I hope someday they add OpenCL <-> CUDA direct interop (not through OpenGL/DirectX interop API)