1. 21 Dec, 2016 13 commits
    • Siva Rajamanickam's avatar
      kokkoskernels (tpetra/kernels): · 839fa8ba
      Siva Rajamanickam authored
         Lawyer stuff. Making copyright headers consistent etc.
      839fa8ba
    • crtrott's avatar
    • Jonathan Hu's avatar
      MueLu: rebasing gold files · ce09cf3b
      Jonathan Hu authored
      This should fix some failing nightly tests.
      ce09cf3b
    • crtrott's avatar
      KokkosKernels: SPMV Benchmark fix to more reader issues · 0b7ef2e4
      crtrott authored
      Can now handle pattern files (i.e. no values for the matrix are contained)
      and symmetric matrices with 0s on the diagonal.
      0b7ef2e4
    • Mark Hoemmen's avatar
      Tpetra: Fix #946 (idot test failure) · a9688972
      Mark Hoemmen authored
      @trilinos/tpetra
      
      Tpetra::Details::idot exercises the sendbuf == recvbuf case of
      Tpetra::Details::iallreduce.  That case had poor test coverage.  It
      worked fine for MPI_VERSION >= 3, but the MPI_VERSION < 3
      implementation was not using MPI_IN_PLACE when it should have.  This
      commit fixes that.  As a result, the test now passes.  I tested it
      with both OpenMPI 1.6.5 (MPI_VERSION < 3) and OpenMPI 1.10.1
      (MPI_VERSION >= 3).
      
      I also made sure that Tpetra::Details::idot builds correctly with
      Tpetra::MultiVector inputs and rank-1 Kokkos::View output.
      
      Build/Test Cases Summary
      Enabled Packages: TpetraCore
      Disabled Packages: PyTrilinos,Claps,TriKota
      0) MPI_RELEASE_DEBUG_SHARED_PT => passed: passed=116,notpassed=0 (3.02 min)
      a9688972
    • gregvw's avatar
      Added header to barrier objective test · f57b70a2
      gregvw authored
      f57b70a2
    • gregvw's avatar
    • crtrott's avatar
      KokkosKernels: fix initialization of vector in SPMV Benchmark · cacbf23b
      crtrott authored
      The initialization of input and output vector assumed square matrices.
      Both y and x were initialized in a loop of numCols length.
      cacbf23b
    • Denis Ridzal's avatar
      Minor changes to thermal fluids. Fixed the PDE Jacobian through · c80ea85b
      Denis Ridzal authored
      pressure pinning.  The physics are still off.
      c80ea85b
    • Roscoe A. Bartlett's avatar
      Run with bash -e and allow skip of load_ci_sems_dev_env.sh (#482, #945, #946) · dde3ea7d
      Roscoe A. Bartlett authored
      Now run with '#!/bin/bash -e' to abort if anything fails before running
      checkin-test.py.  That way, if there is any problem, such as not being able to
      load the SEMS env, then the script will crash and avoid pushing with the wrong
      env (see #946 for an example).
      
      I also added the env var TRILINOS_CHECKIN_TEST_SEMS_SKIP_MODULE_LOAD to allow
      people in extreme cases to skip sourcing load_ci_sems_dev_env.sh.  This can be
      used to allow people to push from RHEL 7 where the standard SEMS CI dev env
      does not work correctly out of the box (see #945 for an example).
      
      Build/Test Cases Summary
      Enabled Packages:
      Disabled Packages: PyTrilinos,Claps,TriKota
      Enabled all Packages
      0) MPI_RELEASE_DEBUG_SHARED_PT => passed: passed=2259,notpassed=0 (18.84 min)
      Other local commits for this build/test group: 0485bbf8
      dde3ea7d
    • Roscoe A. Bartlett's avatar
      Purge modules before loading and add sems-git/2.10.1 (#482, #246) · 0485bbf8
      Roscoe A. Bartlett authored
      It turns out that if you have some SEMS modules loaded manaully before
      sourcing load_sems_dev_env.sh, then some module loads will fail and the user
      will not have the right env.  That can lead to bad pre-push CI testing (see
      
      Therefore, to make these scripts more robust, now a 'module purge' is done
      unconditionally.  Given the structure of the SEMS modules, I don't see a
      robust way to avoid doing a 'module purge'.  But note that one can just open a
      new shell and then source this to be safe.
      
      The unload_sems_dev_env.sh also does an unconditional 'module purge' and wipes
      out the var TRILINOS_SEMS_DEV_ENV_LOADED.
      
      I also added the module sems-git/2.10.1 so as to make the checkin-test-sems.sh
      script work more robustly.
      0485bbf8
    • Roscoe A. Bartlett's avatar
      Set SEMS ParMETIS libs explicitly (#945) · 0960c822
      Roscoe A. Bartlett authored
      For some versions of CMake (2.8.12 and 3.7.0) and GCC (4.7.2), CMake will
      refuse to find the SEMS-installed ParMETIS libs with FIND_LIBRARY().  I have
      not seen this problem with any other TPL other than the SEMS-installed version
      of ParMETIS.
      
      I have pinged Kitware about this issue and am working to get Brad King access
      to our env so that he can reproduce this issue.  But until this can be fixed,
      this is a workaround to allow us to use newer CMake versions with the SEMS
      env.  This should allow the usage of the checkin-test-sems.sh script on RHEL 7
      systems with the SEMS env (see #945).
      
      Build/Test Cases Summary
      Enabled Packages:
      Disabled Packages: PyTrilinos,Claps,TriKota
      Enabled all Packages
      0) MPI_RELEASE_DEBUG_SHARED_PT => passed: passed=2259,notpassed=0 (87.86 min)
      0960c822
    • crtrott's avatar
      KokkosKernels: add spmv performnace test · 158c38a1
      crtrott authored
      This test can be compiled without cmake and can include MKL, CuSparse and a
      custom local kernel in addition to the Kokkos kernels spmv call.
      It also includes a basic matrix-market reader as well as a binary format
      reader and writer.
      
      Addresses #948
      158c38a1
  2. 20 Dec, 2016 5 commits
  3. 19 Dec, 2016 3 commits
  4. 18 Dec, 2016 3 commits
    • Jenkins's avatar
      Updating testbed configure scripts. · 5f1c389c
      Jenkins authored
      This adds proper blas/lapack handling.
      5f1c389c
    • Mark Hoemmen's avatar
      Tpetra: Add run-time-length sorts (progress on #834 and #941) · cb72be81
      Mark Hoemmen authored
      @trilinos/tpetra For #834 and #941, we need sort routines that take the
      array length as a run-time parameter, and have the KOKKOS_FUNCTION
      marking.  This commit adds those routines, and tests for them.
      
      Build/Test Cases Summary
      Enabled Packages: TpetraCore
      Disabled Packages: FEI,PyTrilinos,Moertel,STK,SEACAS,ThreadPool,OptiPack,ShyLU,Rythmos,NOX
      0) MPI_RELEASE_DEBUG_SHARED_PT => Test case MPI_RELEASE_DEBUG_SHARED_PT was not run! => Does not affect push readiness! (-1.00 min)
      1) MPI_DEBUG_THREADS => passed: passed=118,notpassed=0 (2.75 min)
      2) SERIAL_RELEASE_THREADS => passed: passed=86,notpassed=0 (2.60 min)
      Other local commits for this build/test group: 79ed3ecc, d39aa900
      cb72be81
    • Mark Hoemmen's avatar
      Tpetra: Add keys-only sort routines (progress on #834) · 79ed3ecc
      Mark Hoemmen authored
      @trilinos/tpetra Make progress on #834, by adding keys-only sort
      routines for arrays of small, compile-time-constant sizes, and a
      test for those routines.
      79ed3ecc
  5. 17 Dec, 2016 3 commits
  6. 16 Dec, 2016 13 commits