy BGRA render target acquired a duplicate shader export mapping");
+
+	const uint32_t shader[] = {
+	    EncodeExp0(0x00, 0xf),
+	    EncodeExp1(0, 1, 2, 3), // MRT0
+	    0xbf810000u,
+	};
+	ShaderPixelInputInfo identity_info;
+	ShaderPixelInputInfo reversed_info;
+	reversed_info.target_export_mapping[0] = format.export_mapping;
+
+	ShaderRecompiler::CompileOptions options;
+	options.stage            = ShaderType::Pixel;
+	options.pixel_input_info = &identity_info;
+	ShaderRecompiler::CompileResult identity_result;
+	std::string                     error;
+	Check(ShaderRecompiler::TryRecompile(shader, options, &identity_result, &error), error.c_str());
+	Check(SpirvInstructionOpcodeCount(identity_result.spirv, 79u) == 0u,
+	      "identity MRT export unexpectedly added a component shuffle");
+
+	options.pixel_input_info = &reversed_info;
+	ShaderRecompiler::CompileResult reversed_result;
+	Check(ShaderRecompiler::TryRecompile(shader, options, &reversed_result, &error), error.c_str());
+	Check(SpirvContainsVectorShuffle(reversed_result.spirv, {3u, 2u, 1u, 0u}),
+	      "reverse MRT export did not emit a WZYX component shuffle");
+	CheckSpirvBinaryValidates(reversed_result.spirv);
+
+	HW::PixelShaderInfo regs {};
+	Check(ShaderGetIdPS(&regs, &identity_info, false) !=
+	          ShaderGetIdPS(&regs, &reversed_info, false),
+	      "pixel shader cache identity omitted the render-target export mapping");
+
+	regs.ps_regs.data_addr = reinterpret_cast<uint64_t>(shader);
+	regs.ps_regs.chksum    = 0xf16ab6f000000001ull;
+	ShaderMappedData mapped {};
+	mapped.code_size_bytes = sizeof(shader);
+	ShaderMapUserData(regs.ps_regs.data_addr, mapped);
+	HW::ShaderRegisters sh {};
+	ShaderVertexInputInfo vs_info {};
+	vs_info.stage.program = std::make_shared<ShaderRecompiler::IR::Program>();
+	std::array<Prospero::ColorComponentMapping, 8> mappings {};
+	mappings[0] = format.export_mapping;
+	ShaderPixelInputInfo      compiled_info {};
+	std::span<const uint32_t> compiled_spirv;
+	Check(ShaderCompileInfoPS(&regs, &sh, ShaderLaneMaskMode::NativeWave, &vs_info, mappings,
+	                          &compiled_info, &compiled_spirv) &&
+	          compiled_info.target_export_mapping[0].IsIdentity(),
+	      "inactive reverse MRT mapping was not normalized out of the shader cache key");
+	sh.target_output_mode[0] = 4;
+	Check(ShaderCompileInfoPS(&regs, &sh, ShaderLaneMaskMode::NativeWave, &vs_info, mappings,
+	                          &compiled_info, &compiled_spirv) &&
+	          compiled_info.target_export_mapping[0] == format.export_mapping,
+	      "active reverse MRT mapping was lost before shader specialization");
+}
+
 void TestNewShaderRecompilerEarlyZDisabledWhenPixelKillEnabled() {
 	constexpr uint32_t ExecutionModeEarlyFragmentTests = 9;
 
@@ -6510,10 +6605,11 @@ void TestPixelProgramCacheDescriptorSetIdentity() {
 			}
 			ShaderVertexInputInfo vs_info {};
 			vs_info.stage.program = std::move(vs_program);
+			const std::array<Prospero::ColorComponentMapping, 8> identity_mappings {};
 			ShaderPixelInputInfo      ps_info {};
 			std::span<const uint32_t> spirv;
 			Check(ShaderCompileInfoPS(&regs, &sh, ShaderLaneMaskMode::NativeWave, &vs_info,
-			                          &ps_info, &spirv),
+			                          identity_mappings, &ps_info, &spirv),
 			      "pixel program-cache transition failed to compile");
 			const auto expected_set = has_vs_descriptors ? 1u : 0u;
 			Check(ps_info.descriptor_set == expected_set && ps_info.stage.program != nullptr &&
@@ -6537,9 +6633,11 @@ void TestPixelProgramCacheDescriptorSetIdentity() {
 	auto compile_mask_mode = [&](ShaderLaneMaskMode mode) {
 		ShaderVertexInputInfo vs_info {};
 		vs_info.stage.program = std::make_shared<ShaderRecompiler::IR::Program>();
+		const std::array<Prospero::ColorComponentMapping, 8> identity_mappings {};
 		ShaderPixelInputInfo      ps_info {};
 		std::span<const uint32_t> spirv;
-		Check(ShaderCompileInfoPS(&mask_regs, &sh, mode, &vs_info, &ps_info, &spirv),
+		Check(ShaderCompileInfoPS(&mask_regs, &sh, mode, &vs_info, identity_mappings, &ps_info,
+		                          &spirv),
 		      "pixel lane-mask cache transition failed to compile");
 		Check(ps_info.stage.program != nullptr && ps_info.stage.program->lane_mask_mode == mode,
 		      "pixel program cache reused a different lane-mask lowering");
@@ -6715,6 +6813,7 @@ int main() {
 	TestNewShaderRecompilerVertexExportUsesLaneExecMask();
 	TestNewShaderRecompilerPerInvocationMasks();
 	TestNewShaderRecompilerExpPixelOutputs();
+	TestRenderTargetReverseFloat16ExportMapping();
 	TestNewShaderRecompilerEarlyZDisabledWhenPixelKillEnabled();
 	TestScalarProvenanceRealWideMoveLowering();
 	TestScalarProvenanceRealCarryAndScalarLoads();