I also have some updates, and they preserve the ability to work with MPI-2 and MPI-3 implementations. Bill William Gropp Director, Parallel Computing Institute Thomas M. Siebel Chair in Computer Science University of Illinois Urbana-Champaign On Aug 22, 2014, at 5:07 PM, James <[email protected]> wrote:
That seems to have helped: I'm down from ~1600 lines of error messages to under 500.
I'll probably have to do the rest by hand, and will try to make a patch file of the changes
James
On Fri, 22 Aug 2014 13:49:14 -0700, Balaji, Pavan <[email protected]> wrote:
Can you try the patch from this ticket?
https://trac.mpich.org/projects/mpich/ticket/2090
— Pavan
On Aug 22, 2014, at 3:47 PM, James <[email protected]> wrote:
Hi,
(Apologies if this is the wrong place to post, but all the links on the MPE page seem to redirect to MPICH.)
I am trying to build the MPE profiling libraries to work with MPICH. However, I am getting about a thousand lines of compile error messages, because some MPE routines use K&R-style function declarations, which don't match up with the prototypes in mpi.h
00106 log_mpi_core.c: In function 'MPI_Allgather': 00107 log_mpi_core.c:1550:8: error: argument 'sendbuf' doesn't match prototype 00108 void * sendbuf; 00109 ^ 00110 In file included from log_mpi_core.c:10:0: 00111 /opt/mpich/include/mpi.h:1002:5: error: prototype declaration 00112 int MPI_Allgather(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, 00113 ^
Before I go and try to fix all these myself, could I ask 1) Has it already been done somewhere? or 2) Has it not been done because there's now something better than MPE for tracing/profiling?
To be clear, I'm not interested in profiling MPI itself, but what the program is doing between calls to MPI routines.
Thanks, James
_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss