Skip to content

Latest commit

 

History

History
144 lines (93 loc) · 4.85 KB

INSTALL.rst

File metadata and controls

144 lines (93 loc) · 4.85 KB

Compiling and installing DFTB+

Requirements

In order to install DFTB+, you need the following software components:

  • A Fortran 2003 compliant compiler
  • A C-compiler
  • GNU make (version >= 3.79.1)
  • LAPACK/BLAS libraries (or compatible equivalents)
  • Optionally: ScaLAPACK (version 2.0 or later) and an MPI aware Fortran compiler, if you want to build the MPI-parallelised version of the code
  • Optionally: M4 preprocessor, if you want to build the MPI-parallelised version of the code
  • Optionally: the ARPACK or ARPACK-ng library for excited state DFTB functionality; the DftD3 dispersion library (if you need this dispersion model).

In order to execute the tests and compare them against precalculated results, you will additionally need:

  • Python (version >= 2.6) with NumPy

Obtaining the source

The source code of the last stable release can be downloaded from the DFTB+ homepage.

Alternatively you can clone the public git repository. The tagged revisions correspond to stable releases, while the master branch contains the latest development version. As the project uses git-submodules, those must be additionally downloaded

git clone https://github.com/dftbplus/dftbplus.git
cd dftbplus
git submodule update --init --recursive

Some optinal software components are not distributed with the DFTB+ source code, but can be included during the DFTB+ compilation if they are not installed on your system. We recommend you download these components by using the get_opt_externals utility, e.g.:

./utils/get_opt_externals

This will download all license compatible optional external components.

If you wish also to download and integrate some optional components with conflicting licenses (e.g. the DftD3 library), you must explicitly specify it:

./utils/get_opt_externals ALL

Note: if you include components with conflicting licenses into your build, you are only allowed to use the resulting binary for your personal research, and are not permitted to distribute it.

For more information see the detailed help for this tool by issuing ./utils/get_opt_externals -h.

Compile

  • Look at the makefiles in the sys/ folder and find the one closest to your system. The suffix of the makefiles indicate the architecture, operating system and the compiler they have been written for. Copy the most suitable makefile to make.arch in the root directory of the source tree, e.g.:

    cp sys/make.x86_64-linux-gnu make.arch

  • Adjust the settings in make.arch according to your system. Note that there are often separate settings in make.arch for compiling with and without MPI. The code is also usually compiled with openMP enabled.

  • Open the file make.config and check the configuration options set there. In this file binary choices are defined as either 0 (false) or 1 (true).

  • Build the binaries by issuing

    make

    in the root directory of your source tree. DFTB+ can be built in parallel, so you may use the -j option of make to specify the number of parallel build processes, e.g.:

    make -j4

    The build takes place in a separate directory _build. You can customise the build directory location within the make.config file (variable BUILDDIR), but the default location is inside the root of the source tree.

  • The code can be compiled with distributed memory parallelism (MPI), but for smaller shared memory machines, you may find that the performance is better when using OpenMP parallelism only and an optimised thread aware BLAS library.

Testing DFTB+

  • After successful compilation, execute the tests with

    make test

    You may also run the tests in parallel (option -j) in order to speed this up. If you use parallel testing, ensure that the number of OpenMP threads is reduced accordingly. As an example, assuming 4 cores in your workstation you could use:

    make -j2 test TEST_OMP_THREADS=2

    for an OpenMP compiled binary.

    If you want to test the MPI enabled binary with more than one MPI-processes, you can set the TEST_MPI_PROCS variable accordingly e.g:

    make test TEST_MPI_PROCS=2

    Testing with hybrid (MPI/OpenMP) parallelisation can be specified by setting both, the TEST_MPI_PROCS and TEST_OMP_THREADS variables, e.g:

    make test TEST_MPI_PROCS=2 TEST_OMP_THREADS=2

    Note that efficient production use of the code in this mode may require process affinity (settings will depend on your specific mpi implementation).

  • The compiled executables can be copied into an installation directory by

    make install

    where the destination directory can be configured in the make.config file (set by the variable INSTALLDIR).