$\renewcommand{\AA}{\text{Å}}$

3.8. Packages with extra build options

When building with some packages, additional steps may be required, in addition to

CMake build	Traditional make
cmake -D PKG_NAME=yes	make yes-name

as described on the Build_package page.

For a CMake build there may be additional optional or required variables to set.

Changed in version 10Sep2025.

The traditional build system with GNU make no longer supports packages that require extra steps in the lammps/lib directory.

This is the list of packages that may require additional steps.

ADIOS	APIP	COLVARS	COMPRESS	ELECTRODE	GPU
H5MD	INTEL	KIM	KOKKOS	LEPTON	MACHDYN
MDI	MISC	ML-HDNNP	ML-IAP	ML-PACE	ML-POD
ML-QUIP	MOLFILE	NETCDF	OPENMP	OPT	PLUMED
PYTHON	QMMM	RHEO	SCAFACOS	VORONOI	VTK

3.8.1. COMPRESS package

To build with this package you must have the zlib compression library available on your system to build dump styles with a /gz suffix. There are also styles using the Zstandard library which have a ‘/zstd’ suffix. The zstd library version must be at least 1.4. Older versions use an incompatible API and thus LAMMPS will fail to compile.

If CMake cannot find the zlib library or include files, you can set these variables:

-D ZLIB_INCLUDE_DIR=path    # path to zlib.h header file
-D ZLIB_LIBRARY=path        # path to libz.a (.so) file

Support for Zstandard compression is auto-detected and for that CMake depends on the pkg-config tool to identify the necessary flags to compile with this library, so the corresponding libzstandard.pc file must be in a folder where pkg-config can find it, which may require adding it to the PKG_CONFIG_PATH environment variable.

3.8.2. GPU package

To build with this package, you must choose options for precision and which GPU hardware to build for. The GPU package currently supports three different types of back ends: OpenCL, CUDA and HIP.

CMake build

-D GPU_API=value             # value = opencl (default) or cuda or hip
-D GPU_PREC=value            # precision setting
                             # value = double or mixed (default) or single
-D GPU_ARCH=value            # primary GPU hardware choice for GPU_API=cuda
                             # value = sm_XX (see below, default is sm_75)
-D GPU_DEBUG=value           # enable debug code in the GPU package library,
                             # mostly useful for developers
                             # value = yes or no (default)
-D HIP_PATH=value            # value = path to HIP installation. Must be set if
                             # GPU_API=HIP
-D HIP_ARCH=value            # primary GPU hardware choice for GPU_API=hip
                             # value depends on selected HIP_PLATFORM
                             # default is 'gfx906' for HIP_PLATFORM=amd and 'sm_75' for
                             # HIP_PLATFORM=nvcc
-D HIP_USE_DEVICE_SORT=value # enables GPU sorting
                             # value = yes (default) or no
-D CUDPP_OPT=value           # use GPU binning with CUDA (should be off for modern GPUs)
                             # enables CUDA Performance Primitives, must be "no" for
                             # CUDA_MPS_SUPPORT=yes
                             # value = yes or no (default)
-D CUDA_MPS_SUPPORT=value    # enables some tweaks required to run with active
                             # nvidia-cuda-mps daemon
                             # value = yes or no (default)
-D CUDA_BUILD_MULTIARCH=value  # enables building CUDA kernels for all supported GPU
                               # architectures
                               # value = yes (default) or no
-D USE_STATIC_OPENCL_LOADER=value  # downloads/includes OpenCL ICD loader library,
                                   # no local OpenCL headers/libs needed
                                   # value = yes (default) or no

The GPU package supports 3 precision modes: single, double, and mixed, with the latter being the default. In the double precision mode, atom positions, forces and energies are stored, computed and accumulated in double precision. In the mixed precision mode, forces and energies are accumulated in double precision while atom coordinates are stored and arithmetic operations are performed in single precision. In the single precision mode, all are stored, executed and accumulated in single precision.

To specify the precision mode (output to the screen before LAMMPS runs for verification), set GPU_PREC to one of single, double, or mixed.

Some accelerators or OpenCL implementations only support single precision. This mode should be used with care and appropriate validation as the errors can scale with system size in this implementation. This can be useful for accelerating test runs when setting up a simulation for production runs on another machine. In the case where only single precision is supported, either LAMMPS must be compiled with -DFFT_SINGLE to use PPPM with GPU acceleration or GPU acceleration should be disabled for PPPM (e.g. suffix off or pair/only as described in the LAMMPS documentation).

GPU_ARCH settings for different GPU hardware is as follows:

sm_30 for Kepler (supported since CUDA 5 and until CUDA 10.x)
sm_35 or sm_37 for Kepler (supported since CUDA 5 and until CUDA 11.x)
sm_50 or sm_52 for Maxwell (supported since CUDA 6)
sm_60 or sm_61 for Pascal (supported since CUDA 8)
sm_70 for Volta (supported since CUDA 9)
sm_75 for Turing (supported since CUDA 10)
sm_80 or sm_86 for Ampere (supported since CUDA 11, sm_86 since CUDA 11.1)
sm_89 for Lovelace (supported since CUDA 11.8)
sm_90 or sm_90a for Hopper (supported since CUDA 12.0)
sm_100 or sm_103 for Blackwell B100/B200/B300 (supported since CUDA 12.8)
sm_120 for Blackwell B20x/B40 (supported since CUDA 12.8)
sm_121 for Blackwell (supported since CUDA 12.9)

A more detailed list can be found, for example, at Wikipedia’s CUDA article

CMake can detect which version of the CUDA toolkit is used and thus will try to include support for all major GPU architectures supported by this toolkit. Thus the GPU_ARCH setting is merely an optimization, to have code for the preferred GPU architecture directly included rather than having to wait for the JIT compiler of the CUDA driver to translate it. This behavior can be turned off (e.g. to speed up compilation) by setting CUDA_ENABLE_MULTIARCH to no.

When compiling for CUDA or HIP with CUDA, version 8.0 or later of the CUDA toolkit is required and a GPU architecture of Kepler or later, which must also be supported by the CUDA toolkit in use and the CUDA driver in use. When compiling for OpenCL, OpenCL version 1.2 or later is required and the GPU must be supported by the GPU driver and OpenCL runtime bundled with the driver.

Please note that the GPU library accesses the CUDA driver library directly, so it needs to be linked with the CUDA driver library (libcuda.so) that ships with the Nvidia driver. If you are compiling LAMMPS on the head node of a GPU cluster, this library may not be installed, so you may need to copy it over from one of the compute nodes (best into this directory). Recent versions of the CUDA toolkit starting from CUDA 9 provide a dummy libcuda.so library (typically under $(CUDA_HOME)/lib64/stubs), that can be used for linking.

To support the CUDA multi-process server (MPS) you can set the define -DCUDA_MPS_SUPPORT. Please note that in this case you must not use the CUDA performance primitives and thus set the variable CUDPP_OPT to empty.

If you are compiling for OpenCL, the default setting is to download, build, and link with a static OpenCL ICD loader library and standard OpenCL headers. This way no local OpenCL development headers or library needs to be present and only OpenCL compatible drivers need to be installed to use OpenCL. If this is not desired, you can set USE_STATIC_OPENCL_LOADER to no.

If GERYON_NUMA_FISSION is defined at build time (-DGPU_DEBUG=no), LAMMPS will consider separate NUMA nodes on GPUs or accelerators as separate devices. For example, a 2-socket CPU would appear as two separate devices for OpenCL (and LAMMPS would require two MPI processes to use both sockets with the GPU library - each with its own device ID as output by ocl_get_devices). OpenCL version 1.2 or later is required.

If you are compiling with HIP, note that before running CMake you will have to set appropriate environment variables. Some variables such as HCC_AMDGPU_TARGET (for ROCm <= 4.0) or CUDA_PATH are necessary for hipcc and the linker to work correctly.

When compiling for HIP ROCm, GPU sorting with -D HIP_USE_DEVICE_SORT=on requires installing the hipcub library (https://github.com/ROCmSoftwarePlatform/hipCUB). The HIP CUDA-backend additionally requires cub (https://nvlabs.github.io/cub). Setting -DDOWNLOAD_CUB=yes will download and compile CUB.

The GPU library has some multi-thread support using OpenMP. If LAMMPS is built with -D BUILD_OMP=on this will also be enabled.

For a debug build, set GPU_DEBUG to be yes.

Added in version 3Aug2022.

Using the CHIP-SPV implementation of HIP is supported. It allows one to run HIP code on Intel GPUs via the OpenCL or Level Zero back ends. To use CHIP-SPV, you must set -DHIP_USE_DEVICE_SORT=OFF in your CMake command-line as CHIP-SPV does not yet support hipCUB. As of Summer 2022, the use of HIP for Intel GPUs is experimental. You should only use this option in preparations to run on Aurora system at Argonne.

# AMDGPU target (ROCm <= 4.0)
export HIP_PLATFORM=hcc
export HIP_PATH=/path/to/HIP/install
export HCC_AMDGPU_TARGET=gfx906
cmake -D PKG_GPU=on -D GPU_API=HIP -D HIP_ARCH=gfx906 -D CMAKE_CXX_COMPILER=hipcc ..
make -j 4

# AMDGPU target (ROCm >= 4.1)
export HIP_PLATFORM=amd
export HIP_PATH=/path/to/HIP/install
cmake -D PKG_GPU=on -D GPU_API=HIP -D HIP_ARCH=gfx906 -D CMAKE_CXX_COMPILER=hipcc ..
make -j 4

# CUDA target (not recommended, use GPU_API=cuda)
# !!! DO NOT set CMAKE_CXX_COMPILER !!!
export HIP_PLATFORM=nvcc
export HIP_PATH=/path/to/HIP/install
export CUDA_PATH=/usr/local/cuda
cmake -D PKG_GPU=on -D GPU_API=HIP -D HIP_ARCH=sm_70 ..
make -j 4

# SPIR-V target (Intel GPUs)
export HIP_PLATFORM=spirv
export HIP_PATH=/path/to/HIP/install
export CMAKE_CXX_COMPILER=<hipcc/clang++>
cmake -D PKG_GPU=on -D GPU_API=HIP ..
make -j 4

3.8.3. KIM package

To build with this package, the KIM library with API v2 must be downloaded and built on your system. It must include the KIM models that you want to use with LAMMPS.

If you would like to use the kim query command, you also need to have libcurl installed with the matching development headers and the curl-config tool.

If you would like to use the kim property command, you need to build LAMMPS with the PYTHON package installed and linked to Python 3.6 or later. See the PYTHON package build info for more details on this. After successfully building LAMMPS with Python, you also need to install the kim-property Python package, which can be easily done using pip as pip install kim-property, or from the conda-forge channel as conda install kim-property if LAMMPS is built in Conda. More detailed information is available at: kim-property installation.

In addition to installing the KIM API, it is also necessary to install the library of KIM models (interatomic potentials). See Obtaining KIM Models to learn how to install a pre-build binary of the OpenKIM Repository of Models. See the list of all KIM models here: https://openkim.org/browse/models

(Also note that when downloading and installing from source the KIM API library with all its models, may take a long time (tens of minutes to hours) to build. Of course you only need to do that once.)

-D DOWNLOAD_KIM=value           # download OpenKIM API v2 for build
                                # value = no (default) or yes
-D LMP_DEBUG_CURL=value         # set libcurl verbose mode on/off
                                # value = off (default) or on
-D LMP_NO_SSL_CHECK=value       # tell libcurl to not verify the peer
                                # value = no (default) or yes
-D KIM_EXTRA_UNITTESTS=value    # enables extra unit tests
                                # value = no (default) or yes

If DOWNLOAD_KIM is set to yes (or on), the KIM API library will be downloaded and built inside the CMake build directory. If the KIM library is already installed on your system (in a location where CMake cannot find it), you may need to set the PKG_CONFIG_PATH environment variable so that libkim-api can be found, or run the command source kim-api-activate.

Extra unit tests can only be available if they are explicitly requested (KIM_EXTRA_UNITTESTS is set to yes (or on)) and the prerequisites are met. See KIM Extra unit tests for more details on this.

Debugging OpenKIM web queries in LAMMPS

If LMP_DEBUG_CURL is set, the libcurl verbose mode will be turned on, and any libcurl calls within the KIM web query display a lot of information about libcurl operations. You hardly ever want this set in production use, you will almost always want this when you debug or report problems.

The libcurl library performs peer SSL certificate verification by default. This verification is done using a CA certificate store that the SSL library can use to make sure the peer’s server certificate is valid. If SSL reports an error (“certificate verify failed”) during the handshake and thus refuses further communicate with that server, you can set LMP_NO_SSL_CHECK to override that behavior. When LAMMPS is compiled with LMP_NO_SSL_CHECK set, libcurl does not verify the peer and connection attempts will succeed regardless of the names in the certificate. This option is insecure. As an alternative, you can specify your own CA cert path by setting the environment variable CURL_CA_BUNDLE to the path of your choice. A call to the KIM web query would get this value from the environment variable.

KIM Extra unit tests (CMake only)

During development, testing, or debugging, if unit testing is enabled in LAMMPS, one can also enable extra tests on KIM commands by setting the KIM_EXTRA_UNITTESTS to yes (or on).

Enabling the extra unit tests have some requirements,

It requires to have internet access.
It requires to have libcurl installed with the matching development headers and the curl-config tool.
It requires to build LAMMPS with the PYTHON package installed and linked to Python 3.6 or later. See the PYTHON package build info for more details on this.
It requires to have kim-property Python package installed, which can be easily done using pip as pip install kim-property, or from the conda-forge channel as conda install kim-property if LAMMPS is built in Conda. More detailed information is available at: kim-property installation.
It is also necessary to install the following KIM models:
- EAM_Dynamo_MendelevAckland_2007v3_Zr__MO_004835508849_000
- EAM_Dynamo_ErcolessiAdams_1994_Al__MO_123629422045_005
- LennardJones612_UniversalShifted__MO_959249795837_003
See Obtaining KIM Models to learn how to install a pre-built binary of the OpenKIM Repository of Models or see Installing KIM Models to learn how to install the specific KIM models.

3.8.4. KOKKOS package

Using the KOKKOS package requires choosing several settings. You have to select whether you want to compile with parallelization on the host and whether you want to include offloading of calculations to a device (e.g. a GPU). The default setting is to have no host parallelization and no device offloading. In addition, you can select the hardware architecture to select the instruction set. Since most hardware is backward compatible, you may choose settings for an older architecture to have an executable that will run on this and newer architectures.

Note

If you run Kokkos on a different GPU architecture than what LAMMPS was compiled with, there will be a delay during device initialization while the just-in-time compiler is recompiling all GPU kernels for the new hardware. This is, however, only supported for GPUs of the same major hardware version and different minor hardware versions, e.g. 5.0 and 5.2 but not 5.2 and 6.0. LAMMPS will abort with an error message indicating a mismatch, if the major version differs.

The settings discussed below have been tested with LAMMPS and are confirmed to work. Kokkos is an active project with ongoing improvements and projects working on including support for additional architectures. More information on Kokkos can be found on the Kokkos GitHub project.

Available Architecture settings

These are the possible choices for the Kokkos architecture ID. They must be specified in uppercase.

Arch-ID	HOST or GPU	Description
NATIVE	HOST	Local machine
AMDAVX	HOST	AMD chip
ARMV80	HOST	ARMv8.0 Compatible CPU
ARMV81	HOST	ARMv8.1 Compatible CPU
ARMV8_THUNDERX	HOST	ARMv8 Cavium ThunderX CPU
ARMV8_THUNDERX2	HOST	ARMv8 Cavium ThunderX2 CPU
A64FX	HOST	ARMv8.2 with SVE Support
ARMV9_GRACE	HOST	ARMv9 NVIDIA Grace CPU
SNB	HOST	Intel Sandy/Ivy Bridge CPUs
HSW	HOST	Intel Haswell CPUs
BDW	HOST	Intel Broadwell Xeon E-class CPUs
ICL	HOST	Intel Ice Lake Client CPUs (AVX512)
ICX	HOST	Intel Ice Lake Xeon Server CPUs (AVX512)
SKL	HOST	Intel Skylake Client CPUs
SKX	HOST	Intel Skylake Xeon Server CPUs (AVX512)
KNC	HOST	Intel Knights Corner Xeon Phi
KNL	HOST	Intel Knights Landing Xeon Phi
SPR	HOST	Intel Sapphire Rapids Xeon Server CPUs (AVX512)
POWER8	HOST	IBM POWER8 CPUs
POWER9	HOST	IBM POWER9 CPUs
ZEN	HOST	AMD Zen architecture
ZEN2	HOST	AMD Zen2 architecture
ZEN3	HOST	AMD Zen3 architecture
ZEN4	HOST	AMD Zen4 architecture
ZEN5	HOST	AMD Zen5 architecture
RISCV_SG2042	HOST	SG2042 (RISC-V) CPUs
RISCV_RVA22V	HOST	RVA22V (RISC-V) CPUs
KEPLER30	GPU	NVIDIA Kepler generation CC 3.0
KEPLER32	GPU	NVIDIA Kepler generation CC 3.2
KEPLER35	GPU	NVIDIA Kepler generation CC 3.5
KEPLER37	GPU	NVIDIA Kepler generation CC 3.7
MAXWELL50	GPU	NVIDIA Maxwell generation CC 5.0
MAXWELL52	GPU	NVIDIA Maxwell generation CC 5.2
MAXWELL53	GPU	NVIDIA Maxwell generation CC 5.3
PASCAL60	GPU	NVIDIA Pascal generation CC 6.0
PASCAL61	GPU	NVIDIA Pascal generation CC 6.1
VOLTA70	GPU	NVIDIA Volta generation CC 7.0
VOLTA72	GPU	NVIDIA Volta generation CC 7.2
TURING75	GPU	NVIDIA Turing generation CC 7.5
AMPERE80	GPU	NVIDIA Ampere generation CC 8.0
AMPERE86	GPU	NVIDIA Ampere generation CC 8.6
ADA89	GPU	NVIDIA Ada generation CC 8.9
HOPPER90	GPU	NVIDIA Hopper generation CC 9.0
BLACKWELL100	GPU	NVIDIA Blackwell generation CC 10.0
BLACKWELL120	GPU	NVIDIA Blackwell generation CC 12.0
AMD_GFX906	GPU	AMD GPU MI50/60
AMD_GFX908	GPU	AMD GPU MI100
AMD_GFX90A	GPU	AMD GPU MI200
AMD_GFX940	GPU	AMD GPU MI300
AMD_GFX942	GPU	AMD GPU MI300
AMD_GFX942_APU	GPU	AMD APU MI300A
AMD_GFX1030	GPU	AMD GPU V620/W6800
AMD_GFX1100	GPU	AMD GPU RX7900XTX
AMD_GFX1103	GPU	AMD APU Phoenix
INTEL_GEN	GPU	SPIR64-based devices, e.g. Intel GPUs, using JIT
INTEL_DG1	GPU	Intel Iris XeMAX GPU
INTEL_GEN9	GPU	Intel GPU Gen9
INTEL_GEN11	GPU	Intel GPU Gen11
INTEL_GEN12LP	GPU	Intel GPU Gen12LP
INTEL_XEHP	GPU	Intel GPU Xe-HP
INTEL_PVC	GPU	Intel GPU Ponte Vecchio
INTEL_DG2	GPU	Intel GPU DG2

This list was last updated for version 4.6.2 of the Kokkos library.

For multicore CPUs using OpenMP, set these 2 variables.

-D Kokkos_ARCH_HOSTARCH=yes  # HOSTARCH = HOST from list above
-D Kokkos_ENABLE_OPENMP=yes
-D BUILD_OMP=yes

Please note that enabling OpenMP for KOKKOS requires that OpenMP is also enabled for the rest of LAMMPS.

For Intel KNLs using OpenMP, set these variables:

-D Kokkos_ARCH_KNL=yes
-D Kokkos_ENABLE_OPENMP=yes

For NVIDIA GPUs using CUDA, set these variables:

-D Kokkos_ARCH_HOSTARCH=yes   # HOSTARCH = HOST from list above
-D Kokkos_ARCH_GPUARCH=yes    # GPUARCH = GPU from list above
-D Kokkos_ENABLE_CUDA=yes
-D Kokkos_ENABLE_OPENMP=yes

This will also enable executing FFTs on the GPU, either via the internal KISSFFT library, or - by preference - with the cuFFT library bundled with the CUDA toolkit, depending on whether CMake can identify its location.

For AMD or NVIDIA GPUs using HIP, set these variables:

-D Kokkos_ARCH_HOSTARCH=yes   # HOSTARCH = HOST from list above
-D Kokkos_ARCH_GPUARCH=yes    # GPUARCH = GPU from list above
-D Kokkos_ENABLE_HIP=yes
-D Kokkos_ENABLE_OPENMP=yes

This will enable FFTs on the GPU, either by the internal KISSFFT library or with the hipFFT wrapper library, which will call out to the platform-appropriate vendor library: rocFFT on AMD GPUs or cuFFT on NVIDIA GPUs.

For Intel GPUs using SYCL, set these variables:

-D Kokkos_ARCH_HOSTARCH=yes   # HOSTARCH = HOST from list above
-D Kokkos_ARCH_GPUARCH=yes    # GPUARCH = GPU from list above
-D Kokkos_ENABLE_SYCL=yes
-D Kokkos_ENABLE_OPENMP=yes
-D FFT_KOKKOS=MKL_GPU

This will enable FFTs on the GPU using the oneMKL library.

To simplify compilation, seven preset files are included in the cmake/presets folder, kokkos-serial.cmake, kokkos-openmp.cmake, kokkos-cuda.cmake, kokkos-cuda-nowrapper.cmake, kokkos-hip.cmake, kokkos-sycl-nvidia.cmake, and kokkos-sycl-intel.cmake. They will enable the KOKKOS package and enable some hardware choices. For GPU support those preset files may need to be customized to match the hardware used. For some platforms, e.g. CUDA, the Kokkos library will try to auto-detect a suitable configuration. So to compile with CUDA device parallelization with some common packages enabled, you can do the following:

mkdir build-kokkos-cuda
cd build-kokkos-cuda
cmake -C ../cmake/presets/basic.cmake \
      -C ../cmake/presets/kokkos-cuda-nowrapper.cmake ../cmake
cmake --build .

The kokkos-openmp.cmake preset can be combined with any of the others, but it is not possible to combine multiple GPU acceleration settings (CUDA, HIP, SYCL) into a single executable.

Choose which hardware to support in Makefile.machine via KOKKOS_DEVICES and KOKKOS_ARCH settings. See the src/MAKE/OPTIONS/Makefile.kokkos* files for examples.

For multicore CPUs using OpenMP:

KOKKOS_DEVICES = OpenMP
KOKKOS_ARCH = HOSTARCH          # HOSTARCH = HOST from list above

For Intel KNLs using OpenMP:

KOKKOS_DEVICES = OpenMP
KOKKOS_ARCH = KNL

For NVIDIA GPUs using CUDA:

KOKKOS_DEVICES = Cuda
KOKKOS_ARCH = HOSTARCH,GPUARCH  # HOSTARCH = HOST from list above that is
                                #            hosting the GPU
                                # GPUARCH = GPU from list above
KOKKOS_CUDA_OPTIONS = "enable_lambda"
FFT_INC = -DFFT_CUFFT           # enable use of cuFFT (optional)
FFT_LIB = -lcufft               # link to cuFFT library

For GPUs, you also need the following lines in your Makefile.machine before the CC line is defined. They tell mpicxx to use an nvcc compiler wrapper, which will use nvcc for compiling CUDA files and a C++ compiler for non-Kokkos, non-CUDA files.

# For OpenMPI
KOKKOS_ABSOLUTE_PATH = $(shell cd $(KOKKOS_PATH); pwd)
export OMPI_CXX = $(KOKKOS_ABSOLUTE_PATH)/config/nvcc_wrapper
CC = mpicxx

# For MPICH and derivatives
KOKKOS_ABSOLUTE_PATH = $(shell cd $(KOKKOS_PATH); pwd)
CC = mpicxx -cxx=$(KOKKOS_ABSOLUTE_PATH)/config/nvcc_wrapper

For AMD or NVIDIA GPUs using HIP:

KOKKOS_DEVICES = HIP
KOKKOS_ARCH = HOSTARCH,GPUARCH  # HOSTARCH = HOST from list above that is
                                #            hosting the GPU
                                # GPUARCH = GPU from list above
FFT_INC = -DFFT_HIPFFT          # enable use of hipFFT (optional)
FFT_LIB = -lhipfft              # link to hipFFT library

For Intel GPUs using SYCL:

KOKKOS_DEVICES = SYCL
KOKKOS_ARCH = HOSTARCH,GPUARCH  # HOSTARCH = HOST from list above that is
                                #            hosting the GPU
                                # GPUARCH = GPU from list above
FFT_INC = -DFFT_KOKKOS_MKL_GPU  # enable use of oneMKL for Intel GPUs (optional)
                                # link to oneMKL FFT library
FFT_LIB = -lmkl_sycl_dft -lmkl_intel_ilp64 -lmkl_tbb_thread -mkl_core -ltbb

Advanced KOKKOS compilation settings

There are other allowed options when building with the KOKKOS package that can improve performance or assist in debugging or profiling. Below are some examples that may be useful in combination with LAMMPS. For the full list (which keeps changing as the Kokkos package itself evolves), please consult the Kokkos library documentation.

As alternative to using multi-threading via OpenMP (-DKokkos_ENABLE_OPENMP=on or KOKKOS_DEVICES=OpenMP) it is also possible to use Posix threads directly (-DKokkos_ENABLE_PTHREAD=on or KOKKOS_DEVICES=Pthread). While binding of threads to individual or groups of CPU cores is managed in OpenMP with environment variables, you need assistance from either the “hwloc” or “libnuma” library for the Pthread thread parallelization option. To enable use with CMake: -DKokkos_ENABLE_HWLOC=on or -DKokkos_ENABLE_LIBNUMA=on; and with conventional make: KOKKOS_USE_TPLS=hwloc or KOKKOS_USE_TPLS=libnuma.

The CMake option -DKokkos_ENABLE_LIBRT=on or the makefile setting KOKKOS_USE_TPLS=librt enables the use of a more accurate timer mechanism on many Unix-like platforms for internal profiling.

The CMake option -DKokkos_ENABLE_DEBUG=on or the makefile setting KOKKOS_DEBUG=yes enables printing of run-time debugging information that can be useful. It also enables runtime bounds checking on Kokkos data structures. As to be expected, enabling this option will negatively impact the performance and thus is only recommended when developing a Kokkos-enabled style in LAMMPS.

The CMake option -DKokkos_ENABLE_CUDA_UVM=on or the makefile setting KOKKOS_CUDA_OPTIONS=enable_lambda,force_uvm enables the use of CUDA “Unified Virtual Memory” (UVM) in Kokkos. UVM allows to transparently use RAM on the host to supplement the memory used on the GPU (with some performance penalty) and thus enables running larger problems that would otherwise not fit into the RAM on the GPU.

The CMake option -D KOKKOS_PREC=value sets the floating point precision of the calculations, where value can be one of: double (FP64, default) or mixed (FP64 for accumulation of forces, energy, and virial, FP32 otherwise) or single (FP32). Similarly the makefile settings -DLMP_KOKKOS_DOUBLE_DOUBLE (default), -DLMP_KOKKOS_SINGLE_DOUBLE, and -DLMP_KOKKOS_SINGLE_SINGLE set double, mixed, single precision respectively. When using reduced precision (single or mixed), the simulation should be carefully checked to ensure it is stable and that energy is acceptably conserved.

The CMake option -D KOKKOS_LAYOUT=value sets the array layout of Kokkos views (e.g. forces, velocities, etc.) on GPUs, where value can be one of: legacy (mostly LayoutRight, default) or default (mostly LayoutLeft). Similarly the makefile settings -DLMP_KOKKOS_LAYOUT_LEGACY (default) and -DLMP_KOKKOS_LAYOUT_DEFAULT set legacy or default layouts respectively. Using the default layout (LayoutLeft) can give speedup on GPUs for some models, but a slowdown for others. LayoutRight is always used for positions on GPUs since it has been found to be faster, and when compiling exclusively for CPUs.

3.8.5. LEPTON package

To build with this package, you must build the Lepton library which is included in the LAMMPS source distribution in the lib/lepton folder.

This is the recommended build procedure for using Lepton in LAMMPS. No additional settings are normally needed besides -D PKG_LEPTON=yes.

On x86 hardware the Lepton library will also include a just-in-time compiler for faster execution. This is auto detected but can be explicitly disabled by setting -D LEPTON_ENABLE_JIT=no (or enabled by setting it to yes).

3.8.6. MACHDYN package

To build with this package, you must download the Eigen3 library. Eigen3 is a template library, so you do not need to build it.

-D DOWNLOAD_EIGEN3            # download Eigen3, value = no (default) or yes
-D EIGEN3_INCLUDE_DIR=path    # path to Eigen library (only needed if a
                              # custom location)

If DOWNLOAD_EIGEN3 is set, the Eigen3 library will be downloaded and inside the CMake build directory. If the Eigen3 library is already on your system (in a location where CMake cannot find it), set EIGEN3_INCLUDE_DIR to the directory the Eigen3 include file is in.

3.8.7. ML-IAP package

Building the ML-IAP package requires including the ML-SNAP package. There will be an error message if this requirement is not satisfied. Using the mliappy model also requires enabling Python support, which in turn requires to include the PYTHON package and requires to have the cython software installed and with it a working cythonize command. This feature requires compiling LAMMPS with Python version 3.6 or later.

-D MLIAP_ENABLE_PYTHON=value   # enable mliappy model (default is autodetect)

Without this setting, CMake will check whether it can find a suitable Python version and the cythonize command and choose the default accordingly. During the build procedure the provided .pyx file(s) will be automatically translated to C++ code and compiled. Please do not run cythonize manually in the src/ML-IAP folder, as that can lead to compilation errors if Python support is not enabled. If you did it by accident, please remove the generated .cpp and .h files.

The build uses the lib/python/Makefile.mliap_python file in the compile/link process to add a rule to update the files generated by the cythonize command in case the corresponding .pyx file(s) were modified. You may need to modify lib/python/Makefile.lammps if the LAMMPS build fails.

To enable building the ML-IAP package with Python support enabled, you need to add -DMLIAP_PYTHON to the LMP_INC variable in your machine makefile. You may have to manually run the cythonize command on .pyx file(s) in the src folder, if this is not automatically done during installing the ML-IAP package. Please do not run cythonize in the src/ML-IAP folder, as that can lead to compilation errors if Python support is not enabled. If you did this by accident, please remove the generated .cpp and .h files.

3.8.8. OPT package

No additional settings are needed besides -D PKG_OPT=yes

The compiler flag -restrict must be used to build LAMMPS with the OPT package when using Intel compilers. It should be added to the CCFLAGS line of your Makefile.machine. See src/MAKE/OPTIONS/Makefile.opt for an example.

3.8.9. PYTHON package

Building with the PYTHON package requires you have a the Python development headers and library available on your system, which needs to be Python version 3.6 or later. See lib/python/README for additional details.

-D Python_EXECUTABLE=path   # path to Python executable to use

Without this setting, CMake will guess the default Python version on your system. To use a different Python version, you can either create a virtualenv, activate it and then run cmake. Or you can set the Python_EXECUTABLE variable to specify which Python interpreter should be used. Note note that you will also need to have the development headers installed for this version, e.g. python3-devel.

3.8.10. VORONOI package

To build with this package, you must download and build the Voro++ library or install a binary package provided by your operating system.

-D DOWNLOAD_VORO=value    # download Voro++ for build
                          # value = no (default) or yes
-D VORO_LIBRARY=path      # Voro++ library file
                          # (only needed if at custom location)
-D VORO_INCLUDE_DIR=path  # Voro++ include directory
                          # (only needed if at custom location)

If DOWNLOAD_VORO is set, the Voro++ library will be downloaded and built inside the CMake build directory. If the Voro++ library is already on your system (in a location CMake cannot find it), VORO_LIBRARY is the filename (plus path) of the Voro++ library file, not the directory the library file is in. VORO_INCLUDE_DIR is the directory the Voro++ include file is in.

3.8.11. ADIOS package

The ADIOS package requires the ADIOS I/O library, version 2.3.1 or newer. Make sure that you have ADIOS built either with or without MPI to match if you build LAMMPS with or without MPI. ADIOS compilation settings for LAMMPS are automatically detected, if the PATH and LD_LIBRARY_PATH environment variables have been updated for the local ADIOS installation and the instructions below are followed for the respective build systems.

-D ADIOS2_DIR=path        # path is where ADIOS 2.x is installed
-D PKG_ADIOS=yes

Turn on the ADIOS package before building LAMMPS. If the ADIOS 2.x software is installed in PATH, there is nothing else to do:

make yes-adios

otherwise, set ADIOS2_DIR environment variable when turning on the package:

ADIOS2_DIR=path make yes-adios   # path is where ADIOS 2.x is installed

3.8.12. APIP package

The APIP package depends on the library of the ML-PACE package. The code for the library can be found at: https://github.com/ICAMS/lammps-user-pace/

No additional settings are needed besides -D PKG_APIP=yes and -D PKG_ML-PACE=yes. One can use a local version of the ML-PACE library instead of automatically downloading the library as described here.

3.8.13. COLVARS package

This package enables the use of the Colvars module included in the LAMMPS source distribution.

This is the recommended build procedure for using Colvars in LAMMPS. No additional settings are normally needed besides -D PKG_COLVARS=yes. The following CMake variables are available.

-D PKG_COLVARS=yes          # enable the package itself
-D COLVARS_LEPTON=yes       # use the Lepton library for custom expression (on by defaul)
-D COLVARS_DEBUG=no         # eneable debugging message (verbose, off by default)

3.8.14. ELECTRODE package

This package depends on the KSPACE package.

-D PKG_ELECTRODE=yes          # enable the package itself
-D PKG_KSPACE=yes             # the ELECTRODE package requires KSPACE
-D USE_INTERNAL_LINALG=value  #

Features in the ELECTRODE package are dependent on code in the KSPACE package so the latter one must be enabled.

The ELECTRODE package also requires LAPACK (and BLAS) and CMake can identify their locations and pass that info to the ELECTRODE build script. But on some systems this may cause problems when linking or the dependency is not desired. Try enabling USE_INTERNAL_LINALG in those cases to use the bundled linear algebra library and work around the limitation.

3.8.15. ML-PACE package

This package requires a library that can be downloaded and built in lib/pace or somewhere else, which must be done before building LAMMPS with this package. The code for the library can be found at: https://github.com/ICAMS/lammps-user-pace/

Instead of including the ML-PACE package directly into LAMMPS, it is also possible to skip this step and build the ML-PACE package as a plugin using the CMake script files in the examples/PACKAGE/pace/plugin folder and then load this plugin at runtime with the plugin command.

By default the library will be downloaded from the git repository and built automatically when the ML-PACE package is enabled with -D PKG_ML-PACE=yes. The location for the sources may be customized by setting the variable PACELIB_URL when configuring with CMake (e.g. to use a local archive on machines without internet access). Since CMake checks the validity of the archive with md5sum you may also need to set PACELIB_MD5 if you provide a different library version than what is downloaded automatically.

3.8.16. ML-POD package

No additional settings are needed besides -D PKG_ML-POD=yes.

3.8.17. ML-QUIP package

To build with this package, you must download and build the QUIP library. It can be obtained from GitHub. For support of GAP potentials, additional files with specific licensing conditions need to be downloaded and configured. The automatic download will from within CMake will download the non-commercial use version.

-D DOWNLOAD_QUIP=value       # download QUIP library for build
                             # value = no (default) or yes
-D QUIP_LIBRARY=path         # path to libquip.a
                             # (only needed if a custom location)
-D USE_INTERNAL_LINALG=value # Use the internal linear algebra library
                             # instead of LAPACK
                             # value = no (default) or yes

CMake will try to download and build the QUIP library from GitHub, if it is not found on the local machine. This requires to have git installed. It will use the same compilers and flags as used for compiling LAMMPS. Currently this is only supported for the GNU and the Intel compilers. Set the QUIP_LIBRARY variable if you want to use a previously compiled and installed QUIP library and CMake cannot find it.

The QUIP library requires LAPACK (and BLAS) and CMake can identify their locations and pass that info to the QUIP build script. But on some systems this triggers a (current) limitation of CMake and the configuration will fail. Try enabling USE_INTERNAL_LINALG in those cases to use the bundled linear algebra library and work around the limitation.

3.8.18. PLUMED package

Before building LAMMPS with this package, you must first build PLUMED. PLUMED can be built as part of the LAMMPS build or installed separately from LAMMPS using the generic PLUMED installation instructions. The PLUMED package has been tested to work with Plumed versions 2.4.x, to 2.9.x and will error out, when trying to run calculations with a different version of the Plumed kernel.

PLUMED can be linked into MD codes in three different modes: static, shared, and runtime. With the “static” mode, all the code that PLUMED requires is linked statically into LAMMPS. LAMMPS is then fully independent from the PLUMED installation, but you have to rebuild/relink it in order to update the PLUMED code inside it. With the “shared” linkage mode, LAMMPS is linked to a shared library that contains the PLUMED code. This library should preferably be installed in a globally accessible location. When PLUMED is linked in this way the same library can be used by multiple MD packages. Furthermore, the PLUMED library LAMMPS uses can be updated without the need for a recompile of LAMMPS for as long as the shared PLUMED library is ABI-compatible.

The third linkage mode is “runtime” which allows the user to specify which PLUMED kernel should be used at runtime by using the PLUMED_KERNEL environment variable. This variable should point to the location of the libplumedKernel.so dynamical shared object, which is then loaded at runtime. This mode of linking is particularly convenient for doing PLUMED development and comparing multiple PLUMED versions as these sorts of comparisons can be done without recompiling the hosting MD code. All three linkage modes are supported by LAMMPS on selected operating systems (e.g. Linux) and using either CMake or traditional make build. The “static” mode should be the most portable, while the “runtime” mode support in LAMMPS makes the most assumptions about operating system and compiler environment. If one mode does not work, try a different one, switch to a different build system, consider a global PLUMED installation or consider downloading PLUMED during the LAMMPS build.

Instead of including the PLUMED package directly into LAMMPS, it is also possible to skip this step and build the PLUMED package as a plugin using the CMake script files in the examples/PACKAGE/plumed/plugin folder and then load this plugin at runtime with the plugin command.

When the -D PKG_PLUMED=yes flag is included in the cmake command you must ensure that the GNU Scientific Library (GSL) <https://www.gnu.org/software/gsl/> is installed in locations that are accessible in your environment. There are then two additional variables that control the manner in which PLUMED is obtained and linked into LAMMPS.

-D DOWNLOAD_PLUMED=value   # download PLUMED for build
                           # value = no (default) or yes
-D PLUMED_MODE=value       # Linkage mode for PLUMED
                           # value = static (default), shared,
                           #         or runtime

If DOWNLOAD_PLUMED is set to yes, the PLUMED library will be downloaded (the version of PLUMED that will be downloaded is hard-coded to a vetted version of PLUMED, usually a recent stable release version) and built inside the CMake build directory. If DOWNLOAD_PLUMED is set to “no” (the default), CMake will try to detect and link to an installed version of PLUMED. For this to work, the PLUMED library has to be installed into a location where the pkg-config tool can find it or the PKG_CONFIG_PATH environment variable has to be set up accordingly. PLUMED should be installed in such a location if you compile it using the default make; make install commands.

The PLUMED_MODE setting determines the linkage mode for the PLUMED library. The allowed values for this flag are “static” (default), “shared”, or “runtime”. If you want to switch the linkage mode, just re-run CMake with a different setting. For a discussion of PLUMED linkage modes, please see above. When DOWNLOAD_PLUMED is enabled the static linkage mode is recommended.

3.8.19. H5MD package

To build with this package you must have the HDF5 software package installed on your system, which should include the h5cc compiler and the HDF5 library.

No additional settings are needed besides -D PKG_H5MD=yes.

This should auto-detect the H5MD library on your system. Several advanced CMake H5MD options exist if you need to specify where it is installed. Use the ccmake (terminal window) or cmake-gui (graphical) tools to see these options and set them interactively from their user interfaces.

3.8.20. ML-HDNNP package

To build with the ML-HDNNP package it is required to download and build the external n2p2 library v2.1.4 (or higher). The LAMMPS build process offers an automatic download and compilation of n2p2 or allows you to choose the installation directory of n2p2 manually. Please see the boxes below for the CMake and traditional build system for detailed information.

In case of a manual installation of n2p2 you only need to build the n2p2 core library libnnp and interface library libnnpif. When using GCC it should suffice to execute make libnnpif in the n2p2 src directory. For more details please see lib/hdnnp/README and the n2p2 build documentation.

-D DOWNLOAD_N2P2=value    # download n2p2 for build
                          # value = no (default) or yes
-D N2P2_DIR=path          # n2p2 base directory
                          # (only needed if a custom location)

If DOWNLOAD_N2P2 is set, the n2p2 library will be downloaded and built inside the CMake build directory. If the n2p2 library is already on your system (in a location CMake cannot find it), set the N2P2_DIR to path where n2p2 is located. If n2p2 is located directly in lib/hdnnp/n2p2 it will be automatically found by CMake.

3.8.21. INTEL package

To build with this package, you must choose which hardware you want to build for, either x86 CPUs or Intel KNLs in offload mode. You should also typically install the OPENMP package, as it can be used in tandem with the INTEL package to good effect, as explained on the INTEL package page.

When using Intel compilers version 16.0 or later is required. You can also use the GNU or Clang compilers and they will provide performance improvements over regular styles and OPENMP styles, but less so than with the Intel compilers. Please also note, that some compilers have been found to apply memory alignment constraints incompletely or incorrectly and thus can cause segmentation faults in otherwise correct code when using features from the INTEL package.

-D INTEL_ARCH=value     # value = cpu (default) or knl
-D INTEL_LRT_MODE=value # value = threads, none, or c++17

Choose which hardware to compile for in Makefile.machine via the following settings. See src/MAKE/OPTIONS/Makefile.intel_cpu* and Makefile.knl files for examples. and src/INTEL/README for additional information.

For CPUs:

OPTFLAGS =  -xHost -O2 -fp-model fast=2 -no-prec-div -qoverride-limits -qopt-zmm-usage=high
CCFLAGS =   -g -qopenmp -DLAMMPS_MEMALIGN=64 -no-offload -fno-alias -ansi-alias -restrict $(OPTFLAGS)
LINKFLAGS = -g -qopenmp $(OPTFLAGS)
LIB =       -ltbbmalloc

For KNLs:

OPTFLAGS =  -xMIC-AVX512 -O2 -fp-model fast=2 -no-prec-div -qoverride-limits
CCFLAGS =   -g -qopenmp -DLAMMPS_MEMALIGN=64 -no-offload -fno-alias -ansi-alias -restrict $(OPTFLAGS)
LINKFLAGS = -g -qopenmp $(OPTFLAGS)
LIB =       -ltbbmalloc

In Long-range thread mode (LRT) a modified verlet style is used, that operates the Kspace calculation in a separate thread concurrently to other calculations. This has to be enabled in the package intel command at runtime. With the setting “threads” it used the pthreads library, while “c++17” will use the built-in thread support of C++17 compilers. The option “none” skips compilation of this feature. The default is to use “threads” if pthreads is available and otherwise “none”.

Best performance is achieved with Intel hardware, Intel compilers, as well as the Intel TBB and MKL libraries. However, the code also compiles, links, and runs with other compilers / hardware and without TBB and MKL.

3.8.22. MDI package

-D DOWNLOAD_MDI=value    # download MDI Library for build
                         # value = no (default) or yes

3.8.23. MISC package

The fix imd style in this package can be run either synchronously (communication with IMD clients is done in the main process) or asynchronously (the fix spawns a separate thread that can communicate with IMD clients concurrently to the LAMMPS execution).

-D LAMMPS_ASYNC_IMD=value  # Run IMD server asynchronously
                           # value = no (default) or yes

To enable asynchronous mode the -DLAMMPS_ASYNC_IMD define needs to be added to the LMP_INC variable in the Makefile.machine you are using. For example:

LMP_INC = -DLAMMPS_ASYNC_IMD -DLAMMPS_MEMALIGN=64

3.8.24. MOLFILE package

-D MOLFILE_INCLUDE_DIR=path   # (optional) path where VMD molfile
                              # plugin headers are installed
-D PKG_MOLFILE=yes

Using -D PKG_MOLFILE=yes enables the package, and setting -D MOLFILE_INCLUDE_DIR allows to provide a custom location for the molfile plugin header files. These should match the ABI of the plugin files used, and thus one typically sets them to include folder of the local VMD installation in use. LAMMPS ships with a couple of default header files that correspond to a popular VMD version, usually the latest release.

3.8.25. NETCDF package

To build with this package you must have the NetCDF library installed on your system.

No additional settings are needed besides -D PKG_NETCDF=yes.

This should auto-detect the NETCDF library if it is installed on your system at standard locations. Several advanced CMake NETCDF options exist if you need to specify where it was installed. Use the ccmake (terminal window) or cmake-gui (graphical) tools to see these options and set them interactively from their user interfaces.

3.8.26. OPENMP package

No additional settings are required besides -D PKG_OPENMP=yes. If CMake detects OpenMP compiler support, the OPENMP code will be compiled with multi-threading support enabled, otherwise as optimized serial code.

To enable multi-threading support in the OPENMP package (and other styles supporting OpenMP) the following compile and link flags must be added to your Makefile.machine file. See src/MAKE/OPTIONS/Makefile.omp for an example.

CCFLAGS: -fopenmp               # for GNU and Clang Compilers
CCFLAGS: -qopenmp -restrict     # for Intel compilers on Linux
LINKFLAGS: -fopenmp             # for GNU and Clang Compilers
LINKFLAGS: -qopenmp             # for Intel compilers on Linux

For other platforms and compilers, please consult the documentation about OpenMP support for your compiler.

Adding OpenMP support on macOS

Apple offers the Xcode package and IDE for compiling software on macOS, so you have likely installed it to compile LAMMPS. Their compiler is based on Clang, but while it is capable of processing OpenMP directives, the necessary header files and OpenMP runtime library are missing. The R developers have figured out a way to build those in a compatible fashion. One can download them from https://mac.r-project.org/openmp/. Simply adding those files as instructed enables the Xcode C++ compiler to compile LAMMPS with -D BUILD_OMP=yes.

3.8.27. QMMM package

For using LAMMPS to do QM/MM simulations via the QMMM package you need to build LAMMPS as a library. A LAMMPS executable with fix qmmm included can be built, but will not be able to do a QM/MM simulation on as such. You must also build a QM code - currently only Quantum ESPRESSO (QE) is supported - and create a new executable which links LAMMPS and the QM code together. Details are given in the lib/qmmm/README file. It is also recommended to read the instructions for linking with LAMMPS as a library for background information. This requires compatible Quantum Espresso and LAMMPS versions. The current interface and makefiles have last been verified to work in February 2020 with Quantum Espresso versions 6.3 to 6.5.

When using CMake, building a LAMMPS library is required and it is recommended to build a shared library, since any libraries built from the sources in the lib folder (including the essential libqmmm.a) are not included in the static LAMMPS library and are (currently) not installed, while their code is included in the shared LAMMPS library. Thus a typical command to configure building LAMMPS for QMMM would be:

cmake -C ../cmake/presets/basic.cmake -D PKG_QMMM=yes \
    -D BUILD_LIB=yes -DBUILD_SHARED_LIBS=yes ../cmake

After completing the LAMMPS build and also configuring and compiling Quantum ESPRESSO with external library support (via “make couple”), go back to the lib/qmmm folder and follow the instructions on the README file to build the combined LAMMPS/QE QM/MM executable (pwqmmm.x) in the lib/qmmm folder.

3.8.28. RHEO package

This package depends on the BPM package.

-D PKG_RHEO=yes               # enable the package itself
-D PKG_BPM=yes                # the RHEO package requires BPM
-D USE_INTERNAL_LINALG=value  # prefer internal LAPACK if true

Some features in the RHEO package are dependent on code in the BPM package so the latter one must be enabled as well.

The RHEO package also requires LAPACK (and BLAS) and CMake can identify their locations and pass that info to the RHEO build script. But on some systems this may cause problems when linking or the dependency is not desired. By using the setting -D USE_INTERNAL_LINALG=yes when running the CMake configuration, you will select compiling and linking the bundled linear algebra library and work around the limitations.

3.8.29. SCAFACOS package

To build with this package, you must download and build the ScaFaCoS Coulomb solver library

-D DOWNLOAD_SCAFACOS=value    # download ScaFaCoS for build, value = no (default) or yes
-D SCAFACOS_LIBRARY=path      # ScaFaCos library file (only needed if at custom location)
-D SCAFACOS_INCLUDE_DIR=path  # ScaFaCoS include directory (only needed if at custom location)

If DOWNLOAD_SCAFACOS is set, the ScaFaCoS library will be downloaded and built inside the CMake build directory. If the ScaFaCoS library is already on your system (in a location CMake cannot find it), SCAFACOS_LIBRARY is the filename (plus path) of the ScaFaCoS library file, not the directory the library file is in. SCAFACOS_INCLUDE_DIR is the directory the ScaFaCoS include file is in.

3.8.30. VTK package

To build with this package you must have the VTK library installed on your system.

No additional settings are needed besides -D PKG_VTK=yes.

This should auto-detect the VTK library if it is installed on your system at standard locations. Several advanced VTK options exist if you need to specify where it was installed. Use the ccmake (terminal window) or cmake-gui (graphical) tools to see these options and set them interactively from their user interfaces.