[lammps-users] segmentation fault due to restart command

hi everyone

i’ve been trying to run the 20Dec06 version of LAMMPS compiled with the Portland Group C++ compiler on an AMD Opteron cluster running Linux. I keep getting segmentation faults that seem to be due to “restart” commands - if i include a “restart” command, the calculation crashes and a restart file which cannot be read by restart2data, but if i leave out the “restart” command, the calculation runs smoothly.

i get this result if i use my own input and data files and also with the *.rhodo input and data files supplied in the “bench” directory with LAMMPS. if i try to use the restart2data command on the restart file produced, i get the following error message:

Reading restart file …
ERROR: Unknown improper style

interestingly, if i compile LAMMPS with the serial (rather than parallel MPI compiler) and use no optimization flag (in the place of the “-O” optimization flag), the calculation runs fine and a readable restart file is produced. on the other hand, if i use no optimization flag with the parellel compiler, i still have the problem with the “restart” command.

i also have this problem when i compile the 24Jul06 version of LAMMPS with the Portland Group compiler on the same cluster, but everything works fine when i compile the same version of LAMMPS with the PathScale compiler (unfortunately the licence for this compiler on the cluster has expired so i can no longer use it).

any help with solving this problem would be greatly appreciated.


This doesn't sound like a LAMMPS problem, since it goes away
w/out optimization and with a different compiler. It also
sounds like the problem is when the restart file is written, not
when it's read. You could add a line to your Makefile.foo to
compile the write_restart.cpp file w/out optimization.

I think you'd need to find where the restart file is corrupted (by
to make progess on this.