Dear amber developpers,
We are trying to run EVB calculations on a 2-VBstate system containing 74
atom types on several cores.
We have changed the rdparm.F90 and parms.F90 files to remove the limit of
60 atom types :
- in parms.F90 :
line 12 : BC_PARMR 46150 (instead of 24560)
line 30-35 : size of arrays cn1, cn2, cn3, cn4, cn5 : 5050 instead of 1830
- in rdparm.F90
line 108 nttyp > 5050 (instead of 1830)
However it does not seem to be enough since we first get velocities
exceeding vlimit and ultimately a crash due to a segmentation fault :
node19:00308] *** Process received signal ***
[node19:00309] *** Process received signal ***
[node19:00309] Signal: Segmentation fault (11)
[node19:00309] Signal code: Address not mapped (1)
[node19:00309] Failing at address: 0x20a7fc634
We tested our version of sander.MPI (compiled with gnu) on various systems
to understand where the problem came from exactly :
- it runs fine on our 69-atom-type system if a single VB state is used
(standard run with one topology)
- EVB calculations run correctly if we ask for only 2 cores (one for each
VB state) but crash as soon as we ask for more than 2 cores
- EVB calculations using the same executable run properly for another system
containing less than 60 atom types.
So our problem seems to be limited to EVB calculations on more than 1 core
per VB state, on a system having more than 60 atom types.
We tried to compile with the -debug flag in the configuration, but it does
not seem to give us a more detailed output. It may be a problem with the
way we compiled it, but so far we thus do not have any more information
regarding where the problem could come from.
Does someone have any idea how to solve our problem ?
Thank you in advance for your help,
Elise Duboué-Dijon
PhD student at the Ecole Normale Supérieure, Paris
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Mar 04 2013 - 15:30:02 PST
