add nvidia benchmark

benchmarks/lammps-nvidia/.gitignore

@@ -0,0 +1,4 @@
+lammps.sif
+log.lammps
+out.*
+logs/

benchmarks/lammps-nvidia/README.md

@@ -0,0 +1,3 @@
+'run.sh' and 'install.sh' assume that conda is on the path. You may need to modify to load conda.
+
+`submit.sh` is a simple test script to submit on Bridges-2.

benchmarks/lammps-nvidia/get_performance.py

@@ -0,0 +1,32 @@
+import os
+import json
+import yaml
+
+out = {}
+nextline = False
+logfile = 'log.lammps'
+path = ''
+if not os.path.isfile(logfile):
+    path = 'benchmarks/lammps-allegro'
+f = open(os.path.join(path,logfile))
+for line in f.readlines():
+    if 'Performance' in line:
+        print(line)
+        out['raw1'] = line
+        _, p1, u1, p2, u2 = line.split(' ')
+        out['simulation length'] = float(p1)
+        out['simulation length units'] = u1[:-1]
+        out['timesteps/s'] = float(p2)
+        nextline = True
+    elif nextline:
+        print(line)
+        out['raw2'] = line
+        out['utilization'] = line.split(' ')[0]
+        nextline = False
+
+f.close()
+
+with open(
+    os.path.join(path,'out.yaml'),'w'
+) as fl:
+    yaml.dump(out,fl)

benchmarks/lammps-nvidia/in.lj.txt

@@ -0,0 +1,30 @@
+# 3d Lennard-Jones melt
+
+variable	x index 1
+variable	y index 1
+variable	z index 1
+
+variable	xx equal 20*$x
+variable	yy equal 20*$y
+variable	zz equal 20*$z
+
+units		lj
+atom_style	atomic
+
+lattice		fcc 0.8442
+region		box block 0 ${xx} 0 ${yy} 0 ${zz}
+create_box	1 box
+create_atoms	1 box
+mass		1 1.0
+
+velocity	all create 1.44 87287 loop geom
+
+pair_style	lj/cut 2.5
+pair_coeff	1 1 1.0 1.0 2.5
+
+neighbor	0.3 bin
+neigh_modify	delay 0 every 20 check no
+
+fix		1 all nve
+
+run		100

benchmarks/lammps-nvidia/install.sh

@@ -0,0 +1,7 @@
+chmod +x run.sh
+
+mkdir logs
+
+conda create -n env python=3.9
+conda activate env
+conda install pyyaml

benchmarks/lammps-nvidia/run.sh

@@ -0,0 +1,19 @@
+#!/usr/bin/bash
+
+set -e; set -o pipefail
+
+# Build SIF, if it doesn't exist
+if [[ ! -f lammps.sif ]]; then
+    singularity build lammps.sif docker://nvcr.io/hpc/lammps:29Oct2020
+fi
+
+readonly gpus_per_node=$(( SLURM_NTASKS / SLURM_JOB_NUM_NODES  ))
+
+echo "Running Lennard Jones 8x8x8 example on ${SLURM_NTASKS} GPUS..."
+# echo "Running 2NN MEAM example on ${SLURM_NTASKS} GPUS..."
+srun --mpi=pmi2 \
+singularity run --nv -B ${PWD}:/host_pwd  lammps.sif \
+lmp -k on g ${gpus_per_node} -sf kk -pk kokkos cuda/aware on neigh full comm device binsize 2.8 -var x 8 -var y 8 -var z 8 -in /host_pwd/in.lj.txt
+
+conda activate env
+python get_performance.py

benchmarks/lammps-nvidia/submit.sh

@@ -0,0 +1,16 @@
+#!/usr/bin/bash
+
+#SBATCH -t 00-01:00
+#SBATCH -J lammps_benchmark
+#SBATCH -o logs/output.%j
+#SBATCH -e logs/error.%j
+#SBATCH -p GPU-shared
+#SBATCH -N 1
+#SBATCH --ntasks=1
+#SBATCH --gpus=v100-32:1
+#SBATCH --mem-per-gpu=50000
+#SBATCH --mail-user=eannevel@andrew.cmu.edu
+
+module load anaconda3
+
+./run.sh