Build from source

Requirements

To build Fluidfft from source, ones needs:

  • A modern Python (>=3.9) with Python headers and pip

  • A decent amount of RAM (at least few GB available).

  • A C++ compiler fully compliant with the C++-11 standard (currently not Intel compilers)

Get the source

Fluidfft development uses the revision control software Mercurial with modern Mercurial extensions like Evolve and Topic. The main repository is hosted here in Heptapod.

There are other ways to get the source but we are going here to assume that you can install Mercurial. It can be useful when working with Fluidfft source to fully setup Mercurial with these extensions and learn a bit of Mercurial. Then, the Fluidfft repository can be cloned with

hg clone https://foss.heptapod.net/fluiddyn/fluidsim

Why Mercurial/Heptapod and not simply Git/Github?

We consider that modern Mercurial is really great, even better in some aspects than Git. Moreover, we do not think that it is a good thing that the whole open-source ecosystem depends on Github, a close-source project owned by Microsoft.

Thanks to Octobus and Clever Cloud for providing https://foss.heptapod.net!

Installing from the repository

Simple installation from source

We recommend to create a clean virtual environment, for example with:

cd fluidsim
python3 -m venv .venv
. .venv/bin/activate
pip install pip -U

Then, let us use pip to install the local project:

pip install . -v

Note

-v toggles pip verbose mode so that we see the compilation log and can check that everything goes well.

Moreover, the build (which uses Meson) can be controlled through environment variables (for the C++ compilation) and options. The particular build options for Fluidfft are defined in the file meson.options which contains:

option(
    'transonic-backend',
    type: 'string',
    value: 'pythran,python,numba',
    description:
      'pythran,python,numba (default), cython, numpy, numba; ' +
      'or comma separated value representing multi-backends',
)
option(
    'native',
    type: 'boolean',
    value: false,
    description: 'Performance oriented and not portable build',
)
option(
    'use-xsimd',
    type: 'boolean',
    value: true,
    description: 'Turns on xsimd vectorization',
)
option(
    'pythran-complex-hook',
    type: 'combo',
    choices: ['os-dependent', 'true', 'false'],
    value: 'os-dependent',
    description: 'Pythran complex_hook option',
)

To choose a value different from the default value, one can use this ugly syntax:

pip install . -v --config-settings=setup-args=-Dtransonic-backend=python
# or
pip install . -v -C setup-args=-Dtransonic-backend=python

Let’s decompose this syntax!

There are 3 levels:

  • --config-settings / -C is a pip option to pass configuration to the PEP 517 build backend (for Fluidfft [meson-python]).

  • setup-args is understood by meson-python

  • transonic-backend is a Fluidfft build option. But one needs to add the -D for Meson!

Important

One can activate a performance oriented and not portable build using:

pip install . -v -C setup-args=-Dnative=true

Note

Recent versions of pip allow one to specify different options like so:

pip install . -v \
  -C setup-args=-Dtransonic-backend=python \
  -C setup-args=-Duse-xsimd=false

Of course, one can also change values of other buildin Meson options.

Warning

Meson builds Fluidfft binaries in parallel. This speedups the build process a lot on most computers. However, it can be a very bad idea on computers with not enough memory. One can control the number of processes launched in parallel with:

pip install . -v -C compile-args=-j2

Another example to set the optimization level

The default optimization level is -O3. One can change that with:

pip install . -v -C setup-args=-Doptimization=2

Advanced compilation configuration

The environment variables CC, CXX, CFLAGS, CXXFLAGS and LDFLAGS are honored.

Note that Fluidfft builds are not sensible to the ~/.pythranrc file!

FAQ

  • How to know which compilers and compilation flags are used? How to check if XSIMD was indeed used?

    One can study the file build/cp39/compile_commands.json.

  • How to differentiate a native build from a regular build to produce binaries usable on other computers?

    By default the produced wheels should be portable. There is the native build option to target the exact CPU used for compilation.

  • How to produce a wheel for other architectures (cross-compilation)?

    ???

Setup a full developer environment with editable installation

Let us first present the tools used for Fluidfft development.

  • PDM is a modern Python package and dependency manager,

  • Meson is an open source build system (in particular used by Scipy),

  • Nox is a command-line tool that automates testing in multiple Python environments,

  • Pytest is the most popular testing framework for Python,

  • pip is the official package installer for Python,

  • Pythran is an ahead of time compiler for a subset of the Python language, with a focus on scientific computing,

  • Transonic is a pure Python package to accelerate modern Python-Numpy code with different accelerators (in particular Pythran).

Fluidfft is built with Meson. We use PDM for Fluidfft development. Pytest and Nox are used for testing. We use Pythran through Transonic to accelerate some numerical kernels written in Python.

Standard Python from Python.org

We present here how one can build Fluidfft from source like the main developers and users.

Install PDM

A first step is to install PDM as an external independant application. I (Pierre Augier) usually use pipx for that but there are other methods.

python3 -m pip install pipx
pipx install pdm -U

Installing in editable mode is a bit particular with Meson, since editable installations are incompatible with isolated builds, meaning that all build dependencies have to be installed in the main virtual environment! Fortunatelly, it’s not too difficult with PDM. From the root directory of the repository, just run:

pdm install --no-self

This command creates a virtual environment and installs all build and runtime dependencies. You can then activate this environment and build/install Fluidfft with:

. .venv/bin/activate
pip install -e . -v --no-build-isolation --no-deps

Conda-based Python with conda-forge and Pixi

One can use Pixi to setup a developer environment based on conda-forge and compile from source. From the root directory of Fluidfft repository, just run:

pixi run install-editable

Then, pip is available and previous commands should work.

Advice for developers

Run the tests

You can run some unit tests by running make tests (shortcut for fluidsim-test -v) or make tests_mpi (shortcut for mpirun -np 2 fluidsim-test -v). Alternatively, you can also run pytest from the root directory or from any of the source directories.

About using Pythran to compile functions

When developing with Pythran, it can be useful to have a ~/.pythranrc file, with for example something like (see the dedicated section in Pythran documentation):

[pythran]
complex_hook = True

[compiler]
CXX=clang++
CC=clang

Note however, that Fluidfft build does not take into account this file! Instead there is a build option pythran-complex-hook and one can use environment variables to change the C++ compilation (performed with Meson).

Set the MESONPY_EDITABLE_VERBOSE mode

It can be useful to set this environment variable when using the editable mode.

export MESONPY_EDITABLE_VERBOSE=1

See https://meson-python.readthedocs.io/en/latest/how-to-guides/editable-installs.html#verbose-mode