There are DX/GL interop functions in the CUDA API. But, they are a bit tricky. Unless you really have a strong, specific need for something only available in CUDA, you are better off using compute shaders.