Hi Matthew,
Using the debuguer I finaly found the problem. It is related to MPI. In
src/sys/objects/pinit.c line 779, petsc test the availability of
PETSC_HAVE_MPI_INIT_THREAD and this is set to True beccause my OpenMPI version
is compiled with --enable-mpi-thread-multiple.
However the call to MPI_Init_thread(argc,args,MPI_THREAD_FUNNELED,&provided)
hangs and freeze the application.
Unsetting PETSC_HAVE_MPI_INIT_THREAD in petsc solves the problem.
I remember a HPC seminar on BULLX systems where they give us some informations
about known problems whith the MPI_Init_thread call in openMPI. May be I should
use a more recent version of OpenMPI. This also explain why I had this problem
with previous versions of petsc (same OpenMPI environment).
None of our codes mixes OpenMP and MPI.... so I never fall in this situation in
production and petsc always behave fine (excepted for make test).
Is there a way to turn off PETSC_HAVE_MPI_INIT_THREAD at configure time for
Petsc ? I've manualy removed it in the generated petscconf.h file before
compiling Petsc in debug mode but I don't think it is the best way to to this.....
Thanks for your help in running separatly the test codes.
Patrick
--
===================================================================
| Equipe M.O.S.T. | |
| Patrick BEGOU | mailto:[email protected] |
| LEGI | |
| BP 53 X | Tel 04 76 82 51 35 |
| 38041 GRENOBLE CEDEX | Fax 04 76 82 52 71 |
===================================================================
