Hi all,

I've just implemented a screenshot function for my engine, using dx11.
This all works fine, also when the rendertarget has MSAA applied, then I use resolvesubresource into an additional rendertarget.

Basically, without AA:

- copy primary RT to RT with usage STAGING and CPU access (staging texture)
- map the RT, copy the pixel data and unmap

And with AA:

- copy primary RT with AA to temporary 1 without AA (usage default, no CPU access) (resolve texture)
- copy temporary RT without AA to new RT with usage STAGING and CPU access (staging texture)
- map the RT, copy the pixel data and unmap

So it all works fine, I applied a branch for MSAA enabled/ disabled.

My question; according to the MSDN documentation, "Resolvesubresource", should only work when 'source' has >1 samples and 'dest' has 1 sample (in the desc). But somehow the process with the resolve texture, including using 'resolvesubresource', also works when I don't have MSAA enabled. So both source + dest has 1 sample.

I would expect the D3D debugger/ logging to give an error or at least a warning.
Below the code in which this applies (don't mind it not being optimized/ cleaned up yet, just for illustration)

std::vector<char> DX11RenderTarget::GetRGBByteArray() const
{
	CComPtr<ID3D11Device> myDev;
	CComPtr<ID3D11DeviceContext> myDevContext;
	mBuffer->GetDevice(&myDev);
	myDev->GetImmediateContext(&myDevContext);

	// resolve texture
	D3D11_TEXTURE2D_DESC tempRTDesc;
	tempRTDesc.Width				= mWidth;
	tempRTDesc.Height				= mHeight;
	tempRTDesc.MipLevels			= 1;
	tempRTDesc.ArraySize			= 1;
	tempRTDesc.Format				= mDXGIFormat;
	tempRTDesc.BindFlags			= 0;
	tempRTDesc.MiscFlags			= 0;
	tempRTDesc.SampleDesc.Count		= 1;
	tempRTDesc.SampleDesc.Quality	= 0;
	tempRTDesc.Usage			= D3D11_USAGE_DEFAULT;
	tempRTDesc.CPUAccessFlags	= 0;

	CComPtr<ID3D11Texture2D> tempTex;
	CComPtr<ID3D11Texture2D> anotherTempTex;
	myDev->CreateTexture2D(&tempRTDesc, 0, &tempTex);			// add check if(FAILED), logging

	myDevContext->ResolveSubresource(tempTex, 0, mBuffer, 0, GetDXGIFormat(mBufferFormat));		// format = DXGI_FORMAT_R8G8B8A8_UNORM_SRGB ; not flexible now

	// staging texture
	tempRTDesc.Usage				= D3D11_USAGE_STAGING;
	tempRTDesc.CPUAccessFlags		= D3D11_CPU_ACCESS_READ;

	myDev->CreateTexture2D(&tempRTDesc, 0, &anotherTempTex);			// add check if(FAILED), logging
	myDevContext->CopyResource(anotherTempTex, tempTex);	

	// map, copy pixels and unmap
	D3D11_MAPPED_SUBRESOURCE loadPixels;
	myDevContext->Map(anotherTempTex, 0, D3D11_MAP_READ, 0, &loadPixels);

 	std::vector<char> image(mWidth*mHeight*3);

	char *texPtr = (char*)loadPixels.pData;
	int currPos = 0;
	for(size_t row=0;row<mHeight;++row)
	{
		texPtr = (char*)loadPixels.pData + loadPixels.RowPitch * row;

		for(size_t j=0;j<mWidth*4;j+=4)		// RGBA, skip Alpha
		{
			image[currPos] = texPtr[j];
			image[currPos+1] = texPtr[j+1];
			image[currPos+2] = texPtr[j+2];
			currPos += 3;
		}
	}
	myDevContext->Unmap(anotherTempTex, 0);

	return image;
}

I've known this to work in the past as well. It's probably just for convenience that it was made to work with 1 -> 1 Resolves. It may even boil down to a 'Copy' somewhere further down the stack. I wouldn't complain though!