Currently we do not provide any FFT routine, assuming the user can rely on PyLops' ones by simply wrapping them into a pylops_mpi.MPILinearOperator (or pylops_mpi.MPIVStack and similar). This however does not allow to either perform a 1D-FFT to a ND-array distributed over the axis where we want to perform FFT or a ND-FFT over multiple axes, one of which is distributed.
mpi4py-fft provide such functionalities, which we could benefit from. The integration should be rather straighforward apart from the fact that we may need to test how our pylops_mpi.DistributedArray object and their mpi4py_fft.DistArray can talk to each other without incurring in any further data-copy in the process.
Currently we do not provide any FFT routine, assuming the user can rely on PyLops' ones by simply wrapping them into a
pylops_mpi.MPILinearOperator(orpylops_mpi.MPIVStackand similar). This however does not allow to either perform a 1D-FFT to a ND-array distributed over the axis where we want to perform FFT or a ND-FFT over multiple axes, one of which is distributed.mpi4py-fftprovide such functionalities, which we could benefit from. The integration should be rather straighforward apart from the fact that we may need to test how ourpylops_mpi.DistributedArrayobject and theirmpi4py_fft.DistArraycan talk to each other without incurring in any further data-copy in the process.