[Top][All Lists]

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

Re: [gnugo-devel] how influence influences owl

From: Arend Bayer
Subject: Re: [gnugo-devel] how influence influences owl
Date: Tue, 29 Jan 2002 15:42:48 +0100 (CET)

> > my K6-II 400, I realized that I might actually be so kind to my gcc to tell
> > him about the processor. He returned the favor of adding the flag -m486 in
> > engine/Makefile by producing a huge performance gain, 
> I don't quite understand. What exactly did you do?
> Isn't your AMD processor a 686? How would adding a -m486 flag help in this
> case?
I manually added -m486 to the GCC flags in the makefiles (GNUGO_ALL_WARNINGS/
GNUGO_SOME_WARNINGS=...). Then I recompiled.
According to gcc --version, I have egcs-2.91.66 (on GNU/Linux), which does
not have specific K6-processor type option. According to what I found with
a quick google search, the recommended -m option for gcc+K6-II is then
-m486; the alternatives are '-mpentium' or `-mpentiumpro', which are
apparently designed for the Intel processors and, according to what I read,
not beneficial for the AMD K6-II. Next time I will run regressions without
-m486 again to see the exact performance gain.

If I had set an environment variable via 'export CFLAGS=-m486', then
configure would have added this in all makefiles automatically. (That's
at least what ./configure --help claims.)

I would assume you have optimizing compiler flags before, e.g. for
tournament versions?

> > Among these FAILS are some owl failures which I do not understand. Can
> > someone more owl-knowledgeable explain me how the influence function affects
> > the owl code, and e.g. change the moves tried? That would be very helpful.
> Through the escape values. Look at pattern A229. Whether or not this
> is matched would depend on the influence function. See the function
> compute_owl_escape_values() in owl.c, which calls the influence code.

So maybe I should fall back to standard influence when the escape
influence is computed.


reply via email to

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