void FDrawEvent::Start(FRHICommandList& InRHICmdList, const FColor& Color, const TCHAR* Fmt, ...)
{
check(IsInParallelRenderingThread() || IsInRHIThread());
{
va_list ptr;
va_start(ptr, Fmt);
TCHAR TempStr[256];
// Build the string in the temp buffer
FCString::GetVarArgs(TempStr, ARRAY_COUNT(TempStr), ARRAY_COUNT(TempStr) - 1, Fmt, ptr);
InRHICmdList.PushEvent(TempStr);
RHICmdList = &InRHICmdList;
}
}
void FDrawEventRHIExecute::Start(IRHIComputeContext& InRHICommandContext, FColor Color, const TCHAR* Fmt, ...)
{
check(IsInParallelRenderingThread() || IsInRHIThread() || (!GRHIThread && IsInRenderingThread()));
{
va_list ptr;
va_start(ptr, Fmt);
TCHAR TempStr[256];
// Build the string in the temp buffer
FCString::GetVarArgs(TempStr, ARRAY_COUNT(TempStr), ARRAY_COUNT(TempStr) - 1, Fmt, ptr);
RHICommandContext = &InRHICommandContext;
RHICommandContext->RHIPushEvent(TempStr, Color);
}
}
/**
* Creates a bound shader state instance which encapsulates a decl, vertex shader, and pixel shader
* @param VertexDeclaration - existing vertex decl
* @param StreamStrides - optional stream strides
* @param VertexShader - existing vertex shader
* @param HullShader - existing hull shader
* @param DomainShader - existing domain shader
* @param PixelShader - existing pixel shader
* @param GeometryShader - existing geometry shader
*/
FBoundShaderStateRHIRef FD3D12DynamicRHI::RHICreateBoundShaderState(
FVertexDeclarationRHIParamRef VertexDeclarationRHI,
FVertexShaderRHIParamRef VertexShaderRHI,
FHullShaderRHIParamRef HullShaderRHI,
FDomainShaderRHIParamRef DomainShaderRHI,
FPixelShaderRHIParamRef PixelShaderRHI,
FGeometryShaderRHIParamRef GeometryShaderRHI
)
{
SCOPE_CYCLE_COUNTER(STAT_D3D12CreateBoundShaderStateTime);
checkf(GIsRHIInitialized && GetRHIDevice()->GetCommandListManager().IsReady(), (TEXT("Bound shader state RHI resource was created without initializing Direct3D first")));
#if D3D12_SUPPORTS_PARALLEL_RHI_EXECUTE
// Check for an existing bound shader state which matches the parameters
FBoundShaderStateRHIRef CachedBoundShaderState = GetCachedBoundShaderState_Threadsafe(
VertexDeclarationRHI,
VertexShaderRHI,
PixelShaderRHI,
HullShaderRHI,
DomainShaderRHI,
GeometryShaderRHI
);
if(CachedBoundShaderState.GetReference())
{
// If we've already created a bound shader state with these parameters, reuse it.
return CachedBoundShaderState;
}
#else
check(IsInRenderingThread() || IsInRHIThread());
// Check for an existing bound shader state which matches the parameters
FCachedBoundShaderStateLink* CachedBoundShaderStateLink = GetCachedBoundShaderState(
VertexDeclarationRHI,
VertexShaderRHI,
PixelShaderRHI,
HullShaderRHI,
DomainShaderRHI,
GeometryShaderRHI
);
if(CachedBoundShaderStateLink)
{
// If we've already created a bound shader state with these parameters, reuse it.
return CachedBoundShaderStateLink->BoundShaderState;
}
#endif
else
{
SCOPE_CYCLE_COUNTER(STAT_D3D12NewBoundShaderStateTime);
return new FD3D12BoundShaderState(VertexDeclarationRHI, VertexShaderRHI, PixelShaderRHI, HullShaderRHI, DomainShaderRHI, GeometryShaderRHI);
}
}
FUniformBufferRHIRef FD3D11DynamicRHI::RHICreateUniformBuffer(const void* Contents,const FRHIUniformBufferLayout& Layout,EUniformBufferUsage Usage)
{
check(IsInRenderingThread() || IsInRHIThread());
FD3D11UniformBuffer* NewUniformBuffer = nullptr;
const uint32 NumBytes = Layout.ConstantBufferSize;
if (NumBytes > 0)
{
// Constant buffers must also be 16-byte aligned.
check(Align(NumBytes,16) == NumBytes);
check(Align(Contents,16) == Contents);
check(NumBytes <= D3D11_REQ_CONSTANT_BUFFER_ELEMENT_COUNT * 16);
check(NumBytes < (1 << NumPoolBuckets));
SCOPE_CYCLE_COUNTER(STAT_D3D11UpdateUniformBufferTime);
if (IsPoolingEnabled())
{
TRefCountPtr<ID3D11Buffer> UniformBufferResource;
FRingAllocation RingAllocation;
if (!RingAllocation.IsValid())
{
// Find the appropriate bucket based on size
const uint32 BucketIndex = GetPoolBucketIndex(NumBytes);
TArray<FPooledUniformBuffer>& PoolBucket = UniformBufferPool[BucketIndex];
if (PoolBucket.Num() > 0)
{
// Reuse the last entry in this size bucket
FPooledUniformBuffer FreeBufferEntry = PoolBucket.Pop();
check(IsValidRef(FreeBufferEntry.Buffer));
UniformBufferResource = FreeBufferEntry.Buffer;
checkf(FreeBufferEntry.CreatedSize >= NumBytes, TEXT("%u %u %u %u"), NumBytes, BucketIndex, FreeBufferEntry.CreatedSize, GetPoolBucketSize(NumBytes));
DEC_DWORD_STAT(STAT_D3D11NumFreeUniformBuffers);
DEC_MEMORY_STAT_BY(STAT_D3D11FreeUniformBufferMemory, FreeBufferEntry.CreatedSize);
}
// Nothing usable was found in the free pool, create a new uniform buffer
if (!IsValidRef(UniformBufferResource))
{
D3D11_BUFFER_DESC Desc;
// Allocate based on the bucket size, since this uniform buffer will be reused later
Desc.ByteWidth = GetPoolBucketSize(NumBytes);
// Use D3D11_USAGE_DYNAMIC, which allows multiple CPU writes for pool reuses
// This is method of updating is vastly superior to creating a new constant buffer each time with D3D11_USAGE_IMMUTABLE,
// Since that inserts the data into the command buffer which causes GPU flushes
Desc.Usage = D3D11_USAGE_DYNAMIC;
Desc.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
Desc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
Desc.MiscFlags = 0;
Desc.StructureByteStride = 0;
VERIFYD3D11RESULT(Direct3DDevice->CreateBuffer(&Desc, NULL, UniformBufferResource.GetInitReference()));
UpdateBufferStats(UniformBufferResource, true);
}
check(IsValidRef(UniformBufferResource));
D3D11_MAPPED_SUBRESOURCE MappedSubresource;
// Discard previous results since we always do a full update
VERIFYD3D11RESULT(Direct3DDeviceIMContext->Map(UniformBufferResource, 0, D3D11_MAP_WRITE_DISCARD, 0, &MappedSubresource));
check(MappedSubresource.RowPitch >= NumBytes);
FMemory::Memcpy(MappedSubresource.pData, Contents, NumBytes);
Direct3DDeviceIMContext->Unmap(UniformBufferResource, 0);
}
NewUniformBuffer = new FD3D11UniformBuffer(this, Layout, UniformBufferResource, RingAllocation);
}
else
{
// No pooling
D3D11_BUFFER_DESC Desc;
Desc.ByteWidth = NumBytes;
Desc.Usage = D3D11_USAGE_IMMUTABLE;
Desc.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
Desc.CPUAccessFlags = 0;
Desc.MiscFlags = 0;
Desc.StructureByteStride = 0;
D3D11_SUBRESOURCE_DATA ImmutableData;
ImmutableData.pSysMem = Contents;
ImmutableData.SysMemPitch = ImmutableData.SysMemSlicePitch = 0;
TRefCountPtr<ID3D11Buffer> UniformBufferResource;
VERIFYD3D11RESULT(Direct3DDevice->CreateBuffer(&Desc,&ImmutableData,UniformBufferResource.GetInitReference()));
NewUniformBuffer = new FD3D11UniformBuffer(this, Layout, UniformBufferResource, FRingAllocation());
}
}
else
{
// This uniform buffer contains no constants, only a resource table.
NewUniformBuffer = new FD3D11UniformBuffer(this, Layout, nullptr, FRingAllocation());
}
if (Layout.Resources.Num())
{
int32 NumResources = Layout.Resources.Num();
FRHIResource** InResources = (FRHIResource**)((uint8*)Contents + Layout.ResourceOffset);
NewUniformBuffer->ResourceTable.Empty(NumResources);
NewUniformBuffer->ResourceTable.AddZeroed(NumResources);
for (int32 i = 0; i < NumResources; ++i)
{
check(InResources[i]);
NewUniformBuffer->ResourceTable[i] = InResources[i];
}
NewUniformBuffer->RawResourceTable.Empty(NumResources);
NewUniformBuffer->RawResourceTable.AddZeroed(NumResources);
}
return NewUniformBuffer;
}
void* FD3D12DynamicRHI::LockBuffer(FRHICommandListImmediate* RHICmdList, BufferType* Buffer, uint32 Offset, uint32 Size, EResourceLockMode LockMode)
{
FD3D12LockedData LockedData;
// Determine whether the buffer is dynamic or not.
const bool bIsDynamic = (Buffer->GetUsage() & BUF_AnyDynamic) ? true : false;
void* Data = nullptr;
if (bIsDynamic)
{
check(LockMode == RLM_WriteOnly);
TRefCountPtr<FD3D12ResourceLocation> newLocation = new FD3D12ResourceLocation(GetRHIDevice());
// Allocate a new resource
Data = GetRHIDevice()->GetDefaultUploadHeapAllocator().AllocUploadResource(Buffer->ResourceLocation->GetEffectiveBufferSize(), Buffer->BufferAlignment, newLocation);
// If on the RenderThread, queue up a command on the RHIThread to rename this buffer at the correct time
if (ShouldDeferBufferLockOperation(RHICmdList))
{
new (RHICmdList->AllocCommand<FRHICommandRenameUploadBuffer<BufferType>>()) FRHICommandRenameUploadBuffer<BufferType>(Buffer, newLocation);
}
else
{
Buffer->Rename(newLocation);
}
}
else
{
FD3D12Resource* pResource = Buffer->ResourceLocation->GetResource();
// Locking for read must occur immediately so we can't queue up the operations later.
if (LockMode == RLM_ReadOnly)
{
// If the static buffer is being locked for reading, create a staging buffer.
TRefCountPtr<FD3D12Resource> StagingBuffer;
VERIFYD3D11RESULT(GetRHIDevice()->GetResourceHelper().CreateBuffer(D3D12_HEAP_TYPE_READBACK, Offset + Size, StagingBuffer.GetInitReference()));
LockedData.StagingResource = StagingBuffer;
// Copy the contents of the buffer to the staging buffer.
{
const auto& pfnCopyContents = [&]()
{
FD3D12CommandContext& DefaultContext = GetRHIDevice()->GetDefaultCommandContext();
FD3D12CommandListHandle& hCommandList = DefaultContext.CommandListHandle;
FScopeResourceBarrier ScopeResourceBarrierSource(hCommandList, pResource, pResource->GetDefaultResourceState(), D3D12_RESOURCE_STATE_COPY_SOURCE, 0);
// Don't need to transition upload heaps
DefaultContext.numCopies++;
hCommandList->CopyBufferRegion(
StagingBuffer->GetResource(),
0,
pResource->GetResource(),
Offset, Size);
DefaultContext.FlushCommands(true);
};
if (ShouldDeferBufferLockOperation(RHICmdList))
{
// Sync when in the render thread implementation
check(IsInRHIThread() == false);
RHICmdList->ImmediateFlush(EImmediateFlushType::FlushRHIThread);
pfnCopyContents();
}
else
{
check(IsInRHIThread());
pfnCopyContents();
}
}
// Map the staging buffer's memory for reading.
VERIFYD3D11RESULT(StagingBuffer->GetResource()->Map(0, nullptr, &Data));
}
else
{
// If the static buffer is being locked for writing, allocate memory for the contents to be written to.
TRefCountPtr<FD3D12ResourceLocation> UploadBufferLocation = new FD3D12ResourceLocation(GetRHIDevice());
Data = GetRHIDevice()->GetDefaultFastAllocator().Allocate(Offset + Size, D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT, UploadBufferLocation);
// Keep track of the underlying resource for the upload heap so it can be referenced during Unmap.
LockedData.UploadHeapLocation = UploadBufferLocation;
}
}
check(Data);
// Add the lock to the lock map.
LockedData.SetData(Data);
LockedData.Pitch = Offset + Size;
// Add the lock to the lock map.
FD3D12LockedKey LockedKey(Buffer);
AddToOutstandingLocks(LockedKey, LockedData);
// Return the offset pointer
return (void*)((uint8*)LockedData.GetData() + Offset);
}
void* FD3D12DynamicRHI::LockBuffer(FRHICommandListImmediate* RHICmdList, BufferType* Buffer, uint32 Offset, uint32 Size, EResourceLockMode LockMode)
{
#if STATS
LockBufferCalls++;
SCOPE_CYCLE_COUNTER(STAT_D3D12LockBufferTime);
INC_DWORD_STAT_BY(STAT_D3D12LockBufferCalls, LockBufferCalls);
#endif
FD3D12LockedResource& LockedData = Buffer->LockedData;
check(LockedData.bLocked == false);
FD3D12Device* Device = GetRHIDevice();
FD3D12Adapter& Adapter = GetAdapter();
// Determine whether the buffer is dynamic or not.
const bool bIsDynamic = (Buffer->GetUsage() & BUF_AnyDynamic) ? true : false;
void* Data = nullptr;
if (bIsDynamic)
{
check(LockMode == RLM_WriteOnly);
BufferType* CurrentBuffer = Buffer;
// Update all of the resources in the LDA chain
while (CurrentBuffer)
{
// Allocate a new resource
// If on the RenderThread, queue up a command on the RHIThread to rename this buffer at the correct time
if (ShouldDeferBufferLockOperation(RHICmdList))
{
FRHICommandRenameUploadBuffer<BufferType>* Command = new (RHICmdList->AllocCommand<FRHICommandRenameUploadBuffer<BufferType>>()) FRHICommandRenameUploadBuffer<BufferType>(CurrentBuffer, Device);
Data = Adapter.GetUploadHeapAllocator().AllocUploadResource(Buffer->GetSize(), Buffer->BufferAlignment, Command->NewResource);
}
else
{
FD3D12ResourceLocation Location(CurrentBuffer->GetParentDevice());
Data = Adapter.GetUploadHeapAllocator().AllocUploadResource(Buffer->GetSize(), Buffer->BufferAlignment, Location);
CurrentBuffer->Rename(Location);
}
CurrentBuffer = CurrentBuffer->GetNextObject();
}
}
else
{
FD3D12Resource* pResource = Buffer->ResourceLocation.GetResource();
// Locking for read must occur immediately so we can't queue up the operations later.
if (LockMode == RLM_ReadOnly)
{
LockedData.bLockedForReadOnly = true;
// If the static buffer is being locked for reading, create a staging buffer.
FD3D12Resource* StagingBuffer = nullptr;
const GPUNodeMask Node = Device->GetNodeMask();
VERIFYD3D12RESULT(Adapter.CreateBuffer(D3D12_HEAP_TYPE_READBACK, Node, Node, Offset + Size, &StagingBuffer));
// Copy the contents of the buffer to the staging buffer.
{
const auto& pfnCopyContents = [&]()
{
FD3D12CommandContext& DefaultContext = Device->GetDefaultCommandContext();
FD3D12CommandListHandle& hCommandList = DefaultContext.CommandListHandle;
FScopeResourceBarrier ScopeResourceBarrierSource(hCommandList, pResource, pResource->GetDefaultResourceState(), D3D12_RESOURCE_STATE_COPY_SOURCE, 0);
// Don't need to transition upload heaps
DefaultContext.numCopies++;
hCommandList->CopyBufferRegion(
StagingBuffer->GetResource(),
0,
pResource->GetResource(),
Offset, Size);
hCommandList.UpdateResidency(StagingBuffer);
hCommandList.UpdateResidency(pResource);
DefaultContext.FlushCommands(true);
};
if (ShouldDeferBufferLockOperation(RHICmdList))
{
// Sync when in the render thread implementation
check(IsInRHIThread() == false);
RHICmdList->ImmediateFlush(EImmediateFlushType::FlushRHIThread);
pfnCopyContents();
}
else
{
check(IsInRHIThread());
pfnCopyContents();
}
}
LockedData.ResourceLocation.AsStandAlone(StagingBuffer, Size);
Data = LockedData.ResourceLocation.GetMappedBaseAddress();
}
else
{
// If the static buffer is being locked for writing, allocate memory for the contents to be written to.
Data = Device->GetDefaultFastAllocator().Allocate<FD3D12ScopeLock>(Size, D3D12_TEXTURE_DATA_PLACEMENT_ALIGNMENT, &LockedData.ResourceLocation);
}
}
LockedData.LockedOffset = Offset;
LockedData.LockedPitch = Size;
LockedData.bLocked = true;
// Return the offset pointer
check(Data != nullptr);
return Data;
}