[Wrf-users] WRF 2.0.2 on dual Opteron with Portland 5.2-4

[Craig Arthur]

Hi all,

I am attempting to compile WRF V2.0.2 on a dual AMD Opteron running SUSE
Linux 9.2 with Portland Group compilers (V5.2-4). I have had success in
compiling em_quarter_ss and executing this across both processors using OMP
directives.

However when I compile em_real, WRF crashes near the first OMP directive
(though not always in the same spot as found using 'pgdbg') if I attempt to
run on two CPU's (OMP_NUM_THREADS=2). However it runs without failure on one
CPU (OMP_NUM_THREADS=1). I have not compiled with the addtional AMD Core
Math Libraries (ACML) which are designed for the AMD architecture. The
compilation logs reveal nothing untoward through the compilation stage.

I've appended the PGI config to configure.defaults which were posted on the
Portland website (www.pgroup.com.resources/wrf/wrfv2_pgi52.htm), with the
additional modification that I added '$(OMPCPP)' to the 'POUND_DEF'
definition (I found that no executable would run on dual-CPU's without
this). The single-threaded configuration compiles and runs cleanly.

My query is, has anyone come across the problem of em_real not running, but
the idealised cases running when used in an multi-processor environment. Are
there any suggestions for troubleshooting this compilation issue?


Additional info:
Machine: AMD 2Ghz dual Opteron running SuSE Linux Pro 9.2 kernel 2.6.8-24
(as installed)
Fortran compiler: Portland Group 5.2-4 (installed without ACML)
netCDF: 3.6.0
NCAR Graphics: 4.3.1





Craig Arthur
Meteorologist / Computer Modelling
The Weather Company
Level 1/7 West Street
North Sydney, NSW 2060
Ph:  +61 2 9965 9257
Mob: 0409 779 478
Fax: +61 2 9955 1536
Email: carthur at theweather.com.au