预测查询

  • 项目

D3D12PredicationQueries 示例演示了使用 DirectX 12 查询堆和预测进行遮挡剔除。 本演练介绍了扩展 HelloConstBuffer 示例以处理谓词查询所需的其他代码。

创建深度模具描述符堆和遮挡查询堆

在 LoadPipeline 方法中创建一个深度模具描述符堆。

              // Describe and create a depth stencil view (DSV) descriptor heap.
              D3D12_DESCRIPTOR_HEAP_DESC dsvHeapDesc = {};
              dsvHeapDesc.NumDescriptors = 1;
              dsvHeapDesc.Type = D3D12_DESCRIPTOR_HEAP_TYPE_DSV;
              dsvHeapDesc.Flags = D3D12_DESCRIPTOR_HEAP_FLAG_NONE;
              ThrowIfFailed(m_device->CreateDescriptorHeap(&dsvHeapDesc, IID_PPV_ARGS(&m_dsvHeap)));
调用流程 参数
D3D12_DESCRIPTOR_HEAP_DESC
D3D12_DESCRIPTOR_HEAP_TYPE
[D3D12_DESCRIPTOR_HEAP_FLAG](/windows/desktop/api/d3d12/ne-d3d12-d3d12_descriptor_heap_flags)
CreateDescriptorHeap

 

在 LoadAssets 方法中，为遮挡查询创建一个堆。

     // Describe and create a heap for occlusion queries.
              D3D12_QUERY_HEAP_DESC queryHeapDesc = {};
              queryHeapDesc.Count = 1;
              queryHeapDesc.Type = D3D12_QUERY_HEAP_TYPE_OCCLUSION;
              ThrowIfFailed(m_device->CreateQueryHeap(&queryHeapDesc, IID_PPV_ARGS(&m_queryHeap)));
调用流程 参数
D3D12_QUERY_HEAP_DESC D3D12_QUERY_HEAP_TYPE
CreateQueryHeap

 

启用 alpha 值混合处理

该示例绘制了两个四边形，并演示了一个二进制遮挡查询。 前面的四边形动态显示在屏幕上，后面的四边形偶尔会被遮挡。 在 LoadAssets 方法中，为此示例启用了 alpha 值混合处理，以便我们能看到 D3D 在哪些情况下将后面的四边形视为被遮挡。

     // Enable alpha blending so we can visualize the occlusion query results.
              CD3DX12_BLEND_DESC blendDesc(CD3DX12_DEFAULT);
              blendDesc.RenderTarget[0] =
              {
                     TRUE, FALSE,
                     D3D12_BLEND_SRC_ALPHA, D3D12_BLEND_INV_SRC_ALPHA, D3D12_BLEND_OP_ADD,
                     D3D12_BLEND_ONE, D3D12_BLEND_ZERO, D3D12_BLEND_OP_ADD,
                     D3D12_LOGIC_OP_NOOP,
                     D3D12_COLOR_WRITE_ENABLE_ALL,
              };
调用流程 参数
CD3DX12_BLEND_DESC
CD3DX12_DEFAULT
[D3D12_BLEND](/windows/desktop/api/d3d12/ne-d3d12-d3d12_blend)
[D3D12_BLEND_OP](/windows/desktop/api/d3d12/ne-d3d12-d3d12_blend_op)
[D3D12_LOGIC_OP](/windows/desktop/api/d3d12/ne-d3d12-d3d12_logic_op)
[D3D12_COLOR_WRITE_ENABLE](/windows/desktop/api/d3d12/ne-d3d12-d3d12_color_write_enable)

 

禁用颜色和深度写入

遮挡查询是通过呈现四边形来执行的，该四边形覆盖的区域与我们想要测试其可见性的四边形相同。 在更复杂的场景中，查询可能是一个包围盒，而不是简单的四边形。 无论是哪种情况，都会创建一个新的管道状态，禁止写入呈现目标和 Z 缓冲区，以便遮挡查询本身不会影响呈现通道的可见输出。

在 LoadAssets 方法中，禁用遮挡查询状态的颜色写入和深度写入。

 // Disable color writes and depth writes for the occlusion query's state.
              psoDesc.BlendState.RenderTarget[0].RenderTargetWriteMask = 0;
              psoDesc.DepthStencilState.DepthWriteMask = D3D12_DEPTH_WRITE_MASK_ZERO;

              ThrowIfFailed(m_device->CreateGraphicsPipelineState(&psoDesc, IID_PPV_ARGS(&m_queryState)));
调用流程 参数
D3D12_GRAPHICS_PIPELINE_STATE_DESC D3D12_DEPTH_WRITE_MASK
CreateGraphicsPipelineState

 

创建用于存储查询结果的缓冲区

需要在 LoadAssets 方法中创建一个缓冲区来存储查询结果。 每个查询在 GPU 内存中需要 8 个字节的空间。 此示例仅执行一个查询，为简单起见，它创建了一个恰好具有此大小的缓冲区（即使此函数调用将分配 64K 页的 GPU 内存也是如此；大多数实际应用可能会创建一个更大的缓冲区）。

 // Create the query result buffer.
              CD3DX12_HEAP_PROPERTIES heapProps(D3D12_HEAP_TYPE_DEFAULT);
              auto queryBufferDesc = CD3DX12_RESOURCE_DESC::Buffer(8);
              ThrowIfFailed(m_device->CreateCommittedResource(
                     &heapProps,
                     D3D12_HEAP_FLAG_NONE,
                     &queryBufferDesc,
                     D3D12_RESOURCE_STATE_GENERIC_READ,
                     nullptr,
                     IID_PPV_ARGS(&m_queryResult)
                     ));
调用流程 参数
CreateCommittedResource
CD3DX12_HEAP_PROPERTIES
[D3D12_HEAP_TYPE](/windows/desktop/api/d3d12/ne-d3d12-d3d12_heap_type)
[D3D12_HEAP_FLAG](/windows/desktop/api/d3d12/ne-d3d12-d3d12_heap_flags)
[CD3DX12_RESOURCE_DESC](cd3dx12-resource-desc.md)
[D3D12_RESOURCE_STATES](/windows/desktop/api/d3d12/ne-d3d12-d3d12_resource_states)

 

绘制四边形并执行和解析遮挡查询

完成设置后，主循环会在 PopulateCommandLists 方法中更新。

1. 从后到前绘制象限，使透明度效果正常工作。 从后向前绘制四边形取决于前一帧查询的结果，这是一种相当常见的技术。 2. 更改 PSO 以禁用呈现目标和深度模具写入。 3. 执行封闭查询。 4. 解决封闭查询。 
       // Draw the quads and perform the occlusion query.
       {
              CD3DX12_GPU_DESCRIPTOR_HANDLE cbvFarQuad(m_cbvHeap->GetGPUDescriptorHandleForHeapStart(), m_frameIndex * CbvCountPerFrame, m_cbvSrvDescriptorSize);
              CD3DX12_GPU_DESCRIPTOR_HANDLE cbvNearQuad(cbvFarQuad, m_cbvSrvDescriptorSize);

              m_commandList->IASetPrimitiveTopology(D3D_PRIMITIVE_TOPOLOGY_TRIANGLESTRIP);
              m_commandList->IASetVertexBuffers(0, 1, &m_vertexBufferView);

              // Draw the far quad conditionally based on the result of the occlusion query
              // from the previous frame.
              m_commandList->SetGraphicsRootDescriptorTable(0, cbvFarQuad);
              m_commandList->SetPredication(m_queryResult.Get(), 0, D3D12_PREDICATION_OP_EQUAL_ZERO);
              m_commandList->DrawInstanced(4, 1, 0, 0);

              // Disable predication and always draw the near quad.
              m_commandList->SetPredication(nullptr, 0, D3D12_PREDICATION_OP_EQUAL_ZERO);
              m_commandList->SetGraphicsRootDescriptorTable(0, cbvNearQuad);
              m_commandList->DrawInstanced(4, 1, 4, 0);

              // Run the occlusion query with the bounding box quad.
              m_commandList->SetGraphicsRootDescriptorTable(0, cbvFarQuad);
              m_commandList->SetPipelineState(m_queryState.Get());
              m_commandList->BeginQuery(m_queryHeap.Get(), D3D12_QUERY_TYPE_BINARY_OCCLUSION, 0);
              m_commandList->DrawInstanced(4, 1, 8, 0);
              m_commandList->EndQuery(m_queryHeap.Get(), D3D12_QUERY_TYPE_BINARY_OCCLUSION, 0);

              // Resolve the occlusion query and store the results in the query result buffer
              // to be used on the subsequent frame.
              m_commandList->ResourceBarrier(1, &CD3DX12_RESOURCE_BARRIER::Transition(m_queryResult.Get(), D3D12_RESOURCE_STATE_GENERIC_READ, D3D12_RESOURCE_STATE_COPY_DEST));
              m_commandList->ResolveQueryData(m_queryHeap.Get(), D3D12_QUERY_TYPE_BINARY_OCCLUSION, 0, 1, m_queryResult.Get(), 0);
              m_commandList->ResourceBarrier(1, &CD3DX12_RESOURCE_BARRIER::Transition(m_queryResult.Get(), D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_GENERIC_READ));
       }
调用流程 参数
CD3DX12_GPU_DESCRIPTOR_HANDLE GetGPUDescriptorHandleForHeapStart
IASetPrimitiveTopology D3D_PRIMITIVE_TOPOLOGY
IASetVertexBuffers
SetGraphicsRootDescriptorTable
SetPredication D3D12_PREDICATION_OP
DrawInstanced
SetPredication D3D12_PREDICATION_OP
SetGraphicsRootDescriptorTable
DrawInstanced
SetGraphicsRootDescriptorTable
SetPipelineState
BeginQuery D3D12_QUERY_TYPE
DrawInstanced
EndQuery D3D12_QUERY_TYPE
ResourceBarrier
CD3DX12_RESOURCE_BARRIER
[D3D12_RESOURCE_STATES](/windows/desktop/api/d3d12/ne-d3d12-d3d12_resource_states)
ResolveQueryData D3D12_QUERY_TYPE
ResourceBarrier
CD3DX12_RESOURCE_BARRIER
[D3D12_RESOURCE_STATES](/windows/desktop/api/d3d12/ne-d3d12-d3d12_resource_states)

 

运行示例

未遮挡：

两个框均未被遮挡

遮挡：

完全遮挡一个框

部分遮挡：

一个框被部分遮挡

D3D12 代码演练

预测

查询