[PyCUDA] passing GPUArray slices to a kernel