Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Perft 5,6 {Fastest program is List}

Author: Sune Fischer

Date: 14:39:33 06/05/02

Go up one level in this thread


On June 05, 2002 at 11:00:30, Uri Blass wrote:

>The point is that people may use the fast perft to debug their program by
>comparing the result of the perft that is based on their move generator with the
>result of the fast perft.
>
>Doing it with a function that calculates perft slower can take more time.

You only need the perft numbers once for every testposition, so speed is
irrelevant.
I would probably not even trust my own numbers, I use Sjeng and Beowulf to get
the "right" numbers (if they match), then I test if I get the same.

To improve speed I would have to take a lot of stuff out of my movegen (like the
incremental key generation and evaluation), but I want perft to be a permanent
function in the program so that is not an option.
The alternative is to make seperate makemove/unmakemove functions, but then
perft is no longer usable as a debugging tool, which is its main/only(!)
purpose.

I remember we had this discussion before, we tried to start a small competition,
but nobody wanted to strip their program naked just to run a fast perft.
Is it possible you are now part of that group? ;)

>Uri
-S.



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.