Advanced Installation Options
This guide helps you install ABACUS with advanced features. Please make sure to read the easy-installation guide before. ABACUS use exchange-correlation functionals by default. However, for some functionals (such as HSE hybrid functional), Libxc is required. Dependency: Libxc >= 5.1.7 . Note: Building Libxc from source with Makefile does NOT support using it in CMake here. Please compile Libxc with CMake instead. If Libxc is not installed in standard path (i.e. installed with a custom prefix path), you can set If DeePKS feature is required for DeePKS-kit, the following prerequisites and steps are needed: C++ compiler, supporting C++14 (GCC >= 5 is sufficient) CMake >= 3.18 LibTorch with cxx11 ABI supporting CPU CMake will try to download Libnpy if it cannot be found locally. Note: This part is only required if you want to load a trained DeeP Potential and run molecular dynamics with that. To train the DeeP Potential with DP-GEN, no extra prerequisite is needed and please refer to this page for ABACUS interface with DP-GEN. If the Deep Potential model is employed in Molecule Dynamics calculations, the following prerequisites and steps are needed: The new EXX implementation depends on two external libraries: These two libraries are added as submodules in the deps folder. Set If you prefer using manually downloaded libraries, provide To build tests for ABACUS, define After building and installing, unit tests can be performed with To run a subset of unit test, use To build performance tests for ABACUS, define Google Benchmark requires Google Test to build and run the tests. When setting To build NVIDIA GPU support for ABACUS, define Note: This flag is enabled by default. It will get better performance than the standard implementation on To build math functions from source code, instead of using c++ standard implementation, define Currently supported math functions: ABACUS supports the PEXSI library for gamma only LCAO calculations. PEXSI version 2.0.0 is tested to work with ABACUS, please always use the latest version of PEXSI. To build ABACUS with PEXSI support, you need to compile PEXSI (and its dependencies) first. Please refer to the PEXSI Installation Guide for more details. Note that PEXSI requires ParMETIS and SuperLU_DIST. After compiling PEXSI, you can set Note: We suggest using CMake to configure and compile. To compile the ABACUS program using legacy For example, below is a case where the Intel C++ compiler, Intel MPI and CEREAL are used, along with Intel MKL library. The file Makefile.vars can be set as follows: When Another example is where the Gnu C++ compiler, MPICH, OPENBLAS, ScaLAPACK, ELPA and CEREAL are used: When Except modifying ABACUS now support full version and pw version. Use Besides, libxc and deepks are optional libs to compile abacus. They will be used when or After modifying the After the compilation finishes without error messages (except perhaps for some warnings), an executable program The program compiled using the above instructions do not link with LIBXC and use exchange-correlation functionals as written in the ABACUS program. However, for some functionals (such as HSE hybrid functional), LIBXC is required. To compile ABACUS with LIBXC, you need to define directly. To compile ABACUS with DEEPKS, you need to define directly. Note: This part is only required if you want to load a trained DeeP Potential and run molecular dynamics with that. To train the DeeP Potential with DP-GEN, no extra prerequisite is needed and please refer to this page for ABACUS interface with DP-GEN. To compile ABACUS with DeePMD-kit, you need to define directly. To use new EXX, you need two libraries: LibRI and LibComm and need to define directly.Build with Libxc
Libxc_DIR
to the corresponding directory.cmake -B build -DLibxc_DIR=~/libxc
Build with DeePKS
cmake -B build -DENABLE_DEEPKS=1 -DTorch_DIR=~/libtorch/share/cmake/Torch/ -Dlibnpy_INCLUDE_DIR=~/libnpy/include
Build with DeePMD-kit
cmake -B build -DDeePMD_DIR=~/deepmd-kit -DTensorFlow_DIR=~/tensorflow
deepmd_c
/deepmd_cc
and tensorflow_cc
libraries would be called according to DeePMD_DIR
and TensorFlow_DIR
, which is showed in detail in this page.Build with LibRI and LibComm
-DENABLE_LIBRI=ON
to build with these two libraries.-DLIBRI_DIR=${path to your LibRI folder} -DLIBCOMM_DIR=${path to your LibComm folder}
.Build Unit Tests
BUILD_TESTING
flag. You can also specify path to local installation of Googletest by setting GTEST_DIR
flags. If not found in local, the configuration process will try to download it automatically.cmake -B build -DBUILD_TESTING=1
ctest
.ctest -R <test-match-pattern>
to perform tests with name matched by given pattern.Build Performance Tests
ENABLE_GOOGLEBENCH
flag. You can also specify the path to a local installation of Google Benchmark by setting BENCHMARK_DIR
flags. If not found locally, the configuration process will try to download it automatically.cmake -B build -DENABLE_GOOGLEBENCH=1
ENABLE_GOOGLEBENCH
to ON, BUILD_TESTING
is automatically enabled. After building and installing, performance tests can be executed with ctest
.Build with CUDA support
Extra prerequisites
USE_CUDA
flag. You can also specify path to local installation of CUDA Toolkit by setting CMAKE_CUDA_COMPILER
flags.cmake -B build -DUSE_CUDA=1 -DCMAKE_CUDA_COMPILER=${path to cuda toolkit}/bin/nvcc
Build math library from source
gcc
and clang
. But it will be disabled when using Intel Compiler
since the math functions will get wrong results and the performance is also unexpectly poor.USE_ABACUS_LIBM
flag.sin
, cos
, sincos
, exp
, cexp
cmake -B build -DUSE_ABACUS_LIBM=1
Build with PEXSI support
ENABLE_PEXSI
to ON
. If the libraries are not installed in standard paths, you can set PEXSI_DIR
, ParMETIS_DIR
and SuperLU_DIST_DIR
to the corresponding directories.cmake -B build -DENABLE_PEXSI=ON -DPEXSI_DIR=${path to PEXSI installation directory} -DParMETIS_DIR=${path to ParMETIS installation directory} -DSuperLU_DIST_DIR=${path to SuperLU_DIST installation directory}
Build ABACUS with make
make
, users need to specify the location of the compilers, headers and libraries in source/Makefile.vars
:# This is the Makefile of ABACUS API
#======================================================================
# Users set
#======================================================================
CXX = mpiicpc
# mpiicpc: compile intel parallel version
# icpc: compile intel sequential version
# make: ELPA_DIR, ELPA_INCLUDE_DIR, CEREAL_DIR must also be set.
# make pw: nothing need to be set except LIBXC_DIR
#
# mpicxx: compile gnu parallel version
# g++: compile gnu sequential version
# make: FFTW_DIR, OPENBLAS_LIB_DIR, SCALAPACK_LIB_DIR, ELPA_DIR, ELPA_INCLUDE_DIR, CEREAL_DIR must also be set.
# make pw: FFTW_DIR, OPENBLAS_LIB_DIR must be set.
# GPU = OFF #We do not support GPU yet
# OFF: do not use GPU
# CUDA: use CUDA
#======================================================================
#-------------------- FOR INTEL COMPILER ----------------------------
## ELPA_DIR should contain an include folder and lib/libelpa.a
## CEREAL_DIR should contain an include folder.
#----------------------------------------------------------------------
ELPA_DIR = /usr/local/include/elpa-2021.05.002
ELPA_INCLUDE_DIR = ${ELPA_DIR}/elpa
CEREAL_DIR = /usr/local/include/cereal
##------------------- FOR GNU COMPILER ------------------------------
## FFTW_DIR should contain lib/libfftw3.a.
## OPENBLAS_LIB_DIR should contain libopenblas.a.
## SCALAPACK_LIB_DIR should contain libscalapack.a
## All three above will only be used when CXX=mpicxx or g++
## ELPA_DIR should contain an include folder and lib/libelpa.a
## CEREAL_DIR should contain an include folder.
##---------------------------------------------------------------------
# FFTW_DIR = /public/soft/fftw_3.3.8
# OPENBLAS_LIB_DIR = /public/soft/openblas/lib
# SCALAPACK_LIB_DIR = /public/soft/openblas/lib
# ELPA_DIR = /public/soft/elpa_21.05.002
# ELPA_INCLUDE_DIR = ${ELPA_DIR}/include/elpa-2021.05.002
# CEREAL_DIR = /public/soft/cereal
##------------------- OPTIONAL LIBS ---------------------------------
## To use DEEPKS: set LIBTORCH_DIR and LIBNPY_DIR
## To use LIBXC: set LIBXC_DIR which contains include and lib/libxc.a (>5.1.7)
## To use DeePMD: set DeePMD_DIR and TensorFlow_DIR
## To use LibRI: set LIBRI_DIR and LIBCOMM_DIR
##---------------------------------------------------------------------
# LIBTORCH_DIR = /usr/local
# LIBNPY_DIR = /usr/local
# LIBXC_DIR = /public/soft/libxc
# DeePMD_DIR = ${deepmd_root}
# TensorFlow_DIR = ${tensorflow_root}
# LIBRI_DIR = /public/software/LibRI
# LIBCOMM_DIR = /public/software/LibComm
##---------------------------------------------------------------------
# NP = 14 # It is not supported. use make -j14 or make -j to parallelly compile
# DEBUG = OFF
# Only for developers
# ON: use gnu compiler and check segmental defaults
# OFF: nothing
#======================================================================
CXX = mpiicpc #(or CXX = icpc)
ELPA_DIR = /public/soft/elpa_21.05.002
ELPA_INCLUDE_DIR = ${ELPA_DIR}/include/elpa-2021.05.002
CEREAL_DIR = /public/soft/cereal
CXX=mpiicpc
, a parallel version will be compiled. When CXX=icpc
, a sequential version will be compiled.CXX = mpicxx/g++
FFTW_DIR = /public/soft/fftw_3.3.8
OPENBLAS_LIB_DIR = /public/soft/openblas/lib
SCALAPACK_LIB_DIR = /public/soft/openblas/lib
ELPA_DIR = /public/soft/elpa_21.05.002
ELPA_INCLUDE_DIR = ${ELPA_DIR}/include/elpa-2021.05.002
CEREAL_DIR = /public/soft/cereal
CXX=mpicxx
, a parallel version will be compiled. When CXX=g++
, a sequential version will be compiled.Makefile.vars
, you can also directly usemake CXX=mpiicpc ELPA_DIR=/public/soft/elpa_21.05.002 \
ELPA_INCLUDE_DIR=${ELPA_DIR}/include/elpa-2021.05.002 \
CEREAL_DIR=/public/soft/cereal
make
or make abacus
to compile full version which supports LCAO calculations. Use make pw
to compile pw version which only supports pw calculations. For pw version, make pw CXX=mpiicpc
, you do not need to provide any libs. For make pw CXX=mpicxx
, you need provide FFTW_DIR
and OPENBLAS_LIB_DIR
.LIBXC_DIR
is defined likeLIBXC_DIR = /public/soft/libxc
LIBTORCH_DIR
and LIBNPY_DIR
likeLIBTORCH_DIR = /usr/local
LIBNPY_DIR = /usr/local
Makefile.vars
file, execute make
or make -j12
to build the program.ABACUS.mpi
will be created in directory bin/
.Add Libxc Support
LIBXC_DIR
in the file Makefile.vars
or usemake LIBXC_DIR=/pulic/soft/libxc
Add DeePKS Support
LIBTORCH_DIR
and LIBNPY_DIR
in the file Makefile.vars
or usemake LIBTORCH_DIR=/opt/libtorch/ LIBNPY_DIR=/opt/libnpy/
Add DeePMD-kit Support
DeePMD_DIR
and TensorFlow_DIR
in the file Makefile.vars
or usemake DeePMD_DIR=~/deepmd-kit TensorFlow_DIR=~/tensorflow
deepmd_c
/deepmd_cc
and tensorflow_cc
libraries would be called according to DeePMD_DIR
and TensorFlow_DIR
, which is showed in detail in this page.Add LibRI Support
LIBRI_DIR
and LIBCOMM_DIR
in the file Makefile.vars
or usemake LIBRI_DIR=/public/software/LibRI LIBCOMM_DIR=/public/software/LibComm