PETSC installation issues

[ryan.c...@uconn.edu]

Greetings,

I am having trouble installing Petsc with the given configuration options in the scripts/update_and_rebuild_petsc.sh. I have attached the confiugure.log. I am running Arch linux - kernel 4.20.12, mpi: gcc 8.2.1, mpif77/90: gnu fortran gcc 8.2.1. 

The main problem is that I keep getting errors for superlu_dist and scalapack that the downloaded file cannot be installed with the given configuration. Are there some other flags I can specify to compile and link the packages with Petsc?

[Alexander Lindsay]

Ryan, are you using a system provided MPI and/or gfortran? The error is this:

Executing: mpif90 -c -o /tmp/petsc-VuReVk/config.setCompilers/conftest.o -I/tmp/petsc-VuReVk/config.compilers -I/tmp/petsc-VuReVk/config.types -I/tmp/petsc-VuReVk/config.libraries -I/tmp/petsc-VuReVk/config.setCompilers -fPIC -fopenmp  -g -O  -mfp16-format=ieee  /tmp/petsc-VuReVk/config.setCompilers/conftest.F90
Possible ERROR while running compiler: exit code 256
stderr:
gfortran: error: unrecognized command line option ‘-mfp16-format=ieee’

The `-mfp16-format` flag comes from PETSc's configure scripts

--
You received this message because you are subscribed to the Google Groups "moose-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to moose-users...@googlegroups.com.
Visit this group at https://groups.google.com/group/moose-users.
To view this discussion on the web visit https://groups.google.com/d/msgid/moose-users/09180386-e88b-4776-94cc-1f0cc00e00f6%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[ryan.c...@uconn.edu]

Yes, I'm using the Arch Linux mpi:
https://www.archlinux.org/packages/extra/x86_64/openmpi/

Are there specific flags I need for petsc?

[eDgar]

Dear Ryan,

I managed to compile (see attachment; I hope that helps), but I'm having some issues https://groups.google.com/forum/#!topic/moose-users/q0p1-249eDI

I attach the binary to install openmpi, if you want to trust me. Otherwise, checkout that version from the packages web-page. You can also find setmooseenv.sh in https://www.notabug.org/broncodev/spider.

I hope it helps. Disclaimer: I am a regular user, not developer.

[Fande Kong]

Hi Ray,

Any reason you are not going to work with the default PETSc? PETSc-submodule right now is for only experienced users and developers. We we haven't push that for general users yet.

Fande,

--

You received this message because you are subscribed to the Google Groups "moose-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to moose-users...@googlegroups.com.
Visit this group at https://groups.google.com/group/moose-users.

To view this discussion on the web visit https://groups.google.com/d/msgid/moose-users/a4ace05f-0617-4d62-b533-43106ace2c73%40googlegroups.com.

[Fande Kong]

The first choice is alway to use MOOSE environment  packages, if possible.

[ryan.c...@uconn.edu]

Hi Fande, doesn't libmesh look for PETSc? 

[ryan.c...@uconn.edu]

Thanks eDgar,  I'll give this a shot and see where I end up. I thought that openmpi might be a version too high, but I wasn't sure if there is a way to remedy with a flag. 

Best,
Ryan

[ryan.c...@uconn.edu]

I tried removing the petsc (rm -rf moose/petsc) then using the "setmooseenv.sh" script, but I still get an error that "superlu_dist" could not be used. 

Fande, how can I use the MOOSE Environment package for PETSc?

Best,

Ryan

[Fande Kong]

Hi Ryan,

Could you look at the webpage at https://mooseframework.org/getting_started/index.html?

It introduces how to setup an environment and compile moose.

Fande 

--

You received this message because you are subscribed to the Google Groups "moose-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to moose-users...@googlegroups.com.
Visit this group at https://groups.google.com/group/moose-users.

To view this discussion on the web visit https://groups.google.com/d/msgid/moose-users/0fdcd0f3-3027-4ca0-a361-a28a040d5487%40googlegroups.com.

[Martin Luethi]

Ryan

ryan.c...@uconn.edu writes:
> I tried removing the petsc (rm -rf moose/petsc) then using the
> "setmooseenv.sh" script, but I still get an error that "superlu_dist" could
> not be used.

I had exactly this issue, which somehow seems to be related to openmpi4.

By using mpich automatic install from PETSC configuration with the options

--download-mpich=1 \
--with-mpich=1 \

PETSC compiles and works flawlessly.

HTH, Martin

[Alexander Lindsay]

Martin, thank you for sharing that. Fande, he's running on Arch Linux. We provide deb or rpm packages, so that's not really going to work for him. Ryan, I would try including the configure options that Martin shared when configuring PETSc and see where you get. Sorry for the difficulty thus far!

--
You received this message because you are subscribed to the Google Groups "moose-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to moose-users...@googlegroups.com.
Visit this group at https://groups.google.com/group/moose-users.

To view this discussion on the web visit https://groups.google.com/d/msgid/moose-users/87pnrbkg0m.fsf%40geo.uzh.ch.

[Martin Luethi]

Hi

Arch Linux changed openmpi on Feb 14th to version 4, which apparently
gives these problems. So, either you downgrade to version 3.1.3
(downgrader, or just get that version from the archlinux archive and
install it by hand, procedure is explained on wiki.archlinux.org),
or you let PETSC install mpich the way I indicated.

Best, Martin

[eDgar]

Hello Ryan,

1. I want to remark that I am not part of the MOOSE developing team, and that setmooseenv.sh is possibly unknown to them

2. I think that they would like to see what is the error that you get when compiling

3. I am guessing that you read the PDF that I attached on the installation

   - Are you using OpenMPI 3.x.x?

   - Did you  `source setmooseenv.sh` ? Does your prompt has   (venv) $ >    ?

   - Did you use the petsc-git package from AUR (not the regular petsc)?

   - Could you show the command that you use to configure? (with the full configuration flags)?

   - Did you remove sundials and party from the external packages?

I hope you manage to compile PETSc. It was a headache for me in the past.

[eDgar]

Also, make sure that you really delete any previous installation of PETSc (such as /opt/petsc, /opt/moose/petsc, etc.). There may be interfering files.

[ryan.c...@uconn.edu]

Hi Martin, thanks so much. Do I have to uninstall the current openmpi to allow PETSc to install the compatible version?

So far I have a new error:
/usr/bin/ld: cannot find -l-L/usr/lib/gcc/x86_64-pc-linux-gnu/8.2.1 collect2: error: ld returned 1 exit status

Is this a gcc version problem too? The 8.2.1 folder exists.

[ryan.c...@uconn.edu]

I've made some progress. 

PETSc installs with mpich if I remove the C, CXX, F, F90, and F77 flags 'fPIC -fopenmp'. Then libmesh installs without a problem. 

When I run the moose run_tests, all of the 50 tests fail. Are the flags the root of the problem?

Best,

Ryan

[Jed Brown]

Joining this late, but I build PETSc daily on Arch Linux. The original
error is due to an apparently stale fblaslapack. The "openblas-lapack"
package (in AUR) is a fine way to get an optimized BLAS and LAPACK;
PETSc will find it automatically.

A space is missing. More context would be needed.

[ryan.c...@uconn.edu]

After a week not getting to this issue, I'm back and solved the install. 

To close out this thread an help future Arch linux users, here are my findings:

1. openmpi 4.0 is not compatible with the petsc and subsequently libmesh installs

a. the AUR mpich was missing some headers that prevented libmesh from compiling

b. best course: follow the mpich-3.2 installation instructions explicitly. Remember to keep a sane work environment. I usually use fish (as opposed to bash) so I started a separate terminal with bash and cleared the variables in the mooseframework getting started instructions

c. there's no need to install gcc on Arch. just use the default pacman install

2. There was a stale superlu_dist git package in the PETSc that seems to be working now with petsc-3.9.4

3. After these considerations, the mpich and PETSc installations went well and the update_and_rebuild_libmesh.sh script worked perfectly. Now the 1951 tests ran in 1005.0 seconds and all passed. 

Thanks so much for the guidance. 

-Ryan

[eDgar]

On Saturday, March 23, 2019 at 5:08:40 PM UTC, ryan.c...@uconn.edu wrote:

After a week not getting to this issue, I'm back and solved the install. 

Congratulations!

 

To close out this thread an help future Arch linux users, here are my findings:

1. openmpi 4.0 is not compatible with the petsc and subsequently libmesh installs

Yep, I guess you missed my comments on that.

 

a. the AUR mpich was missing some headers that prevented libmesh from compiling

Oh, that's interesting. It's extra work, but, if you have the time, shoot an e-mail to the maintainer.

 

b. best course: follow the mpich-3.2 installation instructions explicitly. Remember to keep a sane work environment. I usually use fish (as opposed to bash) so I started a separate terminal with bash and cleared the variables in the mooseframework getting started instructions

That's the purpose of having my setmooseenv.sh

 

c. there's no need to install gcc on Arch. just use the default pacman install

2. There was a stale superlu_dist git package in the PETSc that seems to be working now with petsc-3.9.4

I didn't really use it, but will try it.

 

3. After these considerations, the mpich and PETSc installations went well and the update_and_rebuild_libmesh.sh script worked perfectly. Now the 1951 tests ran in 1005.0 seconds and all passed. 

Thanks so much for the guidance. 

Thanks for reporting! I also hope to see more Parabola and Arch users around :) .

 

-Ryan

[eDgar]

Did you get MOOSE to compile with Python 3, by the way?

On Saturday, March 23, 2019 at 5:08:40 PM UTC, ryan.c...@uconn.edu wrote:

[ryan.c...@uconn.edu]

Hi Edgar,

1. openmpi 4.0 is not compatible with the petsc and subsequently libmesh installs

> Yep, I guess you missed my comments on that.

I saw the comment, that's how I knew to try something else. 

 

a. the AUR mpich was missing some headers that prevented libmesh from compiling

> Oh, that's interesting. It's extra work, but, if you have the time, shoot an e-mail to the maintainer.

Good idea, but I would have to reproduce the bug. I forget which headers were missing unfortunately.

> Did you get MOOSE to compile with Python 3, by the way?

No, I have Python 2 and 3, so the install went with Python 2. Would be nice to switch the install to 3.