[703] | 1 | #!/bin/bash |
---|
| 2 | |
---|
| 3 | # This script provides site-specific functions |
---|
| 4 | # called by the site-independent scripts |
---|
| 5 | # create_runs.sh and submit.sh |
---|
| 6 | |
---|
| 7 | # NB : we must use the queue 'normal' because there seems to be a limit |
---|
| 8 | # on the number of jobs that can be submitted to queue 'test' |
---|
| 9 | |
---|
| 10 | #----------------------------- split_XXX ---------------------------- |
---|
| 11 | |
---|
| 12 | function split_mpi_32() { |
---|
| 13 | setvar nbp 33 nsplit_i 2 nsplit_j 4 |
---|
| 14 | } |
---|
| 15 | function split_serial_40() { |
---|
| 16 | setvar nbp 41 nsplit_i 2 nsplit_j 2 |
---|
| 17 | } |
---|
| 18 | function split_mpi_40() { |
---|
| 19 | setvar nbp 41 nsplit_i 2 nsplit_j 3 |
---|
| 20 | } |
---|
| 21 | function split_mpi_80() { |
---|
| 22 | setvar nbp 81 nsplit_i 4 nsplit_j 3 |
---|
| 23 | } |
---|
| 24 | function split_mpi_omp_40() { |
---|
| 25 | setvar nbp 41 nsplit_i 2 nsplit_j 2 omp_level_size 4 |
---|
| 26 | } |
---|
| 27 | |
---|
| 28 | #----------------- job submission ------------------ |
---|
| 29 | |
---|
| 30 | function submit_job_X64_CURIE() # $1=script $* = SLURM OPTIONS |
---|
| 31 | { |
---|
| 32 | TMP=$(mktemp) |
---|
| 33 | ccc_msub $1 > $TMP |
---|
| 34 | cat $TMP 1>&2 |
---|
| 35 | cat $TMP | awk '{print $NF }' |
---|
| 36 | rm -f $TMP |
---|
| 37 | } |
---|
| 38 | |
---|
| 39 | #------------------------------ job_XXX ----------------------------- |
---|
| 40 | |
---|
| 41 | # Serial jobs |
---|
| 42 | function job_serial() { # EXP_NAME |
---|
| 43 | job_gen $1 1 1 |
---|
| 44 | } |
---|
| 45 | function job_serial_40() { # EXP_NAME |
---|
| 46 | job_serial $1 |
---|
| 47 | } |
---|
| 48 | |
---|
| 49 | # MPI jobs |
---|
| 50 | function job_mpi() { #EXP_NAME NB_MPI |
---|
| 51 | job_gen $1 $2 1 |
---|
| 52 | } |
---|
| 53 | function job_mpi_32() { # EXP_NAME |
---|
| 54 | job_mpi $1 80 |
---|
| 55 | } |
---|
| 56 | function job_mpi_40() { # EXP_NAME |
---|
| 57 | job_mpi $1 60 |
---|
| 58 | } |
---|
| 59 | function job_mpi_80() { # EXP_NAME |
---|
| 60 | job_mpi $1 120 |
---|
| 61 | } |
---|
| 62 | |
---|
| 63 | # MPI-OMP jobs |
---|
| 64 | function job_mpi_omp() { #EXP_NAME NB_MPI NB_OMP |
---|
| 65 | job_gen $1 $2 $3 |
---|
| 66 | } |
---|
| 67 | function job_mpi_omp_40() { # EXP_NAME |
---|
| 68 | job_mpi_omp $1 40 4 |
---|
| 69 | } |
---|
| 70 | |
---|
| 71 | # Generic |
---|
| 72 | function job_gen() { # EXP_NAME MPI_TASKS OMP_TASKS MPIRUN |
---|
| 73 | MPI=$2 |
---|
| 74 | OMP=$3 |
---|
| 75 | MPI_OMP=$((MPI*OMP)) |
---|
| 76 | NODES=$(((MPI_OMP+63)/64)) |
---|
| 77 | VMEM=$((2*MPI_OMP)) |
---|
| 78 | cat <<EOF |
---|
| 79 | #!/bin/bash |
---|
| 80 | #PBS -N $1 |
---|
| 81 | #PBS -q std |
---|
| 82 | #PBS -n |
---|
| 83 | #PBS -l nodes=$NODES:ppn=64 |
---|
| 84 | #PBS -l walltime=00:15:00 |
---|
| 85 | #PBS -l mem=${VMEM}gb |
---|
| 86 | #PBS -l vmem=${VMEM}gb |
---|
| 87 | |
---|
| 88 | #####comment:PBS -l mem=31922327552,ncpus=60 |
---|
| 89 | #Jobs start in the HOME directory, cd to submitted directory |
---|
| 90 | cd "\$PBS_O_WORKDIR" |
---|
| 91 | |
---|
| 92 | ############################################################ |
---|
| 93 | NCPU=\$(wc -l < \$PBS_NODEFILE) |
---|
| 94 | awk "NR % $OMP == 0" \$PBS_NODEFILE | head -n $MPI > nodelist.txt |
---|
| 95 | |
---|
| 96 | echo ------------------JOB SUMMARY-------------------------- |
---|
| 97 | echo 'PBS OUT: Job allocated on: '\${NCPU}' cpu(s)' |
---|
| 98 | echo 'PBS OUT: Job running on the following node(s): ' |
---|
| 99 | cat nodelist.txt |
---|
| 100 | echo ------------------------------------------------------- |
---|
| 101 | echo PBS OUT: Job running on: \$PBS_O_HOST |
---|
| 102 | echo PBS OUT: Originating queue: \$PBS_O_QUEUE |
---|
| 103 | echo PBS OUT: Executing queue: \$PBS_QUEUE |
---|
| 104 | echo PBS OUT: Work directory: \$PBS_O_WORKDIR |
---|
| 105 | echo PBS OUT: Execution mode: \$PBS_ENVIRONMENT |
---|
| 106 | echo PBS OUT: Job identifier: \$PBS_JOBID |
---|
| 107 | echo PBS OUT: Job name: \$PBS_JOBNAME |
---|
| 108 | echo PBS OUT: Node file: \$PBS_NODEFILE |
---|
| 109 | echo PBS OUT: Current home directory: \$PBS_O_HOME |
---|
| 110 | echo PBS OUT: PATH= \$PBS_O_PATH |
---|
| 111 | echo ------------------------------------------------------- |
---|
| 112 | |
---|
| 113 | |
---|
| 114 | export OMP_NUM_THREADS=$3 |
---|
| 115 | export OMP_STACKSIZE=128M |
---|
| 116 | ulimit -s unlimited |
---|
| 117 | |
---|
| 118 | $(cat $ROOT/DYNAMICO/arch/arch-ifort_CICLAD.env) |
---|
| 119 | |
---|
| 120 | rm -rf gcm.log logs *.nc netcdf |
---|
| 121 | mpirun -np $MPI -cpus-per-proc \$OMP_NUM_THREADS -bycore -bind-to-core -report-bindings ./gcm.exe > gcm.log 2>&1 |
---|
| 122 | $4 >> gcm.log |
---|
| 123 | date >> gcm.log |
---|
| 124 | |
---|
| 125 | mkdir -p netcdf |
---|
| 126 | cp gcm.log *.def netcdf |
---|
| 127 | mv *.nc netcdf |
---|
| 128 | |
---|
| 129 | mkdir -p logs |
---|
| 130 | cp *.xml logs |
---|
| 131 | cp -pr $(readlink -e gcm.exe) logs |
---|
| 132 | cp -pr $(readlink -e build/src) logs |
---|
| 133 | mv xios_client_*.err xios_client_*.out gcm.log logs |
---|
| 134 | |
---|
| 135 | EOF |
---|
| 136 | } |
---|
| 137 | |
---|
| 138 | #------------------------------ post-processing job ----------------------------- |
---|
| 139 | |
---|
| 140 | function job_post_X64_CURIE() # LIST |
---|
| 141 | { |
---|
| 142 | DEPLIST=afterany |
---|
| 143 | for ID in $* ; do |
---|
| 144 | DEPLIST="$DEPLIST:$ID" |
---|
| 145 | done |
---|
| 146 | cat <<EOF |
---|
| 147 | #!/bin/bash |
---|
| 148 | ## Request name |
---|
| 149 | #MSUB -r testing |
---|
| 150 | #MSUB -q standard |
---|
| 151 | #MSUB -A gen7548 |
---|
| 152 | #MSUB -E "--dependency=$DEPLIST" |
---|
| 153 | ## Number of tasks (=MPI processes) to use |
---|
| 154 | #MSUB -n 1 |
---|
| 155 | ## Elapsed time limit in seconds |
---|
| 156 | #MSUB -T 600 |
---|
| 157 | ## Quality of Service required (long [3 days], normal [1 day], test [30 min]) |
---|
| 158 | #MSUB -Q normal |
---|
| 159 | |
---|
| 160 | export OMP_NUM_THREADS=1 |
---|
| 161 | # this script is submitted from $ROOT/logs |
---|
| 162 | cd \${BRIDGE_MSUB_PWD}/.. |
---|
| 163 | |
---|
| 164 | $(cat $ROOT/DYNAMICO/arch/arch-X64_CURIE.env) |
---|
| 165 | module load python |
---|
| 166 | |
---|
| 167 | bash/post.sh |
---|
| 168 | EOF |
---|
| 169 | } |
---|