From Montana Tech High Performance Computing

Revision as of 15:11, 9 February 2018 by Bdeng (talk | contribs)

LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator

LAMMPS is a popular molecular dynamics program from Sandia National Laboratories:

This article explains the LAMMPS usage on MTech HPC and how to compile LAMMPS for advanced users.

LAMMPS has been installed for all users. The installed LAMMPS includes all optional packages except KIM, REAX, VORONOI, USER-CUDA, and USER_OMP. The excluded packages are either deprecated or require additional libraries. The LAMMPS examples, potentials and other files are installed at /opt/lammps/.

There are two versions of LAMMPS installed: executable without GPU package and executables with GPU package. The CPU LAMMPS executable is lmp_openmpi, while the GPU LAMMPS executables are lmp_gpu_single_single (all single precision), lmp_gpu_single_double (mix of single and double precision), and lmp_gpu_double_double (all double precision). Any of the GPU executables include all the functionalities of the CPU executable, but the GPU executables require the CUDA library to run. The GPU LAMMPS is compiled with CUDA 8.0. Detailed information regarding the GPU calculation can be found at LAMMPS website.

Loading LAMMPS

To check available installed LAMMPS version:

module avail lammps

To load LAMMPS module:

module load lammps openmpi

The above command will load the default version of LAMMPS and OpenMPI. The LAMMPS is compiled with OpenMPI 1.6.4. You may experience poor performance if other version of OpenMPI is loaded.

To check the loaded modules:

module list


To use lammps:

mpiexec -n 4 lmp_openmpi < lammps_input.txt

The above command uses 4 CPU processes to run LAMMPS input script lammps_input.txt.

Sample batch files for LAMMPS

For LAMMPS regular users who do not use GPU computing:

#PBS -l nodes=2:ppn=32
#PBS -l walltime=02:00:00
module purge
module load lammps openmpi
mpiexec -n 64 lmp_openmpi < lammps_input.txt

For LAMMPS GPU users:

#PBS -l nodes=1:ppn=4
#PBS -l feature=gpunode
#PBS -l walltime=02:00:00
module purge
module load lammps openmpi cuda
mpiexec -n 4 lmp_gpu_double_double < in.gpu.phosphate

Compiling LAMMPS

The following part highlights some important steps for compiling LAMMPS source code on MTech HPC. You are suggested read the LAMMPS official installation manual before continues:

Compiling LAMMPS from source does not require root privilege. Compilation prerequisite: g++, gfortran, openmpi, jpeg/png lib, and/or CUDA.

Compile LAMMPS

1. Load gcc, openmpi and cuda

module load gcc openmpi cuda

2. Make a new directory called ‘package’ at your home directory, and go to that directory:

mkdir package
cd package

3. At your package directory, download the LAMMPS source code (stable release)


4. Extract the LAMMPS package, and go to the created folder

tar zxvf lammps_stable.tar.gz
cd lammps-1Feb14

5. Build the lib

cd lib
5.1 Build atc
cd atc
make -f Makefile.mpic++ -j 32
cd ..
5.2 Build awpmd
cd awpmd
make -f Makefile.openmpi -j 32
cd ..
5.3 Build colvars
cd awpmd
make -f Makefile.g++ -j 32
cd ..
5.4 Build linalg
cd linalg
make -f Makefile.gfortran -j 32
cd ..
5.5 Build meam
cd meam
make -f Makefile.gfortran (do not use parallel compiling -j)
cd ..
5.6 Build poems
cd poems
make -f Makefile.g++ -j 32
cd ..
5.7 Build gpu
cd gpu
Edit the file of Makefile.linux.
CUDA_ARCH = -arch=sm_35
You may also edit the precision flagger (CUDA_PRECISION = -D_SINGLE_DOUBLE etc.)
make -f Makefile.linux
cd ..

6. Go to the LAMMPS src folder, and include all packages

cd ../../src
make yes-all
make no-voronoi (requires additional package)
make no-kim (requires addition package)
make no-user-omp (error in the code)
make no-user-cuda (no longer maintained)
make no-reax (deprecated, uses user-reaxc instead)
At this point, you may exclude other packages you do not wish to include.
Check package status
make package-status

7. Go to the LAMMPS MAKE folder, and edit the file Makefile.openmpi

Modify the line of LMP_INC as
Modify the JPEG/PNG lib as
JPG_LIB = -ljpeg -lpng
Modify the lines of FFTW as
FFT_LIB = -lfftw3

8. Build the LAMMPS

make openmpi -j 32
The created binary file is lmp_openmpi

9. Run some test

9.1 Simple computation with LJ potential (this example does not require any external potential file)
cd ../../examples/indent
Edit the in.indent file and uncomment the following 3 lines:
dump 2 all image 1000 image.*.jpg type type &
zoom 1.6 adiam 1.5
dump_modify 2 pad 5
Run the simulation
mpiexec -n 4 ../../src/lmp_openmpi < in.indent
Check the folder for the created jpg snapshots
9.2 MEAM test (this will test the FORTRAN part and the external potential file)
Go to examples/meam, and run
mpiexec -n 4 ../../src/lmp_openmpi < in.meam
9.3 gpu test
Go to examples/gpu, and run
mpiexec -n 4 ../../src/lmp_openmpi < in.gpu.phosphate
../../src/lmp_openmpi < in.gpu.rhodo