1
0
mirror of https://github.com/paboyle/Grid.git synced 2024-09-20 17:25:37 +01:00
Grid/systems/Aurora/tests/reproN.pbs
2024-03-19 14:36:21 +00:00

79 lines
1.9 KiB
Bash

#!/bin/bash
#PBS -l select=16:system=sunspot,place=scatter
#PBS -A LatticeQCD_aesp_CNDA
#PBS -l walltime=02:00:00
#PBS -N reproN
#PBS -k doe
#export OMP_PROC_BIND=spread
#unset OMP_PLACES
module load oneapi/eng-compiler/2023.05.15.003
module load mpich/51.2/icc-all-deterministic-pmix-gpu
# 56 cores / 6 threads ~9
export OMP_NUM_THREADS=6
export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0
export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_BUFFER_SZ=1048576
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_THRESHOLD=131072
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16
export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16
export MPICH_OFI_NIC_POLICY=GPU
export MPIR_CVAR_ALLREDUCE_DEVICE_COLLECTIVE=0
export MPIR_CVAR_REDUCE_DEVICE_COLLECTIVE=0
export MPIR_CVAR_ALLREDUCE_INTRA_ALGORITHM=recursive_doubling
unset MPIR_CVAR_CH4_COLL_SELECTION_TUNING_JSON_FILE
unset MPIR_CVAR_COLL_SELECTION_TUNING_JSON_FILE
unset MPIR_CVAR_CH4_POSIX_COLL_SELECTION_TUNING_JSON_FILE
cd $PBS_O_WORKDIR
NN=`cat $PBS_NODEFILE | wc -l`
echo $PBS_NODEFILE
cat $PBS_NODEFILE
echo $NN nodes in node file
for n in `eval echo {1..$NN}`
do
cd $PBS_O_WORKDIR
THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
echo Node $n is $THIS_NODE
DIR=repro.$PBS_JOBID/node-$n-$THIS_NODE
mkdir -p $DIR
cd $DIR
echo $THIS_NODE > nodefile
CMD="mpiexec -np 12 -ppn 12 -envall --hostfile nodefile \
../../gpu_tile_compact.sh \
../../Test_dwf_mixedcg_prec --mpi 1.2.2.3 --grid 32.64.64.96 \
--shm-mpi 0 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 6000 --debug-stdout --log Message --comms-overlap"
$CMD &
done
wait
for n in ` eval echo {1..$NN} `
do
THIS_NODE=`head -n$n $PBS_NODEFILE | tail -n1 `
DIR=repro.$PBS_JOBID/node-$n-$THIS_NODE
cd $DIR
grep Oops Grid.stderr.* > failures.$PBS_JOBID
rm core.*
done