Undefined symbols cesm2.1.1 on a Mac

[mark@atmos_colostate_edu]

I'm attempting to build a cesm2.1.1 case on a 2019 Mac Pro running Catalina with the Intel compiler v19.1.0.166. I added my machine specfics to config_machines.xml and config_compilers.xml. The case is:

create_newcase --case cam6_fv2deg_mar2022 --res f19_f19_mg17 --compset F2000climo --machine tundra --run-unsupported

I didn't add any source code changes to SourceMods, nor any changes to the user_nl_xxx files.

All of the individual model components get built successfully, but then I get undefined symbols in the link step. The first of the errors are surprising to me as they are seemingly benign pieces of code from the CLM:

Undefined symbols for architecture x86_64:
"_clm_varorb_mp_eccen_", referenced from:
_lnd_comp_mct_mp_lnd_run_mct_ in libclm.a(lnd_comp_mct.o)
_lnd_comp_mct_mp_lnd_init_mct_ in libclm.a(lnd_comp_mct.o)
_clm_initializemod_mp_initialize2_ in libclm.a(clm_initializeMod.o)
"_clm_varorb_mp_lambm0_", referenced from:
_lnd_comp_mct_mp_lnd_run_mct_ in libclm.a(lnd_comp_mct.o)
_lnd_comp_mct_mp_lnd_init_mct_ in libclm.a(lnd_comp_mct.o)
_clm_initializemod_mp_initialize2_ in libclm.a(clm_initializeMod.o)
"_clm_varorb_mp_mvelpp_", referenced from:
_lnd_comp_mct_mp_lnd_run_mct_ in libclm.a(lnd_comp_mct.o)
_lnd_comp_mct_mp_lnd_init_mct_ in libclm.a(lnd_comp_mct.o)
_clm_initializemod_mp_initialize2_ in libclm.a(clm_initializeMod.o)
"_clm_varorb_mp_obliqr_", referenced from:
_lnd_comp_mct_mp_lnd_run_mct_ in libclm.a(lnd_comp_mct.o)
_lnd_comp_mct_mp_lnd_init_mct_ in libclm.a(lnd_comp_mct.o)
_clm_driver_mp_clm_drv_ in libclm.a(clm_driver.o)
_clm_initializemod_mp_initialize2_ in libclm.a(clm_initializeMod.o)

Any ideas?

Thanks,
Mark Branson

 

[jedwards]

Hi Mark,

I don't see any problems with the config_machines.xml but the lnd.bldlog that you provided is empty. I also looked at the
cice errors and can't make sense of them. It looks like it's looking for a routine ice_scam_mp_scmlat but the closest thing to
that is a module variable scmlat in the cice ice_scam.F90 file which was apparently compiled correctly.

 

[mark@atmos_colostate_edu]

Oops, sorry about that. Here's a new set of log files where the lnd bld log is complete.
There's also a multitude of rrtmg undefined symbols that also don't make any sense either.

 

[jedwards]

I wonder if it has something to do with the ar command on Mac. At the end of each.bldlog I see:

ar: creating archive /Users/mark/projects/scratch/cam6_fv2deg_mar2022/bld/lib/libatm.a

I'm used to seeing more verbose output, listing each file added to the library. I don't know, I'm afraid you've
got me stumped - have you tried with gnu compilers on this system?

 

[mark@atmos_colostate_edu]

Well, if it's any help, here's the output from doing the nm command on the libclm.a that got created on my Mac. I've also attached a corresponding one that I got from compiling the same compset with the Intel compiler on Cheyenne.
I've haven't tried the gnu compilers on this Mac desktop. I actually figured Intel would be easy since it's the default compiler on Cheyenne. But I suppose I could download the gnu suite of compilers and give them a try.

 

[jedwards]

There is a generic mac machine called homebrew defined in config_machines.xml
I don't have any idea why intel19 isn't working - you could also try the latest intel compiler
since it no longer requires a license.

 

[mark@atmos_colostate_edu]

I was finally able to solve the problem. Turns out it's a Mac-specific issue having to do with modules that only declare variables and don't contain any subroutines or functions. You have to do a ranlib -c on all of the component libraries. So I was finally able to get it to compile successfully with the Intel compiler.

Here's a link to the useful article that I found:

Undefined symbols for Fortran module variables in static library on OSX. Problem (and solution)

If you program in Fortran on the Mac, you might meet this odd problem. You have a module test.f90 containing nothing but public variables

stefanoborini.com

--Mark

 

[jedwards]

Awesome -- thanks for the pointer.