1
0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-19 00:07:05 +01:00

Almost working on Aurora

This commit is contained in:
2024-09-23 09:43:50 +00:00
parent e637fbacae
commit 02c8178f16
7 changed files with 118 additions and 98 deletions

View File

@ -11,17 +11,16 @@
cd $PBS_O_WORKDIR
source ../sourceme.sh
module load pti-gpu
#module load pti-gpu
#cat $PBS_NODEFILE
cp $PBS_NODEFILE nodefile
export OMP_NUM_THREADS=4
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
@ -34,22 +33,26 @@ export MPICH_OFI_NIC_POLICY=GPU
# 12 ppn, 2 nodes, 24 ranks
#
CMD="mpiexec -np 24 -ppn 12 -envall \
./gpu_tile_compact.sh \
./gpu_tile.sh \
./Benchmark_comms_host_device --mpi 2.2.2.3 --grid 24.32.32.24 \
--shm-mpi 0 --shm 2048 --device-mem 32000 --accelerator-threads 32"
$CMD | tee 2node.comms
#$CMD | tee 2node.comms.hbm
CMD="mpiexec -np 24 -ppn 12 -envall \
./gpu_tile_compact.sh \
./Benchmark_dwf_fp32 --mpi 2.2.2.3 --grid 32.32.64.48 \
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32 "
$CMD | tee 2node.32.32.64.48.dwf
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32 --comms-overlap --debug-signals"
#for f in 1 2 3 4 5 6 7 8
for f in 1
do
echo $CMD
$CMD | tee 2node.32.32.64.48.dwf.hbm.$f
done
CMD="mpiexec -np 24 -ppn 12 -envall \
./gpu_tile_compact.sh \
./gpu_tile.sh \
./Benchmark_dwf_fp32 --mpi 2.2.2.3 --grid 64.64.64.96 \
--shm-mpi 1 --shm 2048 --device-mem 32000 --accelerator-threads 32 "
$CMD | tee 2node.64.64.64.96.dwf
--shm-mpi 0 --shm 2048 --device-mem 32000 --accelerator-threads 32 --comms-overlap"
#$CMD | tee 2node.64.64.64.96.dwf.hbm