[Mvapich-discuss] Segmentation fault if MPI_Finalize is called without freeing objects with attributes attached

[Subramoni, Hari]

Hi, Mose.

Sorry for the delay in getting back to you.

Can you please try the attached patch/workaround and see if it resolves the issues?

You can apply the patch as follows after placing the patch file in the top directory of your MVAPICH source code.

patch -p1< mv2_datatype_attr_patch.txt

Best,
Hari.

-----Original Message-----
From: Giordano, Mose <m.giordano at ucl.ac.uk> 
Sent: Monday, January 2, 2023 10:02 AM
To: Simon Byrne <simonbyrne at gmail.com>; mvapich-discuss at lists.osu.edu; Subramoni, Hari <subramoni.1 at osu.edu>
Subject: Re: [Mvapich-discuss] Segmentation fault if MPI_Finalize is called without freeing objects with attributes attached

!-------------------------------------------------------------------|
  This Message Is From an External Sender
  This message came from outside your organization.
|-------------------------------------------------------------------!

Hi Hari,

> Could you please let us know if this is a single node run or a multi-node run?

I can reproduce with both single- and multi-node (tested with two) runs.

> Could you also let us know how MVAPICH was configured (output of mpiname -a) and what compiler versions were used?

I can provide the information you requested.  We can reproduce the reported bug with both MVAPICH 2.3.6 built with GCC 11.1.0 and 2.3.7 built with GCC 12.1.0.  Configurations:

MVAPICH2 2.3.6 Mon March 29 22:00:00 EST 2021 ch3:mrail

Compilation
CC: gcc    -DNDEBUG -DNVALGRIND -O2
CXX: g++   -DNDEBUG -DNVALGRIND -O2
F77: gfortran -fallow-argument-mismatch  -O2
FC: gfortran   -O2

Configuration
--prefix=/lustre/software/mvapich2/gcc11/2.3.6 --with-knem=/opt/knem-1.1.3.90mlnx1 --with-hcoll=/opt/mellanox/hcoll --enable-fortran=all --enable-cxx --with-file-system=lustre --with-slurm=/cm/shared/apps/slurm/current --with-pm=slurm --with-pmi=pmi1 --with-device=ch3:mrail --with-rdma=gen2

----------

MVAPICH2 2.3.7 Wed March 02 22:00:00 EST 2022 ch3:mrail

Compilation
CC: gcc    -DNDEBUG -DNVALGRIND -O2
CXX: g++   -DNDEBUG -DNVALGRIND -O2
F77: gfortran -w -fallow-argument-mismatch -O2  -O2
FC: gfortran   -O2

Configuration
--prefix=/lustre/software/mvapich2/gcc12/2.3.7 --with-knem=/opt/knem-1.1.4.90mlnx1 --with-hcoll=/opt/mellanox/hcoll --enable-fortran=all --enable-cxx --with-file-system=lustre --with-slurm=/cm/shared/apps/slurm/current --with-pm=slurm --with-device=ch3:mrail --with-pmi=pmi1 -with-rdma=gen2


Best,
Mosè
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: mv2_datatype_attr_patch.txt
URL: <http://lists.osu.edu/pipermail/mvapich-discuss/attachments/20230113/4765753e/attachment-0010.txt>