Preliminary Distributed FFT Benchmarking Results on ROCm MPI

[matinraayai]

This issue is to keep track of the preliminary FFT benchmarks on ROCm GPUs, now that we have a ROCm-aware MPI.

Tasks:

1. Build MPICH and OpenMPI with ROCm support
2. Fix the AMDGPU.jl Base.unsafe_wrap PR: Preliminary fix for Base.unsafe_wrap JuliaGPU/AMDGPU.jl#583
3. Benchmark HeFFTe
4. Benchmark PencilFFTs.jl (after improvements)

@sanatgp while I try to land the unsafe_wrap PR, you can use my fork of AMDGPU.jl if you want to benchmark PencilFFTs.jl.