在 DirectX 11 计算着色器中加载 Texture2D 数据

Question

我试图从 DirectX11 计算着色器中的 texture2d 读取一些数据，但是，即使纹理对象填充了相同的浮点数，texture2D 对象的 'Load' 函数仍然返回 0。

这是一个 160 * 120 的 texture2d DXGI_FORMAT_R32G32B32A32_FLOAT。以下代码是我创建此资源的方式：

HRESULT TestResources(ID3D11Device* pd3dDevice, ID3D11DeviceContext* pImmediateContext) {
    float *test = new float[4 * 80 * 60 * 4]; // 80 * 60, 4 channels, 1 big texture contains 4 80 * 60 subimage
    for (int i = 0; i < 4 * 80 * 60 * 4; i++) test[i] = 0.7f;

    HRESULT hr = S_OK;
    D3D11_TEXTURE2D_DESC RTtextureDesc;
    ZeroMemory(&RTtextureDesc, sizeof(D3D11_TEXTURE2D_DESC));
    RTtextureDesc.Width = 160;
    RTtextureDesc.Height = 120;
    RTtextureDesc.MipLevels = 1;
    RTtextureDesc.ArraySize = 1;
    RTtextureDesc.Format = DXGI_FORMAT_R32G32B32A32_FLOAT;

    RTtextureDesc.SampleDesc.Count = 1;
    RTtextureDesc.SampleDesc.Quality = 0;
    RTtextureDesc.Usage = D3D11_USAGE_DYNAMIC;
    RTtextureDesc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
    RTtextureDesc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
    RTtextureDesc.MiscFlags = 0;

    D3D11_SUBRESOURCE_DATA InitData;
    InitData.pSysMem = test;
    InitData.SysMemPitch = sizeof(float) * 4;
    V_RETURN(pd3dDevice->CreateTexture2D(&RTtextureDesc, &InitData, &m_pInputTex2Ds));
    //V_RETURN(pd3dDevice->CreateTexture2D(&RTtextureDesc, NULL, &m_pInputTex2Ds));

    D3D11_SHADER_RESOURCE_VIEW_DESC SRViewDesc;
    ZeroMemory(&SRViewDesc, sizeof(SRViewDesc));
    SRViewDesc.Format = RTtextureDesc.Format;
    SRViewDesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2D;
    SRViewDesc.Texture2D.MostDetailedMip = 0;
    SRViewDesc.Texture2D.MipLevels = 1;
    V_RETURN(pd3dDevice->CreateShaderResourceView(m_pInputTex2Ds, &SRViewDesc, &m_pInputTexSRV));

    delete[] test;
    return hr;
}

然后我尝试运行使用 X = Y = 2 和 Z = 1 发送，如下所示：

void ComputeShaderReduction::ExecuteComputeShader(ID3D11DeviceContext* pd3dImmediateContext, UINT uInputNum, ID3D11UnorderedAccessView** ppUAVInputs, UINT X, UINT Y, UINT Z) {
    pd3dImmediateContext->CSSetShader(m_pComputeShader, nullptr, 0);

    pd3dImmediateContext->CSSetShaderResources(0, 1, &m_pInputTexSRV); // test code

    pd3dImmediateContext->CSSetUnorderedAccessViews(0, uInputNum, ppUAVInputs, nullptr);
    //pd3dImmediateContext->CSSetUnorderedAccessViews(0, 1, &m_pGPUOutUAVs, nullptr);

    pd3dImmediateContext->UpdateSubresource(m_pConstBuf, 0, nullptr, &m_ConstBuf, 0, 0);
    pd3dImmediateContext->CSSetConstantBuffers(0, 1, &m_pConstBuf);
    pd3dImmediateContext->Dispatch(X, Y, Z);

    pd3dImmediateContext->CSSetShader(nullptr, nullptr, 0);

    ID3D11UnorderedAccessView* ppUAViewnullptr[1] = { nullptr };
    pd3dImmediateContext->CSSetUnorderedAccessViews(0, 1, ppUAViewnullptr, nullptr);

    ID3D11ShaderResourceView* ppSRVnullptr[1] = { nullptr };
    pd3dImmediateContext->CSSetShaderResources(0, 1, ppSRVnullptr);

    ID3D11Buffer* ppCBnullptr[1] = { nullptr };
    pd3dImmediateContext->CSSetConstantBuffers(0, 1, ppCBnullptr);
}

我写了一个非常简单的 CS 着色器来尝试获取 texture2d 中的数据并从中取出。所以，计算着色器看起来像这样：

#define subimg_dim_x 80
#define subimg_dim_y 60

Texture2D<float4> BufferIn : register(t0);
StructuredBuffer<float> Test: register(t1);
RWStructuredBuffer<float> BufferOut : register(u0);

groupshared float sdata[subimg_dim_x];

[numthreads(subimg_dim_x, 1, 1)]
void CSMain(uint3 DTid : SV_DispatchThreadID,
            uint3 threadIdx : SV_GroupThreadID,
            uint3 groupIdx  : SV_GroupID) {

    sdata[threadIdx.x] = 0.0;
    GroupMemoryBarrierWithGroupSync();

    if (threadIdx.x == 0) {
        float4 num = BufferIn.Load(uint3(groupIdx.x, groupIdx.y, 1));
        //BufferOut[groupIdx.y * 2 + groupIdx.x] = 2.0; //This one gives me 2.0 as output in the console
        BufferOut[groupIdx.y * 2 + groupIdx.x] = num.x; //This one keeps giving me 0.0 and in the texture, r = g = b = a = 0.7 or x = y = z = w = 0.7, so it suppose to print 0.7 in the console.
    }
    GroupMemoryBarrierWithGroupSync();
}

我认为我在 CPU 端打印 CS 着色器结果的方式是正确的。

void ComputeShaderReduction::CopyToCPUBuffer(ID3D11Device* pdevice, ID3D11DeviceContext* pd3dImmediateContext, ID3D11Buffer* pGPUOutBufs) {
    D3D11_BUFFER_DESC desc;
    ZeroMemory(&desc, sizeof(desc));
    pGPUOutBufs->GetDesc(&desc);
    desc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
    desc.Usage = D3D11_USAGE_STAGING;
    desc.BindFlags = 0;
    desc.MiscFlags = 0;
    if (!m_pCPUOutBufs && SUCCEEDED(pdevice->CreateBuffer(&desc, nullptr, &m_pCPUOutBufs))) {
        pd3dImmediateContext->CopyResource(m_pCPUOutBufs, pGPUOutBufs);
    }
    else pd3dImmediateContext->CopyResource(m_pCPUOutBufs, pGPUOutBufs);

    D3D11_MAPPED_SUBRESOURCE MappedResource;
    float *p;
    pd3dImmediateContext->Map(m_pCPUOutBufs, 0, D3D11_MAP_READ, 0, &MappedResource);
    p = (float*)MappedResource.pData;
    for (int i = 0; i < 4; i++) printf("%d %f\n", i, p[i]);
    pd3dImmediateContext->Unmap(m_pCPUOutBufs, 0);
    printf("\n");
}

绑定无人机的缓冲区只有4个元素。所以，如果我的 texture2d 中的所有浮点数都是 0.7，我应该在 CopyToCPUBuffer 函数中打印 4 个 0.7s 而不是 0.0s.

有谁知道我的代码中可能有什么问题，或者有人可以为我提供一个完整的示例或教程来说明如何在计算着色器中正确读取 DirectX 11 texture2d 的数据吗？

提前致谢。

Answer 1

开头以下是错误的。输入数据的间距是每行纹理的字节数，而不是每个像素。

InitData.SysMemPitch = sizeof(float) * 4;

其次：

float4 num = BufferIn.Load(uint3(groupIdx.x, groupIdx.y, 1));

您正在尝试从纹理的第 2 个 mip 加载数据，它只有 1 个 mip 级别。

在 DirectX 11 计算着色器中加载 Texture2D 数据

Loading Texture2D data in DirectX 11 Compute Shader

directx

directx-11

texture2d

compute-shader