Open MPI User's Mailing List Archives

Subject: [OMPI users] Valgrind suppression not so suppressed
From: David Mathog (mathog_at_[hidden])
Date: 2010-12-15 16:04:33

• Next message: Tom Rosmond: "[OMPI users] MPI-IO problem"
• Previous message: Ralph Castain: "Re: [OMPI users] Issue with : mca_oob_tcp_peer_recv_connect_ack on SGI Altix"

OMPI 1.4.3
Valgrind 3.5.0

Trying to use valgrind on a program and getting a ton of MPI related
noise, totally swamping the memory problems in the program itself.

Looked at the FAQ and used the suppression file referred to there:

mpirun -np 2 -host newsaf.cluster \
 valgrind \
  --leak-check=full \
  --suppressions=/opt/ompi143/share/openmpi/openmpi-valgrind.supp \
 hmmfetch \
   --mpisplit \
   /usr/common/BLASTDB/PFAMDIR3/Pfam-A.hmm CagE_TrbE_VirB

To count up all of these messages, piped with:
   2>&1 | grep PMPI_Init | wc

which returned 92 for the command above. Leaving off the

  --leak-check=full

reduces the count to 4 (as well as the usefulness of valgrind for
finding memory leaks!) Those counts are unchanged when

  --suppressions=/opt/ompi143/share/openmpi/openmpi-valgrind.supp

is omitted, suggesting that it is ineffective on this combination of
OMPI and Valgrind.

Is there a suppression file for these versions that will shut down all
messages under PMPI_INit but still allow the messages from the program
being tested???

Thanks,

David Mathog
mathog_at_[hidden]
Manager, Sequence Analysis Facility, Biology Division, Caltech

• Next message: Tom Rosmond: "[OMPI users] MPI-IO problem"
• Previous message: Ralph Castain: "Re: [OMPI users] Issue with : mca_oob_tcp_peer_recv_connect_ack on SGI Altix"