|
| 1 | +#!/bin/bash |
| 2 | + |
| 3 | +source ${CHIMBUKO_CONFIG} |
| 4 | + |
| 5 | +#export HG_LOG_LEVEL=debug |
| 6 | +#export HG_LOG_SUBSYS=cls |
| 7 | +#export FI_LOG_LEVEL=debug |
| 8 | + |
| 9 | +rank=${SLURM_PROCID} |
| 10 | + |
| 11 | +narg=$# |
| 12 | +if [[ $narg -lt 1 ]]; then |
| 13 | + echo "Require at least one argument: the number of application tasks/ranks" |
| 14 | + exit 1 |
| 15 | +fi |
| 16 | + |
| 17 | +app_tasks=$1 |
| 18 | +echo "rank \"${rank}\" app_tasks \"${app_tasks}\" node \"${SLURMD_NODENAME}\"" |
| 19 | +if [[ ${rank} -ge ${app_tasks} ]]; then |
| 20 | + #Service task |
| 21 | + if [[ ${SLURM_LOCALID} -eq 0 ]]; then |
| 22 | + echo "Launching service task on node ${SLURMD_NODENAME} local ID ${SLURM_LOCALID}" |
| 23 | + echo "Environment:" |
| 24 | + printenv |
| 25 | + |
| 26 | + exec ${chimbuko_services} 2>&1 | tee services.log |
| 27 | + fi |
| 28 | +else |
| 29 | + echo "Driver rank ${rank} waiting for service spin-up" |
| 30 | + while [ ! -f chimbuko/vars/chimbuko_ad_cmdline.var ]; do sleep 1; done |
| 31 | + ad_opts=$(cat chimbuko/vars/chimbuko_ad_opts.var) |
| 32 | + ad_cmd="driver ${TAU_ADIOS2_ENGINE} ${TAU_ADIOS2_PATH} ${TAU_ADIOS2_FILE_PREFIX}-${EXE_NAME} ${ad_opts} -rank ${rank}" |
| 33 | + |
| 34 | + a=1 |
| 35 | + args=("$@") |
| 36 | + |
| 37 | + preop="" |
| 38 | + |
| 39 | + while [ $a -lt $narg ]; do |
| 40 | + arga=${args[$a]} |
| 41 | + if [[ "$arga" == "--core_bind" ]]; then |
| 42 | + ap1=$(( a + 1 )) |
| 43 | + bnd=${args[$ap1]} |
| 44 | + |
| 45 | + rbnd="None" |
| 46 | + cnt=0 |
| 47 | + for i in $(echo $bnd | tr ":" "\n") |
| 48 | + do |
| 49 | + if [[ $cnt -eq $rank ]]; then |
| 50 | + rbnd=$i |
| 51 | + fi |
| 52 | + cnt=$(( cnt + 1 )) |
| 53 | + done |
| 54 | + |
| 55 | + if [[ "$rbnd" == "None" ]]; then |
| 56 | + echo "Could not parse binding for rank ${rank}" |
| 57 | + exit 1 |
| 58 | + fi |
| 59 | + |
| 60 | + cmask=`python3 <<EOF |
| 61 | +import sys |
| 62 | +
|
| 63 | +#Provide a comma-separated list of CPUs |
| 64 | +#Output: binary and hex masks corresponding to the list |
| 65 | +bnd="${rbnd}" |
| 66 | +lst=bnd.split(',') |
| 67 | +
|
| 68 | +val=0 |
| 69 | +for v in lst: |
| 70 | + val = val | (1<<int(v)) |
| 71 | +
|
| 72 | +print(hex(val)) |
| 73 | +EOF` |
| 74 | + |
| 75 | + echo "Binding rank ${rank} to ${rbnd} with core mask ${cmask}" |
| 76 | + preop="taskset ${cmask} ${preop}" |
| 77 | + a=$(( a + 2 )) |
| 78 | + else |
| 79 | + echo "Unrecognized argument \"$arga\"" |
| 80 | + exit 1 |
| 81 | + fi |
| 82 | + done #end while |
| 83 | + |
| 84 | + exec ${preop} ${ad_cmd} 2>&1 | tee chimbuko/logs/ad.${SLURM_PROCID}.log |
| 85 | +fi |
0 commit comments