I've been working on the following COPRs:
https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi3.1/ https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi4.0/
And collecting notes on problems here:
https://docs.google.com/spreadsheets/d/1jBK_lnsOMcAsgGj8qr0-1Db6wtiegrFQExF2...
https://docs.google.com/spreadsheets/d/1U2YobgRn592CFoiKBxCFvCNTf0vqozTyVYoB...
Many of the 4.0 builds fail because I've left in (for now) the default dropping of the MPI1 API. That affects a bunch of things. I've filed some upstream bugs on many of the issues.
There are some worrying "Illegal instruction" failures on Fedora Rawhide x86_64 builds for a number of packages that I haven't yet tracked down.
Hi, Why does scalapack need mpich-devel-static if using OpenMPI? Might it be possible to add the MPI compiler wrappers as well? Benson
On Fri, Nov 30, 2018, at 3:24 AM, Orion Poplawski wrote:
I've been working on the following COPRs:
https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi3.1/ https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi4.0/
And collecting notes on problems here:
https://docs.google.com/spreadsheets/d/1jBK_lnsOMcAsgGj8qr0-1Db6wtiegrFQExF2...
https://docs.google.com/spreadsheets/d/1U2YobgRn592CFoiKBxCFvCNTf0vqozTyVYoB...
Many of the 4.0 builds fail because I've left in (for now) the default dropping of the MPI1 API. That affects a bunch of things. I've filed some upstream bugs on many of the issues.
There are some worrying "Illegal instruction" failures on Fedora Rawhide x86_64 builds for a number of packages that I haven't yet tracked down.
-- Orion Poplawski Manager of NWRA Technical Systems 720-772-5637 NWRA, Boulder/CoRA Office FAX: 303-415-9702 3380 Mitchell Lane orion@nwra.com Boulder, CO 80301 https://www.nwra.com/ _______________________________________________ scitech mailing list -- scitech@lists.fedoraproject.org To unsubscribe send an email to scitech-leave@lists.fedoraproject.org Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/scitech@lists.fedoraproject.or...
I'm simply rebuilding the full packages from Rawhide - I'm not disabling mpich though I suppose I might if needed.
I don't know what you mean by "Might it be possible to add the MPI compiler wrappers as well?" openmpi-devel contains them:
/usr/lib64/openmpi/bin/mpiCC /usr/lib64/openmpi/bin/mpic++ /usr/lib64/openmpi/bin/mpicc /usr/lib64/openmpi/bin/mpicxx /usr/lib64/openmpi/bin/mpif77 /usr/lib64/openmpi/bin/mpif90 /usr/lib64/openmpi/bin/mpifort
and so forth.
On 11/30/18 9:14 AM, Benson Muite wrote:
Hi, Why does scalapack need mpich-devel-static if using OpenMPI? Might it be possible to add the MPI compiler wrappers as well? Benson
On Fri, Nov 30, 2018, at 3:24 AM, Orion Poplawski wrote:
I've been working on the following COPRs:
https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi3.1/ https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi4.0/
And collecting notes on problems here:
https://docs.google.com/spreadsheets/d/1jBK_lnsOMcAsgGj8qr0-1Db6wtiegrFQExF2...
https://docs.google.com/spreadsheets/d/1U2YobgRn592CFoiKBxCFvCNTf0vqozTyVYoB...
Many of the 4.0 builds fail because I've left in (for now) the default dropping of the MPI1 API. That affects a bunch of things. I've filed some upstream bugs on many of the issues.
There are some worrying "Illegal instruction" failures on Fedora Rawhide x86_64 builds for a number of packages that I haven't yet tracked down.
-- Orion Poplawski Manager of NWRA Technical Systems 720-772-5637 NWRA, Boulder/CoRA Office FAX: 303-415-9702 3380 Mitchell Lane orion@nwra.com Boulder, CO 80301 https://www.nwra.com/ _______________________________________________ scitech mailing list -- scitech@lists.fedoraproject.org To unsubscribe send an email to scitech-leave@lists.fedoraproject.org Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/scitech@lists.fedoraproject.or...
scitech mailing list -- scitech@lists.fedoraproject.org To unsubscribe send an email to scitech-leave@lists.fedoraproject.org Fedora Code of Conduct: https://getfedora.org/code-of-conduct.html List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/scitech@lists.fedoraproject.or...
Hi,
Thanks for maintaining these packages. Usually the compilers do not show up as the location is not usually one which is on the default path. /usr/local/bin (I think /usr/bin is not for Fedora packages) is probably a better place for the compiler wrappers, since this is usually on the default Path, but /usr/lib64/openmpi/bin/ is not on the default path.
Benson
On 12/4/18 10:38 AM, Benson Muite wrote:
Hi,
Thanks for maintaining these packages. Usually the compilers do not show up as the location is not usually one which is on the default path. /usr/local/bin (I think /usr/bin is not for Fedora packages) is probably a better place for the compiler wrappers, since this is usually on the default Path, but /usr/lib64/openmpi/bin/ is not on the default path.
Benson
You use environment modules to load the proper MPI environment, e.g.:
module load mpi/openmpi-x86_64
Hi Orion,
On Tue, Dec 4, 2018, at 6:48 PM, Orion Poplawski wrote:
On 12/4/18 10:38 AM, Benson Muite wrote:
Hi,
Thanks for maintaining these packages. Usually the compilers do not show up as the location is not usually one which is on the default path. /usr/local/bin (I think /usr/bin is not for Fedora packages) is probably a better place for the compiler wrappers, since this is usually on the default Path, but /usr/lib64/openmpi/bin/ is not on the default path.
Benson
You use environment modules to load the proper MPI environment, e.g.:
module load mpi/openmpi-x86_64
Thanks.
Orion Poplawski orion@nwra.com writes:
I've been working on the following COPRs:
https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi3.1/ https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi4.0/
And collecting notes on problems here:
https://docs.google.com/spreadsheets/d/1jBK_lnsOMcAsgGj8qr0-1Db6wtiegrFQExF2...
Could we use Fedora resources? I'd rather not be used by Google.
I realized recently that EPEL7 should have openmpi3 variants, as RHEL7 has both 1.10 and 3.0. Perhaps that's the first thing to look at? I assume EPEL is more used for MPI than Fedora, though I don't know how much use will be made of the distributed packages as most people don't believe in packaging :-(. (I've made a start with what I maintain, and built the current Scalasca toolset for RHEL8 beta with OMPI 3.1.)
On 12/4/18 10:07 AM, Dave Love wrote:
Orion Poplawski orion@nwra.com writes:
I've been working on the following COPRs:
https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi3.1/ https://copr.fedorainfracloud.org/coprs/g/scitech/openmpi4.0/
And collecting notes on problems here:
https://docs.google.com/spreadsheets/d/1jBK_lnsOMcAsgGj8qr0-1Db6wtiegrFQExF2...
Could we use Fedora resources? I'd rather not be used by Google.
If Fedora has as an online spreadsheet, sure. But I'm not interested in dealing with wiki formatting.
I realized recently that EPEL7 should have openmpi3 variants, as RHEL7 has both 1.10 and 3.0. Perhaps that's the first thing to look at? I assume EPEL is more used for MPI than Fedora, though I don't know how much use will be made of the distributed packages as most people don't believe in packaging :-(. (I've made a start with what I maintain, and built the current Scalasca toolset for RHEL8 beta with OMPI 3.1.)
I'd completely forgotten about that - definitely worth looking at.
My main purpose for the COPRs was to test readiness for updating Rawhide, distributing for older releases a secondary benefit.
scitech@lists.fedoraproject.org
-
Benson Muite -
Dave Love -
Orion Poplawski