[Wrf-users] Segfaults with Intel 2017 Compiler

[Tyle, Kevin R]

Hi Francisco,

I experienced the crashes in a two-domain simulation with sf_surface_physics = 2 (also tried 4) and sf_urban_physics = 2 (also tried 1) in both domains.  For debugging purposes, I then ran on only one domain.

Cheers,

Kevin

_____________________________________________
Kevin Tyle, Manager of Departmental Computing
Dept. of Atmospheric & Environmental Sciences
University at Albany
Earth Science 235, 1400 Washington Avenue
Albany, NY 12222
Email: ktyle at albany.edu<mailto:ktyle at albany.edu>
Phone: 518-442-4578
_____________________________________________

From: Francisco Salamanca Palou [mailto:fsalaman at asu.edu]
Sent: Friday, July 21, 2017 1:03 PM
To: Tyle, Kevin R <ktyle at albany.edu>
Cc: wrf-users at ucar.edu; wrfhelp at ucar.edu
Subject: Re: [Wrf-users] Segfaults with Intel 2017 Compiler

Hi Kevin,
You must run with the same sf_urban_physics option in all the domains, otherwise the model will crash.
I hope this help you
Best
Francisco

On Fri, Jul 21, 2017 at 9:57 AM, Tyle, Kevin R <ktyle at albany.edu<mailto:ktyle at albany.edu>> wrote:
Thanks to those who responded.  One of the suggestions was to increase the value of the debugging parameter in the namelist and that was a big help.

After recompiling with debugging and traceback, the true cause of the segfault revealed itself.  If sf_urban_physics is non zero AND a Noah land surface scheme is chosen, WRF will crash; not only with Intel 17.0.4 but also 14.0.1 (and the latest 18 beta).  Looks to be a problem with uninitialized variables when module_sf_noahdrv calls the particular sf_urban scheme.

When WRF is run using the PGI compiler (2017), crashes are much less likely, but still occur on occasion.

The documentation for WRF says that the urban physics schemes will work if a Noah LSM is chosen … perhaps this is a problem that was thought to be fixed but appears not to be.

--Kevin

_____________________________________________
Kevin Tyle, Manager of Departmental Computing
Dept. of Atmospheric & Environmental Sciences
University at Albany
Earth Science 235, 1400 Washington Avenue
Albany, NY 12222
Email: ktyle at albany.edu<mailto:ktyle at albany.edu>
Phone: 518-442-4578<tel:(518)%20442-4578>
_____________________________________________

From: Tyle, Kevin R
Sent: Monday, July 17, 2017 4:05 PM
To: 'wrf-users at ucar.edu<mailto:wrf-users at ucar.edu>' <wrf-users at ucar.edu<mailto:wrf-users at ucar.edu>>
Subject: Segfaults with Intel 2017 Compiler

Hi,

I’ve experienced segfaults in various WRF versions (3.7.1 and 3.9) that appear tied to executables built using the current (17.0.4) version of the Intel compilers.

With debugging set at a value of 100 in namelist.input, I consistently see “CALL rrtmg_lw” as the last thing written to the rsl.error.0000 file prior to the segfault.

Longwave schemes 4 (3.7.1 and 3.9) and 24 (3.9) have been tried with similar behavior.

This is using openmpi (have tried the latest 1.10 and 2.1 releases).  Stacksize is unlimited.

Intel compiler version 14.0.1 does not show this problem, nor does the latest PGI 2017 community release.

Any ideas or suggestions would be appreciated!

--Kevin

_____________________________________________
Kevin Tyle, Manager of Departmental Computing
Dept. of Atmospheric & Environmental Sciences
University at Albany
Earth Science 235, 1400 Washington Avenue
Albany, NY 12222
Email: ktyle at albany.edu<mailto:ktyle at albany.edu>
Phone: 518-442-4578<tel:(518)%20442-4578>
_____________________________________________


_______________________________________________
Wrf-users mailing list
Wrf-users at ucar.edu<mailto:Wrf-users at ucar.edu>
http://mailman.ucar.edu/mailman/listinfo/wrf-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/wrf-users/attachments/20170721/cb064d70/attachment-0001.html