[jawilson2]

Never tried numba. I write all of our cuda stuff by hand anyway, and wrap that into cython from c++ where needed.