Hmm, that isn't good. I will take a look and get a patch ready for 1.8.2. -Nathan
On Mon, Apr 28, 2014 at 03:04:24PM +0000, VandeVondele Joost wrote: > Hi, > > using openmpi 1.7.5 I observe quick growth of memory allocated by our > program, and compiling it with '-O1 -g -fno-omit-frame-pointer > -fsanitize=leak' using gcc 4.9, they seem to originate from within openmpi. > As an example: > > Direct leak of 32000 byte(s) in 200 object(s) allocated from: > #0 0x7fbe57737793 in __interceptor_calloc > ../../../../gcc/libsanitizer/lsan/lsan_interceptors.cc:89 > #1 0x7fbe58526375 in mca_coll_base_comm_select > ../../../../ompi/mca/coll/base/coll_base_comm_select.c:144 > #2 0x7fbe584d664b in ompi_comm_activate > ../../ompi/communicator/comm_cid.c:692 > #3 0x7fbe584d2c7f in ompi_comm_split ../../ompi/communicator/comm.c:582 > #4 0x7fbe58501231 in PMPI_Comm_split > /data/vjoost/openmpi-1.7.5/build/ompi/mpi/c/profile/pcomm_split.c:68 > #5 0x7fbe587b6e84 in ompi_comm_split_f > /data/vjoost/openmpi-1.7.5/build/ompi/mpi/fortran/mpif-h/profile/pcomm_split_f.c:74 > #6 0x1c0382e in __message_passing_MOD_mp_comm_split > /data/vjoost/clean/cp2k/cp2k/src/common/message_passing.F:1592 > > which corresponds to (coll_base_comm_select.c:144) : > > /* TODO -- remove me for 1.9.x */ > comm->c_coll.neigh = calloc (1, sizeof (*comm->c_coll.neigh)); > > I'm wondering if this variable gets freed upon freeing the communicator. > > Thanks in advance, > > Joost > _______________________________________________ > 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/2014/04/14629.php
pgpms5e7GyQTT.pgp
Description: PGP signature