Project

General

Profile

Bug #2474

Double free/corruption for gmx solvate when topol.tpr explicitly listed as input argument and gmx built with mpi

Added by Viveca Lindahl 4 months ago. Updated about 1 month ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
analysis tools
Target version:
Affected version - extra info:
2018.2-dev-20180403-0d904bb
Affected version:
Difficulty:
uncategorized
Close

Description

gmx solvate -cp conf.gro -cs topol.tpr -o conf.gro -p topol.top

leads to a double free/memory corruption error. For '-cs topol.tpr' --> '-cs' avoids the segmentation fault but running 'valgrind -v' shows that the issue is still there (see valgrind.log).

I am using a gmx binary built with mpi. Building without mpi avoids the error.

setup.tgz (177 KB) setup.tgz files for reproducing Viveca Lindahl, 04/03/2018 02:45 PM
valgrind.log (48.6 KB) valgrind.log valgrind -v output Viveca Lindahl, 04/03/2018 02:51 PM

History

#1 Updated by Paul Bauer 4 months ago

Hej, which version of gcc and openmpi did you use for compiling?
I don't see something like this when compiling with gcc-7.3, openmpi-2.1.2 and version 2018.2-dev-20180403-0d904bb

#2 Updated by Viveca Lindahl 4 months ago

Paul Bauer wrote:

Hej, which version of gcc and openmpi did you use for compiling?
I don't see something like this when compiling with gcc-7.3, openmpi-2.1.2 and version 2018.2-dev-20180403-0d904bb

Good point. I used some relatively ancient versions: gcc-4.8.4 and openmpi-1.6.5.

#3 Updated by Paul Bauer 4 months ago

I confirmed it with gcc-4.8.4 and openmpi-1.8.4.

#4 Updated by Viveca Lindahl 4 months ago

I don't see something like this when compiling with gcc-7.3, openmpi-2.1.2 and version 2018.2-dev-20180403-0d904bb

means valgrind does not detect anything or no seg fault?

#5 Updated by Paul Bauer 4 months ago

Viveca Lindahl wrote:

I don't see something like this when compiling with gcc-7.3, openmpi-2.1.2 and version 2018.2-dev-20180403-0d904bb

means valgrind does not detect anything or no seg fault?

I finds some unrelated stuff, and does not segfault.

#6 Updated by Paul Bauer 3 months ago

this one might be related as well to #2479

#7 Updated by Mark Abraham about 1 month ago

  • Target version changed from 2018.2 to 2018.3

I'm struggling to reproduce this. I did get a hard segfault once, so I think there's a problem, but we don't have a backtrace to work from. Memory debugging with ddt didn't help find anything.

Also available in: Atom PDF