Greetings!
I'm trying to get a BHIST run going with cesm 2.2, and am running into some compiler errors I haven't seen before. I'm using the intel compiler suite on linux, and have a machine configuration set up that works for F2000Climo among others, so the general configuration should be OK. I suspect BHIST is pulling in some other components I haven't needed to build before, which is tickling some compiler configuration issue.
Here is the case setup and end of the build output including the error:
I can get things to compile successfully by editing cesm2/cime/src/share/util/shr_vmath_mod.F90 and adding
to turn off all of the optimized vd* functions. From the code, it looks like these should be defined for the Intel compiler, though, so I'm puzzled that this is necessary.
Does anyone have ideas about what might be going wrong here? Maybe there is some extra ifort flag I need to add somewhere?
Does anyone have a sense of how much performance I'm likely to leave on the table by disabling these? (No need to create more CO2 by failing to optimize my compiler settings... :)
Thanks for looking!
-Rob
I'm trying to get a BHIST run going with cesm 2.2, and am running into some compiler errors I haven't seen before. I'm using the intel compiler suite on linux, and have a machine configuration set up that works for F2000Climo among others, so the general configuration should be OK. I suspect BHIST is pulling in some other components I haven't needed to build before, which is tickling some compiler configuration issue.
Here is the case setup and end of the build output including the error:
$ create_newcase --case 0716_BHIST_10day --compset BHIST --res f09_g17
$ cd 0716_BHIST_10day
$ ./xmlchange STOP_OPTION=ndays,STOP_N=10,REST_OPTION=ndays,REST_N=3
$ ./case.setup && ./case.build && time ./case.submit
....
cism built in 187.321711 seconds
Component atm build complete with 555 warnings
cam built in 204.794333 seconds
Building cesm from /data/cesm/opt/ncar/cesm2/cime/src/drivers/mct/cime_config/buildexe with output to /data/home/jrvb/timings/0716_BHIST_10day/bld/cesm.bldlog.210716-183558
/data/cesm/opt/ncar/cesm2/cime/src/share/util/shr_vmath_mod.F90:44: undefined reference to `vdsqrt_'
/data/cesm/opt/ncar/cesm2/cime/src/share/util/shr_vmath_mod.F90:105: undefined reference to `vdexp_'
/data/cesm/opt/ncar/cesm2/cime/src/share/util/shr_vmath_mod.F90:130: undefined reference to `vddiv_'
/data/cesm/opt/ncar/cesm2/cime/src/share/util/shr_vmath_mod.F90:164: undefined reference to `vdln_'
/data/cesm/opt/ncar/cesm2/cime/src/share/util/shr_vmath_mod.F90:194: undefined reference to `vdsin_'
/data/cesm/opt/ncar/cesm2/cime/src/share/util/shr_vmath_mod.F90:222: undefined reference to `vdcos_'
Component cesm exe build complete with 29 warnings
ERROR: BUILD FAIL: buildexe failed, cat /data/home/jrvb/timings/0716_BHIST_10day/bld/cesm.bldlog.210716-183558
I can get things to compile successfully by editing cesm2/cime/src/share/util/shr_vmath_mod.F90 and adding
#define NO_SHR_VMATH 1
to turn off all of the optimized vd* functions. From the code, it looks like these should be defined for the Intel compiler, though, so I'm puzzled that this is necessary.
Does anyone have ideas about what might be going wrong here? Maybe there is some extra ifort flag I need to add somewhere?
Does anyone have a sense of how much performance I'm likely to leave on the table by disabling these? (No need to create more CO2 by failing to optimize my compiler settings... :)
Thanks for looking!
-Rob