Mirrors
/
Cemu
mirror of https://github.com/cemu-project/Cemu.git
1
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Latte: Small refactor for shader decompiler 

Latte: Small refactor for shader decompiler
This commit is contained in:
Exzap 2023-03-01 22:16:57 +01:00
parent 4a564e2447
commit f8ea594447
9 changed files with 67 additions and 149 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
src/Cafe/HW/Latte/Core/LatteShader.cpp
View File

@ -549,7 +549,7 @@ uint64 LatteSHRC_CalcVSAuxHash(LatteDecompilerShader* vertexShader, uint32* cont
// hash stride for streamout buffers
for (uint32 i = 0; i < LATTE_NUM_STREAMOUT_BUFFER; i++)
{
if(!vertexShader->streamoutBufferWriteMask2[i])
if(!vertexShader->streamoutBufferWriteMask[i])
continue;
uint32 bufferStride = contextRegisters[mmVGT_STRMOUT_VTX_STRIDE_0 + i * 4];
auxHash = std::rotl<uint64>(auxHash, 7);
@ -617,7 +617,7 @@ LatteDecompilerShader* LatteShader_CreateShaderFromDecompilerOutput(LatteDecompi
// copy texture info
shader->textureUnitMask2 = decompilerOutput.textureUnitMask;
// copy streamout info
shader->streamoutBufferWriteMask2 = decompilerOutput.streamoutBufferWriteMask;
shader->streamoutBufferWriteMask = decompilerOutput.streamoutBufferWriteMask;
shader->hasStreamoutBufferWrite = decompilerOutput.streamoutBufferWriteMask.any();
// copy uniform offsets
// for OpenGL these are retrieved in _prepareSeparableUniforms()
@ -746,9 +746,7 @@ LatteDecompilerShader* LatteShader_CompileSeparableVertexShader(uint64 baseHash,
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(), options, &decompilerOutput);
LatteDecompiler_DecompileVertexShader(_shaderBaseHash_vs, LatteGPUState.contextRegister, vertexShaderPtr, vertexShaderSize, fetchShader, options, &decompilerOutput);
LatteDecompilerShader* vertexShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, baseHash, true, 0, LatteGPUState.contextRegister);
vsAuxHash = vertexShader->auxHash;
if (vertexShader->hasError == false)
@ -779,9 +777,7 @@ LatteDecompilerShader* LatteShader_CompileSeparableGeometryShader(uint64 baseHas
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, options, &decompilerOutput);
LatteDecompiler_DecompileGeometryShader(_shaderBaseHash_gs, LatteGPUState.contextRegister, geometryShaderPtr, geometryShaderSize, geometryCopyShader, geometryCopyShaderSize, _activeVertexShader->ringParameterCount, options, &decompilerOutput);
LatteDecompilerShader* geometryShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, baseHash, true, 0, LatteGPUState.contextRegister);
if (geometryShader->hasError == false)
{
@ -810,7 +806,7 @@ LatteDecompilerShader* LatteShader_CompileSeparablePixelShader(uint64 baseHash,
LatteShader_GetDecompilerOptions(options, LatteConst::ShaderType::Pixel, usesGeometryShader);
LatteDecompilerOutput_t decompilerOutput{};
LatteDecompiler_DecompilePixelShader(baseHash, LatteGPUState.contextRegister, pixelShaderPtr, pixelShaderSize, LatteGPUState.contextNew.GetSpecialStateValues(), options, &decompilerOutput);
LatteDecompiler_DecompilePixelShader(baseHash, LatteGPUState.contextRegister, pixelShaderPtr, pixelShaderSize, options, &decompilerOutput);
LatteDecompilerShader* pixelShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, baseHash, true, 0, LatteGPUState.contextRegister);
psAuxHash = pixelShader->auxHash;
LatteShader_DumpShader(_shaderBaseHash_ps, psAuxHash, pixelShader);

10
src/Cafe/HW/Latte/Core/LatteShaderCache.cpp
View File

@ -651,15 +651,12 @@ bool LatteShaderCache_readSeparableVertexShader(MemStreamReader& streamReader, u
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(), options, &decompilerOutput);
LatteDecompiler_DecompileVertexShader(shaderBaseHash, lcr->GetRawView(), vertexShaderData.data(), vertexShaderData.size(), fetchShader, options, &decompilerOutput);
LatteDecompilerShader* vertexShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, shaderBaseHash, false, shaderAuxHash, lcr->GetRawView());
// compile
LatteShader_DumpShader(shaderBaseHash, shaderAuxHash, vertexShader);
LatteShader_DumpRawShader(shaderBaseHash, shaderAuxHash, SHADER_DUMP_TYPE_VERTEX, vertexShaderData.data(), vertexShaderData.size());
LatteShaderCache_loadOrCompileSeparableShader(vertexShader, shaderBaseHash, shaderAuxHash);
catchOpenGLError();
LatteSHRC_RegisterShader(vertexShader, shaderBaseHash, shaderAuxHash);
return true;
}
@ -696,7 +693,7 @@ bool LatteShaderCache_readSeparableGeometryShader(MemStreamReader& streamReader,
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, options, &decompilerOutput);
LatteDecompiler_DecompileGeometryShader(shaderBaseHash, lcr->GetRawView(), geometryShaderData.data(), geometryShaderData.size(), geometryCopyShaderData.data(), geometryCopyShaderData.size(), vsRingParameterCount, options, &decompilerOutput);
LatteDecompilerShader* geometryShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, shaderBaseHash, false, shaderAuxHash, lcr->GetRawView());
// compile
LatteShader_DumpShader(shaderBaseHash, shaderAuxHash, geometryShader);
@ -734,13 +731,12 @@ bool LatteShaderCache_readSeparablePixelShader(MemStreamReader& streamReader, ui
LatteShader_GetDecompilerOptions(options, LatteConst::ShaderType::Pixel, usesGeometryShader);
// decompile pixel shader
LatteDecompilerOutput_t decompilerOutput{};
LatteDecompiler_DecompilePixelShader(shaderBaseHash, lcr->GetRawView(), pixelShaderData.data(), pixelShaderData.size(), lcr->GetSpecialStateValues(), options, &decompilerOutput);
LatteDecompiler_DecompilePixelShader(shaderBaseHash, lcr->GetRawView(), pixelShaderData.data(), pixelShaderData.size(), options, &decompilerOutput);
LatteDecompilerShader* pixelShader = LatteShader_CreateShaderFromDecompilerOutput(decompilerOutput, shaderBaseHash, false, shaderAuxHash, lcr->GetRawView());
// compile
LatteShader_DumpShader(shaderBaseHash, shaderAuxHash, pixelShader);
LatteShader_DumpRawShader(shaderBaseHash, shaderAuxHash, SHADER_DUMP_TYPE_PIXEL, pixelShaderData.data(), pixelShaderData.size());
LatteShaderCache_loadOrCompileSeparableShader(pixelShader, shaderBaseHash, shaderAuxHash);
catchOpenGLError();
LatteSHRC_RegisterShader(pixelShader, shaderBaseHash, shaderAuxHash);
return true;
}

6
src/Cafe/HW/Latte/Core/LatteStreamoutGPU.cpp
View File

@ -101,16 +101,16 @@ void LatteStreamout_PrepareDrawcall(uint32 count, uint32 instanceCount)
if (geometryShader)
{
#ifdef CEMU_DEBUG_ASSERT
cemu_assert_debug(vertexShader->streamoutBufferWriteMask2.any() == false);
cemu_assert_debug(vertexShader->streamoutBufferWriteMask.any() == false);
#endif
for (sint32 i = 0; i < LATTE_NUM_STREAMOUT_BUFFER; i++)
if (geometryShader->streamoutBufferWriteMask2[i])
if (geometryShader->streamoutBufferWriteMask[i])
streamoutWriteMask |= (1 << i);
}
else
{
for (sint32 i = 0; i < LATTE_NUM_STREAMOUT_BUFFER; i++)
if (vertexShader->streamoutBufferWriteMask2[i])
if (vertexShader->streamoutBufferWriteMask[i])
streamoutWriteMask |= (1 << i);
}
activeStreamoutOperation.streamoutWriteMask = streamoutWriteMask;

37
src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.cpp
View File

@ -1071,34 +1071,24 @@ void LatteDecompiler_InitContext(LatteDecompilerShaderContext& dCtx, const Latte
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;
output->shaderType = shaderType;
}
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_DecompileVertexShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, struct LatteFetchShader* fetchShader, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output)
{
cemu_assert_debug(fetchShader);
cemu_assert_debug((programSize & 3) == 0);
performanceMonitor.gpuTime_shaderCreate.beginMeasuring();
// prepare decompiler context
LatteDecompilerShaderContext shaderContext = { 0 };
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;
shaderContext.fetchShader = fetchShader;
// prepare shader (deprecated)
LatteDecompilerShader* shader = new LatteDecompilerShader();
shader->shaderType = LatteConst::ShaderType::Vertex;
shader->compatibleFetchShader = shaderContext.fetchShaderList[0];
shaderContext.shaderType = LatteConst::ShaderType::Vertex;
LatteDecompilerShader* shader = new LatteDecompilerShader(LatteConst::ShaderType::Vertex);
shader->compatibleFetchShader = shaderContext.fetchShader;
output->shaderType = LatteConst::ShaderType::Vertex;
shaderContext.shader = shader;
output->shader = shader;
@ -1112,20 +1102,16 @@ 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, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output)
void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint8* gsCopyProgramData, uint32 gsCopyProgramSize, 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, options, output, LatteConst::ShaderType::Geometry, shaderBaseHash, contextRegisters);
// prepare shader
LatteDecompilerShader* shader = new LatteDecompilerShader();
shaderContext.output = output;
shader->shaderType = LatteConst::ShaderType::Geometry;
LatteDecompilerShader* shader = new LatteDecompilerShader(LatteConst::ShaderType::Geometry);
shader->ringParameterCountFromPrevStage = vsRingParameterCount;
shaderContext.shaderType = LatteConst::ShaderType::Geometry;
output->shaderType = LatteConst::ShaderType::Geometry;
shaderContext.shader = shader;
output->shader = shader;
@ -1147,7 +1133,7 @@ void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* cont
performanceMonitor.gpuTime_shaderCreate.endMeasuring();
}
void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint32* hleSpecialState, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output)
void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output)
{
cemu_assert_debug((programSize & 3) == 0);
performanceMonitor.gpuTime_shaderCreate.beginMeasuring();
@ -1156,10 +1142,7 @@ void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* context
LatteDecompiler_InitContext(shaderContext, options, output, LatteConst::ShaderType::Pixel, shaderBaseHash, contextRegisters);
shaderContext.contextRegisters = contextRegisters;
// prepare shader
LatteDecompilerShader* shader = new LatteDecompilerShader();
shaderContext.output = output;
shader->shaderType = LatteConst::ShaderType::Pixel;
shaderContext.shaderType = LatteConst::ShaderType::Pixel;
LatteDecompilerShader* shader = new LatteDecompilerShader(LatteConst::ShaderType::Pixel);
output->shaderType = LatteConst::ShaderType::Pixel;
shaderContext.shader = shader;
output->shader = shader;

16
src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompiler.h
View File

@ -148,6 +148,8 @@ struct LatteDecompilerShaderResourceMapping
struct LatteDecompilerShader
{
LatteDecompilerShader(LatteConst::ShaderType shaderType) : shaderType(shaderType) {}
LatteDecompilerShader* next;
LatteConst::ShaderType shaderType;
uint64 baseHash;
@ -167,21 +169,21 @@ struct LatteDecompilerShader
Latte::E_DIM textureUnitDim[LATTE_NUM_MAX_TEX_UNITS]; // dimension of texture unit, from the currently set texture
bool textureIsIntegerFormat[LATTE_NUM_MAX_TEX_UNITS]{};
// analyzer stage (uniforms)
uint8 uniformMode; // determines how uniforms are managed within the shader (see GPU7_DECOMPILER_UNIFORM_MODE_* constants)
uint8 uniformMode; // determines how uniforms are managed within the shader (see LATTE_DECOMPILER_UNIFORM_MODE_* constants)
uint64 uniformDataHash64[2]; // used to avoid redundant calls to glUniform*
std::vector<LatteDecompilerRemappedUniformEntry_t> list_remappedUniformEntries;
// analyzer stage (textures)
std::bitset<LATTE_NUM_MAX_TEX_UNITS> textureUnitMask2;
uint16 textureUnitSamplerAssignment[LATTE_NUM_MAX_TEX_UNITS]; // GPU7_SAMPLER_NONE means undefined
uint16 textureUnitSamplerAssignment[LATTE_NUM_MAX_TEX_UNITS]; // LATTE_DECOMPILER_SAMPLER_NONE means undefined
bool textureUsesDepthCompare[LATTE_NUM_MAX_TEX_UNITS];
// analyzer stage (pixel outputs)
uint32 pixelColorOutputMask; // from LSB to MSB, 1 bit per written output. 1 if written (indices of color attachments, may differ from export index inside the pixel shader)
uint32 pixelColorOutputMask; // from LSB to MSB, 1 bit per written output. 1 if written (indices of color attachments)
// analyzer stage (geometry shader parameters/inputs)
uint32 ringParameterCount;
uint32 ringParameterCountFromPrevStage; // used in geometry shader to hold VS ringParameterCount
// analyzer stage (misc)
std::bitset<LATTE_NUM_STREAMOUT_BUFFER> streamoutBufferWriteMask2;
std::bitset<LATTE_NUM_STREAMOUT_BUFFER> streamoutBufferWriteMask;
bool hasStreamoutBufferWrite;
// output code
class StringBuf* strBuf_shaderSource{nullptr};
@ -283,9 +285,9 @@ struct LatteDecompilerOutput_t
struct LatteDecompilerSubroutineInfo;
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);
void LatteDecompiler_DecompileVertexShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, struct LatteFetchShader* fetchShader, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output);
void LatteDecompiler_DecompileGeometryShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, uint8* gsCopyProgramData, uint32 gsCopyProgramSize, uint32 vsRingParameterCount, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output);
void LatteDecompiler_DecompilePixelShader(uint64 shaderBaseHash, uint32* contextRegisters, uint8* programData, uint32 programSize, LatteDecompilerOptions& options, LatteDecompilerOutput_t* output);
// specialized shader parsers

15
src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerAnalyzer.cpp
View File

@ -630,7 +630,7 @@ namespace LatteDecompiler
if (decompilerContext->shaderType == LatteConst::ShaderType::Geometry && decompilerContext->analyzer.outputPointSize && decompilerContext->analyzer.writesPointSize == false)
decompilerContext->hasUniformVarBlock = true; // uf_pointSize
if (decompilerContext->analyzer.useSSBOForStreamout &&
(decompilerContext->shaderType == LatteConst::ShaderType::Vertex && decompilerContext->usesGeometryShader == false) ||
(decompilerContext->shaderType == LatteConst::ShaderType::Vertex && !decompilerContext->options->usesGeometryShader) ||
(decompilerContext->shaderType == LatteConst::ShaderType::Geometry))
{
decompilerContext->hasUniformVarBlock = true; // uf_verticesPerInstance and uf_streamoutBufferBase*
@ -735,7 +735,7 @@ void LatteDecompiler_analyze(LatteDecompilerShaderContext* shaderContext, LatteD
// analyze render state
shaderContext->analyzer.isPointsPrimitive = shaderContext->contextRegistersNew->VGT_PRIMITIVE_TYPE.get_PRIMITIVE_MODE() == Latte::LATTE_VGT_PRIMITIVE_TYPE::E_PRIMITIVE_TYPE::POINTS;
shaderContext->analyzer.hasStreamoutEnable = shaderContext->contextRegisters[mmVGT_STRMOUT_EN] != 0; // set if the shader is used for transform feedback operations
if (shaderContext->shaderType == LatteConst::ShaderType::Vertex && shaderContext->usesGeometryShader == false)
if (shaderContext->shaderType == LatteConst::ShaderType::Vertex && !shaderContext->options->usesGeometryShader)
shaderContext->analyzer.outputPointSize = shaderContext->analyzer.isPointsPrimitive;
else if (shaderContext->shaderType == LatteConst::ShaderType::Geometry)
{
@ -746,10 +746,9 @@ void LatteDecompiler_analyze(LatteDecompilerShaderContext* shaderContext, LatteD
// analyze input attributes for vertex/geometry shader
if (shader->shaderType == LatteConst::ShaderType::Vertex || shader->shaderType == LatteConst::ShaderType::Geometry)
{
for (sint32 f = 0; f < shaderContext->fetchShaderCount; f++)
if(shaderContext->fetchShader)
{
LatteFetchShader* parsedFetchShader = (LatteFetchShader*)shaderContext->fetchShaderList[f];
LatteFetchShader* parsedFetchShader = shaderContext->fetchShader;
for(auto& bufferGroup : parsedFetchShader->bufferGroups)
{
for (sint32 i = 0; i < bufferGroup.attribCount; i++)
@ -938,9 +937,9 @@ void LatteDecompiler_analyze(LatteDecompilerShaderContext* shaderContext, LatteD
// analyze input attributes again (if shader has relative GPR read)
if(shaderContext->analyzer.usesRelativeGPRRead && (shader->shaderType == LatteConst::ShaderType::Vertex || shader->shaderType == LatteConst::ShaderType::Geometry) )
{
for (sint32 f = 0; f < shaderContext->fetchShaderCount; f++)
if(shaderContext->fetchShader)
{
LatteFetchShader* parsedFetchShader = (LatteFetchShader*)shaderContext->fetchShaderList[f];
LatteFetchShader* parsedFetchShader = shaderContext->fetchShader;
for(auto& bufferGroup : parsedFetchShader->bufferGroups)
{
for (sint32 i = 0; i < bufferGroup.attribCount; i++)
@ -1077,7 +1076,7 @@ void LatteDecompiler_analyze(LatteDecompilerShaderContext* shaderContext, LatteD
if(list_subroutineAddrs.empty() == false)
forceLogDebug_printf("Todo - analyze shader subroutine CF stack");
// TF mode
if (shaderContext->useTFViaSSBO && shaderContext->output->streamoutBufferWriteMask.any())
if (shaderContext->options->useTFViaSSBO && shaderContext->output->streamoutBufferWriteMask.any())
{
shaderContext->analyzer.useSSBOForStreamout = true;
}

97
src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitGLSL.cpp
View File

@ -3363,7 +3363,7 @@ void _emitCFRingWriteCode(LatteDecompilerShaderContext* shaderContext, LatteDeco
if ((cfInstruction->memWriteCompMask&(1 << i)) == 0)
continue;
if (shaderContext->useTFViaSSBO)
if (shaderContext->options->useTFViaSSBO)
{
uint32 u32Offset = streamWrite->exportArrayBase + i;
src->addFmt("sb_buffer[sbBase{} + {}]", streamWrite->bufferIndex, u32Offset);
@ -3483,7 +3483,7 @@ void _emitStreamWriteCode(LatteDecompilerShaderContext* shaderContext, LatteDeco
if ((cfInstruction->memWriteCompMask&(1 << i)) == 0)
continue;
if (shaderContext->useTFViaSSBO)
if (shaderContext->options->useTFViaSSBO)
{
uint32 u32Offset = cfInstruction->exportArrayBase + i;
src->addFmt("sb_buffer[sbBase{} + {}]", streamoutBufferIndex, u32Offset);
@ -3831,7 +3831,7 @@ void LatteDecompiler_emitGLSLHelperFunctions(LatteDecompilerShaderContext* shade
//fCStr_shaderSource->add("float mul_nonIEEE(float a, float b){float r = a*b;r = intBitsToFloat(floatBitsToInt(r)&(((floatBitsToInt(a) != 0) && (floatBitsToInt(b) != 0))?0xFFFFFFFF:0));return r;}" STR_LINEBREAK); works
if( LatteGPUState.glVendor == GLVENDOR_NVIDIA && !ActiveSettings::DumpShadersEnabled())
fCStr_shaderSource->add("float mul_nonIEEE(float a, float b){return mix(0.0, a*b, (a != 0.0) && (b != 0.0));}" _CRLF); // compiles faster on Nvidia and also results in lower RAM usage
fCStr_shaderSource->add("float mul_nonIEEE(float a, float b){return mix(0.0, a*b, (a != 0.0) && (b != 0.0));}" _CRLF); // compiles faster on Nvidia and also results in lower RAM usage (OpenGL)
else
fCStr_shaderSource->add("float mul_nonIEEE(float a, float b){ if( a == 0.0 || b == 0.0 ) return 0.0; return a*b; }" _CRLF);
}
@ -3841,51 +3841,6 @@ void LatteDecompiler_emitGLSLHelperFunctions(LatteDecompilerShaderContext* shade
}
}
void _addPixelShaderExtraDebugInfo(LatteDecompilerShaderContext* shaderContext, StringBuf* fCStr_shaderSource)
{
#ifdef CEMU_DEBUG_ASSERT
fCStr_shaderSource->add("// Color buffers:" _CRLF);
for(uint32 i=0; i<8; i++)
{
uint32 regColorBuffer = shaderContext->contextRegisters[mmCB_COLOR0_BASE+i];
uint32 regColorSize = shaderContext->contextRegisters[mmCB_COLOR0_SIZE+i];
uint32 regColorInfo = shaderContext->contextRegisters[mmCB_COLOR0_INFO+i];
uint32 regColorView = shaderContext->contextRegisters[mmCB_COLOR0_VIEW+i];
MPTR colorBufferPhysMem = regColorBuffer;
if( regColorBuffer == MPTR_NULL )
continue;
uint32 colorBufferFormat = (regColorInfo>>2)&0x3F; // format
uint32 colorBufferTileMode = 0;
colorBufferTileMode = (regColorInfo >> 8) & 0xF;
switch ( (regColorInfo >> 12) & 7 )
{
case 4:
colorBufferFormat |= 0x100;
break;
case 1:
colorBufferFormat |= 0x200;
break;
case 5:
colorBufferFormat |= 0x300;
break;
case 6:
colorBufferFormat |= 0x400;
break;
case 7:
colorBufferFormat |= 0x800;
break;
default:
break;
}
uint32 colorBufferWidth = (regColorSize>>0)&0xFFFF;
uint32 colorBufferHeight = (regColorSize>>16)&0xFFFF;
fCStr_shaderSource->addFmt("// Color{}: {}x{} at 0x{:08x} fmt {:04x} tm {}" _CRLF, i, colorBufferWidth, colorBufferHeight, colorBufferPhysMem, colorBufferFormat, colorBufferTileMode);
}
#endif
}
#include "Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitGLSLHeader.hpp"
void LatteDecompiler_emitAttributeImport(LatteDecompilerShaderContext* shaderContext, LatteParsedFetchShaderAttribute_t& attrib)
@ -3954,18 +3909,14 @@ void LatteDecompiler_emitGLSLShader(LatteDecompilerShaderContext* shaderContext,
src->add("#extension GL_ARB_texture_gather : enable" _CRLF);
src->add("#extension GL_ARB_separate_shader_objects : enable" _CRLF);
if (shaderContext->analyzer.hasStreamoutWrite || shaderContext->usesGeometryShader )
if (shaderContext->analyzer.hasStreamoutWrite || shaderContext->options->usesGeometryShader )
src->add("#extension GL_ARB_enhanced_layouts : enable" _CRLF);
// debug info
src->addFmt("// shader %08x%08x" _CRLF, (uint32)(shaderContext->shaderBaseHash >> 32), (uint32)(shaderContext->shaderBaseHash & 0xFFFFFFFF));
src->addFmt("// shader {:016x}" _CRLF, shaderContext->shaderBaseHash);
#ifdef CEMU_DEBUG_ASSERT
src->addFmt("// usesIntegerValues: {}" _CRLF, shaderContext->analyzer.usesIntegerValues?"true":"false");
src->addFmt(_CRLF);
if( shader->shaderType == LatteConst::ShaderType::Pixel )
_addPixelShaderExtraDebugInfo(shaderContext, src);
#endif
// header part (definitions for inputs and outputs)
LatteDecompiler::emitHeader(shaderContext);
@ -3982,7 +3933,6 @@ void LatteDecompiler_emitGLSLShader(LatteDecompilerShaderContext* shaderContext,
{
if (shaderContext->analyzer.usesRelativeGPRRead || (shaderContext->analyzer.gprUseMask[i / 8] & (1 << (i & 7))) != 0)
{
//fCStr_appendFormatted(fCStr_shaderSource, "ivec4 R{}i, R{}i, R{}i, R{}i;" STR_LINEBREAK, i*4+0, i*4+1, i*4+2, i*4+3);
if (shaderContext->typeTracker.genIntReg)
src->addFmt("ivec4 R{}i = ivec4(0);" _CRLF, i);
else if (shaderContext->typeTracker.genFloatReg)
@ -4035,7 +3985,6 @@ void LatteDecompiler_emitGLSLShader(LatteDecompilerShaderContext* shaderContext,
src->add("bool predResult = true;" _CRLF);
if(shaderContext->analyzer.modifiesPixelActiveState )
{
// cemu_assert_debug(shaderContext->analyzer.activeStackMaxDepth == 0);
src->addFmt("bool activeMaskStack[{}];" _CRLF, shaderContext->analyzer.activeStackMaxDepth+1);
src->addFmt("bool activeMaskStackC[{}];" _CRLF, shaderContext->analyzer.activeStackMaxDepth+2);
for (sint32 i = 0; i < shaderContext->analyzer.activeStackMaxDepth; i++)
@ -4058,8 +4007,11 @@ void LatteDecompiler_emitGLSLShader(LatteDecompilerShaderContext* shaderContext,
}
}
// helper variables for cube maps (todo: Only emit when used)
src->addFmt("vec3 cubeMapSTM;" _CRLF);
src->addFmt("int cubeMapFaceId;" _CRLF);
if (shaderContext->analyzer.hasRedcCUBE)
{
src->add("vec3 cubeMapSTM;" _CRLF);
src->add("int cubeMapFaceId;" _CRLF);
}
for(sint32 i=0; i<LATTE_NUM_MAX_TEX_UNITS; i++)
{
if(!shaderContext->output->textureUnitMask[i])
@ -4106,25 +4058,18 @@ void LatteDecompiler_emitGLSLShader(LatteDecompilerShaderContext* shaderContext,
cemu_assert_unimplemented();
}
if (shaderContext->fetchShaderCount == 1)
LatteFetchShader* parsedFetchShader = shaderContext->fetchShader;
for(auto& bufferGroup : parsedFetchShader->bufferGroups)
{
LatteFetchShader* parsedFetchShader = shaderContext->fetchShaderList[0];
for(auto& bufferGroup : parsedFetchShader->bufferGroups)
{
for(sint32 i=0; i<bufferGroup.attribCount; i++)
LatteDecompiler_emitAttributeImport(shaderContext, bufferGroup.attrib[i]);
}
for (auto& bufferGroup : parsedFetchShader->bufferGroupsInvalid)
{
// these attributes point to non-existent buffers
// todo - figure out how the hardware actually handles this, currently we assume the input values are zero
for (sint32 i = 0; i < bufferGroup.attribCount; i++)
LatteDecompiler_emitAttributeImport(shaderContext, bufferGroup.attrib[i]);
}
for(sint32 i=0; i<bufferGroup.attribCount; i++)
LatteDecompiler_emitAttributeImport(shaderContext, bufferGroup.attrib[i]);
}
else
for (auto& bufferGroup : parsedFetchShader->bufferGroupsInvalid)
{
cemu_assert_unimplemented();
// these attributes point to non-existent buffers
// todo - figure out how the hardware actually handles this, currently we assume the input values are zero
for (sint32 i = 0; i < bufferGroup.attribCount; i++)
LatteDecompiler_emitAttributeImport(shaderContext, bufferGroup.attrib[i]);
}
}
else if (shader->shaderType == LatteConst::ShaderType::Pixel)
@ -4172,7 +4117,7 @@ void LatteDecompiler_emitGLSLShader(LatteDecompilerShaderContext* shaderContext,
continue;
}
if (shaderContext->usesGeometryShader)
if (shaderContext->options->usesGeometryShader)
{
// import from geometry shader
if (shaderContext->typeTracker.defaultDataType == LATTE_DECOMPILER_DTYPE_SIGNED_INT)
@ -4216,7 +4161,7 @@ void LatteDecompiler_emitGLSLShader(LatteDecompilerShaderContext* shaderContext,
// vertex shader should write renderstate point size at the end if required but not modified by shader
if (shaderContext->analyzer.outputPointSize && shaderContext->analyzer.writesPointSize == false)
{
if (shader->shaderType == LatteConst::ShaderType::Vertex && shaderContext->usesGeometryShader == false)
if (shader->shaderType == LatteConst::ShaderType::Vertex && shaderContext->options->usesGeometryShader == false)
src->add("gl_PointSize = uf_pointSize;" _CRLF);
}
// end of shader main

14
src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerEmitGLSLHeader.hpp
View File

@ -95,7 +95,7 @@ namespace LatteDecompiler
}
if (decompilerContext->analyzer.outputPointSize && decompilerContext->analyzer.writesPointSize == false)
{
if ((decompilerContext->shaderType == LatteConst::ShaderType::Vertex && !decompilerContext->usesGeometryShader) ||
if ((decompilerContext->shaderType == LatteConst::ShaderType::Vertex && !decompilerContext->options->usesGeometryShader) ||
decompilerContext->shaderType == LatteConst::ShaderType::Geometry)
{
uniformCurrentOffset = (uniformCurrentOffset + 3)&~3;
@ -135,7 +135,7 @@ namespace LatteDecompiler
}
// define uf_verticesPerInstance + uf_streamoutBufferBaseX
if (decompilerContext->analyzer.useSSBOForStreamout &&
(shader->shaderType == LatteConst::ShaderType::Vertex && decompilerContext->usesGeometryShader == false) ||
(shader->shaderType == LatteConst::ShaderType::Vertex && decompilerContext->options->usesGeometryShader == false) ||
(shader->shaderType == LatteConst::ShaderType::Geometry) )
{
shaderSrc->add("uniform int uf_verticesPerInstance;" _CRLF);
@ -298,7 +298,7 @@ namespace LatteDecompiler
if (decompilerContext->shaderType == LatteConst::ShaderType::Vertex || decompilerContext->shaderType == LatteConst::ShaderType::Geometry)
{
if (decompilerContext->usesGeometryShader)
if (decompilerContext->options->usesGeometryShader)
src->add("#define V2G_LAYOUT layout(location = 0)" _CRLF);
}
}
@ -322,7 +322,7 @@ namespace LatteDecompiler
src->add("#define XFB_BLOCK_LAYOUT(__bufferIndex, __stride, __location) layout(xfb_buffer = __bufferIndex, xfb_stride = __stride)" _CRLF);
src->add("#define SET_POSITION(_v) gl_Position = _v\r\n");
if (decompilerContext->usesGeometryShader)
if (decompilerContext->options->usesGeometryShader)
src->add("#define V2G_LAYOUT" _CRLF);
}
else if (decompilerContext->shaderType == LatteConst::ShaderType::Pixel)
@ -430,7 +430,7 @@ namespace LatteDecompiler
{
auto src = decompilerContext->shaderSource;
// per-vertex output (VS or GS)
if ((decompilerContext->shaderType == LatteConst::ShaderType::Vertex && decompilerContext->usesGeometryShader == false) ||
if ((decompilerContext->shaderType == LatteConst::ShaderType::Vertex && !decompilerContext->options->usesGeometryShader) ||
(decompilerContext->shaderType == LatteConst::ShaderType::Geometry))
{
src->add("out gl_PerVertex" _CRLF);
@ -441,7 +441,7 @@ namespace LatteDecompiler
src->add("};" _CRLF);
}
// varyings (variables passed from vertex to pixel shader, only if geometry stage is disabled
if (decompilerContext->usesGeometryShader == false)
if (decompilerContext->options->usesGeometryShader == false)
{
if (decompilerContext->shaderType == LatteConst::ShaderType::Vertex)
{
@ -537,7 +537,7 @@ namespace LatteDecompiler
// streamout buffer (transform feedback)
if ((decompilerContext->shaderType == LatteConst::ShaderType::Vertex || decompilerContext->shaderType == LatteConst::ShaderType::Geometry) && decompilerContext->analyzer.hasStreamoutEnable)
{
if (decompilerContext->useTFViaSSBO)
if (decompilerContext->options->useTFViaSSBO)
{
if (decompilerContext->analyzer.useSSBOForStreamout && decompilerContext->analyzer.hasStreamoutWrite)
{

7
src/Cafe/HW/Latte/LegacyShaderDecompiler/LatteDecompilerInternal.h
View File

@ -150,11 +150,10 @@ struct LatteDecompilerShaderContext
uint32* contextRegisters; // deprecated
struct LatteContextRegister* contextRegistersNew;
uint64 shaderBaseHash;
StringBuf* shaderSource; // move to output struct
StringBuf* shaderSource;
std::vector<LatteDecompilerCFInstruction> cfInstructions;
// fetch shader (required for vertex shader)
LatteFetchShader* fetchShaderList[32];
sint32 fetchShaderCount;
LatteFetchShader* fetchShader{};
// geometry copy shader (only present when geometry shader is active)
LatteParsedGSCopyShader* parsedGSCopyShader;
// state
@ -217,8 +216,6 @@ struct LatteDecompilerShaderContext
sint32 currentBindingPointVK{};
// misc
bool usesGeometryShader; // for VS
bool useTFViaSSBO;
std::vector<LatteDecompilerSubroutineInfo> list_subroutines;
};