[devel] Оптимизированные для i686 и выше библиотеки.

[Alexey Tourbin]

On Fri, Sep 03, 2010 at 12:16:49AM +0400, Konstantin Pavlov wrote:
> On Fri, Sep 03, 2010 at 12:07:31AM +0400, Alexey Tourbin wrote:
> > Hey, was it really even a single percent?  Didn't you, like, cheat?
> > Assuming that you didn't, have you ever heard of confidence intervals or
> > beta-errors?
> Результаты я получил не одним измерением и они статистически значимы.

There are major pitfalls.  Statistics is hard.  For example, you might
remember, recently I posted "quick survey" of hash functions.  My point
was that djb performed a bit better than jenkins.  I've made another
attempt just recently.  You know what.  Jenkins handle sparse sets a lot
better than djb.  It's only about very dense sets that djb might be a bit
advantageous.

So, things might grow complicated, really.  Of course, that's not your
fault.  And I'm not trying to make them complicated, either.  That's
just what things really are.-(

Anyway, when someone tells you he's got one percent an improvement.
The right thing to do is to ask, are you a major in statistical physics?