Compute Pipeline(컴퓨트 파이프라인)

// Root Signature 1.0 학습 예시: u0 UAV table 하나. D3D12_DESCRIPTOR_RANGE range = {}; range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV; range.NumDescriptors = 1; range.BaseShaderRegister = 0; D3D12_ROOT_PARAMETER param = {}; param.ParameterType = D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE; param.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL; param.DescriptorTable = {1, &range}; D3D12_ROOT_SIGNATURE_DESC desc = {}; desc.NumParameters = 1; desc.pParameters = &param; ComPtr<ID3DBlob> signature, error; HRESULT hr = D3D12SerializeRootSignature(&desc, D3D_ROOT_SIGNATURE_VERSION_1, signature.GetAddressOf(), error.GetAddressOf()); if (error) OutputDebugStringA(static_cast<const char*>(error->GetBufferPointer())); ThrowIfFailed(hr); ComPtr<ID3D12RootSignature> rootSignature; ThrowIfFailed(device->CreateRootSignature(0, signature->GetBufferPointer(), signature->GetBufferSize(), IID_PPV_ARGS(&rootSignature)));

// HLSL: main을 cs_5_0 등 이 예제에 맞는 프로파일로 컴파일합니다. RWTexture2D<float4> outputTexture : register(u0); [numthreads(16, 16, 1)] void main(uint3 localID : SV_GroupThreadID, uint3 globalID : SV_DispatchThreadID) { uint width, height; outputTexture.GetDimensions(width, height); if (globalID.x >= width || globalID.y >= height) return; outputTexture[globalID.xy] = float4( float(localID.x) / 16.0, float(localID.y) / 16.0, float(globalID.x) / float(width), 1.0); } // SV_GroupID는 그룹 번호, SV_GroupThreadID는 그룹 안의 좌표입니다. // SV_GroupIndex는 그룹 안의 스레드를 일렬로 센 번호입니다.

// compShader는 성공적으로 컴파일한 CS 바이트코드입니다. D3D12_COMPUTE_PIPELINE_STATE_DESC psoDesc = {}; psoDesc.pRootSignature = rootSignature.Get(); psoDesc.CS = {compShader->GetBufferPointer(), compShader->GetBufferSize()}; ComPtr<ID3D12PipelineState> pipelineState; ThrowIfFailed(device->CreateComputePipelineState(&psoDesc, IID_PPV_ARGS(&pipelineState))); // 실패를 출력만 하고 Dispatch로 진행하지 않습니다.

// d3dx12.h 사용. width/height > 0, 백버퍼와 같은 크기·포맷·샘플 수를 전제합니다. D3D12_DESCRIPTOR_HEAP_DESC heapDesc = {}; heapDesc.NumDescriptors = 1; heapDesc.Type = D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV; heapDesc.Flags = D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE; ComPtr<ID3D12DescriptorHeap> uavHeap; ThrowIfFailed(device->CreateDescriptorHeap(&heapDesc, IID_PPV_ARGS(&uavHeap))); const auto props = CD3DX12_HEAP_PROPERTIES(D3D12_HEAP_TYPE_DEFAULT); const auto texDesc = CD3DX12_RESOURCE_DESC::Tex2D( DXGI_FORMAT_R8G8B8A8_UNORM, width, height, 1, 1, 1, 0, D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS); ComPtr<ID3D12Resource> outputTexture; ThrowIfFailed(device->CreateCommittedResource(&props, D3D12_HEAP_FLAG_NONE, &texDesc, D3D12_RESOURCE_STATE_UNORDERED_ACCESS, nullptr, IID_PPV_ARGS(&outputTexture))); D3D12_UNORDERED_ACCESS_VIEW_DESC uavDesc = {}; uavDesc.Format = texDesc.Format; uavDesc.ViewDimension = D3D12_UAV_DIMENSION_TEXTURE2D; const auto uavCPU = uavHeap->GetCPUDescriptorHandleForHeapStart(); const auto uavGPU = uavHeap->GetGPUDescriptorHandleForHeapStart(); device->CreateUnorderedAccessView(outputTexture.Get(), nullptr, &uavDesc, uavCPU); // 이 예제는 descriptor 1개를 사용합니다. CPU/GPU handle은 같은 슬롯을 가리킵니다.

// DIRECT Queue/List에서 Compute와 백버퍼 복사를 함께 기록하는 함수 내부입니다. // 이전 GPU 사용 완료 뒤 allocator Reset. backBuffer는 현재 백버퍼 raw pointer. ThrowIfFailed(commandAllocator->Reset()); ThrowIfFailed(commandList->Reset(commandAllocator, pipelineState.Get())); commandList->SetComputeRootSignature(rootSignature.Get()); ID3D12DescriptorHeap* heaps[] = {uavHeap.Get()}; commandList->SetDescriptorHeaps(1, heaps); commandList->SetComputeRootDescriptorTable(0, uavGPU); const auto groups = [](UINT value) { return value / 16 + (value % 16 != 0); }; commandList->Dispatch(groups(width), groups(height), 1); D3D12_RESOURCE_BARRIER beforeCopy[] = { CD3DX12_RESOURCE_BARRIER::Transition(outputTexture.Get(), D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE), CD3DX12_RESOURCE_BARRIER::Transition(backBuffer, D3D12_RESOURCE_STATE_PRESENT, D3D12_RESOURCE_STATE_COPY_DEST) }; commandList->ResourceBarrier(2, beforeCopy); commandList->CopyResource(backBuffer, outputTexture.Get()); D3D12_RESOURCE_BARRIER afterCopy[] = { CD3DX12_RESOURCE_BARRIER::Transition(outputTexture.Get(), D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_UNORDERED_ACCESS), CD3DX12_RESOURCE_BARRIER::Transition(backBuffer, D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_PRESENT) }; commandList->ResourceBarrier(2, afterCopy); ThrowIfFailed(commandList->Close()); // 이어서 ExecuteCommandLists, Present, Fence 처리를 수행합니다.