From Sources
============

From-source installation is directed at those, who wish to carry out :sup:`13`\ C-MFA on complex models and heavy
machinery or develop new features associated with labeling simulation. This approach is arguably the most difficult, but
allows to get every ounce of performance out of ``13CFLUX``. Only installing the C++ backend itself is only recommended
for benchmarking and development, since it does not offer an higher-level function such optimization or inference.

As prerequisite, you need to install

- C++17 capable compiler (supported: *gcc*, *clang*)
- *CMake* (>= 3.18)
- *Boost*: log, date time, unit test framework (>= 1.65, build with -fPIC)
- *FluxML* (with --enable-pic to build the Python package)
- *SUNDIALS* (>= 6.6.0, < 7.0.0)
- *OpenMP* (>= 3.0, optional)
- *Python dev tools* (>= 3.9, only required to build the Python package)
- *ipopt* (>= 3.11, only required to build the Python package)

Python package
--------------

The most difficult task is to install ``boost`` with position-independent code. Unfortunately, this is a requirement by
``pybind``. For more information, consider `this StackOverflow thread <https://stackoverflow.com/questions/27848105/how-to-force-compilation-of-boost-to-use-fpic>`_.
Having installed the prerequisites, a simple

.. code:: bash

   pip install .

in the source directory should suffice.


Only C++ backend
----------------

Create a build directory for ``cmake`` in the top source directory:

.. code:: bash

   mkdir build && cd build

In the build directory run

.. code:: bash

   cmake ..

to configure *CMake*. You can supply this call with ``-DBUILD_TESTS=ON`` to enable testing or ``-DBUILD_NATIVE=ON`` to
optimize the binary to the compilation machine.

After that you can build *13CFLUX* with

.. code:: bash

   sudo make install