Functions for both CPU usage and callable from GPU kernels

[K20shores]

Soon we are going to move all rate constant calculations into open atmos. For example, this rate constant calculation of Arrhenius reactions: https://github.com/NCAR/micm/blob/615cf5cec1dcfbfb5383d02a68f5d570964ac0fe/include/micm/process/arrhenius_rate_constant.hpp#L84-L88

Functions on a class cannot be used on GPU devices. And, functions that need to run on GPU devices typically need a __device__ attribute. However, this video claims that making a function constexpr allows it to be compiled for the GPU and also be useable on the CPU.

1. Is this true and feasible?
2. Can such functions accept structs of data or do we have to pass native types as function arguments?

[K20shores]

Not pursing this as of now