Unable to get correct result using openMPI about spfft

[AdhocMan]

I believe the issue is, that the input pointer when executing the transforms needs to point to the local data.
So changing
transform.backward(signal.data(), SPFFT_PU_HOST);
to
transform.backward(signal.data() + 2 * (xyIndicesGlobal.size() / commSize) * dimZ, SPFFT_PU_HOST);
should give you the correct result in this example.

[Ondraol]

Hello,

Thank you for reply. You are right. The problem is in backward function input data. I focused too much on MPI, indexing atc. so I missed input data distribution.
I am just not sure about your computation of pointer shift . In this example for 2 * (xyIndicesGlobal.size() / commSize) * dimZ gives 8 for each rank. I think that correct computation in data shift for rank is for this case offset * xyIndicesGlobal.size() * 2.

Thank you very much for your help. I really appreciate it.

[AdhocMan]

I'm glad I could help.
You are right, my suggested change is not entirely correct, since the MPI rank is missing and I assumed a different distribution of excess elements. I think the correct shift for your example might actually be offset * dimZ * 2, because the Z dimension is not split for the input signal. Only the output is split in Z, so essentially the output is transposed.

I'll close this issue for now, but feel free to reopen or create a new issue if you have further questions.