LAMMPS GPU/CPU compilation

Prerequisites:

Not all are necessary but might be useful: 

sudo apt-get update
sudo apt-get upgrade
sudo apt-get install build-essential cmake cmake-curses-gui libopenmpi-dev openmpi-bin libfftw3-dev libblas-dev liblapack-dev pkg-config ffmpeg python3-dev
sudo apt install python3-pip python3.10-venv python3-venv

 Compile OpenMPI: see -> HERE <-

CUDA Toolkit

Install CUDA Toolkit for Linux, x86_64, WSL-Ubuntu, 2.0, deb (local) according to the manual here:
https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=WSL-Ubuntu&target_version=2.0&target_type=deb_local,
or use the copied steps below:

wget https://developer.download.nvidia.com/compute/cuda/repos/wsl-ubuntu/x86_64/cuda-wsl-ubuntu.pin
sudo mv cuda-wsl-ubuntu.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/12.3.1/local_installers/cuda-repo-wsl-ubuntu-12-3-local_12.3.1-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-12-3-local_12.3.1-1_amd64.deb
sudo cp /var/cuda-repo-wsl-ubuntu-12-3-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda-toolkit-12-3

The installation will take some time to finish. On native Ubuntu, you can also use CUDA from the NVIDIA HPC-SDK toolkit, just modify paths in the following text accordingly. However, when I was trying to compile LAMMPS GPU package on WSL2 using it from HPC SDK, I was receiving an error. Only CUDA compiled specifically for the WSL Ubuntu version worked for me.

Add CUDA Toolkit to the PATH system variables:

nano ~/.bashrc

Scroll down to the end of the file and add there:

export PATH=$PATH:/usr/local/cuda-12.3/bin

source ~/.bashrc

Check if CUDA is properly installed:

nvcc --version
which nvcc 

It should give the version of the compiler and the path to it.

LAMMPS GPU Package Installation

git clone -b stable https://github.com/lammps/lammps.git lammps
cd lammps
mkdir build && cd build

Check your GPU architecture, e.g., GTX 1060 is built on Pascal arch, so GPU_ARCH=sm_61 (see docs.lammps.org/Build_extras.html#gpu for GPU_ARCH options). For RTX 2060, use sm_75. For RTX 4090, sm_90. Also choose the precision (GPU_PREC, double or mixed (default) or single). Also turn on certain basic packages (ORIENT, PYTHON, OPENMP, MEAM, MANYBODY)

When using CUDA within NVIDIA HPC SDK (set correct path to bin2c):

sudo cmake -D PKG_GPU=ON -D GPU_API=CUDA -D GPU_PREC=mixed -D GPU_ARCH=sm_75 -D PKG_ORIENT=ON -D PKG_PYTHON=ON -D PKG_OPENMP=ON -D PKG_MEAM=ON -D PKG_MANYBODY=ON -D BUILD_SHARED_LIBS=ON -D BIN2C=/opt/nvidia/hpc_sdk/Linux_x86_64/23.9/cuda/12.2/bin/bin2c ../cmake

When using CUDA compiled for WSL2 (set correct path for bin2c and cuda):

sudo cmake -D PKG_GPU=ON -D GPU_API=CUDA -D GPU_PREC=mixed -D GPU_ARCH=sm_75 -D PKG_ORIENT=ON -D PKG_PYTHON=ON -D PKG_OPENMP=ON -D PKG_MEAM=ON -D PKG_MANYBODY=ON -D BUILD_SHARED_LIBS=ON -D BIN2C=/usr/local/cuda-12.3/bin/bin2c -D CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-12.3 ../cmake

To select additional packages via cmake GUI:

ccmake . 

Check if path for BIN2 (first line) is correct (should be something like /usr/local/cuda-12.3/bin/bin2c). If some error appears regarding the bin2c, include its path in the previous step with: -D BIN2C= /usr/local/cuda-12.3/bin/bin2c

Compile and then install:

make -j 4
make install

Note: if you want to build the LAMMPS package for python3, write:

make install-python

Note: MEAM potential is not yet GPU accelerated. It is not implemented.

Add path to the lmp binary in bashrc:

export PATH=$PATH:/home/lebedmi2/lammps/build

Possibly rename lmp binary if you have more lammps versions:

mv lmp lmp_gpu

Note: When compiling older versions of LAMMPS on GPU, -DCMAKE_LIBRARY_PATH=/usr/local/cuda-12.3/targets/x86_64-linux/lib/stubs had to be added to solve the error: CMake Error: The following variables are used in this project, but they are set to NOTFOUND
Please set them or make sure they are set and tested correctly in the CMake files:
CUDA_CUDA_LIBRARY (ADVANCED)

Test

Prepare your input files and try if the compilation was successfull:

lmp_gpu -sf gpu -pk gpu 1 -in test.in
mpirun -np 8 lmp_gpu -sf gpu -pk gpu 1 -in test.in

List of GPU-Accelerated Interatomic Potentials and Functions

amoeba
atom
beck
born_coul_long_cs
born_coul_long
born_coul_wolf_cs
born_coul_wolf
born
buck_coul
buck_coul_long
buck
charmm
charmm_long
colloid
coul
coul_debye
coul_dsf
coul_long_cs
coul_long
device
dipole_lj
dipole_lj_sf
dipole_long_lj
dpd
eam
ellipsoid_nbor
gauss
gayberne
gayberne_lj
hippo
lj96
lj_class2_long
lj_coul
lj_coul_debye
lj_coul_long
lj_coul_msm
lj_cubic
lj
lj_dsf
lj_expand_coul_long
lj_expand
lj_gromacs
lj_smooth
lj_spica
lj_spica_long
lj_tip4p_long
mie
morse
neighbor_cpu
neighbor_gpu
pppm_d
pppm_f
re_squared
re_squared_lj
soft
sw
table
tersoff
tersoff_mod
tersoff_zbl
ufm
vashishta
yukawa_colloid
yukawa
zbl