Re: [eigen] Specializing max_coeff_visitor for some number types

[ Thread Index | Date Index | More lists.tuxfamily.org/eigen Archives ]


On 18.05.2014 15:44, Marc Glisse wrote:
I changed the CGAL testcase because it is important that only 0 have a
score of 0 (the interval [0,1] should have a non-zero score). It is not
optimal (I would like [0,1]>[0,0] but [1,2]<[1,1]) but good enough (some

Ok, using [0,1] as pivot may be better than [-0,0] because the inverse is [1,inf] as opposed to [-inf,inf]. But I think [-1,1] will be worse than [0,0] in general. But as you said, this would not influence the patch to Eigen.

Would opening a PR in bugzilla help move things forward? Or would it
bury it even deeper? I am not familiar with this project so I don't know
the proper procedure...

Some things about your patch I'm not sure about:
 * Introducing a user-level function "cwiseScore" which I doubt will
   have a purpose aside from pivoting.
 * Would there be a way to avoid the extra coeff-access and abs()
   calculation in case score(Scalar)==abs(Scalar)? I guess the
   performance impact is very minor, but it's avoidable.

I could directly apply a patch of yours (preferably generated using `hg export tip` as described here: http://eigen.tuxfamily.org/index.php?title=Mercurial) -- it will not make much difference if sent via the mailing list, or via bugzilla, or a pull-request via bitbucket.

 * But first I would really like some other dev's opinion on it! *

I also would only apply it to the development branch.

Christoph

--
----------------------------------------------
Dipl.-Inf., Dipl.-Math. Christoph Hertzberg
Cartesium 0.049
Universität Bremen
Enrique-Schmidt-Straße 5
28359 Bremen

Tel: +49 (421) 218-64252
----------------------------------------------



Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/