Building pio2 fails when porting CESM3.0 to a new machine

[lprates]

I am working on porting CESM3.0 (cesm3.0-alphabranch, with describe version output cesm3_0_beta03-285-g8a3a165) to Trillium. I have successfully ported previous versions of CESM (cesm2.1.0 and 2.1.5) to this machine following the DRAC guide for their older machine, Niagara. I want to set up a spectral element CAM run test case using F2000climo and ne30_ne30_mg16 with CESM3, but the build process is consistently failing to build pio2, and any solutions I have found to fix this cause problems downstream. Here is the list of steps that I took to create the case:

First, I downloaded CESM3.0 using

git clone -b cesm3.0-alphabranch https://github.com/ESCOMP/CESM.git /path/to/CESM

Then, I populated the model components with

./bin/git-fleximod update

In ccs_config/machines/config_machines.xml, I added the NODENAME_REGEX for trillium,

<value MACH="trillium">tri*</value>

I then created the ccs_config/machines/trillium directory, where I added config_machines.xml and config_batch.xml files based on the DRAC guide for Niagara. Since I noticed there are no config_compilers.xml files in this version of CESM, I instead created a file called intel_trillium.cmake and tried to make it consistent with the previous config_compilers.xml file, although the inclusion of this file appears to be irrelevant to the problem at hand. I have attached these three files below.

I created the case with the following command:

./create_newcase --case $SCRATCH/CESM/cesm3_0/cases/test_ne30np4_cam6 --compset F2000climo --res ne30_ne30_mg16 --machine trillium --run-unsupported

In the case directory, I made to changes to the xml files and simply ran ./case.setup followed by ./case.build, which resulted in the following error while building pio2:

CMake Error at src/clib/CMakeLists.txt:168 (message):
  Must have PnetCDF and/or NetCDF C libraries

I tried to fix this by pointing the NETCDF_C_PATH environment variable in env_machine_specific.xml to the netcdf-mpi installation rather than netcdf-c++4, like so

<env name="NETCDF_C_PATH">/cvmfs/soft.computecanada.ca/easybuild/software/2023/x86-64-v4/MPI/intel2023/openmpi4/netcdf-mpi/4.9.2/</env>

The result was that pio2 built successfully, but this caused another crash downstream in building cam, with the error

ERROR: yaml-cpp library not found at /scratch/lprates/CESM/cesm3_0/output/test_ne30np4_cam6_compiled_w_c/bld/lib/lib/libyaml-cpp.a

I have also attached the associated log file. I would really appreciate if someone could help me identify how to fix this problem. I believe I have provided a complete list of steps for what I have done so far, but please let me know if any more details are required.

 

[fischer]

Hi Lucas,

Please update to the latest cesm3 beta tag and try again.

git clone -b cesm3_0_beta08 GitHub - ESCOMP/CESM: The Community Earth System Model

There's a new CAM_LINKED_LIBS variable that involves yaml. Hopefully this will fix your issue.

Also a couple of other things. You shouldn't be using the cesm3.0-alphabranch, it can be unstable. Also the ne30_ne30_mg16 is a legacy grid, you should be using ne30_ne30_mg17.

Chris

 

[lprates]

Hi Chris,

Thanks for the reply. I repeated the process from my original post with cesm3_0_beta08 and the new grid. I checked my env_build.xml file and CAM_LINKED_LIBS was set to the default value "-ltuvx -lyaml-cpp -lstdc++", but the build still failed with the same error yaml-cpp error.

- Lucas

 

[fischer]

Hi Lucas,

I'm trying to track down why libyaml-cpp.a isn't being built for you. I am making some progress. But I believe if you try cesm3_0_alpha08k instead of cesm3_0_beta08, you should be able to build. CAM has added a new external, tuv-x, that's suppose to be building the yaml library.

Chris