search

LLNL / hiop

HPC solver for nonlinear optimization problems
Other
210 stars 42 forks source link

Install Ginkgo via Spack for HiOp #656

Open nychiang opened 1 year ago

nychiang commented 1 year ago

I tried to install ginkgo, using the same configuration given in newellVariables.sh, i.e., ginkgo@glu_experimental%gcc@8.3.1+cuda~develtools~full_optimizations~hwloc~ipo~oneapi+openmp~rocm+shared build_type=Release cuda_arch=70 arch=linux-rhel7-power9le

but then I got the following error: [+] /usr/tce/packages/cuda/cuda-11.7.0 (external cuda-11.7.0-uokrglpk6zzlvaq2gpfrfgmgtdikqize) ==> Error: Can't extrapolate a URL for version glu_experimental because package ginkgo defines no URLs

Maybe a customized repository needs to be provided.

@cameronrutherford

cameronrutherford commented 1 year ago

I think you just have the spack package spec a little off. The newell modules here are a little old, and so feel free to copy the newer modules from updated ExaGO builds as needed (will be on GitHub soon...) - https://gitlab.pnnl.gov/exasgd/frameworks/exago/-/blob/develop/buildsystem/spack/newell/modules/dependencies.sh

https://github.com/spack/spack/blob/develop/var/spack/repos/builtin/packages/ginkgo/package.py#L29 - the proper syntax for ginkgo for that version is either ginkgo@1.5.0.glu_experimental, ginkgo@git.glu_experimental=1.5.0 or pulling in an external source code version of ginkgo and installing through spack develop or otherwise.

Hopefully that helps. Sorry for getting to this late, as I anticipated this would require more effort when just skimming my emails...