Trilinos GCC 4.8.4 PR build and running of tests is broken

[bartlettroscoe]

@trilinos/framework, @eric-c-cyr, @rppawlo, @trilinos/nox

Expectations

The GCC 4.8.4 auto PR build should work unless the PR code itself is broken.

Current Behavior

The GCC 4.8.4 auto PR build has broken NOX test builds and all of the tests in every package fail, starting in the auto PR testing iteration build last night:

• https://testing-vm.sandia.gov/cdash/index.php?project=Trilinos&parentid=3627737&filtercount=3&showfilters=1&field1=groupname&compare1=61&value1=Pull%20Request&field2=buildstarttime&compare2=84&value2=NOW&field3=buildstarttime&compare3=83&value3=1%20week%20ago&filtercombine=and

YOu can see the NOX failures at:

• https://testing-vm.sandia.gov/cdash/viewBuildError.php?buildid=3627737

And all of the tests fail because --bind-to none is being used with mpiexec with OpenMP 1.6.5 as shown, for example, at:

• https://testing-vm.sandia.gov/cdash/testDetails.php?test=48792061&build=3627756

which shows the commandline:

"/projects/sems/install/rhel6-x86_64/sems/compiler/gcc/4.8.4/openmpi/1.6.5/bin/mpiexec \"--bind-to\" \"none\" \"-np\" \"1\" \"/scratch/trilinos/workspace/trilinos-folder/Trilinos_pullrequest_gcc_4.8.4/pull_request_test/packages/kokkos-kernels/unit_test/KokkosKernels_blas_serial.exe\" \"--gtest_filter=-serial.gemm_double\"

As described in #2462 and #2788, you can't use those options with OpenMPI 1.6.5. OpenMPI 1.10.1 is the only version in the SEMS env where those options work (see #2462 and #2788).

It is important to fix this ASAP since every PR build will fail and therefore no one will be able to merge any PRs until this gets fixed.

Motivation and Context

I can't merge my PR #2964.

Definition of Done

PR builds are working unless code changes break them.

Possible Solution

Use the source and *.cmake scripts in #2788. That will fix it.

Steps to Reproduce

Your Environment

N.A. This is the auto PR testing having this issue.

Related Issues

• Blocks: Disable some individual Kokkos and KokkosKernels tests on a few more full debug builds and disable the Belos_pseudo_stochastic_pcg_hb_[0,1]_MPI_4 tests on a few more platforms #2964

[mhoemmen]

OpenMPI 1.6.x is way too old for us to use. OpenMPI doesn't even support 1.10.x any more, but at least it's not "ancient."

[prwolfe]

See my comment in #2788

[bartlettroscoe]

Looks like the issue with the GCC 4.8.4 build has been resolved since my last auto PR build for PR #2964 all passed and allowed me to merge.

Should we keep this issue open or close it?

[prwolfe]

@bartlettroscoe - I think the other 2 issues cover this well enough - lets close this one.