Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Problem searching too deep!

Author: Sune Fischer

Date: 03:19:22 05/28/02

Go up one level in this thread


On May 28, 2002 at 05:22:12, Georg v. Zimmermann wrote:

>
>>Oh, remembered what the problem was - the extensions!
>>I also need to put a limit on the number of extended plies.
>>So the margin should be: 1 ply per piece (for the qsearch)
>
>-2 for kings

of course :)

>and added to that the
>>maximum extension depth. Last one is a limit many programs has, not a bad idea
>>either, for searching 60 plies there could be many checks (which is also why I
>>saw one case of depth 71, no way qsearch could go 11 plies with just two pawns).
>
>The problem here is that you set the memory for your arrays at startup. And
>there you have to expect say highest search depth 40+ 30 pieces quiesce + 40
>extensions. I am no low level expert. But loads of arrays[110] must make it
>slower because these are memory regions in between that are almost never used.

This a problem everyone has (AFAIK), there are two way to solving it that I can
think of.
1) put a check into the loops and return if maxply exceeded
2) stop the search at some point and leave a good safety buffer

Neither method is good coding style IMHO.
But I prefer using abit of extra memory (~150 kB) to get that speedup (1 less
"if" per node).

Frankly I'm not worried about being cache inefficient, because I only have 1
global pv array "PV pv[MAXPLIES]", about 150 kB in the end of that array is just
the safety buffer, there is no data I need to access inbetween.

These dayes 150 kB is nothing, I feel I can afford it better than an "if" that's
wasted 99.99% of the time.

Some may dislike it because there is a theoretic posibility of a crash (we all
know Murphy's:), but it's really plague or colera.


Maybe I'll change it later, you never know...:)

>>
>>If the number of estimated pieces is not sat too low, then this should be
>>foolprof, right?
>
>Chess program and foolproof ? oh my :)

Well I can't speak for others ;)

-S.


>Georg



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.