write_rmn broken with MPI

[jaemolihm]

When run with multiple cores, seedname_r.dat file written with write_rmn = .true. contains NaN. When run with 1 core, the result is fine. I used develop branch commit dd450c6.

Example output (diam_r.dat, 8 cores):

 written on 18Jul2024 at 00:41:09 
           8
          93
   -3    1    1    1    1    0.056618    0.106395    0.016255   -0.040207    0.024689   -0.024178
   -3    1    1    2    1************************************************************************
   -3    1    1    3    1         NaN         NaN         NaN         NaN         NaN         NaN
   -3    1    1    4    1         NaN         NaN         NaN         NaN         NaN         NaN

Example output (diam_r.dat, 1 core)

 written on 18Jul2024 at 00:45:46 
           8
          93
   -3    1    1    1    1    0.000000    0.000000   -0.000000   -0.000000   -0.000000    0.000000
   -3    1    1    2    1   -0.000528    0.000000    0.000528   -0.000000    0.001362   -0.000000
   -3    1    1    3    1    0.001362    0.000000   -0.000528   -0.000000   -0.000528   -0.000000
   -3    1    1    4    1   -0.000528   -0.000000    0.001362   -0.000000    0.000528    0.000000
   -3    1    1    5    1   -0.000369   -0.000000    0.000369    0.000000    0.000369    0.000000

mpirun -np 8 pw.x -in scf.in
mpirun -np 1 wannier90.x -pp diam
mpirun -np 8 pw2wannier90.x -in pw2wan.in
mpirun -np 8 wannier90.x diam

Input files
diam.tar.gz

[JeromeCCP9]

Hi Jae-Mo!

Thank you for finding this and sending the inputs to work with!

This revealed an important bug--that the write_rmn code required the full m_matrix, which was not available in parrallel.

I have fixed this by parallelizing the algorithm, in PR "make write_rmn parallel" #510 and hopefully we'll review and merge this rapidly.

Thank you again for your careful issue post!
Yours, Jerome

[jaemolihm]

Hi Jerome,

Thanks a lot for the quick fix! I confirm that the MR fixes the problem.

Best,
Jae-Mo