LAMMPS on Theta
LAMMPS is a general-purpose molecular dynamics software package for massively parallel computers. It is written in an exceptionally clean style that makes it one of the most popular codes for users to extend and it currently has dozens of user-developed extensions.
For details about the code and its usage, see the LAMMPS home page. This page is dedicated to information pertaining to Theta/ThetaGPU at the ALCF.
Using LAMMPS at ALCF
ALCF provides assistance with build instructions, compiling executables, submitting jobs, and providing prebuilt binaries. For questions, contact us at email@example.com.
How to Obtain the Code
LAMMPS is an open-source code, which can be downloaded at http://lammps.sandia.gov/download.html.
Building on Theta
After LAMMPS has been downloaded and unpacked, you should see a directory whose name is of the form lammps-
# theta = Flags for Knights Landing Xeon Phi Processor, Intel compiler, Cray MPI, MKL FFT # module unload libsci # make theta -j 8 SHELL = /bin/sh # --------------------------------------------------------------------- # compiler/linker settings # specify flags and libraries needed for your compiler KOKKOS_DEVICES = OpenMP KOKKOS_ARCH = KNL CC = CC -mkl OPTFLAGS = -xMIC-AVX512 -O3 -fp-model fast=2 -no-prec-div -qoverride-limits CCFLAGS = -g -qopenmp -qno-offload -ansi-alias -restrict CCFLAGS += -DLMP_INTEL_USELRT -DLMP_USE_MKL_RNG $(OPTFLAGS) CCFLAGS += -std=c++11 CCFLAGS += -DLAMMPS_MEMALIGN=64 SHFLAGS = -fPIC DEPFLAGS = -M LINK = $(CC) LINKFLAGS = -g -qopenmp $(OPTFLAGS) -dynamic #LIB = -ltbbmalloc LIB = -L$(TBBROOT)/lib/intel64/gcc4.8 -ltbbmalloc -Wl,-rpath=$(TBBROOT)/lib/intel64/gcc4.8 SIZE = size ARCHIVE = ar ARFLAGS = -rc SHLIBFLAGS = -shared # --------------------------------------------------------------------- # LAMMPS-specific settings, all OPTIONAL # specify settings for LAMMPS features you will use # if you change any -D setting, do full re-compile after "make clean" # LAMMPS ifdef settings # see possible settings in Section 2.2 (step 4) of manual LMP_INC = # MPI library # see discussion in Section 2.2 (step 5) of manual # MPI wrapper compiler/linker can provide this info # can point to dummy MPI library in src/STUBS as in Makefile.serial # use -D MPICH and OMPI settings in INC to avoid C++ lib conflicts # INC = path for mpi.h, MPI compiler settings # PATH = path for MPI library # LIB = name of MPI library MPI_INC = -DMPICH_SKIP_MPICXX -DOMPI_SKIP_MPICXX=1 MPI_PATH = MPI_LIB = # FFT library # see discussion in Section 2.2 (step 6) of manaul # can be left blank to use provided KISS FFT library # INC = -DFFT setting, e.g. -DFFT_FFTW, FFT compiler settings # PATH = path for FFT library # LIB = name of FFT library FFT_INC = -DFFT_MKL -DFFT_SINGLE FFT_PATH = FFT_LIB = -L$(MKLROOT)/lib/intel64 -Wl,--start-group -lmkl_intel_lp64 \ -lmkl_core -lmkl_intel_thread -Wl,--end-group ...
cd lammps-<version>/src make theta -j 8
Running LAMMPS Jobs on Theta
Following is an example executable script “run_lammps.csh” to run LAMMPS on two nodes of Theta with 64 MPI ranks per node. The job can be submitted with command “qsub run_lammps.csh”, where
#!/bin/csh #COBALT -n 2 -t 10 -q debug-cache-quad -A <project_name> -O LAMMPS aprun -n 128 -N 64 -d 1 --cc depth -e OMP_NUM_THREADS=1 -j 1 ./lmp_theta -in lmp.in
When possible, users will want to build LAMMPS executables with the USER-OMP and USER-INTEL packages for best performance on Theta. Following is an example script “run_lammps_intel.csh” to run LAMMPS on two nodes of Theta with 64 MPI ranks per node and two OpenMP threads per rank with the USER-INTEL and USER-OMP packages. The job can be submitted with command “qsub run_lammps_intel.csh.”
#!/bin/csh #COBALT -n 2 -t 10 -q debug-cache-quad -A <project_name> -O LAMMPS aprun -n 128 -N 64 -d 2 --cc depth -e OMP_NUM_THREADS=2 -j 2 ./lmp_theta -in lmp.in -sf hybrid intel omp
Building on ThetaGPU
There are two key packages available in LAMMPS for running on the GPUs available in ThetaGPU: GPU and KOKKOS. Example Makefiles based on recent version of LAMMPS are available for download from ALCF GitHub.
LAMMPS can be built on the ThetaGPU compute nodes with the default software environment and support for the GPU package using the following commands once the Makefiles at the above link are placed appropriately based on instructions in the README.
cp Makefile.gpu_thetagpu lammps-<version>/lib/gpu cp Makefile.thetagpu lammps-<version>/src/MAKE/MACHINES cd lammps-<version>/lib/gpu make -f Makefile.gpu_thetagpu -j 8 cd ../../src make yes-GPU make thetagpu -j 8
LAMMPS can be built with the default software environment and support for the KOKKOS package using the following commands and the Makefile at the above link.
cp Makefile.thetagpu_kokkos lammps-<version>/src/MAKE/MACHINES cd lammps-<version>/src make yes-KOKKOS make thetagpu_kokkos -j 8
Running LAMMPS jobs on ThetaGPU
Following is an example executable script submit_full-node.sh to run LAMMPS a ThetaGPU node using all GPUs for both the GPU and KOKKOS packages. This example is based on the Rhodoposin benchmark using lammps-
After the appropriate command is uncommented, the job can be submitted with “qsub ./submit_full-node.sh”, where "-A Catalyst" in the script is replaced with an appropriate active project allocation.
Note: The preceding 'qsub' command should be executed 1) on the ThetaGPU login nodes or 2) from the Theta login node after executing 'module load cobalt/cobalt-gpu'.
#!/bin/sh #COBALT -n 1 -t 15 -q full-node -A Catalyst # submit job to run on 8 GPUs w/ 8 MPI ranks per GPU and 2 OpenMP threads per rank #env OMP_NUM_THREADS=2 mpirun -np 64 ~/bin/lammps/lammps-git/src/lmp_thetagpu -in in.rhodo -pk gpu 8 -pk omp 0 -sf hybrid gpu omp # KOKKOS package: submit job to run on 1 GPU w/ 1 MPI ranks per GPU and 2 OpenMP threads per rank #!/bin/sh #COBALT -n 1 -t 15 -q full-node -A Catalyst # submit job to run on 8 GPUs w/ 8 MPI ranks per GPU and 2 OpenMP threads per rank #env OMP_NUM_THREADS=2 mpirun -np 64 ~/bin/lammps/lammps-git/src/lmp_thetagpu -in in.rhodo -pk gpu 8 -pk omp 0 -sf hybrid gpu omp # KOKKOS package: submit job to run on 1 GPU w/ 1 MPI ranks per GPU and 2 OpenMP threads per rank mpirun -np 8 ~/bin/lammps/lammps-git/src/lmp_thetagpu -in in.rhodo -k on g 8 -sf kk -pk kokkos neigh half