Fix Dashboard builds that launch too many OpenMP threads, causing test time-outs
Created by: mhoemmen
@trilinos/framework
Some Dashboard builds have failing tests due to time-outs, because the tests spawn too many OpenMP threads. See, for example, this test:
https://testing.sandia.gov/cdash/testDetails.php?test=41662515&build=3135886
which includes the following output:
Kokkos::OpenMP::initialize WARNING: You are likely oversubscribing your CPU cores.
Detected: 72 cores per node.
Detected: 4 MPI_ranks per node.
Requested: 36 threads per process.
ascic141
appears to be the one machine on which this occurs:
https://testing.sandia.gov/cdash/index.php?project=Trilinos&subproject=Tpetra
I'm not sure what would be the best way to fix this, but you could always just set the OMP_NUM_THREADS
environment variable somehow.