pkgs/by-name/mp/mpiCheckPhaseHook/mpi-check-hook.sh

   1 preCheckHooks+=('setupMpiCheck')
   2 preInstallCheckHooks+=('setupMpiCheck')
   3
   4
   5 setupMpiCheck() {
   6   # Find out which MPI implementation we are using
   7   # and set safe defaults that are guaranteed to run
   8   # on any build machine
   9   mpiType="NONE"
  10
  11   # OpenMPI signature
  12   if command ompi_info &> /dev/null; then
  13     mpiType="openmpi"
  14   fi
  15
  16   # MPICH based implementations
  17   if command mpichversion &> /dev/null; then
  18     if [ "$mpiType" != "NONE" ]; then
  19       echo "WARNING: found OpenMPI and MPICH/MVAPICH executables"
  20     fi
  21
  22     version=$(mpichversion)
  23     if [[ "$version" == *"MPICH"* ]]; then
  24       mpiType="MPICH"
  25     fi
  26     if [[ "$version" == *"MVAPICH"* ]]; then
  27       mpiType="MVAPICH"
  28     fi
  29   fi
  30
  31   echo "Found MPI implementation: $mpiType"
  32
  33   case $mpiType in
  34     openmpi)
  35       # Note, that openmpi-5 switched to using PRRTE.
  36       # Thus we need to set PRTE_MCA_* instead of OMPI_MCA_*.
  37       # We keep the openmpi-4 parameters for backward compatability.
  38
  39       # Make sure the test starts even if we have less than the requested amount of cores
  40       export OMPI_MCA_rmaps_base_oversubscribe=1
  41       export PRTE_MCA_rmaps_default_mapping_policy=node:oversubscribe
  42
  43       # Make sure we do not need openssh in the checkPhase
  44       export OMPI_MCA_plm_ssh_agent=false
  45       export PRRTE_MCA_plm_ssh_agent=false
  46
  47       # Disable CPU pinning
  48       export OMPI_MCA_hwloc_base_binding_policy=none
  49       export PRTE_MCA_hwloc_default_binding_policy=none
  50
  51       # OpenMPI get confused by the sandbox environment and spew errors like this (both to stdout and stderr):
  52       #     [hwloc/linux] failed to find sysfs cpu topology directory, aborting linux discovery.
  53       #     [1729458724.473282] [localhost:78   :0]       tcp_iface.c:893  UCX  ERROR scandir(/sys/class/net) failed: No such file or directory
  54       # These messages contaminate test output, which makes the difftest to fail.
  55       # The solution is to use a preset cpu topology file and disable ucx model.
  56
  57       # Disable sysfs cpu topology directory discovery.
  58       export PRTE_MCA_hwloc_use_topo_file="@topology@"
  59       # Use the network model ob1 instead of ucx.
  60       export OMPI_MCA_pml=ob1
  61       ;;
  62     MPICH)
  63       # Fix to make mpich run in a sandbox
  64       export HYDRA_IFACE=lo
  65       # Disable sysfs cpu topology directory discovery.
  66       export HWLOC_XMLFILE="@topology@"
  67       ;;
  68     MVAPICH)
  69       # Disable CPU pinning
  70       export MV2_ENABLE_AFFINITY=0
  71       ;;
  72   esac
  73
  74   # Limit number of OpenMP threads. Default is "all cores".
  75   export OMP_NUM_THREADS=1
  76 }
  77