Loading run_benchmark +15 −17 Original line number Diff line number Diff line Loading @@ -2,16 +2,12 @@ set -e NCPUS="$SLURM_CPUS_PER_TASK" # set fallback when not running under Slurm if [[ "$NCPUS" == "" ]]; then NCPUS=12 fi # maximum number of CPU cores available on the system # (this respects even $SLURM_CPUS_PER_TASK when executed under Slurm) NCPUS=$(nproc) function run_benchmark() { echo "mesh = $(basename $mesh_file)" # set number of OpenMP threads omp_threads=${omp_threads:=1} if [[ "$omp_threads" == "1" ]]; then Loading Loading @@ -48,6 +44,10 @@ function run_benchmark() function run_set() { if (( $omp_threads > $NCPUS )); then return fi for mesh_type in mrizka cube1m; do for id in 1 2 3 4 5; do [[ "$mesh_type" == "mrizka" ]] && id="$id.rcm" Loading @@ -61,13 +61,11 @@ function run_set() #omp_threads=1 #run_set if [[ "$CUDA_VISIBLE_DEVICES" == "" ]]; then solver="./tnl-benchmark-mesh --devices host" for omp_threads in 1 2 4 6 8 $NCPUS; do for omp_threads in 1 2 4 6 8 12; do run_set done else solver="./tnl-benchmark-mesh-cuda --devices cuda" omp_threads=1 run_set fi Loading
run_benchmark +15 −17 Original line number Diff line number Diff line Loading @@ -2,16 +2,12 @@ set -e NCPUS="$SLURM_CPUS_PER_TASK" # set fallback when not running under Slurm if [[ "$NCPUS" == "" ]]; then NCPUS=12 fi # maximum number of CPU cores available on the system # (this respects even $SLURM_CPUS_PER_TASK when executed under Slurm) NCPUS=$(nproc) function run_benchmark() { echo "mesh = $(basename $mesh_file)" # set number of OpenMP threads omp_threads=${omp_threads:=1} if [[ "$omp_threads" == "1" ]]; then Loading Loading @@ -48,6 +44,10 @@ function run_benchmark() function run_set() { if (( $omp_threads > $NCPUS )); then return fi for mesh_type in mrizka cube1m; do for id in 1 2 3 4 5; do [[ "$mesh_type" == "mrizka" ]] && id="$id.rcm" Loading @@ -61,13 +61,11 @@ function run_set() #omp_threads=1 #run_set if [[ "$CUDA_VISIBLE_DEVICES" == "" ]]; then solver="./tnl-benchmark-mesh --devices host" for omp_threads in 1 2 4 6 8 $NCPUS; do for omp_threads in 1 2 4 6 8 12; do run_set done else solver="./tnl-benchmark-mesh-cuda --devices cuda" omp_threads=1 run_set fi