Sparse 3D FFT library with MPI, OpenMP, CUDA / ROCm support

 SpFFT was originally intended for transforms of data with spherical cutoff in
 frequency domain, as required by some computational material science codes.
 For distributed computations, SpFFT uses a slab decomposition in space domain
 and pencil decomposition in frequency domain (all sparse data within a pencil
 must be on one rank). If desired, the library can be compiled without any
 parallelization (MPI, OpenMP, CUDA / ROCm).