diff --git a/src/Cafe/HW/Latte/Core/LatteShader.cpp b/src/Cafe/HW/Latte/Core/LatteShader.cpp index 4ac6c1fe..ccb4c282 100644 --- a/src/Cafe/HW/Latte/Core/LatteShader.cpp +++ b/src/Cafe/HW/Latte/Core/LatteShader.cpp @@ -1,20 +1,25 @@ #include "Cafe/HW/Latte/Core/LatteConst.h" #include "Cafe/HW/Latte/Core/LatteShaderAssembly.h" #include "Cafe/HW/Latte/ISA/RegDefines.h" -#include "Cafe/OS/libs/gx2/GX2.h" // todo - remove dependency #include "Cafe/HW/Latte/ISA/LatteReg.h" #include "Cafe/HW/Latte/Core/LatteShader.h" #include "Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.h" #include "Cafe/HW/Latte/Core/FetchShader.h" #include "Cafe/HW/Latte/Core/LattePerformanceMonitor.h" +#include "Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h" +#include "Cafe/OS/libs/gx2/GX2.h" // todo - remove dependency #include "Cafe/GraphicPack/GraphicPack2.h" #include "util/helpers/StringParser.h" #include "config/ActiveSettings.h" -#include "util/Zir/EmitterGLSL/ZpIREmitGLSL.h" -#include "util/Zir/Core/ZpIRDebug.h" #include "util/containers/flat_hash_map.hpp" #include +// experimental new decompiler (WIP) +#include "util/Zir/EmitterGLSL/ZpIREmitGLSL.h" +#include "util/Zir/Core/ZpIRDebug.h" +#include "Cafe/HW/Latte/Transcompiler/LatteTC.h" +#include "Cafe/HW/Latte/ShaderInfo/ShaderInfo.h" + struct _ShaderHashCache { uint64 prevHash1; @@ -672,10 +677,18 @@ LatteDecompilerShader* LatteShader_CreateShaderFromDecompilerOutput(LatteDecompi return shader; } -#include "Cafe/HW/Latte/Transcompiler/LatteTC.h" -#include "Cafe/HW/Latte/ShaderInfo/ShaderInfo.h" +void LatteShader_GetDecompilerOptions(LatteDecompilerOptions& options, LatteConst::ShaderType shaderType, bool geometryShaderEnabled) +{ + options.usesGeometryShader = geometryShaderEnabled; + options.spirvInstrinsics.hasRoundingModeRTEFloat32 = false; + if (g_renderer->GetType() == RendererAPI::Vulkan) + { + options.useTFViaSSBO = VulkanRenderer::GetInstance()->UseTFViaSSBO(); + options.spirvInstrinsics.hasRoundingModeRTEFloat32 = VulkanRenderer::GetInstance()->HasSPRIVRoundingModeRTE32(); + } +} -LatteDecompilerShader* LatteShader_compileSeparableVertexShader(uint64 baseHash, uint64& vsAuxHash, uint8* vertexShaderPtr, uint32 vertexShaderSize, bool usesGeometryShader, LatteFetchShader* fetchShader) +LatteDecompilerShader* LatteShader_CompileSeparableVertexShader2(uint64 baseHash, uint64& vsAuxHash, uint8* vertexShaderPtr, uint32 vertexShaderSize, bool usesGeometryShader, LatteFetchShader* fetchShader) { /* Analyze shader to gather general information about inputs/outputs */ Latte::ShaderDescription shaderDescription; @@ -725,14 +738,17 @@ LatteDecompilerShader* LatteShader_compileSeparableVertexShader(uint64 baseHash, // compile new vertex shader (relies partially on current state) LatteDecompilerShader* LatteShader_CompileSeparableVertexShader(uint64 baseHash, uint64& vsAuxHash, uint8* vertexShaderPtr, uint32 vertexShaderSize, bool usesGeometryShader, LatteFetchShader* fetchShader) { - // new decompiler - //LatteShader_compileSeparableVertexShader(baseHash, vsAuxHash, vertexShaderPtr, vertexShaderSize, usesGeometryShader, fetchShader); + // new decompiler test + //LatteShader_CompileSeparableVertexShader2(baseHash, vsAuxHash, vertexShaderPtr, vertexShaderSize, usesGeometryShader, fetchShader); // legacy decompiler + LatteDecompilerOptions options; + LatteShader_GetDecompilerOptions(options, LatteConst::ShaderType::Vertex, usesGeometryShader); + LatteDecompilerOutput_t decompilerOutput{}; LatteFetchShader* fetchShaderList[1]; fetchShaderList[0] = fetchShader; - LatteDecompiler_DecompileVertexShader(_shaderBaseHash_vs, LatteGPUState.contextRegister, vertexShaderPtr, vertexShaderSize, fetchShaderList, 1, LatteGPUState.contextNew.GetSpecialStateValues(), usesGeometryShader, &decompilerOutput); + LatteDecompiler_DecompileVertexShader(_shaderBaseHash_vs, LatteGPUState.contextRegister, vertexShaderPtr, vertexShaderSize, fetchShaderList, 1, LatteGPUState.contextNew.GetSpecialStateValues(), options, &decompilerOutput); LatteDecompilerShader* vertexShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, baseHash, true, 0, LatteGPUState.contextRegister); vsAuxHash = vertexShader->auxHash; if (vertexShader->hasError == false) @@ -759,10 +775,13 @@ LatteDecompilerShader* LatteShader_CompileSeparableVertexShader(uint64 baseHash, LatteDecompilerShader* LatteShader_CompileSeparableGeometryShader(uint64 baseHash, uint8* geometryShaderPtr, uint32 geometryShaderSize, uint8* geometryCopyShader, uint32 geometryCopyShaderSize) { + LatteDecompilerOptions options; + LatteShader_GetDecompilerOptions(options, LatteConst::ShaderType::Geometry, true); + LatteDecompilerOutput_t decompilerOutput{}; LatteFetchShader* fetchShaderList[1]; fetchShaderList[0] = _activeFetchShader; - LatteDecompiler_DecompileGeometryShader(_shaderBaseHash_gs, LatteGPUState.contextRegister, geometryShaderPtr, geometryShaderSize, geometryCopyShader, geometryCopyShaderSize, LatteGPUState.contextNew.GetSpecialStateValues(), _activeVertexShader->ringParameterCount, &decompilerOutput); + LatteDecompiler_DecompileGeometryShader(_shaderBaseHash_gs, LatteGPUState.contextRegister, geometryShaderPtr, geometryShaderSize, geometryCopyShader, geometryCopyShaderSize, LatteGPUState.contextNew.GetSpecialStateValues(), _activeVertexShader->ringParameterCount, options, &decompilerOutput); LatteDecompilerShader* geometryShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, baseHash, true, 0, LatteGPUState.contextRegister); if (geometryShader->hasError == false) { @@ -787,8 +806,11 @@ LatteDecompilerShader* LatteShader_CompileSeparableGeometryShader(uint64 baseHas LatteDecompilerShader* LatteShader_CompileSeparablePixelShader(uint64 baseHash, uint64& psAuxHash, uint8* pixelShaderPtr, uint32 pixelShaderSize, bool usesGeometryShader) { + LatteDecompilerOptions options; + LatteShader_GetDecompilerOptions(options, LatteConst::ShaderType::Pixel, usesGeometryShader); + LatteDecompilerOutput_t decompilerOutput{}; - LatteDecompiler_DecompilePixelShader(baseHash, LatteGPUState.contextRegister, pixelShaderPtr, pixelShaderSize, LatteGPUState.contextNew.GetSpecialStateValues(), usesGeometryShader, &decompilerOutput); + LatteDecompiler_DecompilePixelShader(baseHash, LatteGPUState.contextRegister, pixelShaderPtr, pixelShaderSize, LatteGPUState.contextNew.GetSpecialStateValues(), options, &decompilerOutput); LatteDecompilerShader* pixelShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, baseHash, true, 0, LatteGPUState.contextRegister); psAuxHash = pixelShader->auxHash; LatteShader_DumpShader(_shaderBaseHash_ps, psAuxHash, pixelShader); diff --git a/src/Cafe/HW/Latte/Core/LatteShader.h b/src/Cafe/HW/Latte/Core/LatteShader.h index 0fba0322..eb623d85 100644 --- a/src/Cafe/HW/Latte/Core/LatteShader.h +++ b/src/Cafe/HW/Latte/Core/LatteShader.h @@ -94,6 +94,7 @@ extern uint64 _shaderBaseHash_vs; extern uint64 _shaderBaseHash_gs; extern uint64 _shaderBaseHash_ps; +void LatteShader_GetDecompilerOptions(struct LatteDecompilerOptions& options, LatteConst::ShaderType shaderType, bool geometryShaderEnabled); LatteDecompilerShader* LatteShader_CreateShaderFromDecompilerOutput(LatteDecompilerOutput_t& decompilerOutput, uint64 baseHash, bool calculateAuxHash, uint64 optionalAuxHash, uint32* contextRegister); void LatteShader_CreateRendererShader(LatteDecompilerShader* shader, bool compileAsync); diff --git a/src/Cafe/HW/Latte/Core/LatteShaderCache.cpp b/src/Cafe/HW/Latte/Core/LatteShaderCache.cpp index 07a0d461..9a6c3f84 100644 --- a/src/Cafe/HW/Latte/Core/LatteShaderCache.cpp +++ b/src/Cafe/HW/Latte/Core/LatteShaderCache.cpp @@ -641,16 +641,19 @@ bool LatteShaderCache_readSeparableVertexShader(MemStreamReader& streamReader, u return false; if (streamReader.hasError() || !streamReader.isEndOfStream()) return false; - // update PS inputs (influence VS shader outputs) + // update PS inputs (affects VS shader outputs) LatteShader_UpdatePSInputs(lcr->GetRawView()); // get fetch shader LatteFetchShader::CacheHash fsHash = LatteFetchShader::CalculateCacheHash((uint32*)fetchShaderData.data(), fetchShaderData.size()); LatteFetchShader* fetchShader = LatteShaderRecompiler_createFetchShader(fsHash, lcr->GetRawView(), (uint32*)fetchShaderData.data(), fetchShaderData.size()); + // determine decompiler options + LatteDecompilerOptions options; + LatteShader_GetDecompilerOptions(options, LatteConst::ShaderType::Vertex, usesGeometryShader); // decompile vertex shader LatteDecompilerOutput_t decompilerOutput{}; LatteFetchShader* fetchShaderList[1]; fetchShaderList[0] = fetchShader; - LatteDecompiler_DecompileVertexShader(shaderBaseHash, lcr->GetRawView(), vertexShaderData.data(), vertexShaderData.size(), fetchShaderList, 1, lcr->GetSpecialStateValues(), usesGeometryShader, &decompilerOutput); + LatteDecompiler_DecompileVertexShader(shaderBaseHash, lcr->GetRawView(), vertexShaderData.data(), vertexShaderData.size(), fetchShaderList, 1, lcr->GetSpecialStateValues(), options, &decompilerOutput); LatteDecompilerShader* vertexShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, shaderBaseHash, false, shaderAuxHash, lcr->GetRawView()); // compile LatteShader_DumpShader(shaderBaseHash, shaderAuxHash, vertexShader); @@ -688,15 +691,17 @@ bool LatteShaderCache_readSeparableGeometryShader(MemStreamReader& streamReader, return false; // update PS inputs LatteShader_UpdatePSInputs(lcr->GetRawView()); + // determine decompiler options + LatteDecompilerOptions options; + LatteShader_GetDecompilerOptions(options, LatteConst::ShaderType::Geometry, true); // decompile geometry shader LatteDecompilerOutput_t decompilerOutput{}; - LatteDecompiler_DecompileGeometryShader(shaderBaseHash, lcr->GetRawView(), geometryShaderData.data(), geometryShaderData.size(), geometryCopyShaderData.data(), geometryCopyShaderData.size(), lcr->GetSpecialStateValues(), vsRingParameterCount, &decompilerOutput); + LatteDecompiler_DecompileGeometryShader(shaderBaseHash, lcr->GetRawView(), geometryShaderData.data(), geometryShaderData.size(), geometryCopyShaderData.data(), geometryCopyShaderData.size(), lcr->GetSpecialStateValues(), vsRingParameterCount, options, &decompilerOutput); LatteDecompilerShader* geometryShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, shaderBaseHash, false, shaderAuxHash, lcr->GetRawView()); // compile LatteShader_DumpShader(shaderBaseHash, shaderAuxHash, geometryShader); LatteShader_DumpRawShader(shaderBaseHash, shaderAuxHash, SHADER_DUMP_TYPE_GEOMETRY, geometryShaderData.data(), geometryShaderData.size()); LatteShaderCache_loadOrCompileSeparableShader(geometryShader, shaderBaseHash, shaderAuxHash); - catchOpenGLError(); LatteSHRC_RegisterShader(geometryShader, shaderBaseHash, shaderAuxHash); return true; } @@ -724,9 +729,12 @@ bool LatteShaderCache_readSeparablePixelShader(MemStreamReader& streamReader, ui return false; // update PS inputs LatteShader_UpdatePSInputs(lcr->GetRawView()); + // determine decompiler options + LatteDecompilerOptions options; + LatteShader_GetDecompilerOptions(options, LatteConst::ShaderType::Pixel, usesGeometryShader); // decompile pixel shader LatteDecompilerOutput_t decompilerOutput{}; - LatteDecompiler_DecompilePixelShader(shaderBaseHash, lcr->GetRawView(), pixelShaderData.data(), pixelShaderData.size(), lcr->GetSpecialStateValues(), usesGeometryShader, &decompilerOutput); + LatteDecompiler_DecompilePixelShader(shaderBaseHash, lcr->GetRawView(), pixelShaderData.data(), pixelShaderData.size(), lcr->GetSpecialStateValues(), options, &decompilerOutput); LatteDecompilerShader* pixelShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, shaderBaseHash, false, shaderAuxHash, lcr->GetRawView()); // compile LatteShader_DumpShader(shaderBaseHash, shaderAuxHash, pixelShader); diff --git a/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.cpp b/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.cpp index 919c3094..ee7e14bd 100644 --- a/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.cpp +++ b/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.cpp @@ -1066,35 +1066,34 @@ void _LatteDecompiler_Process(LatteDecompilerShaderContext* shaderContext, uint8 _LatteDecompiler_GenerateDataForFastAccess(shaderContext->shader); } -void LatteDecompiler_InitContext(LatteDecompilerShaderContext& dCtx, LatteDecompilerOutput_t* output, LatteConst::ShaderType shaderType, uint64 shaderBaseHash, uint32* contextRegisters) +void LatteDecompiler_InitContext(LatteDecompilerShaderContext& dCtx, const LatteDecompilerOptions& options, LatteDecompilerOutput_t* output, LatteConst::ShaderType shaderType, uint64 shaderBaseHash, uint32* contextRegisters) { dCtx.output = output; dCtx.shaderType = shaderType; + dCtx.options = &options; output->shaderType = shaderType; dCtx.shaderBaseHash = shaderBaseHash; dCtx.contextRegisters = contextRegisters; dCtx.contextRegistersNew = (LatteContextRegister*)contextRegisters; + + // set context parameters (redundant stuff since options can be accessed directly) + dCtx.usesGeometryShader = options.usesGeometryShader; + dCtx.useTFViaSSBO = options.useTFViaSSBO; } -void LatteDecompiler_DecompileVertexShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, LatteFetchShader** fetchShaderList, sint32 fetchShaderCount, uint32* hleSpecialState, bool usesGeometryShader, LatteDecompilerOutput_t* output, bool useTFViaSSBO) +void LatteDecompiler_DecompileVertexShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, struct LatteFetchShader** fetchShaderList, sint32 fetchShaderCount, uint32* hleSpecialState, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output) { cemu_assert_debug((programSize & 3) == 0); performanceMonitor.gpuTime_shaderCreate.beginMeasuring(); // prepare decompiler context LatteDecompilerShaderContext shaderContext = { 0 }; - LatteDecompiler_InitContext(shaderContext, output, LatteConst::ShaderType::Vertex, shaderBaseHash, contextRegisters); + LatteDecompiler_InitContext(shaderContext, options, output, LatteConst::ShaderType::Vertex, shaderBaseHash, contextRegisters); cemu_assert_debug(fetchShaderCount == 1); for (sint32 i = 0; i < fetchShaderCount; i++) { shaderContext.fetchShaderList[i] = fetchShaderList[i]; } shaderContext.fetchShaderCount = fetchShaderCount; - // ugly hack to get tf mode from Vulkan renderer - shaderContext.useTFViaSSBO = useTFViaSSBO; - if (g_renderer->GetType() == RendererAPI::Vulkan) - { - shaderContext.useTFViaSSBO = VulkanRenderer::GetInstance()->useTFViaSSBO(); - } // prepare shader (deprecated) LatteDecompilerShader* shader = new LatteDecompilerShader(); shader->shaderType = LatteConst::ShaderType::Vertex; @@ -1103,7 +1102,6 @@ void LatteDecompiler_DecompileVertexShader(uint64 shaderBaseHash, uint32* contex output->shaderType = LatteConst::ShaderType::Vertex; shaderContext.shader = shader; output->shader = shader; - shaderContext.usesGeometryShader = usesGeometryShader; for (sint32 i = 0; i < LATTE_NUM_MAX_TEX_UNITS; i++) { shader->textureUnitSamplerAssignment[i] = LATTE_DECOMPILER_SAMPLER_NONE; @@ -1114,14 +1112,14 @@ void LatteDecompiler_DecompileVertexShader(uint64 shaderBaseHash, uint32* contex performanceMonitor.gpuTime_shaderCreate.endMeasuring(); } -void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint8* gsCopyProgramData, uint32 gsCopyProgramSize, uint32* hleSpecialState, uint32 vsRingParameterCount, LatteDecompilerOutput_t* output, bool useTFViaSSBO) +void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint8* gsCopyProgramData, uint32 gsCopyProgramSize, uint32* hleSpecialState, uint32 vsRingParameterCount, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output) { cemu_assert_debug((programSize & 3) == 0); performanceMonitor.gpuTime_shaderCreate.beginMeasuring(); // prepare decompiler context LatteDecompilerShaderContext shaderContext = { 0 }; shaderContext.fetchShaderCount = 0; - LatteDecompiler_InitContext(shaderContext, output, LatteConst::ShaderType::Geometry, shaderBaseHash, contextRegisters); + LatteDecompiler_InitContext(shaderContext, options, output, LatteConst::ShaderType::Geometry, shaderBaseHash, contextRegisters); // prepare shader LatteDecompilerShader* shader = new LatteDecompilerShader(); shaderContext.output = output; @@ -1131,7 +1129,6 @@ void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* cont output->shaderType = LatteConst::ShaderType::Geometry; shaderContext.shader = shader; output->shader = shader; - shaderContext.usesGeometryShader = true; if (gsCopyProgramData == NULL) { shader->hasError = true; @@ -1145,24 +1142,18 @@ void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* cont shader->textureUnitSamplerAssignment[i] = LATTE_DECOMPILER_SAMPLER_NONE; shader->textureUsesDepthCompare[i] = false; } - // ugly hack to get tf mode from Vulkan renderer - shaderContext.useTFViaSSBO = useTFViaSSBO; - if (g_renderer->GetType() == RendererAPI::Vulkan) - { - shaderContext.useTFViaSSBO = VulkanRenderer::GetInstance()->useTFViaSSBO(); - } // parse & compile _LatteDecompiler_Process(&shaderContext, programData, programSize); performanceMonitor.gpuTime_shaderCreate.endMeasuring(); } -void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint32* hleSpecialState, bool usesGeometryShader, LatteDecompilerOutput_t* output) +void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint32* hleSpecialState, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output) { cemu_assert_debug((programSize & 3) == 0); performanceMonitor.gpuTime_shaderCreate.beginMeasuring(); // prepare decompiler context LatteDecompilerShaderContext shaderContext = { 0 }; - LatteDecompiler_InitContext(shaderContext, output, LatteConst::ShaderType::Pixel, shaderBaseHash, contextRegisters); + LatteDecompiler_InitContext(shaderContext, options, output, LatteConst::ShaderType::Pixel, shaderBaseHash, contextRegisters); shaderContext.contextRegisters = contextRegisters; // prepare shader LatteDecompilerShader* shader = new LatteDecompilerShader(); @@ -1172,7 +1163,6 @@ void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* context output->shaderType = LatteConst::ShaderType::Pixel; shaderContext.shader = shader; output->shader = shader; - shaderContext.usesGeometryShader = usesGeometryShader; for (sint32 i = 0; i < LATTE_NUM_MAX_TEX_UNITS; i++) { shader->textureUnitSamplerAssignment[i] = LATTE_DECOMPILER_SAMPLER_NONE; diff --git a/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.h b/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.h index 4f0d3ff2..febec1a4 100644 --- a/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.h +++ b/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.h @@ -218,7 +218,7 @@ struct LatteDecompilerShader std::vector entries; }; std::vector list_remappedUniformEntries_register; - std::vector<_RemappedUniformBufferGroup> list_remappedUniformEntries_bufferGroups; + std::vector<_RemappedUniformBufferGroup> list_remappedUniformEntries_bufferGroups; }; struct LatteDecompilerOutputUniformOffsets @@ -250,6 +250,17 @@ struct LatteDecompilerOutputUniformOffsets } }; +struct LatteDecompilerOptions +{ + bool usesGeometryShader{ false }; + // Vulkan-specific + bool useTFViaSSBO{ false }; + struct + { + bool hasRoundingModeRTEFloat32{ false }; + }spirvInstrinsics; +}; + struct LatteDecompilerOutput_t { LatteDecompilerShader* shader; @@ -272,9 +283,9 @@ struct LatteDecompilerOutput_t struct LatteDecompilerSubroutineInfo; -void LatteDecompiler_DecompileVertexShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, struct LatteFetchShader** fetchShaderList, sint32 fetchShaderCount, uint32* hleSpecialState, bool usesGeometryShader, LatteDecompilerOutput_t* output, bool useTFViaSSBO = false); -void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint8* gsCopyProgramData, uint32 gsCopyProgramSize, uint32* hleSpecialState, uint32 vsRingParameterCount, LatteDecompilerOutput_t* output, bool useTFViaSSBO = false); -void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint32* hleSpecialState, bool usesGeometryShader, LatteDecompilerOutput_t* output); +void LatteDecompiler_DecompileVertexShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, struct LatteFetchShader** fetchShaderList, sint32 fetchShaderCount, uint32* hleSpecialState, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output); +void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint8* gsCopyProgramData, uint32 gsCopyProgramSize, uint32* hleSpecialState, uint32 vsRingParameterCount, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output); +void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint32* hleSpecialState, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output); // specialized shader parsers diff --git a/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitGLSLHeader.hpp b/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitGLSLHeader.hpp index 677399c2..a2545d1d 100644 --- a/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitGLSLHeader.hpp +++ b/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitGLSLHeader.hpp @@ -306,6 +306,11 @@ namespace LatteDecompiler { src->add("#define GET_FRAGCOORD() vec4(gl_FragCoord.xy*uf_fragCoordScale.xy,gl_FragCoord.z, 1.0/gl_FragCoord.w)" _CRLF); } + if (decompilerContext->options->spirvInstrinsics.hasRoundingModeRTEFloat32) + { + src->add("#extension GL_EXT_spirv_intrinsics: enable" _CRLF); + src->add("spirv_execution_mode(4462, 32);" _CRLF); // RoundingModeRTE 32 + } src->add("#else" _CRLF); // OpenGL defines src->add("#define ATTR_LAYOUT(__vkSet, __location) layout(location = __location)" _CRLF); diff --git a/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerInternal.h b/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerInternal.h index 3dc3eca6..9e3120e6 100644 --- a/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerInternal.h +++ b/src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerInternal.h @@ -70,8 +70,6 @@ struct LatteDecompilerTEXInstruction uint8 nfa{}; uint8 isSigned{}; }memRead; - // custom shadow function - sint32 shadowFunctionIndex{}; }; struct LatteDecompilerCFInstruction @@ -116,7 +114,7 @@ struct LatteDecompilerCFInstruction ~LatteDecompilerCFInstruction() { - cemu_assert_debug(!(instructionsALU.size() != 0 && instructionsTEX.size() != 0)); // make sure we dont accidentally added the wrong instruction type + cemu_assert_debug(!(instructionsALU.size() != 0 && instructionsTEX.size() != 0)); // make sure we haven't accidentally added the wrong instruction type } #if BOOST_OS_WINDOWS @@ -148,6 +146,7 @@ struct LatteDecompilerShaderContext LatteDecompilerOutput_t* output; LatteDecompilerShader* shader; LatteConst::ShaderType shaderType; + const class LatteDecompilerOptions* options; uint32* contextRegisters; // deprecated struct LatteContextRegister* contextRegistersNew; uint64 shaderBaseHash; @@ -217,10 +216,9 @@ struct LatteDecompilerShaderContext bool hasUniformVarBlock; sint32 currentBindingPointVK{}; - // unsorted + // misc bool usesGeometryShader; // for VS bool useTFViaSSBO; - sint32 currentShadowFunctionIndex; std::vector list_subroutines; }; diff --git a/src/Cafe/HW/Latte/Renderer/Vulkan/RendererShaderVk.cpp b/src/Cafe/HW/Latte/Renderer/Vulkan/RendererShaderVk.cpp index 7c577903..672b9c80 100644 --- a/src/Cafe/HW/Latte/Renderer/Vulkan/RendererShaderVk.cpp +++ b/src/Cafe/HW/Latte/Renderer/Vulkan/RendererShaderVk.cpp @@ -1,16 +1,18 @@ #include "Cafe/HW/Latte/Renderer/Vulkan/RendererShaderVk.h" +#include "Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h" +#include "Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h" +#include "config/ActiveSettings.h" +#include "config/CemuConfig.h" +#include "util/helpers/ConcurrentQueue.h" +#include "Cemu/FileCache/FileCache.h" #include #include -#include "Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h" -#include "Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h" +// required for modifying SPIR-V +#include -#include "config/ActiveSettings.h" -#include "config/CemuConfig.h" -#include "util/helpers/ConcurrentQueue.h" -#include "Cemu/FileCache/FileCache.h" bool s_isLoadingShadersVk{ false }; class FileCache* s_spirvCache{nullptr}; diff --git a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp index 3c85baac..df37ae7e 100644 --- a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp +++ b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp @@ -37,13 +37,13 @@ extern std::atomic_int g_compiling_pipelines; const std::vector kOptionalDeviceExtensions = { - //VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME, VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME, VK_NV_FILL_RECTANGLE_EXTENSION_NAME, VK_EXT_PIPELINE_CREATION_FEEDBACK_EXTENSION_NAME, VK_EXT_FILTER_CUBIC_EXTENSION_NAME, // not supported by any device yet VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME, VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME, + VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME }; const std::vector kRequiredDeviceExtensions = @@ -236,25 +236,51 @@ void VulkanRenderer::DetermineVendor() void VulkanRenderer::GetDeviceFeatures() { + /* Get Vulkan features via GetPhysicalDeviceFeatures2 */ + void* prevStruct = nullptr; VkPhysicalDeviceCustomBorderColorFeaturesEXT bcf{}; bcf.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT; + bcf.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT; + prevStruct = &bcf; VkPhysicalDevicePipelineCreationCacheControlFeaturesEXT pcc{}; pcc.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_CREATION_CACHE_CONTROL_FEATURES_EXT; - pcc.pNext = &bcf; + pcc.pNext = prevStruct; + prevStruct = &pcc; VkPhysicalDeviceFeatures2 physicalDeviceFeatures2{}; physicalDeviceFeatures2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2; - physicalDeviceFeatures2.pNext = &pcc; + physicalDeviceFeatures2.pNext = prevStruct; vkGetPhysicalDeviceFeatures2(m_physicalDevice, &physicalDeviceFeatures2); + /* Get Vulkan device properties and limits */ + VkPhysicalDeviceFloatControlsPropertiesKHR pfcp{}; + prevStruct = nullptr; + if (m_featureControl.deviceExtensions.shader_float_controls) + { + pfcp.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES_KHR; + pfcp.pNext = prevStruct; + prevStruct = &pfcp; + } + + VkPhysicalDeviceProperties2 prop2{}; + prop2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2; + prop2.pNext = prevStruct; + + vkGetPhysicalDeviceProperties2(m_physicalDevice, &prop2); + + /* Determine which subfeatures we can use */ + m_featureControl.deviceExtensions.pipeline_creation_cache_control = pcc.pipelineCreationCacheControl; m_featureControl.deviceExtensions.custom_border_color_without_format = m_featureControl.deviceExtensions.custom_border_color && bcf.customBorderColorWithoutFormat; + m_featureControl.shaderFloatControls.shaderRoundingModeRTEFloat32 = m_featureControl.deviceExtensions.shader_float_controls && pfcp.shaderRoundingModeRTEFloat32; + if(!m_featureControl.shaderFloatControls.shaderRoundingModeRTEFloat32) + cemuLog_log(LogType::Force, "Shader round mode control not available on this device or driver. Some rendering issues might occur."); if (!m_featureControl.deviceExtensions.pipeline_creation_cache_control) { - forceLogDebug_printf("VK_EXT_pipeline_creation_cache_control not supported. Cannot use asynchronous shader and pipeline compilation"); + cemuLog_log(LogType::Force, "VK_EXT_pipeline_creation_cache_control not supported. Cannot use asynchronous shader and pipeline compilation"); // if async shader compilation is enabled show warning message if (GetConfig().async_compile) wxMessageBox(_("The currently installed graphics driver does not support the Vulkan extension necessary for asynchronous shader compilation. Asynchronous compilation cannot be used.\n \nRequired extension: VK_EXT_pipeline_creation_cache_control\n\nInstalling the latest graphics driver may solve this error."), _("Information"), wxOK | wxCENTRE); @@ -270,13 +296,11 @@ void VulkanRenderer::GetDeviceFeatures() forceLog_printf("VK_EXT_custom_border_color not supported. Cannot emulate arbitrary border color"); } } - // retrieve limits - VkPhysicalDeviceProperties2 p2{}; - p2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2; - vkGetPhysicalDeviceProperties2(m_physicalDevice, &p2); - m_featureControl.limits.minUniformBufferOffsetAlignment = std::max(p2.properties.limits.minUniformBufferOffsetAlignment, (VkDeviceSize)4); - m_featureControl.limits.nonCoherentAtomSize = std::max(p2.properties.limits.nonCoherentAtomSize, (VkDeviceSize)4); - cemuLog_log(LogType::Force, fmt::format("VulkanLimits: UBAlignment {0} nonCoherentAtomSize {1}", p2.properties.limits.minUniformBufferOffsetAlignment, p2.properties.limits.nonCoherentAtomSize)); + + // get limits + m_featureControl.limits.minUniformBufferOffsetAlignment = std::max(prop2.properties.limits.minUniformBufferOffsetAlignment, (VkDeviceSize)4); + m_featureControl.limits.nonCoherentAtomSize = std::max(prop2.properties.limits.nonCoherentAtomSize, (VkDeviceSize)4); + cemuLog_log(LogType::Force, fmt::format("VulkanLimits: UBAlignment {0} nonCoherentAtomSize {1}", prop2.properties.limits.minUniformBufferOffsetAlignment, prop2.properties.limits.nonCoherentAtomSize)); } VulkanRenderer::VulkanRenderer() @@ -1025,6 +1049,8 @@ VkDeviceCreateInfo VulkanRenderer::CreateDeviceCreateInfo(const std::vector& data, cons { try { - // g_imgui_textures.emplace_back(texture); std::vector tmp(size.x * size.y * 4); for (size_t i = 0; i < data.size() / 3; ++i) { diff --git a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h index 81c3801d..147b6c15 100644 --- a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h +++ b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h @@ -463,8 +463,14 @@ private: bool external_memory_host = false; // VK_EXT_external_memory_host bool synchronization2 = false; // VK_KHR_synchronization2 bool dynamic_rendering = false; // VK_KHR_dynamic_rendering + bool shader_float_controls = false; // VK_KHR_shader_float_controls }deviceExtensions; + struct + { + bool shaderRoundingModeRTEFloat32{ false }; + }shaderFloatControls; // from VK_KHR_shader_float_controls + struct { bool debug_utils = false; // VK_EXT_DEBUG_UTILS @@ -482,8 +488,8 @@ private: uint32 nonCoherentAtomSize = 256; }limits; - bool debugMarkersSupported = false; // frame debugger is attached - bool disableMultithreadedCompilation = false; // for old nvidia drivers + bool debugMarkersSupported{ false }; // frame debugger is attached + bool disableMultithreadedCompilation{ false }; // for old nvidia drivers }m_featureControl{}; static bool CheckDeviceExtensionSupport(const VkPhysicalDevice device, FeatureControl& info); @@ -936,12 +942,10 @@ private: public: - bool GetDisableMultithreadedCompilation() { return m_featureControl.disableMultithreadedCompilation; } - bool useTFViaSSBO() { return m_featureControl.mode.useTFEmulationViaSSBO; } - bool IsDebugUtilsEnabled() const - { - return m_featureControl.debugMarkersSupported && m_featureControl.instanceExtensions.debug_utils; - } + bool GetDisableMultithreadedCompilation() const { return m_featureControl.disableMultithreadedCompilation; } + bool UseTFViaSSBO() const { return m_featureControl.mode.useTFEmulationViaSSBO; } + bool HasSPRIVRoundingModeRTE32() const { return m_featureControl.shaderFloatControls.shaderRoundingModeRTEFloat32; } + bool IsDebugUtilsEnabled() const { return m_featureControl.debugMarkersSupported && m_featureControl.instanceExtensions.debug_utils; } private: diff --git a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRendererCore.cpp b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRendererCore.cpp index f3104e77..5d826b61 100644 --- a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRendererCore.cpp +++ b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRendererCore.cpp @@ -1579,7 +1579,7 @@ void VulkanRenderer::draw_updateVertexBuffersDirectAccess() if (bufferAddress == MPTR_NULL) { - cemu_assert_unimplemented(); + bufferAddress = 0x10000000; } if (m_state.currentVertexBinding[bufferIndex].offset == bufferAddress) continue;