use amrex::FFT for Gaussian random field generation

[BenWibking]

Describe the proposal
Now that amrex::FFT exists, we can do the FFT to generate Gaussian random fields (e.g., for turbulence initial conditions and driving) directly on the GPU.

Describe alternatives you've considered
Keep using the existing Python script to generate the input turbulence field.

Additional context
AMReX-Codes/amrex#4193

cc @psharda @markkrumholz

[markkrumholz]

I can't tell from the AMReX PR: how / does this handle adaptive grids?

[BenWibking]

I can't tell from the AMReX PR: how / does this handle adaptive grids?

It doesn't. What I had in mind was very high resolution uniform grid turbulence simulations. But you could interpolate using the AMR machinery to fine grids.

[markkrumholz]

I can't tell from the AMReX PR: how / does this handle adaptive grids?

It doesn't. What I had in mind was very high resolution uniform grid turbulence simulations. But you could interpolate using the AMR machinery to fine grids.

I see. OK, then while I think we might want to use this for initial condition generation, I don't think we want to use it for turbulence driving. For turbulence driving you want to have the capability to drive at wavenumbers smaller than are supported on the level 0 grid. Ronan's implementation using Christoph's driving code is working now, and he's preparing the PRs to get the necessary changes incorporated into both Christoph's code and Quokka.

[BenWibking]

Ok, great. Right, this would only be for non-AMR cases.

[markkrumholz]

Understood.