To build xtb from the source in this repository the meson build system can be used.
For a decent Fortran support verson 0.51 of meson or newer is required to build xtb,
additionally the default backend ninja is required with version 1.7 or newer.
To install the meson build system first check your package manager for an up-to-date meson version,
usually this will also install ninja as dependency.
Alternatively you can install the latest version of meson and ninja with pip (or pip3 depending on your system):
pip install meson ninja [--user]If you prefer conda as a package manage you can install meson and ninja from the conda-forge channel.
Make sure to select the conda-forge channel for searching packages.
conda config --add channels conda-forge
conda install meson ninjaThe recommended build for xtb is with Intel Parallel Studio using the Intel Fortran compiler and the Math Kernel Library as default backend.
Precompiled, statically linked xtb binaries for Linux are provided at release page.
The setup for the linear algebra backend defaults to MKL, therefore, only the compilers have to exported before configuring the build:
export FC=ifort CC=icc
meson setup build --buildtype release --optimization 2After the configuration step the build can be performed with ninja:
ninja -C buildNote, ninja will by default use all the threads available on your system.
If you share the build machine with others it might be helpful to reduce the number of concurrent jobs using the -j flag.
xtb can also be compiled with GCC version 8 or later.
For this example we additonally choose to change the linear algebra backend to OpenBLAS, if you have Intel Parallel Studio installed, you can leave out the last argument to get the MKL backend.
export FC=gfortran CC=gcc
meson setup build --buildtype release --optimization 2 -Dlapack=openblasTo compile and run xtb with GCC 7 (especially 7.3.0) the source code has to be patched at several placed, for appropriate patches visit the conda-forge feedstock repository.
The build system will check if the OpenBLAS library provides LAPACK features as well, if this is not the case it will additionally search for LAPACK.
If you are compiling xtb on Darwin platforms, ensure that GCC is the actual GCC and not clang.
The build can be performed just like before:
ninja -C buildAfter successfully building the xtb program ensure that it is working as expected.
Run the testsuite with
ninja -C build testAll tests should pass, otherwise open an issue.
To use xtb in production or to pack a release with precompiled binaries the project should be installed with ninja.
The installation prefix defaults to /usr/local on Linux systems, you might want to adjust this first by configuring your build with
meson configure build --prefix $HOME/.localTo perform the actual installation run
ninja -C build installDepending on the installation prefix and your user rights ninja might ask for the root access to perform the installation.
The installation uses the default install directories of the meson build system To modify the behaviour configure meson with
meson build \
--prefix=/usr \
--bindir=bin \
--libdir=lib \
--includedir=include \
--datadir=share \
--mandir=manThis will result in a directory structure like
usr
├── bin
│ └── xtb
├── include
│ └── xtb.h
├── lib
│ ├── libxtb.a
│ ├── libxtb.so -> libxtb.so.6
│ ├── libxtb.so.6 -> libxtb.so.6.3.2
│ ├── libxtb.so.6.3.2
│ └── pkgconfig
│ └── xtb.pc
└── share
├── man
│ ├── man1
│ │ └── xtb.1
│ └── man7
│ └── xcontrol.7
├── modules
│ └── modulefiles
│ └── xtb
│ └── 6.3.2
└── xtb
├── param_gfn0-xtb.txt
├── param_gfn1-xtb.txt
├── param_gfn2-xtb.txt
└── param_ipea-xtb.txtBy default the compiler optimization should only use -O2, although the program can be safely compiled with -O3 when using Intel compilers.
It is known that binaries build with GCC on -O3 will produce wrong results for certain systems.
To adjust the optimization level to -O3 use
meson configure build --optimization=3When configuring the build the wanted linear algebra backend can be specified with
meson setup build -Dlapack=<option>The following options are available:
backend |
linked against |
mkl-static |
static MKL (default) |
mkl |
shared MKL |
mkl-rt |
MKL real time library |
openblas |
OpenBLAS and if required LAPACK |
netlib |
BLAS and LAPACK |
custom |
|
If you are using the MKL provided by conda-forge you have to link against the netlib backend
To add new source code it should be sufficient to add the new files in the meson.build file in the respective source directory in the src tree.
New directories in the src tree can be included by adding them as subdir to the parent meson.build file.
To kinds of source files are distinguished, all sources for the xtb library are included in the srcs variable, while files used exclusively for the executable are stored in the prog list.
Source files for the testing are defined in the TESTSUITE/meson.build file together with their tests.