Pythran 0.9.2 - koailh

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Pythran 0.9.2 - koailh

Serge Guelton
Hi folks, and sorry for the double posting if any.

It's my great pleasure to announce the release of Pythran 0.9.2, codenamed koailh.

Pythran is an ahead of time compiler for scientific kernels written in Python. It is
backward-compatible with Python (no language extension) and can take advantage of
OpenMP annotations and automatic SIMD instruction generation.

It's been more than four months since last release, so there's quite a lot of changes.
The detailed changelog is available online:

    https://pythran.readthedocs.io/en/latest/
   
Notable Changes
===============

In addition to bugfixes and new function support, the most significant changes are the
support of a special notation to describe optional argument in Pythran annotations:

    #pythran export foo(int, int?)

This notation is equivalent to:

    #pythran export foo(int)
    #pythran export foo(int, int)
   
This release also introduces a new dependency on the beniget package that provides use-def
chains construction for Python code:

    https://github.com/serge-sans-paille/beniget

A significant memory leak when converting extended slice from pythran to python has been
spoted and fixed.

In total, a great deal of 43 issues have been closed!

    https://github.com/serge-sans-paille/pythran/issues?utf8=%E2%9C%93&q=is%3Aissue+is%3Aclosed+closed%3A%3E2019-01-20+

Sample Code
===========

As usual, I like to demonstrate Pythran through a small benchmark that used to be unsupported
and now provides a nice x6 speedup (without xsimd) up to (with xsimd, non-strided version) over
Python reference:

    import numpy as np
    #pythran export f_dist(float64[:,:], float64[:,:])
    #pythran export f_dist(float64[::,::], float64[::,::])
    def f_dist(X1, X2):
        return np.sum(np.abs(X1[:, None, :] - X2), axis=-1)

(from https://stackoverflow.com/questions/55854611/efficient-way-of-vectorizing-distance-calculation/)

Download
========

You can retrieve the package through PyPI:

    https://pypi.org/project/pythran/

Download the source on GitHub:

    https://github.com/serge-sans-paille/pythran

Or using the conda package, now available for Windows in addition to Linux and OSX through conda-forge

    https://anaconda.org/conda-forge/pythran

Thanks
======

The following people have contributed to this version, through bug reports or commits, in no
particular order:

- Yann Diorcet
- Neal Becker
- Ashwin Vishnu
- David Menéndez Hurtado
- Jean Laroche
- Anubhab Haldar
- Thierry Dumont
- "keke ge-smile"
- "garanews"
- Said Hadjout
- Rodrigo Iga
- Pierrick Brunet
- Franck Cornevaux-Juignet

Special thoughts to these last two, they reported bugs in 2013 and 2014, and the associated
fixes only happened in 2019 ^^!

Contribute
==========

If you've been brave enough to read this whole mail, the next step is to support Pythran!
You can donate your time through bug reports and fixes, give a small motivation boost through
a gentle email, or just use it and share the joy.

If you need important development in Pythran, you can drive the roadmap! Send an email to the
mailing list (or to my personnal email if you're too shy).

++
Serge


--
https://mail.python.org/mailman/listinfo/python-announce-list

        Support the Python Software Foundation:
        http://www.python.org/psf/donations/