I forgot to add the VTKM_EXEC_CONT modifier to the make_VecC methods,
and that causes them to fail on CUDA devices.

I wish the compiler would have said something. I was calling one of them
from a VTKM_EXEC method.