[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] remaining patches

From: bump
Subject: Re: [gnugo-devel] remaining patches
Date: Sun, 22 Dec 2002 08:03:32 -0800

Nando wrote:

> The motivation behind almost all the tunings I submitted
> (specially the latest one) was the performance, just because it
> seems to me that it's currently an important issue. From its
> definition only, it is impossible to tell exactly if a pattern
> is bad and/or expensive. For instance, VA44 looks like a very
> good pattern, a very logical move if the optics code missed it.
> But actually, it's success rate is under 3% and it costs about
> 10% owl nodes. I'm currently trying to find a better contraint to
> reduce this negative performance impact.

Maybe you explained this and I missed it but how do you
measure success rate?

> I think submitted owl tunings should always be documented with
> full regression results, i.e. documented breakage and performance
> impact. Maybe I missed it on the list, but I don't recall seeing
> such kind of documentation for evan_3_13.8

I think as we approach 3.4 it is very important to have
a policy like this in place. Before 3.2 as I recall we
did almost *no* owl tuning in the weeks immediately
preceeding the release though we continued to tune


reply via email to

[Prev in Thread] Current Thread [Next in Thread]