The Extreme-scale Scientific Software Stack (E4S)

Manual Installation Instructions

References

Spack General Documentation

Spack Environments Documentation

Spack Build Cache Documentation

E4S Model Environment Repository

E4S 23.11 Model Environments

E4S Homepage

Instructions

If you want to use Spack’s Clingo-based concretizer, you can either let Spack bootstrap Clingo for you, or you can install it manually via pip, in which case Spack will not need to bootstrap it. To install it via pip:

$> pip install --user --upgrade pip
$> pip install --user clingo

Clone Spack, checkout the appropriate release branch, if desired, and source the Spack setup-env.sh script.

$> git clone https://github.com/spack/spack
$> (cd spack && git checkout e4s-23.11)
$> . spack/share/spack/setup-env.sh

Configure Spack to know where the E4S Build Cache is located. You can use either a release specific build cache, or the mixed build cache. The release specific build cache contains only binaries from the particular release, whereas the mixed build cache contains binaries from all releases and from in between releases.

Release-specific (we recommend using this if you can)

$> spack mirror add E4S https://cache.e4s.io/23.11
$> spack buildcache keys -it

Mixed Cache

$> spack mirror add E4S https://cache.e4s.io
$> spack buildcache keys -it

Obtain a copy of the E4S Model Environment for the E4S release you are interested in, and rename it to spack.yaml, if needed. These are found in the E4S Environment Repository. For this example, we will use the E4S 23.11 Model Environment for X86_64 Systems with NVIDIA GPUs.

Tweak the model environment so that it suits your needs: remove packages you aren’t interested in, select the correct GPU variants, specify the microarchitecture target and compiler selections, as needed. Once modified, you can concretize the environment, which you should do from the same directory that contains your environment.

$> time spack -e . concretize -f | tee concretize.log

Start the install, specifying -jN where N reflects an appropriate degree of parallelism to use for the builds, as in make -jN.

$> spack -e . install -jN

Build times can be quite a long time depending on the modifications you make to the environment and the capabilities of the machine you use for building. If you are using an E4S release branch of Spack, and using a minimally modified E4S release environment where the same compilers, operating system, and target architecture are retained, that will maximize your ability to pull pre-built packages from the build cache.

Please contact Sameer Shende at sameer [at] cs.uoregon.edu if you have any questions.