f2py-jit

Just-in-time compilation of Fortran code in Python via f2py.

Check out the documentation for full details.

Quick start

Start from a piece of Fortran, say code.f90

subroutine hello()
  print*, "Hello world!"
end subroutine

Compile the code, import it and execute it

from f2py_jit import jit
f90 = jit('code.f90')
f90.hello()

Do the same but from a python string containing the source block

source = """
subroutine hello()
  print*, "Hello world!"
end subroutine
"""
f90 = jit(source)
f90.hello()

If the Fortran source contains multiple subroutines calling each other, f2py will not perform interprocedural optimizations (at least not by default). f2py_jit can inline the source code before compiling it, and you will get a performace boost [This feature is experimental]

f90 = jit('code.f90', inline=True)

Features

• Compilation of Fortran source blocks as Python strings
• Caching of module builds across executions
• Support for Fortran derived types via f90wrap
• Inlining to improve performance (experimental)

Dependencies

• Fortran compiler (ex. gfortran)
• numpy
• meson (for Python >= 3.11)

The package currently supports Python versions from 3.7 to 3.14.

Important: on Debian-based OS, packages pkg-config and python3-dev are required with Python>=3.11. They must be installed at the OS level ex. via apt.

Note: Python versions >= 3.11 will use the meson backend to build Fortran extensions, which has slower build times than distutils (used by default in versions < 3.11).

Installation

From pip

pip install f2py-jit

To install the package with support for derived types (courtesy of f90wrap

pip install f2py-jit[types]

Note that this requires Python >= 3.8.

From source

git clone https://framagit.org/coslo/f2py-jit.git
cd f2py_jit
pip install .

Credits

Part of this code is adapted from numpy.f2py module by Pearu Peterson, in accordance with the NumPy license.

Authors

Daniele Coslovich: https://www.units.it/daniele.coslovich/