Issue with undefined references when compiling LIS

[mcthreems]

Hello, I am attempting to compile LIS and have gotten to what seems like the last step in the compilation process. The compilation fails here due to several undefined references in some fortran files (the terminal output is attached as "comp_out.txt"). Many of these references include "mpi" in their name, so it might be an issue with the mpi module. Do I need to somehow point the compilation to the mpi module directly or is the issue elsewhere?
comp_out.txt

[jvgeiger]

Hello,

I see an inconsistency in your comp_out.txt file regarding MPI.

First, you must set the compiler environment variables like this:

export LIS_FC=mpif90
export LIS_CC=mpicc

Second, please check your ESMF library. You are using /lfs/matthewmckinney.ui/LISF/esmf-8.7.0/lib/libO/Linux.gfortran.64.mpiuni.default.
The “mpiuni” says that ESMF was installed without MPI support. Please check your /lfs/matthewmckinney.ui/LISF/esmf-8.7.0/lib/libO directory. Do you have a subdirectory named something like Linux.gfortran.64.mpich3.default? If no, then you must reinstall ESMF with MPI support enabled. If yes, then please fix the LIS_MODESMF and LIS_LIBESMF environment variables to use it.

Then please go into your LISF/lis directory. Run

make -C make realclean
./configure
./compile

When you run the configure utility, make sure to enable MPI support in LIS:

Parallelism (0-serial, 1-dmpar, default=1): 1

You also have a problem with HDF-EOS2. Are you reading data in the HDF4 and/or HDF-EOS2 format?

Thanks,
Jim

[jvgeiger]

Let's make sure that everything was installed correctly. What does ls /lfs/software/spack-jb/opt/spack/linux-rocky8-broadwell/gcc-12.1.0/hdf-4.2.15-qlpnnqcfr3e332jlvvdripw5h3g6olc3 report? And what does ls /lfs/software/spack-jb/opt/spack/linux-rocky8-broadwell/gcc-12.1.0/hdf-4.2.15-qlpnnqcfr3e332jlvvdripw5h3g6olc3/lib` report?

[mcthreems]

The first command outputs "bin include lib share", and the second one outputs "libdf.a libhdf4.settings libmfhdf.a".

[jvgeiger]

That looks correct.

Please check the make/configure.lis file. Make sure that the LIB_HDF5 and INC_HDF5 variables correctly points to your HDF5 library. If not, then you should rerun the configure and compile scripts:

make -C make realclean
./configure
./compile -j 16

[mcthreems]

Those variables do point to the correct paths, so should not be causing the issues. I currently speculate that our issue is conflicting installs of the same modules (netcdf, hdf4, etc) between my own account on our cluster and cluster-wide installs, so I am currently trying to fully switch over to using only cluster-wide modules. Once we get those set up and I attempt a compilation with them I will let you know if there are still issues.

[mcthreems]

We have switched over to cluster-wide installs of the necessary modules, but are having some issues with the LD flags. Specifically the flags "-lnetcdf", "-ljpeg", and "-lhdf5_hl" seem to refer to different versions of modules than the ones we are using, but removing them results in several undefined reference errors.

[jvgeiger]

Hello. All I can think of right now is to check your environment settings.

Examples:

1. Make sure that your modules are setting the correct paths.

2. Make sure that your LD_LIBRARY_PATH is set correctly. Do you have multiple different entries for the same library? I write a custom modulefile for managing my development/running environment for LIS. In it, I load all needed modules and set all needed environment variables. I do not set any of these in my .profile file.

3. Make sure that all your libraries were installed consistently. Make sure that your libraries were compiled to use the same dependencies. I have seen issues when using a mix of libraries compiled manually with libraries installed via spack. spack will install dependecies, so lib1 installed manually and lib2 installed with spack end up referring to two different versions of a lowel level dependency.

[mcthreems]

Thank you for the suggestions. We have been working on this on our end as well and have managed to get the compilation to finish, but need to test whether or not it actually works. Is there a simple test we can perform to check if our LIS executable is working?

[mcthreems]

To finish this thread, the original issue was resolved by both fixing the compiler environment variables and then tracking down which modules included the references that the compiler was unable to find.

[jvgeiger]

Hello,

You can try our Public Testcase Walkthrough:

https://nasa-lis.github.io/LISF/public_testcase_walkthrough/public_testcase_walkthrough.html