spack/spack-configs

Share Spack configuration files with other HPC sites

Spack Configs

This is a repository that sites can use to share their configuration files for Spack. You can contribute your own configuration files, or browse around and look at what others have done.

License

Spack is distributed under the terms of both the MIT license and the Apache License (Version 2.0). Users may choose either license, at their option.

All new contributions must be made under both the MIT and Apache-2.0 licenses.

See LICENSE-MIT, LICENSE-APACHE, COPYRIGHT, and NOTICE for details.

SPDX-License-Identifier: (Apache-2.0 OR MIT)

LLNL-CODE-811652

spack-configs's Issues

Error while installing with parallel cluster and graviton 3 (c7g instances)

For c7g instances architecture is neoverse_v1.

During execution of /AWS/parallelcluster/postinstall.sh it looks for file AWS/parallelcluster/packages-neoverse_v1.yaml but this file is missing in the repository.
I think it should be enough copying packages-neoverse_n1.yaml to packages-neoverse_v1.yaml to solve this issue.

NERSC config fails to compile gmp due to libtool version incompatibility

When using the NERSC packages.yaml file in this repo, spack install flang on Cori fails during compilation of gmp:

==> Installing gmp
==> Searching for binary cache of gmp
==> No binary for gmp found: installing from source
==> Using cached archive: /global/gscratch1/sd/friesen/spack/cache/gmp/gmp-6.1.2.tar.bz2
==> Staging archive: /global/u2/f/friesen/spack/var/spack/stage/gmp-6.1.2-c36gbbzmkcbaythbgglvr5tl2xccsf2m/gmp-6.1.2.tar.bz2
==> Created stage in /global/u2/f/friesen/spack/var/spack/stage/gmp-6.1.2-c36gbbzmkcbaythbgglvr5tl2xccsf2m
==> No patches needed for gmp
==> Building gmp [AutotoolsPackage]
==> Executing phase: 'autoreconf'
==> Executing phase: 'configure'
==> Executing phase: 'build'
==> Error: ProcessError: Command exited with status 2:
    'make' '-j64'

66 errors found in build log:
     693     libtool: and run autoconf again.
     694     libtool: Version mismatch error.  This is libtool 2.4.6, but the
     695     libtool: definition of this LT_INIT comes from libtool 2.4.2.
     696     libtool: You should recreate aclocal.m4 with macros from libtool 2
             .4.6
     697     libtool: and run autoconf again.
     698     Makefile:491: recipe for target 'mp_bases.lo' failed

This error does not occur if I do not use this packages.yaml.

AWS postinstall.sh script doesn't work for Ubuntu 20 based parallel clusters

I tried to launch an ubuntu-based parallelcluster, and the postinstall.sh script fails due to python2/python3 confusion. The pip command comes from the python3-pip package (verified with dpkg -S /usr/bin/pip), but /usr/bin/env python finds python 2.7.18, which doesn't have the pyyaml package installed.

Issue with gromacs definitions in the packages.yaml files

The gromacs entries in the packages.yaml file look to reference a couple of variants that no longer exist in the actual package spec, +blas and +lapack. This leads to errors when trying to concretize:

[buildhpc@hpcbuild01]$ spack spec gromacs
==> Error: cannot emit requirements for the solver: variant "blas" not found in package "gromacs" [required from package "gromacs"]

NERSC: rename system config dirs

Edison was decommissioned in May 2019, so the cori_and_edison config dir should be renamed.

I think it makes sense to have a separate config for every system, so we could rename this dir to cori. Edison and Cori had mostly the same software, and it was convenient to have a single cori_and_edison config for both. But that scenario is not a guarantee; the Perlmutter system software may be different than Cori, in which case having a single config set for both systems won't work, and we would need a separate config dir called perlmutter.

packages.yaml non properly configured wheh installing on graviton2

The installation script postinstall.sh when started on graviton2 instance is trying to download packages-neovers_n1.yaml that does not exists.
Please rename packages-graviton2.yaml into packages-neovers_n1.yaml

How to install?

This repo could really use a systematic organizational structure and method of installing configuration files.

For example, let's say I'm a new user of ANL LCRC (Bebop) and I want to get started with Spack. What I would like to be able to do is run something like:

$ git clone https://github.com/spack/spack-configs/ANL/LCRC ~/.spack

and be ready to go. Git has a "sparse checkout" feature that does something like this, but I'm not sure if that lets me make changes and contribute them back to this repo. Perhaps submodules are needed?

acfl is compiling from source at every set-up

Could you please add an option to /AWS/parallelcluster/postinstall.sh, to avoid installing acfl arm compiler like the one already existing for intel compiler ( --no-intel-compiler)

It is taking a lot of time to set-up since it is not included in binary cache

NERSC Perlmutter package.yaml

It seems that the current site_config/package.yaml is a little out of date already.

cray-mpich 8.1.10 -> 8.1.11
no cuda/11.3.0 module, use cudatoolkit/21.9_11.4
nvidia 21.7 -> 21.9
libfabric 1.11.0.4.79 -> 1.11.0.4.75

Crusher e4s compilers.yaml bug

In the e4s spack.yaml for crusher, there is a [email protected] and a gcc@7-os system compiler option. This is great and very useful.

However, these compilers do not load the PrgEnv-gnu module and therefore all of the other system modules are not loaded correctly. As a result, cray-libsci will find the PrgEnv-cray version of cray-libsci rather the the GNU version. This makes all of the dependent packages fail to build with errors of a missing blas/lapack symbols because it cannot find the cray-libsci library because it is looking in the path of cray for the library name with gnu in it.