Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: MoveGen Speed

Author: Uri Blass

Date: 17:43:58 03/14/04

Go up one level in this thread


On March 14, 2004 at 20:03:05, Russell Reagan wrote:

>On March 14, 2004 at 19:27:54, macaroni wrote:
>
>>Hi. I have been re-writing my move generation, converting it into 0x88. What are
>>good ways to test movegen speed? I put it on this position:
>>
>>[D]rnbqkbnr/pppppppp/8/8/8/8/8/RNBQKBNR w KQkq - 0 1
>>
>>and had it just regenerate moves over and over again in the same position. It
>>generated 20 million moves per second in this position, is that good, bad, or
>>not really very important?
>>cheers
>>Tor
>
>If you want to test your move generator, you should also test your functions to
>make and undo moves. The reason is that if you generate legal moves, your
>movegen will be a little slower, but you don't have to check if the move was
>illegal after you play it. If you generate pseudolegal moves, then your move
>generator will be faster, but it will be a little slower to make a move, since
>you have to verify that the move was legal some way or another. Therefore, I
>would recommend perft, which tests all three.
>
>I have to give this recommendation a disclaimer though. You should really only
>compare perft numbers with your own program, and with other programs that you
>are sure don't use any "tricks" to speed up perft calculation. Some people use
>attack tables or hash tables or other things to calculate perft faster, and that
>defeats the purpose of testing the speed of your movegen/make/undo functions.


This attack tables in movei were not generated only to speed up perft
calculation and they have other advantages.

Making the perft faster by attack tables is also not simple and my first perft
was very slow because I did not calculate my attack tables incrementally but
calculated them from scratch after every move.

It was signficantly slower than Crafty and I decided at that time that it is
better if I work on improving the speed of my move generator and not continue to
a chess program that may search less than 1000 nodes per second.

Uri



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.