[pythran] Re: Matrix multiplication in Pythran

  • From: Valerio De Carolis <v.decarolis@xxxxxxxx>
  • To: pythran@xxxxxxxxxxxxx
  • Date: Tue, 14 Apr 2015 20:13:33 +0100

On 14/04/15 17:19, Pierrick Brunet wrote:

Hi Valerio,

The dot function is correctly implemented for "a * b" with dim(a) <= 2
and dim(b) <= 2 in the dot branch
(https://github.com/serge-sans-paille/pythran/pull/417)

It will not be integrated in the next Pythran release because it
requires an up to date nt2 version and Pythran is based on the release
version but it will be integrated in the master branch right after the
release.

I hope you succeed to use Pythran,

Hello Pierrick,

thanks for dealing with this. I managed to extensively use the snippet I posted last time. Of course that will introduce a small performance penalty for the deploys that are loading the Python version of the code.

So far a Pythran feature I enjoy a lot is the transparent loading. It gives me flexibility in testing the optimizations without breaking the development flow.

One thing I'm wondering now to which extent Pythran uses the specialized math routines in NT2. I mean compared to what is happening behind the scenes with Numpy and BLAS/LAPACK. Or I guess it is matter of time when support for numpy.linalg will come.

Shameless plug we are about the send some Pythran optimized code under the surface of the North Sea on-board of one of our AUVs. Maybe a small step but still rewarding to avoid rewriting everything from scratch. :D

Cheers,
V.


Other related posts: