Skip to content

Compiling LAMMPS

LAMMPS is a classical molecular dynamics code, and an acronym for Large-scale Atomic/Molecular Massively Parallel Simulator.[1]

These are instructions for compiling LAMMPS 31Mar2017. These instructions may not work for any other version. Proteus has some version(s) installed: see LAMMPS

It is important to understand the LAMMPS build procedure. Consult the official documentation.[2]

Installing Pre-compiled Version♯

Use the Conda method documented here: https://docs.lammps.org/Install_conda.html

Basic Outline♯

This is a basic outline of the old Makefile-based build process.

  1. Build a "basic" LAMMPS (i.e. without any "packages", standard or user). This basic LAMMPS can be serial or OMP (multithreaded) or MPI. E.g. make mpi
  2. If the basic build is successful, add packages: make yes-package_name
  3. And recompile: make mpi
  4. If using a package causes a build failure, you can try to fix it yourself, or remove the package: make no-package_name

The Cmake build process is recommended for newer versions of LAMMPS. It is much more straightforward. See below.

Source Code♯

Download a tarball from: http://lammps.sandia.gov/download.html#tar

You may also use svn or git to download updates. See instructions elsewhere at the download link above.

The official git repository as of Oct 2016 is https://github.com/lammps/lammps

Intel Composer XE 2015.1.133 + OpenMPI 1.8.1 + Python 3.6.x♯

In addition to the official LAMMPS build documentation, you must be familiar with the Intel Link Line Advisor:

https://software.intel.com/en-us/articles/intel-mkl-link-line-advisor

We will use the Intel MKL and its FFTW3 interface. We will build a hybrid MPI-OpenMP executable, i.e. each MPI process will run multithreaded. The total number of separate MPI processes will be fewer than a pure MPI executable, but each MPI process will run more than one thread so that the total number of computational threads is constant.

Environment for Intel Composer XE♯

Load these modules:

shared
proteus
gcc/4.8.1
sge/univa
intel/composerxe/2015.1.133
proteus-openmpi/intel/2015/1.8.1-mlnx-ofed
python/intel/2015/3.6.2

Other modules will be loaded automatically by the ones above.

For convenience, we define an environment variable for the top level of the LAMMPS source code distribution.

export LAMMPS_SRC=~/src/lammps-16Feb16

Proteus-specific Makefile for Intel Composer XE♯

Download the Proteus-specific makefile from LAMMPS_Makefile.proteus-intel.txt.

and save it to the directory

$LAMMPS_SRC/src/MAKE/MINE/     (you may have to create this directory -- "cd MAKE ; make MINE" )

with the name Makefile.proteus-intel

Makefile Snippet for Python 3.x Integration♯

The distributed makefile snippet for Python integration does not work with a shared Python library. Download the makefile snippet from LAMMPS_Makefile.lammps-python3.txt

if you want to use Python integration. Save it to the directory

$LAMMPS_SRC/lib/python/

with the name Makefile.lammps

Build with Intel Composer XE♯

First, build a "basic" version:

[juser@proteusi01 ~]$ cd $LAMMPS_SRC/src [juser@proteusi01 src]$ make -j 4 proteus-intel >> Make.proteus-intel.out 2>&1 &

Then, add packages, and build incrementally:[3]

[juser@proteusi01 src]$ make yes-python Installing package python [juser@proteusi01 src]$ make -j 4 proteus-intel >> Make.proteus-intel.out 2>&1 &

The pattern is to do:

make yes-PACKAGENAME make -j 4 proteus-intel >> Make.proteus-intel.out 2>&1 &

for whatever "package" you wish to include. The order of build is significant: this ordering may or may not be documented.

And the following packages in order -- the USER-OMP package must be last (n.b. the package OPT seems to cause build errors):

misc manybody mpiio user-intel user-omp

The resulting executable will be:

$LAMMPS_SRC/src/lmp_proteus-intel

GCC 4.8.1 + Open MPI 1.8.1 + Python 2.7.x♯

The basic outline is the same as for building with Intel Composer XE. The makefile is different to handle:

  • different base compiler
  • generic FFTW3 library

Environment for GCC♯

Load these modules:

gcc/4.8.1 proteus-openmpi/gcc/64/1.8.1-mlnx-ofed proteus-fftw3/gcc/64/3.3.3 python/2.7-current

Proteus-specific Makefile for GCC♯

Download the Proteus-specific makefile from LAMMPS_Makefile.proteus-gcc.txt.

and save it to the directory

$LAMMPS_SRC/src/MAKE/MINE/     (you may have to create this directory -- "cd MAKE ; make MINE" )

with the name Makefile.proteus-gcc

Makefile snippet for Python 2.7 integration♯

This is the same as for the Intel compilation.

The distributed makefile snippet for Python integration does not work with a shared Python library. Download the makefile snippet from LAMMPS_Makefile.lammps-python2.txt.

if you want to use Python integration. Save it to the directory

$LAMMPS_SRC/lib/python/

with the name Makefile.lammps

Build with GCC♯

First, build a "basic" version:

[juser@proteusi01 ~]$ cd $LAMMPS_SRC/src [juser@proteusi01 src]$ make -j 4 proteus-gcc >> Make.proteus-gcc.out 2>&1 &

Then, add packages, and build incrementally:[4]

[juser@proteusi01 src]$ make yes-python Installing package python [juser@proteusi01 src]$ make -j 4 proteus-gcc >> Make.proteus-gcc.out 2>&1 &

The pattern is to do:

make yes-PACKAGENAME make -j 4 proteus-gcc >> Make.proteus-gcc.out 2>&1 &

for whatever "package" you wish to include. The order of build is significant: this ordering may or may not be documented.

And the following packages in order -- the USER-OMP package must be last (n.b. the package OPT seems to cause build errors):

misc manybody mpiio user-omp

The resulting executable will be:

$LAMMPS_SRC/src/lmp_proteus-gcc

GCC 4.8.1 + CUDA 6.0♯

We have not yet been successful in getting a CUDA-enabled build. If you wish to work on this, see the relevant official documentation:

The GPU package and the Kokkos package seem to be mutually exclusive. The Kokkos "package" is unlike the usual "USER" packages: it seems to directly modify the base Makefiles. Our recommendation is to not use Kokkos, for now.

Environment♯

You may need to set the CUDA_HOME environment variable.

Packages♯

  • gpu
  • omp
  • misc
  • mpiio
  • manybody
  • python

lib/gpu/Nvidia.makefile♯

Has to be modified to include the full path to the bin2c executable:

BIN2C = /cm/shared/apps/cuda60/toolkit/6.0.37/bin/bin2c

GCC 4.8.1 + CUDA 9.0♯

Use Cmake.

Intel Composer XE 2019u1♯

For building on the new (ic23) Skylake nodes. Use Cmake.

Modules Needed♯

The GCC modulefile is needed because Intel C++ relies on GCC's C++ library.

gcc/7.4.0
intel/composerxe/2019u1
proteus-openmpi/intel/2019/3.1.4
zlib/cloudflare/intel/2019/1.2.8

For Python, use the one in /usr/local:

/usr/local/bin/python3

Using Cmake♯

LAMMPS now provides a Cmake build system. To try it out:

module load cmake cd lammps-stable_16Mar2018/cmake mkdir BUILD cd BUILD ccmake ../cmake

and adjust the configuration.

Intel Composer XE CFLAGS to optimize for architecture of build host:

-O3 -xHost

For the CPU version, make sure to turn on both

BUILD_OMP PKG_USER-OMP

See: https://sourceforge.net/p/lammps/mailman/message/36704824/

FIXME? INTEL_LRT_MODE=none

Official documentation (sparse): https://lammps.sandia.gov/doc/Build_cmake.html

Axel Kohlmeyer recommends using GCC instead of Intel compilers.

Writing Your Own Makefile♯

Currently, it is not difficult to create your own makefile. Just use one of the existing examples in the distribution, in the directory src/MAKE/OPTIONS/. The directory src/MAKE/MACHINES/ contains makefiles for various existing supercomputer facilities. Start with a sample that already works. You can test if it works by running "make sample" in the src directory.

  • Create the directory src/MAKE/MINE
  • Copy one of the examples to src/MAKE/MINE, renaming it, say, to Makefile.myname
  • Modify the first line of Makefile.myname to set a description of the makefile

# myname = MYNAME by My Name

  • Modify at least these definitions and maybe others (e.g. FFT):
    • CC
    • OPTFLAGS
    • CCFLAGS
    • LINK -- generally set to be the same as CC
    • MPI_INC

Running♯

The section on running LAMMPS and benchmarking has been moved to LAMMPS - Running

See Also♯

References♯

[1] LAMMPS official web site

[2] LAMMPS User Documentation - Getting Started

[3] LAMMPS Documentation - Packages

[4] LAMMPS Documentation - Packages