simplify device functions

[fwyzard]

alpaka suggests that device functions should be templated on the accelerator type, e.g.

template <typename TAcc>
ALPAKA_FN_ACC float compute(TAcc const& acc, ...);

However in many cases the full accelerator type is not needed - or used.
For example, math functions care about the back-end (CPU, CUDA, SYCL, etc.) but not about the dimensionality of the kernel (1d, 2d, 3d).

Can we find a way to simplify this kind of functions, possibly without breaking the existing code ?

[psychocoderHPC]

@fwyzard Can you point me in our meeting about alpaka2 to this topic again, please. For math functions, I have a solution.