portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 14:04:32 +00:00
Code Releases Activity

Script updates for current summit

Browse Source
This commit is contained in:
Peter Boyle
2022-06-15 19:19:44 -04:00
parent d10d30dda8
commit 31efa5c4da
2 changed files with 23 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
systems/Summit/config-command
View File

@@ -7,6 +7,7 @@
--enable-setdevice \ --enable-setdevice \
--disable-fermion-reps \ --disable-fermion-reps \
--enable-accelerator=cuda \ --enable-accelerator=cuda \
--disable-accelerator-cshift \
--prefix /ccs/home/paboyle/prefix \ --prefix /ccs/home/paboyle/prefix \
CXX=nvcc \ CXX=nvcc \
LDFLAGS=-L/ccs/home/paboyle/prefix/lib/ \ LDFLAGS=-L/ccs/home/paboyle/prefix/lib/ \

30
systems/Summit/dwf16.lsf
View File

@@ -1,25 +1,39 @@
#!/bin/bash #!/bin/bash
#BSUB -P LGT104 #BSUB -P LGT104
#BSUB -W 2:00 #BSUB -W 0:20
#BSUB -nnodes 16 #BSUB -nnodes 16
#BSUB -J DWF #BSUB -J DWF
export OMP_NUM_THREADS=6 export OMP_NUM_THREADS=6
export PAMI_IBV_ADAPTER_AFFINITY=1 export PAMI_IBV_ADAPTER_AFFINITY=1
export PAMI_ENABLE_STRIPING=1 export PAMI_ENABLE_STRIPING=1
export OPT="--comms-concurrent --comms-overlap "
APP="./benchmarks/Benchmark_comms_host_device --mpi 4.4.4.3 " DIR=.
jsrun --nrs 16 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > comms.16node.log source sourceme.sh
APP="./benchmarks/Benchmark_dwf_fp32 --grid 96.96.96.72 --mpi 4.4.4.3 --shm 2048 --shm-force-mpi 1 --device-mem 8000 --shm-force-mpi 1 $OPT " echo MPICH_SMP_SINGLE_COPY_MODE $MPICH_SMP_SINGLE_COPY_MODE
jsrun --nrs 16 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > dwf.16node.24.log
APP="./benchmarks/Benchmark_dwf_fp32 --grid 128.128.128.96 --mpi 4.4.4.3 --shm 2048 --shm-force-mpi 1 --device-mem 8000 --shm-force-mpi 1 $OPT " VOLS=( 32.32.32.16 32.32.32.64 64.32.32.64 64.32.64.64 64.64.64.64 64.64.64.128 64.64.64.256 64.64.64.512 128.64.64.64.512)
jsrun --nrs 16 -a6 -g6 -c42 -dpacked -b packed:7 --latency_priority gpu-cpu --smpiargs=-gpu $APP > dwf.16node.32.log MPI=( 1.1.1.1 1.1.1.4 2.1.1.4 2.1.2.4 2.2.2.4 2.2.2.8 2.2.2.16 2.2.2.32 4.4.2.32 )
RANKS=( 1 4 8 16 32 64 128 256 1024)
NODES=( 1 1 2 4 8 16 32 64 128)
INTS=( 0 1 2 3 4 5 6 7 8)
for i in 5
do
vol=${VOLS[$i]}
nodes=${NODES[$i]}
mpi=${MPI[$i]}
ranks=${RANKS[$i]}
JSRUN="jsrun --nrs $nodes -a4 -g4 -c42 -dpacked -b packed:10 --latency_priority gpu-cpu --smpiargs=-gpu"
PARAMS=" --accelerator-threads 8 --grid $vol --mpi $mpi --comms-sequential --shm 2048 --shm-mpi 0"
$JSRUN ./benchmarks/Benchmark_dwf_fp32 $PARAMS > run.v${vol}.n${nodes}.m${mpi}.seq.ker
PARAMS=" --accelerator-threads 8 --grid $vol --mpi $mpi --comms-overlap --shm 2048 --shm-mpi 0"
$JSRUN ./benchmarks/Benchmark_dwf_fp32 $PARAMS > run.v${vol}.n${nodes}.m${mpi}.over.ker
done