Skip to content

Switch to CUDA 9.2 build on white/ride (TRIL-215)

Created by: bartlettroscoe

CC: @fryeguy52, @rppawlo, @mhoemmen

Description

This gets an ATDM Trilinos CUDA 9.2 build going on white/ride.

Motivation and Context

We lost the CUDA 8.0 build when 'white' and 'ride' got updated last month. See https://software-sandbox.sandia.gov/jira/browse/TRIL-215.

How Has This Been Tested?

I tested this on 'white' with:

$ bsub -x -Is -q rhel7F -n 16 \
  ./checkin-test-atdm.sh cuda-9.2-debug-Power8-Kepler37 \
 --enable-packages=Kokkos,KokkosKernels,MueLu,Panzer \
 --local-do-all

which returned:

99% tests passed, 1 tests failed out of 235

Subproject Time Summary:
Kokkos           = 1351.83 sec*proc (27 tests)
KokkosKernels    = 1030.02 sec*proc (8 tests)
MueLu            = 2711.91 sec*proc (43 tests)
Panzer           = 7359.42 sec*proc (157 tests)

Total Test time (real) = 1638.34 sec

The following tests FAILED:
	  6 - KokkosCore_UnitTest_PushFinalizeHook_terminate (Timeout)
Errors while running CTest

I also did:

$ time env \
    JOB_NAME=Trilinos-atdm-white-ride-cuda-9.2-debug-panzer \
    WORKSPACE=$PWD \
    Trilinos_PACKAGES=Kokkos,Teuchos \
    CTEST_TEST_TYPE=Experimental \
    CTEST_DO_SUBMIT=ON \
    CTEST_DO_UPDATES=OFF \
    CTEST_START_WITH_EMPTY_BINARY_DIRECTORY=TRUE  \
 ~/Trilinos.base/Trilinos/cmake/ctest/drivers/atdm/smart-jenkins-driver.sh     &> console.out

real    33m45.409s
user    0m6.974s
sys     0m0.769s

which posted to:

These show that there are some test failures but we will submit these new CUDA 9.2 jobs to the "Specialized" CDash group and then set up new Trilinos GitHub issues for the new failures.

Checklist

  • My commit messages mention the appropriate GitHub issue numbers.

Merge request reports