[eigen] Architecture specific performance optimizations

[Christoph Hertzberg <chtz@xxxxxxxxxxxxxxxxxxxxxxxx>]

Hi all,

Gael initiated a discussion on this bugzilla-entry, which I think 
deserves a more general discussion on the list:
http://eigen.tuxfamily.org/bz/show_bug.cgi?id=721#c4
> The problem is that change kills the performance on older architectures for
> which pset1 is very slow. I'm also unsure about ARM/Neon.  So this trigger a
> more general question about the policy we should adopt regarding optimizations:
>
> a) shall we limit the number of variants and favor modern architectures?
>
> b) shall we keep multiple variants and detect the best one at compile time
> through compiler's preprocessors or user defined preprocessor token?

IMO, on the long run, we definitely should focus on (mainstream) 
up-to-date hardware, but still at least keep compatibility to out-dated 
hardware (which we currently do when disabling vectorization entirely).
Also we should ensure to not make too big performance regressions on the 
not-so-cutting-edge hardware.

I think as a first step, we should have some kind of survey on what 
platforms Eigen currently is used, who requires to have top-performance 
(I would assume this correlates with people having up-to-date hardware).
Maybe a more important issue is to find people willing to run 
performance tests on a regular basis (especially for both out-dated and 
cutting-edge hardware).


Christoph






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

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