/**
* Construct the final microcode from the compiled and verified shader source.
* @param ShaderOutput - Where to store the microcode and parameter map.
* @param InShaderSource - GLSL source with input/output signature.
* @param SourceLen - The length of the GLSL source code.
*/
static void BuildShaderOutput(
FShaderCompilerOutput& ShaderOutput,
const FShaderCompilerInput& ShaderInput,
const ANSICHAR* InShaderSource,
int32 SourceLen,
GLSLVersion Version
)
{
const ANSICHAR* USFSource = InShaderSource;
CrossCompiler::FHlslccHeader CCHeader;
if (!CCHeader.Read(USFSource, SourceLen))
{
UE_LOG(LogOpenGLShaderCompiler, Error, TEXT("Bad hlslcc header found"));
}
if (*USFSource != '#')
{
UE_LOG(LogOpenGLShaderCompiler, Error, TEXT("Bad hlslcc header found! Missing '#'!"));
}
FOpenGLCodeHeader Header = {0};
FShaderParameterMap& ParameterMap = ShaderOutput.ParameterMap;
EShaderFrequency Frequency = (EShaderFrequency)ShaderOutput.Target.Frequency;
TBitArray<> UsedUniformBufferSlots;
UsedUniformBufferSlots.Init(false, 32);
// Write out the magic markers.
Header.GlslMarker = 0x474c534c;
switch (Frequency)
{
case SF_Vertex:
Header.FrequencyMarker = 0x5653;
break;
case SF_Pixel:
Header.FrequencyMarker = 0x5053;
break;
case SF_Geometry:
Header.FrequencyMarker = 0x4753;
break;
case SF_Hull:
Header.FrequencyMarker = 0x4853;
break;
case SF_Domain:
Header.FrequencyMarker = 0x4453;
break;
case SF_Compute:
Header.FrequencyMarker = 0x4353;
break;
default:
UE_LOG(LogOpenGLShaderCompiler, Fatal, TEXT("Invalid shader frequency: %d"), (int32)Frequency);
}
static const FString AttributePrefix = TEXT("in_ATTRIBUTE");
static const FString GL_Prefix = TEXT("gl_");
for (auto& Input : CCHeader.Inputs)
{
// Only process attributes for vertex shaders.
if (Frequency == SF_Vertex && Input.Name.StartsWith(AttributePrefix))
{
int32 AttributeIndex = ParseNumber(*Input.Name + AttributePrefix.Len());
Header.Bindings.InOutMask |= (1 << AttributeIndex);
}
// Record user-defined input varyings
else if (!Input.Name.StartsWith(GL_Prefix))
{
FOpenGLShaderVarying Var;
Var.Location = Input.Index;
Var.Varying = ParseIdentifierANSI(Input.Name);
Header.Bindings.InputVaryings.Add(Var);
}
}
// Generate vertex attribute remapping table.
// This is used on devices where GL_MAX_VERTEX_ATTRIBS < 16
if (Frequency == SF_Vertex)
{
uint32 AttributeMask = Header.Bindings.InOutMask;
int32 NextAttributeSlot = 0;
Header.Bindings.VertexRemappedMask = 0;
for (int32 AttributeIndex = 0; AttributeIndex < 16; AttributeIndex++, AttributeMask >>= 1)
{
if (AttributeMask & 0x1)
{
Header.Bindings.VertexRemappedMask |= (1 << NextAttributeSlot);
Header.Bindings.VertexAttributeRemap[AttributeIndex] = NextAttributeSlot++;
}
else
{
Header.Bindings.VertexAttributeRemap[AttributeIndex] = -1;
}
}
}
void BuildResourceTableMapping(
const TMap<FString,FResourceTableEntry>& ResourceTableMap,
const TMap<FString,uint32>& ResourceTableLayoutHashes,
TBitArray<>& UsedUniformBufferSlots,
FShaderParameterMap& ParameterMap,
FShaderResourceTable& OutSRT)
{
check(OutSRT.ResourceTableBits == 0);
check(OutSRT.ResourceTableLayoutHashes.Num() == 0);
// Build resource table mapping
int32 MaxBoundResourceTable = -1;
TArray<uint32> ResourceTableSRVs;
TArray<uint32> ResourceTableSamplerStates;
TArray<uint32> ResourceTableUAVs;
for( auto MapIt = ResourceTableMap.CreateConstIterator(); MapIt; ++MapIt )
{
const FString& Name = MapIt->Key;
const FResourceTableEntry& Entry = MapIt->Value;
uint16 BufferIndex, BaseIndex, Size;
if (ParameterMap.FindParameterAllocation( *Name, BufferIndex, BaseIndex, Size ) )
{
ParameterMap.RemoveParameterAllocation(*Name);
uint16 UniformBufferIndex = INDEX_NONE, UBBaseIndex, UBSize;
if (ParameterMap.FindParameterAllocation(*Entry.UniformBufferName, UniformBufferIndex, UBBaseIndex, UBSize) == false)
{
UniformBufferIndex = UsedUniformBufferSlots.FindAndSetFirstZeroBit();
ParameterMap.AddParameterAllocation(*Entry.UniformBufferName,UniformBufferIndex,0,0);
}
OutSRT.ResourceTableBits |= (1 << UniformBufferIndex);
MaxBoundResourceTable = FMath::Max<int32>(MaxBoundResourceTable, (int32)UniformBufferIndex);
while (OutSRT.ResourceTableLayoutHashes.Num() <= MaxBoundResourceTable)
{
OutSRT.ResourceTableLayoutHashes.Add(0);
}
OutSRT.ResourceTableLayoutHashes[UniformBufferIndex] = ResourceTableLayoutHashes.FindChecked(Entry.UniformBufferName);
auto ResourceMap = FRHIResourceTableEntry::Create(UniformBufferIndex, Entry.ResourceIndex, BaseIndex);
switch( Entry.Type )
{
case UBMT_TEXTURE:
OutSRT.TextureMap.Add(ResourceMap);
break;
case UBMT_SAMPLER:
OutSRT.SamplerMap.Add(ResourceMap);
break;
case UBMT_SRV:
OutSRT.ShaderResourceViewMap.Add(ResourceMap);
break;
case UBMT_UAV:
OutSRT.UnorderedAccessViewMap.Add(ResourceMap);
break;
default:
check(0);
}
}
}
OutSRT.MaxBoundResourceTable = MaxBoundResourceTable;
}
//---------------------------------------------------------------------------
// Configure the chunk map for this device. The empty chunk map is an array
// of bool that maps which chunks consist entirely of unallocated clusters.
//
unsigned __fastcall TImageStore::LoadEmptyChunkMap(unsigned int ChunkSize)
{
unsigned nChunkCount, nClustersPerChunk, nBitmapBytes;
VOLUME_BITMAP_BUFFER *VolumeBitmap;
TBitArray *Bitmap = new TBitArray;
int nExtraClusterBits;
bool bChunkEmpty;
unsigned nEmptyChunkCount = 0;
// IF this isn't a Linux or local Windows partition then we have no business here
if (PartitionType == ptUnknown || PartitionType == ptWindows && FType != isDrive )
return 0;
// In cases where the input size is an exact multiple of the chunk size, we actually want nChunkCount to be one
// higher than the actual number of chunks. In that case, the last real chunk can potentially be empty, and the
// next read will return zero bytes, telling the read thread that it's at EOF.
nChunkCount = (Geometry.Bytes + ChunkSize) / ChunkSize;
if (EmptyChunkMap)
delete EmptyChunkMap;
EmptyChunkMap = new TBitArray();
EmptyChunkMap->Size = nChunkCount;
nChunkSize = ChunkSize;
// Determine the number of clusters in a chunk, and the number of cluster bitmap bytes that are taken by a chunk.
nClustersPerChunk = (ChunkSize + Geometry.BytesPerCluster - 1) / Geometry.BytesPerCluster;
nBitmapBytes = (nClustersPerChunk + 7) / 8;
// Once we know all that, we can determine the size of the volume bitmap buffer that we need for each chunk
// nBitmanBytes is actually one larger than we need because there is one bitmap byte already in the VOLUME_BITMAP_BUFFER
// struct. This is good, though, because FSCTL_GET_VOLUME_BITMAP rounds the start cluster down to a multiple of 8, so we
// may need this extra space.
VolumeBitmap = (VOLUME_BITMAP_BUFFER *)SysGetMem(sizeof(VOLUME_BITMAP_BUFFER) + nBitmapBytes);
// Create the progress bar dialog
Application->CreateForm(__classid(TProgressDialog), &ProgressDialog);
ProgressDialog->Display("Mapping unallocated space...", 0, nChunkCount);
// Iterate through all the chunks and determine if the chunk resides entirely on clusters that are unallocated
for (unsigned n = 0; n < nChunkCount - 1; n++) {
LONGLONG nnStartCluster = (LONGLONG)n * (LONGLONG)ChunkSize / (LONGLONG)Geometry.BytesPerCluster;
DWORD nBytesReturned;
// Ask for a cluster bitmap starting with the cluster this chunk starts on
if (PartitionType == ptWindows) {
int ret = DeviceIoControl(hHandle, FSCTL_GET_VOLUME_BITMAP, &nnStartCluster, sizeof(nnStartCluster), VolumeBitmap, sizeof(VOLUME_BITMAP_BUFFER) + nBitmapBytes, &nBytesReturned, NULL);
if (!ret && GetLastError() != ERROR_MORE_DATA) {
delete EmptyChunkMap;
EmptyChunkMap = NULL;
nEmptyChunkCount = 0;
break;
} // if (!ret && GetLastError() != ERROR_MORE_DATA)
} // if (PartitionType == ptWindows)
else if (PartitionType == ptEXT2) {
bool ret;
VolumeBitmap->StartingLcn.QuadPart = nnStartCluster;
ret = GetEXT2VolumeBitmap(&VolumeBitmap->StartingLcn.QuadPart, VolumeBitmap->Buffer, nBitmapBytes + 1);
if (!ret) {
delete EmptyChunkMap;
EmptyChunkMap = NULL;
nEmptyChunkCount = 0;
break;
} // if (!ret)
} // if (PartitionType == ptEXT2)
nExtraClusterBits = nnStartCluster - VolumeBitmap->StartingLcn.QuadPart;
VolumeBitmap->Buffer[0] &= 0xFF>>nExtraClusterBits;
Bitmap->LoadBuffer(VolumeBitmap->Buffer, nClustersPerChunk + nExtraClusterBits);
bChunkEmpty = !Bitmap->HasSetBit();
if (bChunkEmpty)
nEmptyChunkCount++;
EmptyChunkMap->Bits[n] = bChunkEmpty;
if (!(n & 7))
ProgressDialog->Position = n;
} // for (unsigned n = 0; n < nnChunkCount; n++)
ProgressDialog->Position = nChunkCount;
SysFreeMem(VolumeBitmap);
delete Bitmap;
if (PartitionType == ptEXT2)
GetEXT2VolumeBitmap(NULL, NULL, 0, true);
Seek(0, FILE_BEGIN);
ProgressDialog->Close();
delete ProgressDialog;
// The last chunk must always be read normally so that the resulting image is sized properly
if (EmptyChunkMap)
EmptyChunkMap->Bits[nChunkCount - 1] = false;
return nEmptyChunkCount;
} // void __fastcall TImageStore:LoadChunkMap(unsigned int ChunkSize)
/**
* Construct the final microcode from the compiled and verified shader source.
* @param ShaderOutput - Where to store the microcode and parameter map.
* @param InShaderSource - Metal source with input/output signature.
* @param SourceLen - The length of the Metal source code.
*/
static void BuildMetalShaderOutput(
FShaderCompilerOutput& ShaderOutput,
const FShaderCompilerInput& ShaderInput,
const ANSICHAR* InShaderSource,
int32 SourceLen,
TArray<FShaderCompilerError>& OutErrors
)
{
FMetalCodeHeader Header = {0};
const ANSICHAR* ShaderSource = InShaderSource;
FShaderParameterMap& ParameterMap = ShaderOutput.ParameterMap;
EShaderFrequency Frequency = (EShaderFrequency)ShaderOutput.Target.Frequency;
TBitArray<> UsedUniformBufferSlots;
UsedUniformBufferSlots.Init(false,32);
// Write out the magic markers.
Header.Frequency = Frequency;
#define DEF_PREFIX_STR(Str) \
const ANSICHAR* Str##Prefix = "// @" #Str ": "; \
const int32 Str##PrefixLen = FCStringAnsi::Strlen(Str##Prefix)
DEF_PREFIX_STR(Inputs);
DEF_PREFIX_STR(Outputs);
DEF_PREFIX_STR(UniformBlocks);
DEF_PREFIX_STR(Uniforms);
DEF_PREFIX_STR(PackedGlobals);
DEF_PREFIX_STR(PackedUB);
DEF_PREFIX_STR(PackedUBCopies);
DEF_PREFIX_STR(PackedUBGlobalCopies);
DEF_PREFIX_STR(Samplers);
DEF_PREFIX_STR(UAVs);
DEF_PREFIX_STR(SamplerStates);
DEF_PREFIX_STR(NumThreads);
#undef DEF_PREFIX_STR
// Skip any comments that come before the signature.
while ( FCStringAnsi::Strncmp(ShaderSource, "//", 2) == 0 &&
FCStringAnsi::Strncmp(ShaderSource, "// @", 4) != 0 )
{
while (*ShaderSource && *ShaderSource++ != '\n') {}
}
// HLSLCC first prints the list of inputs.
if (FCStringAnsi::Strncmp(ShaderSource, InputsPrefix, InputsPrefixLen) == 0)
{
ShaderSource += InputsPrefixLen;
// Only inputs for vertex shaders must be tracked.
if (Frequency == SF_Vertex)
{
const ANSICHAR* AttributePrefix = "in_ATTRIBUTE";
const int32 AttributePrefixLen = FCStringAnsi::Strlen(AttributePrefix);
while (*ShaderSource && *ShaderSource != '\n')
{
// Skip the type.
while (*ShaderSource && *ShaderSource++ != ':') {}
// Only process attributes.
if (FCStringAnsi::Strncmp(ShaderSource, AttributePrefix, AttributePrefixLen) == 0)
{
ShaderSource += AttributePrefixLen;
uint8 AttributeIndex = ParseNumber(ShaderSource);
Header.Bindings.InOutMask |= (1 << AttributeIndex);
}
// Skip to the next.
while (*ShaderSource && *ShaderSource != ',' && *ShaderSource != '\n')
{
ShaderSource++;
}
if (Match(ShaderSource, '\n'))
{
break;
}
verify(Match(ShaderSource, ','));
}
}
else
{
// Skip to the next line.
while (*ShaderSource && *ShaderSource++ != '\n') {}
}
}
// Then the list of outputs.
if (FCStringAnsi::Strncmp(ShaderSource, OutputsPrefix, OutputsPrefixLen) == 0)
{
ShaderSource += OutputsPrefixLen;
// Only outputs for pixel shaders must be tracked.
if (Frequency == SF_Pixel)
{
const ANSICHAR* TargetPrefix = "out_Target";
const int32 TargetPrefixLen = FCStringAnsi::Strlen(TargetPrefix);
while (*ShaderSource && *ShaderSource != '\n')
{
// Skip the type.
while (*ShaderSource && *ShaderSource++ != ':') {}
// Handle targets.
if (FCStringAnsi::Strncmp(ShaderSource, TargetPrefix, TargetPrefixLen) == 0)
{
ShaderSource += TargetPrefixLen;
uint8 TargetIndex = ParseNumber(ShaderSource);
Header.Bindings.InOutMask |= (1 << TargetIndex);
}
// Handle depth writes.
else if (FCStringAnsi::Strcmp(ShaderSource, "gl_FragDepth") == 0)
{
Header.Bindings.InOutMask |= 0x8000;
}
// Skip to the next.
while (*ShaderSource && *ShaderSource != ',' && *ShaderSource != '\n')
{
ShaderSource++;
}
if (Match(ShaderSource, '\n'))
{
break;
}
verify(Match(ShaderSource, ','));
}
}
else
{
// Skip to the next line.
while (*ShaderSource && *ShaderSource++ != '\n') {}
}
}
bool bHasRegularUniformBuffers = false;
// Then 'normal' uniform buffers.
if (FCStringAnsi::Strncmp(ShaderSource, UniformBlocksPrefix, UniformBlocksPrefixLen) == 0)
{
ShaderSource += UniformBlocksPrefixLen;
while (*ShaderSource && *ShaderSource != '\n')
{
FString BufferName = ParseIdentifier(ShaderSource);
verify(BufferName.Len() > 0);
verify(Match(ShaderSource, '('));
uint16 UBIndex = ParseNumber(ShaderSource);
if (UBIndex >= Header.Bindings.NumUniformBuffers)
{
Header.Bindings.NumUniformBuffers = UBIndex + 1;
}
UsedUniformBufferSlots[UBIndex] = true;
verify(Match(ShaderSource, ')'));
ParameterMap.AddParameterAllocation(*BufferName, UBIndex, 0, 0);
bHasRegularUniformBuffers = true;
// Skip the comma.
if (Match(ShaderSource, '\n'))
{
break;
}
verify(Match(ShaderSource, ','));
}
Match(ShaderSource, '\n');
}
// Then uniforms.
const uint16 BytesPerComponent = 4;
/*
uint16 PackedUniformSize[OGL_NUM_PACKED_UNIFORM_ARRAYS] = {0};
FMemory::Memzero(&PackedUniformSize, sizeof(PackedUniformSize));
*/
if (FCStringAnsi::Strncmp(ShaderSource, UniformsPrefix, UniformsPrefixLen) == 0)
{
// @todo-mobile: Will we ever need to support this code path?
check(0);
/*
ShaderSource += UniformsPrefixLen;
while (*ShaderSource && *ShaderSource != '\n')
{
uint16 ArrayIndex = 0;
uint16 Offset = 0;
uint16 NumComponents = 0;
FString ParameterName = ParseIdentifier(ShaderSource);
verify(ParameterName.Len() > 0);
verify(Match(ShaderSource, '('));
ArrayIndex = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
Offset = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
NumComponents = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ')'));
ParameterMap.AddParameterAllocation(
*ParameterName,
ArrayIndex,
Offset * BytesPerComponent,
NumComponents * BytesPerComponent
);
if (ArrayIndex < OGL_NUM_PACKED_UNIFORM_ARRAYS)
{
PackedUniformSize[ArrayIndex] = FMath::Max<uint16>(
PackedUniformSize[ArrayIndex],
BytesPerComponent * (Offset + NumComponents)
);
}
// Skip the comma.
if (Match(ShaderSource, '\n'))
{
break;
}
verify(Match(ShaderSource, ','));
}
Match(ShaderSource, '\n');
*/
}
// Packed global uniforms
TMap<ANSICHAR, uint16> PackedGlobalArraySize;
if (FCStringAnsi::Strncmp(ShaderSource, PackedGlobalsPrefix, PackedGlobalsPrefixLen) == 0)
{
ShaderSource += PackedGlobalsPrefixLen;
while (*ShaderSource && *ShaderSource != '\n')
{
ANSICHAR ArrayIndex = 0;
uint16 Offset = 0;
uint16 NumComponents = 0;
FString ParameterName = ParseIdentifier(ShaderSource);
verify(ParameterName.Len() > 0);
verify(Match(ShaderSource, '('));
ArrayIndex = *ShaderSource++;
verify(Match(ShaderSource, ':'));
Offset = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ','));
NumComponents = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ')'));
ParameterMap.AddParameterAllocation(
*ParameterName,
ArrayIndex,
Offset * BytesPerComponent,
NumComponents * BytesPerComponent
);
uint16& Size = PackedGlobalArraySize.FindOrAdd(ArrayIndex);
Size = FMath::Max<uint16>(BytesPerComponent * (Offset + NumComponents), Size);
if (Match(ShaderSource, '\n'))
{
break;
}
// Skip the comma.
verify(Match(ShaderSource, ','));
}
Match(ShaderSource, '\n');
}
// Packed Uniform Buffers
TMap<int, TMap<ANSICHAR, uint16> > PackedUniformBuffersSize;
while (FCStringAnsi::Strncmp(ShaderSource, PackedUBPrefix, PackedUBPrefixLen) == 0)
{
ShaderSource += PackedUBPrefixLen;
FString BufferName = ParseIdentifier(ShaderSource);
verify(BufferName.Len() > 0);
verify(Match(ShaderSource, '('));
uint16 BufferIndex = ParseNumber(ShaderSource);
check(BufferIndex == Header.Bindings.NumUniformBuffers);
verify(Match(ShaderSource, ')'));
ParameterMap.AddParameterAllocation(*BufferName, Header.Bindings.NumUniformBuffers++, 0, 0);
verify(Match(ShaderSource, ':'));
Match(ShaderSource, ' ');
while (*ShaderSource && *ShaderSource != '\n')
{
FString ParameterName = ParseIdentifier(ShaderSource);
verify(ParameterName.Len() > 0);
verify(Match(ShaderSource, '('));
ParseNumber(ShaderSource);
verify(Match(ShaderSource, ','));
ParseNumber(ShaderSource);
verify(Match(ShaderSource, ')'));
if (Match(ShaderSource, '\n'))
{
break;
}
verify(Match(ShaderSource, ','));
}
}
// Packed Uniform Buffers copy lists & setup sizes for each UB/Precision entry
if (FCStringAnsi::Strncmp(ShaderSource, PackedUBCopiesPrefix, PackedUBCopiesPrefixLen) == 0)
{
ShaderSource += PackedUBCopiesPrefixLen;
while (*ShaderSource && *ShaderSource != '\n')
{
FMetalUniformBufferCopyInfo CopyInfo;
CopyInfo.SourceUBIndex = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
CopyInfo.SourceOffsetInFloats = ParseNumber(ShaderSource);
verify(Match(ShaderSource, '-'));
CopyInfo.DestUBIndex = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
CopyInfo.DestUBTypeName = *ShaderSource++;
CopyInfo.DestUBTypeIndex = CrossCompiler::PackedTypeNameToTypeIndex(CopyInfo.DestUBTypeName);
verify(Match(ShaderSource, ':'));
CopyInfo.DestOffsetInFloats = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
CopyInfo.SizeInFloats = ParseNumber(ShaderSource);
Header.UniformBuffersCopyInfo.Add(CopyInfo);
auto& UniformBufferSize = PackedUniformBuffersSize.FindOrAdd(CopyInfo.DestUBIndex);
uint16& Size = UniformBufferSize.FindOrAdd(CopyInfo.DestUBTypeName);
Size = FMath::Max<uint16>(BytesPerComponent * (CopyInfo.DestOffsetInFloats + CopyInfo.SizeInFloats), Size);
if (Match(ShaderSource, '\n'))
{
break;
}
verify(Match(ShaderSource, ','));
}
}
if (FCStringAnsi::Strncmp(ShaderSource, PackedUBGlobalCopiesPrefix, PackedUBGlobalCopiesPrefixLen) == 0)
{
ShaderSource += PackedUBGlobalCopiesPrefixLen;
while (*ShaderSource && *ShaderSource != '\n')
{
FMetalUniformBufferCopyInfo CopyInfo;
CopyInfo.SourceUBIndex = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
CopyInfo.SourceOffsetInFloats = ParseNumber(ShaderSource);
verify(Match(ShaderSource, '-'));
CopyInfo.DestUBIndex = 0;
CopyInfo.DestUBTypeName = *ShaderSource++;
CopyInfo.DestUBTypeIndex = CrossCompiler::PackedTypeNameToTypeIndex(CopyInfo.DestUBTypeName);
verify(Match(ShaderSource, ':'));
CopyInfo.DestOffsetInFloats = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
CopyInfo.SizeInFloats = ParseNumber(ShaderSource);
Header.UniformBuffersCopyInfo.Add(CopyInfo);
uint16& Size = PackedGlobalArraySize.FindOrAdd(CopyInfo.DestUBTypeName);
Size = FMath::Max<uint16>(BytesPerComponent * (CopyInfo.DestOffsetInFloats + CopyInfo.SizeInFloats), Size);
if (Match(ShaderSource, '\n'))
{
break;
}
verify(Match(ShaderSource, ','));
}
}
Header.Bindings.bHasRegularUniformBuffers = bHasRegularUniformBuffers;
// Setup Packed Array info
Header.Bindings.PackedGlobalArrays.Reserve(PackedGlobalArraySize.Num());
for (auto Iterator = PackedGlobalArraySize.CreateIterator(); Iterator; ++Iterator)
{
ANSICHAR TypeName = Iterator.Key();
uint16 Size = Iterator.Value();
Size = (Size + 0xf) & (~0xf);
CrossCompiler::FPackedArrayInfo Info;
Info.Size = Size;
Info.TypeName = TypeName;
Info.TypeIndex = CrossCompiler::PackedTypeNameToTypeIndex(TypeName);
Header.Bindings.PackedGlobalArrays.Add(Info);
}
// Setup Packed Uniform Buffers info
Header.Bindings.PackedUniformBuffers.Reserve(PackedUniformBuffersSize.Num());
for (auto Iterator = PackedUniformBuffersSize.CreateIterator(); Iterator; ++Iterator)
{
int BufferIndex = Iterator.Key();
auto& ArraySizes = Iterator.Value();
TArray<CrossCompiler::FPackedArrayInfo> InfoArray;
InfoArray.Reserve(ArraySizes.Num());
for (auto IterSizes = ArraySizes.CreateIterator(); IterSizes; ++IterSizes)
{
ANSICHAR TypeName = IterSizes.Key();
uint16 Size = IterSizes.Value();
Size = (Size + 0xf) & (~0xf);
CrossCompiler::FPackedArrayInfo Info;
Info.Size = Size;
Info.TypeName = TypeName;
Info.TypeIndex = CrossCompiler::PackedTypeNameToTypeIndex(TypeName);
InfoArray.Add(Info);
}
Header.Bindings.PackedUniformBuffers.Add(InfoArray);
}
// Then samplers.
if (FCStringAnsi::Strncmp(ShaderSource, SamplersPrefix, SamplersPrefixLen) == 0)
{
ShaderSource += SamplersPrefixLen;
while (*ShaderSource && *ShaderSource != '\n')
{
uint16 Offset = 0;
uint16 NumSamplers = 0;
FString ParameterName = ParseIdentifier(ShaderSource);
verify(ParameterName.Len() > 0);
verify(Match(ShaderSource, '('));
Offset = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
NumSamplers = ParseNumber(ShaderSource);
ParameterMap.AddParameterAllocation(
*ParameterName,
0,
Offset,
NumSamplers
);
Header.Bindings.NumSamplers = FMath::Max<uint8>(
Header.Bindings.NumSamplers,
Offset + NumSamplers
);
if (Match(ShaderSource, '['))
{
// Sampler States
do
{
FString SamplerState = ParseIdentifier(ShaderSource);
checkSlow(SamplerState.Len() != 0);
ParameterMap.AddParameterAllocation(
*SamplerState,
0,
Offset,
NumSamplers
);
}
while (Match(ShaderSource, ','));
verify(Match(ShaderSource, ']'));
}
verify(Match(ShaderSource, ')'));
if (Match(ShaderSource, '\n'))
{
break;
}
// Skip the comma.
verify(Match(ShaderSource, ','));
}
}
// Then UAVs (images in Metal)
if (FCStringAnsi::Strncmp(ShaderSource, UAVsPrefix, UAVsPrefixLen) == 0)
{
ShaderSource += UAVsPrefixLen;
while (*ShaderSource && *ShaderSource != '\n')
{
uint16 Offset = 0;
uint16 NumUAVs = 0;
FString ParameterName = ParseIdentifier(ShaderSource);
verify(ParameterName.Len() > 0);
verify(Match(ShaderSource, '('));
Offset = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ':'));
NumUAVs = ParseNumber(ShaderSource);
ParameterMap.AddParameterAllocation(
*ParameterName,
0,
Offset,
NumUAVs
);
Header.Bindings.NumUAVs = FMath::Max<uint8>(
Header.Bindings.NumUAVs,
Offset + NumUAVs
);
verify(Match(ShaderSource, ')'));
if (Match(ShaderSource, '\n'))
{
break;
}
// Skip the comma.
verify(Match(ShaderSource, ','));
}
}
if (FCStringAnsi::Strncmp(ShaderSource, NumThreadsPrefix, NumThreadsPrefixLen) == 0)
{
ShaderSource += NumThreadsPrefixLen;
Header.NumThreadsX = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ','));
Match(ShaderSource, ' ');
Header.NumThreadsY = ParseNumber(ShaderSource);
verify(Match(ShaderSource, ','));
Match(ShaderSource, ' ');
Header.NumThreadsZ = ParseNumber(ShaderSource);
verify(Match(ShaderSource, '\n'));
}
// Build the SRT for this shader.
{
// Build the generic SRT for this shader.
FShaderResourceTable GenericSRT;
BuildResourceTableMapping(ShaderInput.Environment.ResourceTableMap, ShaderInput.Environment.ResourceTableLayoutHashes, UsedUniformBufferSlots, ShaderOutput.ParameterMap, GenericSRT);
// Copy over the bits indicating which resource tables are active.
Header.Bindings.ShaderResourceTable.ResourceTableBits = GenericSRT.ResourceTableBits;
Header.Bindings.ShaderResourceTable.ResourceTableLayoutHashes = GenericSRT.ResourceTableLayoutHashes;
// Now build our token streams.
BuildResourceTableTokenStream(GenericSRT.TextureMap, GenericSRT.MaxBoundResourceTable, Header.Bindings.ShaderResourceTable.TextureMap);
BuildResourceTableTokenStream(GenericSRT.ShaderResourceViewMap, GenericSRT.MaxBoundResourceTable, Header.Bindings.ShaderResourceTable.ShaderResourceViewMap);
BuildResourceTableTokenStream(GenericSRT.SamplerMap, GenericSRT.MaxBoundResourceTable, Header.Bindings.ShaderResourceTable.SamplerMap);
BuildResourceTableTokenStream(GenericSRT.UnorderedAccessViewMap, GenericSRT.MaxBoundResourceTable, Header.Bindings.ShaderResourceTable.UnorderedAccessViewMap);
Header.Bindings.NumUniformBuffers = FMath::Max((uint8)GetNumUniformBuffersUsed(GenericSRT), Header.Bindings.NumUniformBuffers);
}
const int32 MaxSamplers = GetFeatureLevelMaxTextureSamplers(ERHIFeatureLevel::ES3_1);
if (Header.Bindings.NumSamplers > MaxSamplers)
{
ShaderOutput.bSucceeded = false;
FShaderCompilerError* NewError = new(ShaderOutput.Errors) FShaderCompilerError();
NewError->StrippedErrorMessage =
FString::Printf(TEXT("shader uses %d samplers exceeding the limit of %d"),
Header.Bindings.NumSamplers, MaxSamplers);
}
else
{
#if METAL_OFFLINE_COMPILE
// at this point, the shader source is ready to be compiled
FString InputFilename = FPaths::CreateTempFilename(*FPaths::EngineIntermediateDir(), TEXT("ShaderIn"), TEXT(""));
FString ObjFilename = InputFilename + TEXT(".o");
FString ArFilename = InputFilename + TEXT(".ar");
FString OutputFilename = InputFilename + TEXT(".lib");
InputFilename = InputFilename + TEXT(".metal");
// write out shader source
FFileHelper::SaveStringToFile(FString(ShaderSource), *InputFilename);
int32 ReturnCode = 0;
FString Results;
FString Errors;
bool bHadError = true;
// metal commandlines
FString Params = FString::Printf(TEXT("-std=ios-metal1.0 %s -o %s"), *InputFilename, *ObjFilename);
FPlatformProcess::ExecProcess( TEXT("/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/usr/bin/metal"), *Params, &ReturnCode, &Results, &Errors );
// handle compile error
if (ReturnCode != 0 || IFileManager::Get().FileSize(*ObjFilename) <= 0)
{
// FShaderCompilerError* Error = new(OutErrors) FShaderCompilerError();
// Error->ErrorFile = InputFilename;
// Error->ErrorLineString = TEXT("0");
// Error->StrippedErrorMessage = Results + Errors;
}
else
{
Params = FString::Printf(TEXT("r %s %s"), *ArFilename, *ObjFilename);
FPlatformProcess::ExecProcess( TEXT("/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/usr/bin/metal-ar"), *Params, &ReturnCode, &Results, &Errors );
// handle compile error
if (ReturnCode != 0 || IFileManager::Get().FileSize(*ArFilename) <= 0)
{
// FShaderCompilerError* Error = new(OutErrors) FShaderCompilerError();
// Error->ErrorFile = InputFilename;
// Error->ErrorLineString = TEXT("0");
// Error->StrippedErrorMessage = Results + Errors;
}
else
{
Params = FString::Printf(TEXT("-o %s %s"), *OutputFilename, *ArFilename);
FPlatformProcess::ExecProcess( TEXT("/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/usr/bin/metallib"), *Params, &ReturnCode, &Results, &Errors );
// handle compile error
if (ReturnCode != 0 || IFileManager::Get().FileSize(*OutputFilename) <= 0)
{
// FShaderCompilerError* Error = new(OutErrors) FShaderCompilerError();
// Error->ErrorFile = InputFilename;
// Error->ErrorLineString = TEXT("0");
// Error->StrippedErrorMessage = Results + Errors;
}
else
{
bHadError = false;
// Write out the header and compiled shader code
FMemoryWriter Ar(ShaderOutput.Code, true);
uint8 PrecompiledFlag = 1;
Ar << PrecompiledFlag;
Ar << Header;
// load output
TArray<uint8> CompiledShader;
FFileHelper::LoadFileToArray(CompiledShader, *OutputFilename);
// jam it into the output bytes
Ar.Serialize(CompiledShader.GetData(), CompiledShader.Num());
ShaderOutput.NumInstructions = 0;
ShaderOutput.NumTextureSamplers = Header.Bindings.NumSamplers;
ShaderOutput.bSucceeded = true;
}
}
}
if (bHadError)
{
// Write out the header and shader source code.
FMemoryWriter Ar(ShaderOutput.Code, true);
uint8 PrecompiledFlag = 0;
Ar << PrecompiledFlag;
Ar << Header;
Ar.Serialize((void*)ShaderSource, SourceLen + 1 - (ShaderSource - InShaderSource));
ShaderOutput.NumInstructions = 0;
ShaderOutput.NumTextureSamplers = Header.Bindings.NumSamplers;
ShaderOutput.bSucceeded = true;
}
IFileManager::Get().Delete(*InputFilename);
IFileManager::Get().Delete(*ObjFilename);
IFileManager::Get().Delete(*ArFilename);
IFileManager::Get().Delete(*OutputFilename);
#else
// Write out the header and shader source code.
FMemoryWriter Ar(ShaderOutput.Code, true);
uint8 PrecompiledFlag = 0;
Ar << PrecompiledFlag;
Ar << Header;
Ar.Serialize((void*)ShaderSource, SourceLen + 1 - (ShaderSource - InShaderSource));
ShaderOutput.NumInstructions = 0;
ShaderOutput.NumTextureSamplers = Header.Bindings.NumSamplers;
ShaderOutput.bSucceeded = true;
#endif
}
}
