Re: [pythran] Pythran 0.7.4 is out!

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

Re: [pythran] Pythran 0.7.4 is out!

Neal Becker
How do I build pythran3?

On Thu, Jan 7, 2016 at 2:51 PM, serge guelton <[hidden email]> wrote:
(sorry for the double posting, if any)

Dear pythraners and pythonists,

The pythran team (a great total of 2 active developers) is delighted to
announce the release of Pythran 0.7.4, available on the traditional

    - pypi:
    - conda:
    - github:

As usual, here is a (new) code sample, once again adapted from a
stackoverflow question[0] that showcases pythran capability:

    #pythran export check_mask(bool[][], bool[])
    #  ^~~~~~~ non intrusive top-level annotation

    import numpy as np
    #      ^~~~~~ numpy support (partial)
    def check_mask(db, out, mask=[1, 0, 1]):
        for idx, line in enumerate(db):
            target, vector = line[0], line[1:]
            # ^~~~~ type destructuring, array view
            if (mask == np.bitwise_and(mask, vector)).all():
            # ^~~~~~~ optimization of high level construct
                if target == 1:
                    out[idx] = 1
        return out

Compiled with:

    % pythran

And benchmarked with:

    % python -m timeit -s 'n=10e3 ; import numpy as np;db  = np.array(np.random.randint(2, size=(n, 4)), dtype=bool); out = np.zeros(int(n),dtype=bool); from eq import check_mask' 'check_mask(db, out)'

On average, the CPython version runs in 137 msec while the pythran version run in 450us on my laptop :-)

Here is an extract of the changelog:

2016-01-05 Serge Guelton <[hidden email]>

        * IPython's magic for pythran now supports extra compile flags

        * Pythran's C++ output is compatible with Python3 and pythran3 can compile it!

        * More syntax checks (and less template traceback)

        * Improved UI (multiline pythran exports, better

        * Pythonic leaning / bugfixing (this tends to be a permanent item)

        * More generic support for numpy's dtype

        * Simpler install (no more boost.python deps, nor nt2 configuration)

        * Faster compilation (no more boost.python deps, smarter pass manager)

        * Better testing (gcc + clang)

Again, thanks a lot to Pierrick for his continuous top-quality work, and
to the OpenDreamKit[1] project that funded (most of) the recent developments!

Special thanks to @hainm, @nbecker, @pkoch, @fsteinmetz, @Suor for their
feedbacks. *You* give us the motivation to go on!


Those who don't understand recursion are doomed to repeat it

SciPy-User mailing list
[hidden email]