The ABINIT Users Mailing List ( CLOSED )

[James Raynolds <JRaynolds@uamail.albany.edu>]

Dear abinitioners,

        The new version of abinip (version_4.0.3) crashes on a job the the
previous version (version_3.4.3) has no trouble with.  Attached to this
e-mail you will find a short gzipped tar file with the input files and
results.  The results show a successfull run with abinip_3.4.3.  The job is
simply a single point energy for an 8-atom Si unit cell.  I'm asking for 8
processors.

        One possibly important thing I'm seeing is (1) when I run the old
version there is a *LOG*0000, *LOG*0001, up to *LOG*0007, that is a log file
for each of 8 processors numbered from 0 to 7.  (2) When I run the new
version, the numbering only goes from *0001, to *0007, so it looks like it
is only grabbing 7 processors.
Basically the only result (excluding core files and empty output and LOG
files) using the new version (everything else being identical) is the
result.err file with the following message (LoadLeveler is IBM's batch
scheduling utility which runs fine with the previous version of abinip).

ATTENTION: 0031-408  8 tasks allocated by LoadLeveler, continuing...
ERROR: 0031-250  task 1: Trace/BPT trap
ERROR: 0031-250  task 6: Trace/BPT trap
ERROR: 0031-250  task 7: Trace/BPT trap
ERROR: 0031-250  task 5: Trace/BPT trap
ERROR: 0031-250  task 0: Trace/BPT trap
ERROR: 0031-250  task 3: Trace/BPT trap
ERROR: 0031-250  task 2: Trace/BPT trap
ERROR: 0031-250  task 4: Trace/BPT trap

        Please someone take a look at this and let me know if there is a
work-around.  I'm very interested in using the parallelism over bands for
ground state calculations with large unit cells.

Sincerely,
Jim Raynolds

 <<logdir.tar.gz>>

Attachment: logdir.tar.gz
Description: Binary data