Here are attached two patches (one for master, one for v1.10) please consider these as experimental ones : - they cannot hurt - they might not always work - they will likely allocate a bit more memory than necessary - if something goes wrong, it will hopefully be caught soon enough in a new assert clause
Cheers, Gilles On Sat, May 21, 2016 at 9:19 PM, Gilles Gouaillardet <gilles.gouaillar...@gmail.com> wrote: > Tabuchi-san, > > thanks for the report. > > this is indeed a bug i was able to reproduce on my linux laptop (for > some unknown reasons, there is no crash on OS X ) > > ompi_datatype_pack_description_length malloc 88 bytes for the datatype > description, but 96 bytes are required. > this causes a memory corruption with undefined side effects (crash in > MPI_Type_free, or in MPI_Win_unlock) > > iirc, we made some changes to ensure data is always aligned (Sparc > processors require this), and we could have missed > some stuff, and hence malloc less bytes than required. > > > Cheers, > > Gilles > > On Sat, May 21, 2016 at 5:50 PM, Akihiro Tabuchi > <tabu...@hpcs.cs.tsukuba.ac.jp> wrote: >> Hi, >> >> At OpenMPI 1.10.2, MPI_Type_free crashes with a many nested derived type >> after using MPI_Put/Get >> with the datatype as target_datatype. >> The test code is attached. >> In the code, MPI_Type_free crashes if N_NEST >= 4. >> >> This problem occurs at OpenMPI 1.8.5 or later. >> There is no problem at OpenMPI 1.8.4, MPICH 3.2, and MVAPICH 2.1. >> >> Does anyone know about the problem? >> >> Regards, >> Akihiro >> >> -- >> Akihiro Tabuchi >> HPCS Lab, Univ. of Tsukuba >> tabu...@hpcs.cs.tsukuba.ac.jp >> >> _______________________________________________ >> users mailing list >> us...@open-mpi.org >> Subscription: https://www.open-mpi.org/mailman/listinfo.cgi/users >> Link to this post: >> http://www.open-mpi.org/community/lists/users/2016/05/29260.php
datatype_description.master.patch
Description: Binary data
datatype_description.v1.10.patch
Description: Binary data
