Hi
I use the compute ackland/atom command.
It works fine on one processor but there is a segmentation fault (11) error when using 4 proc “processors 2 2 1”
Error message is: Signal: Segmentation fault (11)
Signal code: Address not mapped (1)
The code lines are:
compute phase all ackland/atom
dump 4 all custom 100000 dump1.data id type x y z vx vy vz c_phase
When omitting “c_phase” in the dump command the program works fine both scalar and parallel.
Who could help ?
Thansks a lot in advance
Best regards
Pascal
Pascal BRAULT
Directeur de Recherche CNRS
GREMI UMR7344 CNRS - Université d’Orléans
14, rue d’Issoudun BP6744
F-45067 ORLEANS Cedex2
Tel. 06 31 49 87 90 - Fax. 02 38 41 71 54
Pascal.Brault@…2724…
http://www.univ-orleans.fr/gremi
http://www.univ-orleans.fr/gremi/pascal-brault
“Plasmas pave the future of fuel cells”
N’imprimez ce courriel que si cela est absolument nécessaire
Hi
I use the compute ackland/atom command.
It works fine on one processor but there is a segmentation fault (11) error
when using 4 proc “processors 2 2 1”
Error message is: Signal: Segmentation fault (11)
Signal code: Address not mapped (1)
The code lines are:
compute phase all ackland/atom
dump 4 all custom 100000 dump1.data id type x y z vx vy vz c_phase
When omitting “c_phase” in the dump command the program works fine both
scalar and parallel.
Who could help ?
before anybody can help, you will need to do two important things (as
they are also noted in the LAMMPS manual):
1) download and compile the very latest development version of LAMMPS
(unless you are already using it) and try again. does it still
segfault?
2) prepare a small input deck that can reproduce the segfault (as
small as possible, and so that it runs as fast as possible), and post
them here. that will speed up the debugging process enormously (no
need to guess what could be the trigger for the segfault).
thanks,
axel.
Dear Axel
Thanks a lot for your help.
please always reply to the mailing list and not only individual
people. this way the complete conversation gets archived and others
can learn from it, in case they run into similar problems.
The Lammps version is from 1stFeb 2014.
that is quite old. LAMMPS follows a continuous integration development
scheme, so we only fix bugs in the latest versions. sometimes,
bugfixes can be backported, but that would be up to the individual
user to work it out.
I will write a very small code for checking.
What I have already observed: the problem comes when writing the dump file not for executing compute command.
that is normal. compute styles are only executed when they are
requested. so unless there is some other style or variable that uses
it, it will not trigger any bugs.
And it only fails when using multiprocessor run. It works fine with one processor on the same machine.
Moreover.
When replacing "ackland/atom" by "cna/atom cutoff" all works fine both scalar and parallel.
again, no surprise here. i have a hunch what kind of an issue can be
the cause of this, but i have not been able to trigger it with a
simple input example that i concocted myself. so there has to be
something particular in what you are doing in your simulation that
triggers it.
axel