Unable to compile cam6_3_154 with nvhpc/24.3 on Derecho #1025
Comments
|
@brian-eaton - cam6_3_154 is the gust updates tag. Since you are down that rabbit hole right now, I'm wondering if you might have seen this? |
|
I've taken a look at the cesm.bldlog files from both runs. The compile commandline for med_kind_mod.F90 in the run that fails (in the output posted above) contains while the run that succeeds does not have any
|
|
Hi @brian-eaton , thank you very much for tracking down this issue and providing so many helpful details. On Derecho, once we load the I also did not find where the |
|
Hi @sjsprecious, I believe that PIO is in the ESMF library. So the PIO_INCDIR doesn't need to be set, just the ESMF include dir would need to be set. |
|
The successful run in |
|
Thanks @fischer-ncar for your information. That is good to know. By comparing the changes between
Looking at the changes for Looking at the changes for
Could one of them lead to the different behavior between |
|
@sjsprecious, which cime tag are you using. I was able to get your test to work with the latest cesm2_3_alpha17e tag. That's using cime6.0.238_httpsbranch01. |
|
Hi @fischer-ncar , thanks for your quick reply. I used I switched to However, I checked out the Thus it might be caused by something other than cime? |
|
Hi @brian-eaton and @fischer-ncar , it turns out that by switching to the latest The latest ccs_config tag includes a bug fix for PIO and NVHPC compiler specifically on Derecho (ESMCI/ccs_config_cesm#168), thanks to @jedwards4b . However, it is still a mystery to me why |
|
@brian-eaton I think I find the reason about the failure of
If this makes sense to you, I think the conclusion here is to use the latest |
|
@sjsprecious, Nice work tracking this down. Cheryl will soon be updating CAM's externals and the |
|
Thanks @brian-eaton . That sounds great! |
What happened?
When I checked out the CAM tag
cam6_3_154with theccs_config0.0.99tag, I failed to build aF2000devtest on Derecho. The error message in thecesm.bldlog.*file looks like:This error happens because there is an option
-Iwithout a path appended to it somehow.When I checked out the CAM tag
cam6_3_153with theccs_config0.0.99tag, I was able to build the sameF2000devcase on Derecho successfully.What are the steps to reproduce the bug?
./create_newcase --case /path_to_case --mach derecho --res ne30pg3_ne30pg3_mg17 --compset F2000dev --compiler nvhpc --ngpus-per-node 4 --gpu-type a100 --gpu-offload openacc
cd /path_to_case
./case.setup
./case.build
What CAM tag were you using?
cam6_3_154
What machine were you running CAM on?
CISL machine (e.g. cheyenne)
What compiler were you using?
NVHPC
Path to a case directory, if applicable
/glade/derecho/scratch/sunjian/cam6/case/F2000dev.ne30pg3_ne30pg3_mg17.derecho.nvhpc.gpu04_pcols00016_mpi0128_thread001_rrtmg.aer_bug_154
Will you be addressing this bug yourself?
No
Extra info
The cesm error log for cam6_3_154 could be found on Derecho at
/glade/derecho/scratch/sunjian/cam6_run/F2000dev.ne30pg3_ne30pg3_mg17.derecho.nvhpc.gpu04_pcols00016_mpi0128_thread001_rrtmg.aer_bug_154/bld/cesm.bldlog.240503-115432.The successful build of cam6_3_153 could be found on Derecho at
/glade/derecho/scratch/sunjian/cam6_run/F2000dev.ne30pg3_ne30pg3_mg17.derecho.nvhpc.gpu04_pcols00016_mpi0128_thread001_rrtmg.aer_bug_153.The text was updated successfully, but these errors were encountered: