application called MPI_Abort(MPI_COMM_WORLD, 1) - process 7

[jqhuang]

I compile QE-6.3 with intel2017+impi. When i run epw, i get the CRASH "Error in routine readwfc (20): error opening wfc file", and the logfile "application called MPI_Abort(MPI_COMM_WORLD, 1) - process 7". I have done several tests:

(1) I can perform pw.x and ph.x successfully in parallel. This test indicates the mpirun has no problem.
(2) The epw can be run in serial smoothly. This test seems that the CRASH is related to the impi.
(3)If i decrease ecutwfc in previous steps(scf.in, nscf.in), the epw can be run in parallel. But as we know, the results may be not ideal because the NCPP needs a large ecutwfc in general. This test seems that the CRASH is not related to the impi but has something to do with the memory, but no memory overflow was detected in parallel running with large ecutwfc.

The input files have no problems, since they can be run in another supercomputer. Could any one give me some advices. I'm really desperated for i have tried to resolve this problem for a month but still get nothing.

[sponce]

Hello,

Could you try to compile everything with intel 2017 and openmpi 1.10.7 ?

Alternatively I could also suggest to test the current development version of QE/EPW https://gitlab.com/QEF/q-e
I know that at some point there was an issue with intel2017+impi but it now works (see test-farm).

Best wishes,
Samuel