Using SPACK

HPCW provides spack recipes for all of its applications. To install spack (if necessary) and add the package sources, execute the following:

git clone https://github.com/spack/spack
. spack/share/spack/setup-env.sh
spack repo add --scope site $HPCW_SOURCE_DIR/spack/hpcw

Models "spackified"

The following models are available in HPCW:

- dwarf-p-cloudsc
- ecrad
- ectrans
- icon
- nemo@4.0.5-14480 cfg=BENCH
- nemo@4.0.5-14480 cfg=ORCA2_ICE_PISCES del_keys=key_top add_keys=key_iomput
- nicamdc

There are also recipes for dependencies that do not have existing recipes in current spack releases.

You could get an overview of the dependencies for a model and how it would be concretized with:

spack spec -Il <model_spec>

The model can then be installed with:

spack install $model_spec

Recommended: spack environments

Spack environments allow for a stable and easily repeatable installation of packages. HPCW ships with some example environments that show how they can be used on different systems.

Using the models installed via spack

If the desired application is already available on your system or you installed it via spack, the test cases provided in HPCW can be executed with the help of the build wrapper.

For example, after installing icon via spack, all tests can be executed with the spack-built binary by calling the build wrapper with the -DUSE_SYSTEM_icon option as such:

$HPCW_SOURCE_DIR/toolchains/build-wrapper.sh $HPCW_SOURCE_DIR levante/dkrz-intel.env.sh --with=icon \
    --test --cmake-flags="-DUSE_SYSTEM_icon=on" |& tee icon.log

Prior to calling the build wrapper, ensure that the binary is available in you $PATH, for example by calling spack load.

Tested spack version

All packages are continuously tested on spack version 0.20.1. Later versions likely work as well, but there is no guarantee.

Useful links

• https://spack-tutorial.readthedocs.io/en/lanl19/tutorial_developer_workflows.html
• https://spack.readthedocs.io/en/latest/environments.html