DirectX: rendering to Texture2DArray, and Loading from it in compute shader

Question

I want to render several images to a Texture2DArray, and access the resulting images in a compute shader. However, the Load method in the compute shader returns [0.0, 0.0, 0.0, 0.0] rather than the image data.

Currently, I am doing the following:

1. Setup

I first prepare a set of "render slots" by creating a Texture2DArray using CreateTexture2D with texArrayDesc.ArraySize = numRenderSlots :

texArrayDesc.Width = width;
texArrayDesc.Height = height;
texArrayDesc.MipLevels = 1;
texArrayDesc.ArraySize = numRenderSlots;
texArrayDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
texArrayDesc.Usage = D3D11_USAGE_DEFAULT;
texArrayDesc.BindFlags = D3D11_BIND_RENDER_TARGET | D3D11_BIND_SHADER_RESOURCE;
texArrayDesc.CPUAccessFlags = 0;
m_device->CreateTexture2D(&texArrayDesc, NULL, &m_renderSlotsTexArray);

I then create a ShaderResourceView for this texture array so I can access it from a compute shader:

srvDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
srvDesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2DARRAY;
srvDesc.Texture2DArray.FirstArraySlice = 0;
srvDesc.Texture2DArray.MostDetailedMip = 0;
srvDesc.Texture2DArray.MipLevels = 1;
srvDesc.Texture2DArray.ArraySize = numRenderSlots;
m_device->CreateShaderResourceView(m_renderSlotsTexArray, &srvDesc, &m_renderSlotsSrv);

For each slice in the Texture2DArray, I create a RenderTargetView so I can render to it.

rtvDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
rtvDesc.ViewDimension = D3D11_RTV_DIMENSION_TEXTURE2DARRAY;
rtvDesc.Texture2DArray.MipSlice = 0;
rtvDesc.Texture2DArray.ArraySize = 1;

for (int i = 0; i < numRenderSlots; i++) {

    // Change slot RTV desc to choose correct slice from array
    rtvDesc.Texture2DArray.FirstArraySlice = D3D11CalcSubresource(0, i, 1);

    // Create the RTV for the slot in m_renderSlotsTexArray
    ID3D11RenderTargetView* slotRtv;
    m_device->CreateRenderTargetView(m_renderSlotsTexArray, &rtvDesc, &slotRtv);

    // Add the RenderTargetView to a list
    m_slotRtvs.push_back(slotRtv);
}

2. Usage

I then render a set of different images by setting the associated render slot as a render target. Note that the background color is not [0,0,0,0] . Code simplified:

for (int i = 0; i < numRenderSlots; i++) {
    setupScene(i);
    deviceContext->OMSetRenderTargets(1, &m_slotRtvs[i], depthStencilView);
    deviceContext->ClearRenderTargetView(m_slotRtvs[slotIdx], {0.2,0.2,0.2,0.0});
    deviceContext->ClearDepthStencilView(depthStencilView, D3D11_CLEAR_DEPTH, 1.0f, 0);
    render();
}

I then setup a compute shader with a StructuredBuffer for shader output with associated UnorderedAccessView. I pass this into the shader, along with the SRV of the render slots Texture2DArray. I finally dispatch the shader, to operate over the image in 32x32 sized chunks.

deviceContext->CSSetShader(m_computeShader, NULL, 0);
deviceContext->CSSetShaderResources(0, 1, &renderSlotsSrv);
deviceContext->CSSetUnorderedAccessViews(0, 1, &outputUav, nullptr);
deviceContext->Dispatch(img_width / 32, img_height / 32, numParams);

I then try to access the rendered image data in the compute shader. However, imgs.Load always seems to return [0.0, 0.0, 0.0, 0.0] .

Texture2DArray<float4> imgs : register(t0);
RWStructuredBuffer<float3> output : register(u0);

[numthreads(32,32,1)]
void ComputeShaderMain(
    uint3 gID : SV_GroupID,
    uint3 dtID : SV_DispatchThreadID,
    uint3 tID : SV_GroupThreadID, 
    uint gi : SV_GroupIndex )
{  
    int idx_x = (gID.x * 32 + tID.x);
    int idx_y = (gID.y * 32 + tID.y);
    float4 px = imgs.Load(int4(idx_x,idx_y,gID.z,0));
    ...
    output[outIdx] = float3(px.x, px.y, px.z);
}

I know the render slots are working, as I can access each render slot by doing CopySubresourceRegion into a staging texture to view the byte data. However, this GPU->CPU transfer is what I'm trying to avoid.

Also, I know the compute shader output is working as I can map the output buffer and inspect some basic test data.

---

What am I doing wrong? Thanks in advance.

Answer 1

Solved it. I created the D3D device with debugging enabled and noticed:

D3D11 WARNING: ID3D11DeviceContext::CSSetShaderResources:
Resource being set to CS shader resource slot 0 is still bound on output!

So the problem was that my Texture2DArray was still bound as a render target as I tried to set it as a resource in the compute shader.

Easily fixed by unbinding the render targets like so:

deviceContext->OMSetRenderTargets(0, NULL, NULL);

before using CSSetShaderResources() .

I hope this helps anyone that has a similar issue.