I am getting this error in ABINIT 8.0
--- !ERROR
src_file: m_paw_dmft.F90
src_line: 366
mpi_rank: 21
message: |
Someone changed the k-point parallelism again
This is for a regular NSCF calculation with psp8 pseudos, so this is nothing to do with PAW or dmft, but this is not present using ABINIT 7.0
So it seems someone changed the k-point parallelism?
error in m_paw_dmft
Re: error in m_paw_dmft
I can confirm this bug, which mostly prevents using abinit with mpi...
Re: error in m_paw_dmft
Can you post a complete input file which reproduce the error ?
Thanks
Best regards
Bernard Amadon
PS: Could you give your affiliation, thanks
Thanks
Best regards
Bernard Amadon
PS: Could you give your affiliation, thanks
Bernard Amadon
CEA
France
CEA
France
Re: error in m_paw_dmft
Hi Bernard, thank you for quick reply.
Your answer (below) solved the problem.
Bests
eef
****
Indeed on 64 procs, the code use only 54 procs because it is sufficient for the optimal
parallelism.
(grep nproc logfile).
The bug appears because 64 is different from 54 (I will correct it in the next version).
For the moment, you just have to use 54 procs, and it should work !
So more generally, the workaround is to launch
mpiexec -n NPROC abinit < files > log
then do
grep nproc log
it gives you the nproc actually used.
Then relaunch
mpiexec -n nproc abinit < files > log
*****
Your answer (below) solved the problem.
Bests
eef
****
Indeed on 64 procs, the code use only 54 procs because it is sufficient for the optimal
parallelism.
(grep nproc logfile).
The bug appears because 64 is different from 54 (I will correct it in the next version).
For the moment, you just have to use 54 procs, and it should work !
So more generally, the workaround is to launch
mpiexec -n NPROC abinit < files > log
then do
grep nproc log
it gives you the nproc actually used.
Then relaunch
mpiexec -n nproc abinit < files > log
*****
Re: error in m_paw_dmft
Hi Fileti and Bernard, thanks for the workaround! I just found the same problem today and the solution is working for me too...