Package Details: petsc 3.21.5-1

Git Clone URL: https://aur.archlinux.org/petsc.git (read-only, click to copy)
Package Base: petsc
Description: Portable, extensible toolkit for scientific computation
Upstream URL: https://petsc.org
Keywords: computing scientific
Licenses: BSD
Provides: petsc4py
Submitter: heitzmann
Maintainer: MartinDiehl
Last Packager: MartinDiehl
Votes: 19
Popularity: 0.000003
First Submitted: 2018-02-24 11:36 (UTC)
Last Updated: 2024-09-01 08:58 (UTC)

Dependencies (30)

Sources (2)

Pinned Comments

MartinDiehl commented on 2022-10-06 10:26 (UTC)

@jrohwer

When building PETSc (more specifically, petsc4py), one test (ex100 from ksp) will fail if a previous (major) version is installed. I could not figure out why this happens. The solution would be to build in a clean root (which is a little bit complicated due to dependency on other AUR packages) or simply uninstall the old version before.

Any help to solve this issue is welcomed.

Latest Comments

« First ‹ Previous 1 2 3 4 5 6 7 .. 10 Next › Last »

lahwaacz commented on 2024-02-23 17:17 (UTC)

The tests should behave as before with openmpi 5.0.2-4 where the warnings were disabled (by default): https://gitlab.archlinux.org/archlinux/packaging/packages/openmpi/-/commit/ca64cda9e8dcf1460936132eb73e666741a5c38e

So you might want to remove the following from the PKGBUILD:

export OMPI_MCA_opal_warn_on_missing_libcuda=0
if [ -z "$(ldconfig -p | grep libamdhip64.so)" ] || [ -z "$(ldconfig -p | grep libucc.so)" ]; then
  echo "skipping tests"
else

The petsc check target is still bad though, treating all warnings as errors is very naive...

MartinDiehl commented on 2024-02-23 10:13 (UTC)

@Alad: That was my mistake, installing CUDA should not be necessary. It's fixed now.

Alad commented on 2024-02-23 04:03 (UTC) (edited on 2024-02-23 04:20 (UTC) by Alad)

I managed to get past the dependency issues with trilinos 15 and a chroot build, but now tests ex19 and ex47 fail.

edit: I suppose this is due to the CUDA issue mentioned below. Installing cuda into the chroot worked (arch-nspawn <chroot_dir> pacman -S cuda).

MartinDiehl commented on 2024-02-22 21:56 (UTC)

@Alad: Did you recompile after updating openMPI to 5.x?

Alad commented on 2024-02-18 16:11 (UTC)

I have trilinos installed but get the following error:

*********************************************************************************************
           UNABLE to CONFIGURE with GIVEN OPTIONS (see configure.log for details):
---------------------------------------------------------------------------------------------
  Unable to find ml in default locations!
  Perhaps you can specify with --with-ml-dir=<directory>
  If you do not want ml, then give --with-ml=0
  You might also consider using --download-ml instead
*********************************************************************************************

MartinDiehl commented on 2024-02-10 10:42 (UTC)

Unfortunately, OpenMPI with CUDA shows warnings if CUDA is not available. This causes the failure of PETSc tests. See https://github.com/open-mpi/ompi/issues/12314 for the upstream bug report. For the moment, I recommend to simply disable the tests in PKGBUILD, I'll provide a sustainable solution soon.

lahwaacz commented on 2024-02-09 20:58 (UTC)

The messages like

mca_base_component_repository_open: unable to open mca_accelerator_cuda: libcuda.so.1: cannot open shared object file: No such file or directory (ignored)

etc. are just warnings, not erros. It is possible that there are problems with OpenMPI 5.0, but you need to investigate more to find the actual error.

« First ‹ Previous 1 2 3 4 5 6 7 .. 10 Next › Last »