From 891a366f73a5dfe9a6822c8d884c9b22c82de971 Mon Sep 17 00:00:00 2001 From: Peter Boyle Date: Wed, 6 Mar 2024 01:22:55 +0000 Subject: [PATCH] Repro CG script --- systems/Aurora/tests/repro16.pbs | 40 ++++++++++++++++++++++++++++++++ 1 file changed, 40 insertions(+) create mode 100644 systems/Aurora/tests/repro16.pbs diff --git a/systems/Aurora/tests/repro16.pbs b/systems/Aurora/tests/repro16.pbs new file mode 100644 index 00000000..28030a3d --- /dev/null +++ b/systems/Aurora/tests/repro16.pbs @@ -0,0 +1,40 @@ +#!/bin/bash + +## qsub -q EarlyAppAccess -A Aurora_Deployment -I -l select=1 -l walltime=60:00 + +#PBS -q EarlyAppAccess +#PBS -l select=16 +#PBS -l walltime=01:00:00 +#PBS -A LatticeQCD_aesp_CNDA + +#export OMP_PROC_BIND=spread +#unset OMP_PLACES + +cd $PBS_O_WORKDIR + +source ../sourceme.sh + +cat $PBS_NODEFILE + +export OMP_NUM_THREADS=3 +export MPIR_CVAR_CH4_OFI_ENABLE_GPU_PIPELINE=1 + +#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE +#unset MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE +#unset MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST + +export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_D2H_ENGINE_TYPE=0 +export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_H2D_ENGINE_TYPE=0 +export MPIR_CVAR_GPU_USE_IMMEDIATE_COMMAND_LIST=1 +export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_BUFFER_SZ=1048576 +export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_THRESHOLD=131072 +export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_NUM_BUFFERS_PER_CHUNK=16 +export MPIR_CVAR_CH4_OFI_GPU_PIPELINE_MAX_NUM_BUFFERS=16 +export MPICH_OFI_NIC_POLICY=GPU + +# 12 ppn, 16 nodes, 192 ranks +CMD="mpiexec -np 192 -ppn 12 -envall \ + ./gpu_tile_compact.sh \ + ./Test_dwf_mixedcg_prec --mpi 2.4.4.6 --grid 64.128.128.192 \ + --shm-mpi 1 --shm 4096 --device-mem 32000 --accelerator-threads 32 --seconds 3000" +$CMD