LAMMPS 22Jul2025 Update 3 - CUDA | Research Center for Computational Science

Webpage

https://www.lammps.org

Version

22Jul2025 Update 3

Build Environment

GCC 10.3.1 (gcc-toolset-10)
Open MPI 4.1.8 (CUDA-aware)
GSL 2.8
OpenBLAS 0.3.29
CUDA 12.8 Update 1

Files Required

lammps-stable_22Jul2025_update3.tar.gz
(some of files will be downloaded during the installation procedure below)

Build Procedure

#!/bin/sh
VERSION=2025-Jul22-u3
NAME=lammps-stable_22Jul2025_update3
INSTALL_PREFIX=/apl/lammps/${VERSION}-CUDA
BASEDIR=/home/users/${USER}/Software/LAMMPS/${VERSION}
LAMMPS_TARBALL=${BASEDIR}/${NAME}.tar.gz
WORKDIR=/gwork/users/${USER}
LAMMPS_WORKDIR=${WORKDIR}/${NAME}
FFMPEG_BIN=/apl/ffmpeg/6.1/bin/ffmpeg
VMD_MOLFILE_INC=/home/users/${USER}/Software/VMD/1.9.4/vmd-1.9.4a57/plugins/include
GSL_ROOT=/apl/gsl/2.8
GPU_ARCH=sm_80
PARALLEL=12

#------------------------------------------------------------------
umask 0022
export LANG=C
ulimit -s unlimited
module -s purge
module -s load gcc-toolset/10
module -s load openmpi/4.1.8/gcc13-cuda12.8u1
module -s load gsl/2.8
#module -s load mkl/2023.2.0
module -s load openblas/0.3.29-lp64
module -s load cuda/12.8u1
PYTHONEXE=/usr/bin/python3.6m
PYTHONINC=/usr/include/python3.6m
mkdir -p ${WORKDIR}
cd ${WORKDIR}
if [ -d ${NAME} ]; then
mv ${NAME} lammps_erase
rm -rf lammps_erase &
fi

tar zxf ${LAMMPS_TARBALL}
cd ${NAME}
# for FFT_SINGLE=on
sed -i -e '26i#include "force.h"' unittest/force-styles/test_fix_timestep.cpp
sed -i -e '29i#include "kspace.h"' unittest/force-styles/test_pair_style.cpp
# pass python exe path to MDI cmake
sed -i -e "59i\ -DPython_EXECUTABLE=${PYTHONEXE}" cmake/Modules/Packages/MDI.cmake
mkdir build && cd build
# Disabled PKGs:
# ADIOS, VTK: noavail
# GUI: manually disabled
# KIM: to avoid dependence to libkim-api.so.2
# INTEL: not necessary for gcc build

cmake ../cmake \
-DLAMMPS_MACHINE=rccs-cuda \
-DENABLE_TESTING=on \
-DCMAKE_INSTALL_PREFIX=${INSTALL_PREFIX} \
-DCMAKE_C_COMPILER=mpicc \
-DCMAKE_CXX_COMPILER=mpicxx \
-DCMAKE_Fortran_COMPILER=mpif90 \
-DCMAKE_C_FLAGS_RELEASE="-O3 -DNDEBUG" \
-DCMAKE_CXX_FLAGS_RELEASE="-O3 -DNDEBUG" \
-DCMAKE_Fortran_FLAGS_RELEASE="-O3 -DNDEBUG" \
-DPython_EXECUTABLE=${PYTHONEXE} \
-DPython_INCLUDE_DIR=${PYTHONINC} \
-DGSL_ROOT_DIR=${GSL_ROOT} \
-DBUILD_SHARED_LIBS=on \
-DBUILD_TOOLS=on \
-DBUILD_MPI=on \
-DBUILD_OMP=on \
-DBUILD_LAMMPS_GUI=off \
-DBUILD_WHAM=off \
-DFFT=FFTW3 \
-DFFT_SINGLE=on \
-DFFT_FFTW_THREADS=on \
-DWITH_JPEG=on \
-DWITH_PNG=on \
-DWITH_GZIP=on \
-DWITH_FFMPEG=on \
-DFFMPEG_EXECUTABLE=${FFMPEG_BIN} \
-DPKG_ADIOS=off \
-DPKG_AMOEBA=off \
-DPKG_APIP=on \
-DPKG_ASPHERE=on \
-DPKG_ATC=on \
-DPKG_AWPMD=on \
-DPKG_BOCS=on \
-DPKG_BODY=on \
-DPKG_BPM=on \
-DPKG_BROWNIAN=on \
-DPKG_CG-DNA=on \
-DPKG_CG-SPICA=on \
-DPKG_CLASS2=on \
-DPKG_COLLOID=on \
-DPKG_COLVARS=on \
-DPKG_COMPRESS=on \
-DPKG_CORESHELL=on \
-DPKG_DIELECTRIC=on \
-DPKG_DIFFRACTION=on \
-DPKG_DIPOLE=on \
-DPKG_DPD-BASIC=on \
-DPKG_DPD-MESO=on \
-DPKG_DPD-REACT=on \
-DPKG_DPD-SMOOTH=on \
-DPKG_DRUDE=on \
-DPKG_EFF=on \
-DPKG_ELECTRODE=on \
-DPKG_EXTRA-COMMAND=on \
-DPKG_EXTRA-COMPUTE=on \
-DPKG_EXTRA-DUMP=on \
-DPKG_EXTRA-FIX=on \
-DPKG_EXTRA-MOLECULE=on \
-DPKG_EXTRA-PAIR=on \
-DPKG_FEP=on \
-DPKG_GPU=on \
-DGPU_API=cuda \
-DGPU_ARCH=${GPU_ARCH} \
-DPKG_GRANULAR=on \
-DPKG_H5MD=on \
-DPKG_INTEL=off \
-DPKG_INTERLAYER=on \
-DPKG_KIM=off \
-DDOWNLOAD_KIM=off \
-DPKG_KOKKOS=off \
-DPKG_KSPACE=on \
-DPKG_LATBOLTZ=on \
-DPKG_LEPTON=on \
-DPKG_MACHDYN=on \
-DDOWNLOAD_EIGEN3=on \
-DPKG_MANIFOLD=on \
-DPKG_MANYBODY=on \
-DPKG_MC=on \
-DPKG_MDI=on \
-DDOWNLOAD_MDI=on \
-DPKG_MEAM=on \
-DPKG_MESONT=on \
-DPKG_MGPT=on \
-DPKG_MISC=on \
-DPKG_ML-HDNNP=off \
-DDOWNLOAD_N2P2=off \
-DPKG_ML-IAP=off \
-DPKG_ML-PACE=on \
-DPKG_ML-POD=on \
-DPKG_ML-QUIP=off \
-DDOWNLOAD_QUIP=off \
-DPKG_ML-RANN=on \
-DPKG_ML-SNAP=on \
-DPKG_ML-UF3=on \
-DPKG_MOFFF=on \
-DPKG_MOLECULE=on \
-DPKG_MOLFILE=on \
-DMOLFILE_INCLUDE_DIR=${VMD_MOLFILE_INC} \
-DPKG_NETCDF=on \
-DPKG_OPENMP=on \
-DPKG_OPT=on \
-DPKG_ORIENT=on \
-DPKG_PERI=on \
-DPKG_PHONON=on \
-DPKG_PLUGIN=on \
-DPKG_PLUMED=on \
-DDOWNLOAD_PLUMED=on \
-DPKG_POEMS=on \
-DPKG_PTM=on \
-DPKG_PYTHON=on \
-DPKG_QEQ=on \
-DPKG_QMMM=on \
-DPKG_QTB=on \
-DPKG_REACTION=on \
-DPKG_REAXFF=on \
-DPKG_REPLICA=on \
-DPKG_RHEO=on \
-DPKG_RIGID=on \
-DPKG_SCAFACOS=on \
-DDOWNLOAD_SCAFACOS=on \
-DPKG_SHOCK=on \
-DPKG_SMTBQ=on \
-DPKG_SPH=on \
-DPKG_SPIN=on \
-DPKG_SRD=on \
-DPKG_TALLY=on \
-DPKG_UEF=on \
-DPKG_VORONOI=on \
-DDOWNLOAD_VORO=on \
-DPKG_VTK=off \
-DPKG_YAFF=on \
-DBLA_VENDOR=OpenBLAS \
-DBLA_PREFER_PKGCONFIG=on \
-DCMAKE_BUILD_TYPE=Release

#make -j ${PARALLEL}
make VERBOSE=1 -j ${PARALLEL}
export OMP_NUM_THREADS=2
#make test
make install
cp -a ../examples ${INSTALL_PREFIX}
cd ${INSTALL_PREFIX}
for f in etc/profile.d/*; do
if [ -f $f ]; then
ln -s $f .
fi
done
cd lib64
if [ -f liblammps_rccs-cuda.so ]; then
ln -s liblammps_rccs-cuda.so liblammps.so
fi
if [ -f liblammps_rccs-cuda.so.0 ]; then
ln -s liblammps_rccs-cuda.so.0 liblammps.so.0
fi

Test

Tested with the following script on ccgpu.

#!/bin/sh
VERSION=2025-Jul22-u3
NAME=lammps-stable_22Jul2025_update3
INSTALL_PREFIX=/apl/lammps/${VERSION}-CUDA
BASEDIR=/home/users/${USER}/Software/LAMMPS/${VERSION}
LAMMPS_TARBALL=${BASEDIR}/${NAME}.tar.gz
WORKDIR=/gwork/users/${USER}
LAMMPS_WORKDIR=${WORKDIR}/${NAME}
GPU_ARCH=sm_80
FFMPEG_BIN=/apl/ffmpeg/6.1/bin/ffmpeg
VMD_MOLFILE_INC=/home/users/${USER}/Software/VMD/1.9.4/vmd-1.9.4a57/plugins/include
GSL_ROOT=/apl/gsl/2.8

PARALLEL=12
#------------------------------------------------------------------
umask 0022
export LANG=C
ulimit -s unlimited
export OMP_NUM_THREADS=1
module -s purge
module -s load gcc-toolset/10
module -s load openmpi/4.1.8/gcc13-cuda12.8u1
module -s load gsl/2.8
module -s load openblas/0.3.29-lp64
module -s load cuda/12.8u1
PYTHONEXE=/usr/bin/python3.6m
PYTHONINC=/usr/include/python3.6m
cd ${WORKDIR}
cd ${NAME}
cd build
make test # need to do it separately...

Test Results

38 SimpleCommands failed
- same as the case of Update 1
128:MolPairStyle:coul_long_cs (PairStyle.gpu) failed. Large numerical error. This test did not fail in the case of Update 1.
199:MolPairStyle:lj_cut_dipole_cut (PairStyle.gpu) failed. This test did not fail in the case of Update 1.
- Cuda driver error 1 in call at file '/gwork/users/qf7/lammps-stable_22Jul2025_update3/lib/gpu/geryon/nvd_kernel.h' in line 338.

Installed Packages

APIP ASPHERE ATC AWPMD BOCS BODY BPM BROWNIAN CG-DNA CG-SPICA CLASS2 COLLOID
COLVARS COMPRESS CORESHELL DIELECTRIC DIFFRACTION DIPOLE DPD-BASIC DPD-MESO
DPD-REACT DPD-SMOOTH DRUDE EFF ELECTRODE EXTRA-COMMAND EXTRA-COMPUTE
EXTRA-DUMP EXTRA-FIX EXTRA-MOLECULE EXTRA-PAIR FEP GPU GRANULAR H5MD
INTERLAYER KSPACE LATBOLTZ LEPTON MACHDYN MANIFOLD MANYBODY MC MDI MEAM MESONT
MGPT MISC ML-PACE ML-POD ML-RANN ML-SNAP ML-UF3 MOFFF MOLECULE MOLFILE
NETCDF OPENMP OPT ORIENT PERI PHONON PLUGIN PLUMED POEMS PTM PYTHON QEQ QMMM
QTB REACTION REAXFF REPLICA RHEO RIGID SCAFACOS SHOCK SMTBQ SPH SPIN SRD TALLY
UEF VORONOI YAFF

Notes

While almost the same build procedure as 22Jul2025(CUDA), ML-QUIP was disabled to avoid compilation error.
Compared tp Update 1, additional error occurred on 128:MolPairStyle:coul_long_cs and 199:MolPairStyle:lj_cut_dipole_cut.
- This may be due to significant changes (on pairlist/neighbor-search?) introduced around Update 2.
- More tests failed for CUDA version of Update 2. (Update 2 is not installed, just tested.)