1
0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-15 02:05:37 +00:00
Grid/systems/Aurora/tests/repro128.pbs

42 lines
1.2 KiB
Plaintext
Raw Normal View History

2024-03-12 16:11:33 +00:00
#!/bin/bash
## qsub -q EarlyAppAccess -A Aurora_Deployment -I -l select=1 -l walltime=60:00
#PBS -q EarlyAppAccess
#PBS -l select=128
#PBS -l walltime=02:00:00
#PBS -A LatticeQCD_aesp_CNDA
#export OMP_PROC_BIND=spread
#unset OMP_PLACES
cd $PBS_O_WORKDIR
source ../sourceme.sh
cat $PBS_NODEFILE
export OMP_NUM_THREADS=3
export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1
#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE
#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE
#unset MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST
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
# 12 ppn, 16 nodes, 192 ranks
# 12 ppn, 128 nodes, 1536 ranks
CMD="mpiexec -np 1536 -ppn 12 -envall \
./gpu_tile_compact.sh \
./Test_dwf_mixedcg_prec --mpi 4.4.4.24 --grid 128.128.128.384 \
--shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 7000 --comms-overlap "
$CMD