[Wien] MPI_COMM_DUP function

Peter Blaha pblaha at theochem.tuwien.ac.at
Tue Dec 18 14:00:10 CET 2007


We looked into the problem.

a) What case produced this error ? How many k-points do you have ?
   What is the matrix size (:RKM in scf1 file); how many processors did you use ?

b) sl_init is called for every k-point. So if you have many k-points, it
   could explain your problem. We have modified modules.F and I attach this file
   to this mail. Please replace modules.F in SRC_lapw1 and recompile.

c) If this is a small case with many k-points, the mpi-parallel version is
   NOT suitable, but you should run the k-point parallel one.
   For medium size cases, one would first parallelize the k-points,
   and eventually then only parallelize fine grain on a few nodes.

Regards

Nobuaki Igarashi schrieb:
> Dear Wien2k users,
> 
> I use WIEN2k version 07.3 on HITACHI SR11000 which is AIX computer.
> When I run WIEN2k MPI version using IBM MPI with some data, it would be
> stopped because MPI_COMM_DUP function is called many times.
> 
> This problem occurs in lapw1,
> SRC_lapw1/modules.F
>   +102            CALL SL_INIT(ICTXT, NPROW,npcol)


                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------
-------------- next part --------------
A non-text attachment was scrubbed...
Name: modules.F
Type: text/x-fortran
Size: 50962 bytes
Desc: not available
Url : http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20071218/4c9af44d/modules.bin


More information about the Wien mailing list