Unfortunately its embedded in a much larger code base. I will see if I can isolate it somewhat.

BTW this is a Bcast over an intercomm, not an intracomm, so that might make a different.

In the end I put together a valgrind suppression file which seems to suppress all the MPI errors. It looks like this:
{
   MPI_func_leaks1
   Memcheck:Leak
   ...
   fun:MPL_trmalloc
   ...
}

On Thu, Aug 18, 2016 at 12:49 PM, Guo, Yanfei <yguo@anl.gov> wrote:
Hi Ed,

I tried a couple Bcast unittest with the same MPICH and valgrind version. I could not reproduce the problem on our side. Can you provide an example code that has this problem?

Yanfei Guo
Postdoctoral Researcher
MCS Division, ANL


On 8/18/16, 10:29 AM, "Ed Hartnett" <edwardjameshartnett@gmail.com> wrote:


    I am using MPICH 3.2, built with: ./configure --prefix=/usr/local --enable-g=mem,dbg,meminit
    I am using valgrind 3.11.0.
    When I run valgrind on one of my MPI programs, I get a large number of reported leaks for PMPI_Bcast().
    Am I doing something wrong? I thought that this version of MPICH, when used with configure flag --enable-g=mem,dbg,meminit, would allow valgrind to work on MPI programs without false positives.
    Any suggestion would be helpful.



_______________________________________________
discuss mailing list     discuss@mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss