在 directx12 中将深度缓冲区值读回 CPU

Question

我想将深度缓冲区的内容复制回 CPU 以便能够读取它们。

我按以下方式创建深度缓冲区（视图）：

// Create the depth stencil view.
{
    D3D12_DEPTH_STENCIL_VIEW_DESC depthStencilDesc = {};
    depthStencilDesc.Format = DXGI_FORMAT_D32_FLOAT;
    depthStencilDesc.ViewDimension = D3D12_DSV_DIMENSION_TEXTURE2D;
    depthStencilDesc.Flags = D3D12_DSV_FLAG_NONE;

    D3D12_CLEAR_VALUE depthOptimizedClearValue = {};
    depthOptimizedClearValue.Format = DXGI_FORMAT_D32_FLOAT;
    depthOptimizedClearValue.DepthStencil.Depth = 0.0f;
    depthOptimizedClearValue.DepthStencil.Stencil = 0;

    ThrowIfFailed(m_device->CreateCommittedResource(
        &CD3DX12_HEAP_PROPERTIES(D3D12_HEAP_TYPE_DEFAULT),
        D3D12_HEAP_FLAG_NONE,
        &CD3DX12_RESOURCE_DESC::Tex2D(DXGI_FORMAT_D32_FLOAT, m_width, m_height, 1, 0, 1, 0, D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL),
        D3D12_RESOURCE_STATE_DEPTH_WRITE,
        &depthOptimizedClearValue,
        IID_PPV_ARGS(&m_depthStencil)
    ));

    NAME_D3D12_OBJECT(m_depthStencil);

    m_device->CreateDepthStencilView(m_depthStencil.Get(), &depthStencilDesc, m_dsvHeap->GetCPUDescriptorHandleForHeapStart());
}

对于管道，我对深度缓冲区使用以下描述：

D3D12_DEPTH_STENCIL_DESC depthDesc;
ZeroMemory(&depthDesc, sizeof(D3D12_DEPTH_STENCIL_DESC));
depthDesc.DepthEnable = TRUE;
depthDesc.DepthWriteMask = D3D12_DEPTH_WRITE_MASK_ALL;
depthDesc.DepthFunc = D3D12_COMPARISON_FUNC_ALWAYS;
depthDesc.StencilEnable = FALSE;

到目前为止一切正常，使用 NSight 我可以看到在 renderpass 完成后正确的值存储在深度缓冲区中。

现在我想读回 CPU 的值，如果我使用： D3D12_HEAP_PROPERTIES readbackHeapProperties{ CD3DX12_HEAP_PROPERTIES(D3D12_HEAP_TYPE_READBACK) };

堆类型回读我无法分配与深度缓冲区具有相同描述的缓冲区。但是，如果我使用 CD3DX12_HEAP_PROPERTIES(D3D12_HEAP_TYPE_DEFAULT)，我以后将无法“映射”内存（根据 Microsoft 文档 https://docs.microsoft.com/en-us/windows/win32/direct3d12/readback-data-using-heaps）

其余的复制我会做模拟记录的例子：

D3D12_RESOURCE_DESC desc;
ZeroMemory(&desc, sizeof(desc));
desc = m_depthStencil->GetDesc();

//D3D12_HEAP_PROPERTIES readbackHeapProperties{ CD3DX12_HEAP_PROPERTIES(D3D12_HEAP_TYPE_READBACK) };
//D3D12_RESOURCE_DESC readbackBufferDesc{ CD3DX12_RESOURCE_DESC::Buffer(500 * 500) };

D3D12_RESOURCE_DESC readbackBufferDesc = m_depthStencil->GetDesc();

ComPtr<ID3D12Resource> readbackBuffer;
ThrowIfFailed(m_device->CreateCommittedResource(
//&readbackHeapProperties,
&CD3DX12_HEAP_PROPERTIES(D3D12_HEAP_TYPE_DEFAULT),
D3D12_HEAP_FLAG_NONE,
&readbackBufferDesc,
D3D12_RESOURCE_STATE_COPY_DEST,
nullptr,
__uuidof(readbackBuffer),
&readbackBuffer));

使深度缓冲区进入正确状态。

{
    D3D12_RESOURCE_BARRIER outputBufferResourceBarrier
    {
        CD3DX12_RESOURCE_BARRIER::Transition(
            m_depthStencil.Get(),
            D3D12_RESOURCE_STATE_DEPTH_WRITE,
            D3D12_RESOURCE_STATE_COPY_SOURCE)
    };
    m_commandList->ResourceBarrier(1, &outputBufferResourceBarrier);
}

m_commandList->CopyResource(readbackBuffer.Get(), m_depthStencil.Get());

ThrowIfFailed(m_commandList->Close());

ID3D12CommandList* ppCommandLists[] = { m_commandList.Get() };
m_commandQueue->ExecuteCommandLists(_countof(ppCommandLists), ppCommandLists);



D3D12_RANGE readbackBufferRange{ 0, 500 * 500 }; // outputbuffer size = 500x500 = window size?
FLOAT* pReadbackBufferData{};
ThrowIfFailed(
    readbackBuffer->Map
    (
        0,
        &readbackBufferRange,
        reinterpret_cast<void**>(&pReadbackBufferData)
    )
);

D3D12_RANGE emptyRange{ 0, 0 };
readbackBuffer->Unmap
(
    0,
    &emptyRange
);

将深度缓冲区转换为原始状态。

{
    D3D12_RESOURCE_BARRIER outputBufferResourceBarrier
    {
        CD3DX12_RESOURCE_BARRIER::Transition(
            m_depthStencil.Get(),
            D3D12_RESOURCE_STATE_COPY_SOURCE,
            D3D12_RESOURCE_STATE_DEPTH_WRITE)
    };
    m_commandList->ResourceBarrier(1, &outputBufferResourceBarrier);
}

如有任何帮助，我将不胜感激！

Answer 1

在 DirectX 12 中，您不会创建与 GPU 资源格式相同的 'readback' 缓冲区。相反，您只需创建一个与源字节数相同的一维缓冲区。

'trick'就是你需要和原资源有相同的pitch。

CD3DX12_HEAP_PROPERTIES readBackHeapProperties(D3D12_HEAP_TYPE_READBACK);

// Readback resources must be buffers
D3D12_RESOURCE_DESC bufferDesc = {};
bufferDesc.DepthOrArraySize = 1;
bufferDesc.Dimension = D3D12_RESOURCE_DIMENSION_BUFFER;
bufferDesc.Flags = D3D12_RESOURCE_FLAG_NONE;
bufferDesc.Format = DXGI_FORMAT_UNKNOWN;
bufferDesc.Height = 1;
bufferDesc.Width = srcPitch * m_height; // <<----
bufferDesc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;
bufferDesc.MipLevels = 1;
bufferDesc.SampleDesc.Count = 1;

srcPitch获取自原始资源：

auto depthBufferDesc = CD3DX12_RESOURCE_DESC::Tex2D(DXGI_FORMAT_D32_FLOAT, m_width, m_height, 1, 0, 1, 0, D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL);

UINT64 totalResourceSize = 0;
UINT64 fpRowPitch = 0;
UINT fpRowCount = 0;
m_device->GetCopyableFootprints(
        &depthBufferDesc,
        0,
        1,
        0,
        nullptr,
        &fpRowCount,
        &fpRowPitch,
        &totalResourceSize);

// Round up the srcPitch to multiples of 256
UINT64 srcPitch = (fpRowPitch + 255) & ~0xFFu;

请参阅 DX12 的 DirectX 工具包中的 ScreenGrab。

I also filed an edit for that Microsoft Docs page. It uses an example of a buffer reading back to a buffer, but didn't mention the important detail that the read-back resource is always a buffer.

在 directx12 中将深度缓冲区值读回 CPU

Reading depth buffer values back to CPU in directx12

c++

directx

directx-12