[gmx-users] segmentation fault on trjconv with deshuf

[Erik Lindahl]

Lynne E. Bilston wrote:

> Hi all,
>
> I've used the -shuffle -sort options for a longish run (1ns), and now 
> I want to continue the run from that endpoint. Of course, grompp 
> complaints about the atom names not matching due to the 
> shuffling/sorting.
>
> I wanted to try the deshuffling option of trjconv, as suggested in the 
> list, but using that option in trjconv gives me a segmentation fault, 
> with core dump, as it starts to write the file out. Other options in 
> trjconv seem to work fine.
>
> I'm running on a dual athlon, Redhat 7.2, Gromacs 3.1.3. Everything 
> installed from the rpms off the website (except I recompiled lam, but 
> that's probably not relevant as I'm not using it here...)
>
> Any ideas - either on how to continue my run or the core dump in trjconv?
>
> Thanks,
>
> Lynne 


Hmm... Segfault are always really bad, but I just don't have time to 
debug it right now.

What I usually do when I run in parallel is to create a topology where 
the molecules section at the end is the same order as the shuffled 
system, e.g.


DPPC    10
SOL       230
DPPC     10
SOL        230

etc...

This way you can use this topology instead and never care about undoing 
the shuffling. If you don't remember the molecule distribution you can 
just rerun grompp
with the shuffle option and look at the way it distributes molecules 
over CPUs.

Cheers,

Erik