I was using a supersampled depth buffer during the shadow map pass (something like: screen_width * 4, screen_height * 4) and I wanted to try whether it would be faster to use a multisampled texture instead. However I am having problems resolving it to a non-multisampled texture. Specifically I am getting two errors:

D3D11 ERROR: ID3D11DeviceContext::ResolveSubresource: The Format (0x2d, D24_UNORM_S8_UINT) is never able to resolve multisampled resources. [ RESOURCE_MANIPULATION ERROR #294: DEVICE_RESOLVESUBRESOURCE_FORMAT_INVALID]

D3D11 ERROR: ID3D11DeviceContext::ResolveSubresource: Source Resource must not have the D3D11_BIND_DEPTH_STENCIL flag set. [ RESOURCE_MANIPULATION ERROR #292: DEVICE_RESOLVESUBRESOURCE_SOURCE_INVALID]

The second one seems more menacing since the D3D11_BIND_DEPTH_STENCIL has to be set for me to be able to render into the depth buffer.

The code for shadow render target initialization:

        // Multisampled texture to be resolved
	D3D11_TEXTURE2D_DESC msDepthStencilDesc = {};
	msDepthStencilDesc.Width = (UINT)renderer->screenSize.x;
	msDepthStencilDesc.Height = (UINT)renderer->screenSize.y;
	msDepthStencilDesc.MipLevels = 1;
	msDepthStencilDesc.ArraySize = 1;
	msDepthStencilDesc.Format = DXGI_FORMAT_D24_UNORM_S8_UINT;
	msDepthStencilDesc.SampleDesc.Count = 4;
	msDepthStencilDesc.SampleDesc.Quality = D3D11_STANDARD_MULTISAMPLE_PATTERN;
	msDepthStencilDesc.Usage = D3D11_USAGE_DEFAULT;
	msDepthStencilDesc.BindFlags = D3D11_BIND_DEPTH_STENCIL;
	msDepthStencilDesc.CPUAccessFlags = 0;
	msDepthStencilDesc.MiscFlags = 0;
	ID3D11Texture2D* tmpMSDepthStencilTex;
	HRESULT hr = renderer->device->CreateTexture2D(&msDepthStencilDesc, NULL, &tmpMSDepthStencilTex);
	CHECK_WIN_ERROR(hr, "Error creating Depth Buffer texture\n");

	CD3D11_DEPTH_STENCIL_VIEW_DESC depthStencilViewDesc(D3D11_DSV_DIMENSION_TEXTURE2DMS, msDepthStencilDesc.Format);
	hr = renderer->device->CreateDepthStencilView(tmpMSDepthStencilTex, &depthStencilViewDesc, &renderTarget->depthStencilView);
	CHECK_WIN_ERROR(hr, "Error creating depth stencil view\n");


        // Texture to be sampled in the scene pass
	D3D11_TEXTURE2D_DESC depthStencilDesc = {};
	depthStencilDesc.Width = (UINT)renderer->screenSize.x;
	depthStencilDesc.Height = (UINT)renderer->screenSize.y;
	depthStencilDesc.MipLevels = 1;
	depthStencilDesc.ArraySize = 1;
	depthStencilDesc.Format = DXGI_FORMAT_R24G8_TYPELESS;
	depthStencilDesc.SampleDesc.Count = 1;
	depthStencilDesc.SampleDesc.Quality = 0;
	depthStencilDesc.Usage = D3D11_USAGE_DEFAULT;
	depthStencilDesc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
	depthStencilDesc.CPUAccessFlags = 0;
	depthStencilDesc.MiscFlags = 0;
	ID3D11Texture2D* tmpDepthStencilTex;
	hr = renderer->device->CreateTexture2D(&depthStencilDesc, NULL, &tmpDepthStencilTex);
	CHECK_WIN_ERROR(hr, "Error creating Depth Buffer texture\n");

	D3D11_SHADER_RESOURCE_VIEW_DESC depthShaderResourceViewDesc = {};
	depthShaderResourceViewDesc.Format = DXGI_FORMAT_R24_UNORM_X8_TYPELESS;
	depthShaderResourceViewDesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2D;
	depthShaderResourceViewDesc.Texture2D.MostDetailedMip = 0;
	depthShaderResourceViewDesc.Texture2D.MipLevels = 1;
	hr = renderer->device->CreateShaderResourceView(tmpDepthStencilTex, &depthShaderResourceViewDesc, &renderTarget->depthTexture);
	CHECK_WIN_ERROR(hr, "Error creating depth texture\n");The MS resolve code:

        // Resolve multisample
        ID3D11Resource *multisampleTexture;
	renderTarget->depthStencilView->GetResource(&multisampleTexture);
	ID3D11Resource *texture;
	renderTarget->depthTexture->GetResource(&texture);
	renderer->context->ResolveSubresource(texture, 0, multisampleTexture, 0, DXGI_FORMAT_D24_UNORM_S8_UINT);
	RELEASE_DX_RESOURCE(multisampleTexture);
	RELEASE_DX_RESOURCE(texture);

Thanks for your help.

Hmmm, sounds like you have to do that manually in which case you could do the shadowmapping with a Texture2DMS directly, I guess.

What did you expect it to do anyway? The "default" box filter resolve makes no sense whatsoever for depth values.

Mona: Well I am currently using straight dumb supersampling and increasing the size of the texture to be, say 4*screen_width, 4*screen_height improves the quality of the shadows dramatically, so I thought that using multisampling would have a similar effect. Doesn't sampling a supersampled texture have similar effect to using multisampling?

Pixelshader code:

	for (uint i = 0; i < sampleCount; i++)
	{
		if (!shadowTexture.SampleCmp(depthSampler, shadowTexCoord + poissonDisk[i] / 700.0f,
			calculatedDepth))
		{
			shadow += 0.9f / sampleCount;
		}
	}

Okay I get the idea but then why does increasing the resolution of the shadow map texture improve the quality of the shadows? I thought that when I render into a higher resolution texture and then sample it while rendering to a smaller resolution render target, then if the higher resolution texture is 4 x larger than the render target, a single sample would be an average of the 4 pixels in the larger texture.

But as you guys said this doesn't make sense for depth so why would it improve the quality of the shadows. Hopefully you will get what I'm asking :).