geogrid.exe does not work when recompiled after system upgrade

[rmjes_99]

Hello,

I am using WRF on NERSC's Cori supercomputer, which recently underwent a full operating system upgrade. Since that time, I've updated library paths and recompiled WRF and WPS. All executables re-compile without challenge in WRF, but in WPS (version 4.3.1), executables build without the "executables successfully built" message that I'm used to getting, then geogrid.exe does not work.

I'm configuring for option 37; Cray XC CLE/Linux x86_64, Intel compiler (serial), though have tested option 40 and using in compute nodes as well as the traditional 17 (Intel serial).

When I try to run geogrid.exe, I immediately get a program exception:

Parsed 49 entries in GEOGRID.TBL
Processing domain 1 of 3
Processing XLAT and XLONG
forrtl: severe (168): Program Exception - illegal instruction
Image PC Routine Line Source
geogrid.exe 00000000201A002B for__signal_handl Unknown Unknown
libpthread-2.26.s 0000155549375310 Unknown Unknown Unknown
geogrid.exe 0000000020161861 Unknown Unknown Unknown
geogrid.exe 000000002003938A Unknown Unknown Unknown
geogrid.exe 000000002004194C Unknown Unknown Unknown
geogrid.exe 000000002004A830 Unknown Unknown Unknown
geogrid.exe 000000002000E3F1 Unknown Unknown Unknown
geogrid.exe 000000002000C952 Unknown Unknown Unknown
libc-2.26.so 0000155548FCB3EA __libc_start_main Unknown Unknown
geogrid.exe 000000002000C86A Unknown Unknown Unknown


The only things I've found addressing this online suggest that it's due to MKL flags. I've added -mkl to the LDFLAGS and tested a few other link line options I've found, but not achieved different results when recompiling and attempting to run the executable.


I'm attaching my configure script and compile log.

Does anyone know how I might address this so that geogrid.exe is executable?

Thanks in advance for the help.

 

[kwerner]

Hi,
I see the following in your compile log:

Headers and libraries from cray-libsci/20.09.1 will be ignored because they conflict with -mkl.

and from what I can find online, it seems that this is likely an issue related to your system, and not specifically to WPS. I'd recommend reaching out to a systems administrator at your institution to see if they have any ideas. Let us know if you figure anything out, as it may help someone else in the future. Thanks!

 

[rmjes_99]

Thank you, kwerner.

This has been resolved now. The system administrators prefer WRF be run on KNL nodes (opt 71). In the past, I could compile and run WPS executables without problem as long as the same compiler wrappers were used and modules were loaded / paths correct between WRF and WPS. However, since the recent update this hasn't been the case and WPS is unmatched with WRF. I needed to compile WRF on Haswell nodes (opt 51) then recompile WPS. My executables now run correctly.

Thanks again for your help!

 

[kwerner]

That is great news! Thank you so much for the update. It may be able to help someone else in the future.