Regarding the GS instancing to render to all faces in one pass, there is a nice little sample in the dxsdk showing exactly that.

Yes, you may use GS to output a primitive to multiple render targets (the DX SDK example does exactly that).

However, it is perfectly fine to render each cube map side separately too. The performance difference may be negligible since GS is typically a performance hog if used to output lots of primitives. I prefer slightly rendering each side separately since then you don't need separate vertex shaders for rendering cube maps / rendering normally. Since cube map render target gives you 6 rendertargetviews, you can render them separately by just selecting the desired target.

There is a function in D3DX to save dds files (may work with cubemaps), but MSDN suggests using DirectXTK with similar function since D3DX is almost deprecated.

[edit] It seems that DirectXTK won't save cubemaps correctly.

Cheers!

http://msdn.microsoft.com/en-us/library/windows/desktop/ff476298(v=vs.85).aspx You could try using this function for saving. It may or may not work with cube maps.

The cube map resolution is up to your needs. How many cube maps you are planning to use in every scene? What is the cube map surface format? Are you using HDR rendering? Are you using cube maps for specular reflections?

It is ok to render cubemaps "when needed". It is even possible to only render one side at a time and then swap the cube map when finished rendering.

However, if you are considering supporting SLI rendering, then there might be some things to know.

Cheers!

The way it works with the geometry shader, is, that you project each vertex inside the geometry shader onto the different sides and than output the resulting vertices with a SV_RenderTargetArrayIndex semantic which represents which side the projected vertex you are outputting is on.

Well I've managed to get the rendering done (but using 6 passes) now...

How do I combine these 6 RenderTargets to a single CubeMap ShaderResource ?

Is there any way to do that using each RT's ID3D11Texture* ?

edit:

I tried something like:

	ID3D11Texture2D* cubeFaces[6];
	for(unsigned int i = 0; i < 6; i++)
	{
		this->faces = new CustomRenderTarget(this->device, DXGI_FORMAT_R8G8B8A8_UNORM, this->size, this->size);
		cubeFaces = this->faces->GetRenderTargetTexture();
	}

	D3D11_SUBRESOURCE_DATA subData;
	subData.pSysMem = cubeFaces;
	subData.SysMemPitch = sizeof(XMFLOAT4) * this->size;

	ID3D11Texture2D* cubeMap = NULL;
	D3D11_TEXTURE2D_DESC texDesc;
	texDesc.Width = this->size;
	texDesc.Height = this->size;
	texDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
	texDesc.MipLevels = 0;
	texDesc.ArraySize = 6;
	texDesc.SampleDesc.Count = 1;
	texDesc.Usage = D3D11_USAGE_DEFAULT;
	texDesc.SampleDesc.Quality = 0;
	texDesc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
	texDesc.CPUAccessFlags = 0;
	texDesc.MiscFlags = D3D11_RESOURCE_MISC_TEXTURECUBE;
	
	HRESULT hr = this->device->CreateTexture2D(&texDesc, &subData, &cubeMap);

But I'm getting an error on the last line saying Invalid arguments.

It's all in the sample I mentioned in the first answer in the dxsdk, you should really take a look at it.

You create a texture with 6 slices, and the correct flags.

You create one shader resource view

you create 6 rendertarget views if you want to do the 6 passes manually, each referencing a specific slice. Job done