This seems to be the case only with the TKR interface. All the others are either calling the OMPI version directly (mpif-h), or are calling some other internal (or weak symbol function).
George. On Tue, Aug 25, 2015 at 9:04 AM, Bert Wesarg <bert.wes...@tu-dresden.de> wrote: > On 08/25/2015 02:44 PM, Gilles Gouaillardet wrote: > >> Folks, >> >> I ran some basic tests with IPM profiler-like >> https://github.com/nerscadmin/IPM and found that when fortran calls an >> mpi subroutine, this is accounted twice. >> IPM defines both MPI_* subroutines and their fortran mpi_*_ counterpart. >> since the ompi fortran calls the MPI_* symbol (and not the PMPI_* one), >> and >> IPM does nothing to prevent double accounting, all subroutines are >> accounted twice >> >> what is the rationale for calling MPI_* from fortran instead of PMPI_* ? >> >> basically, I can see three options >> 1. we do nothing, this is an IPM problem, not an Open MPI one >> 2. we change ompi to call the PMPI_* symbols >> 3. we add a configure option to call PMPI_* symbols instead of the MPI_* >> ones >> >> any thoughts ? >> > > One more datapoint, also from a monitor tool (Score-P, as some of you > know): The Open SHMEM part of Open MPI also calls the MPI interface, not > the PMPI. That may result in performance data from MPI calls in SHMEM > applications, which seems weird too. > > Bert > > >> Cheers, >> >> Gilles >> >> >> > -- > Dipl.-Inf. Bert Wesarg > wiss. Mitarbeiter > > Technische Universität Dresden > Zentrum für Informationsdienste und Hochleistungsrechnen (ZIH) > 01062 Dresden > Tel.: +49 (351) 463-42451 > Fax: +49 (351) 463-37773 > E-Mail: bert.wes...@tu-dresden.de > > > _______________________________________________ > devel mailing list > de...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel > Link to this post: > http://www.open-mpi.org/community/lists/devel/2015/08/17843.php >
