Author: Christophe Theron
Date: 11:05:36 06/18/02
Go up one level in this thread
On June 17, 2002 at 20:58:12, Pham Hong Nguyen wrote:
>On June 17, 2002 at 12:37:54, Christophe Theron wrote:
>
>>On June 17, 2002 at 00:59:45, Pham Hong Nguyen wrote:
>>
>>>On June 16, 2002 at 12:30:59, Christophe Theron wrote:
>>>
>>>>On June 16, 2002 at 01:21:15, Pham Hong Nguyen wrote:
>>>>
>>>>>>Chess Tiger for Palm and Chess Tiger for PC share the same engine code.
>>>>>
>>>>>I think Chess Tiger is one of good and lucky engines to move into Palm because
>>>>>of nature of your design. Many other engines will find so many difficulties. For
>>>>>example, some have to rewrite their assembly codes or remove them. Cache for
>>>>>handhelds is also different from PC (so many optimizations based on cache will
>>>>>not work). Some advantages of bitboard design will be lose comparing with other
>>>>>designs.
>>>>
>>>>
>>>>Actually the base engine for Chess Tiger for Palm was the PC version 12.0.
>>>
>>>Curious questions: is it bitboard version or not? In general, what full depth it
>>>could reach for fast game (say 10 min/game, in Palm)? Thanks.
>>
>>
>>
>>Chess Tiger does not use bitboards. That's why it runs fine on 16 and 32 bits
>>processors (and will still run fine on 64 bits ones).
>
>Amazing, because I have read somewhere that you have used bitboard for current
>version and say good bye to non-bitboard world ;)
I guess you are mixing me up with Frans Morsch.
I have read somewhere that he was preparing a bitboard version of Fritz, but I
do not know if it is true.
So I don't want to spread false information.
My program makes no use of bitboards and I have no intention to use them.
The data structure I'm using is flexible, efficient enough for me, and portable
from 16 to 128 bits platforms (and to 8 bits ones if needed).
>>In 10mn/game on my Palm it will reach 5-6-7 plies typically.
>>
>>That means that some variations are cut at 3 or 4 plies, and some others are
>>extended to 20 plies (or more).
>>
>>From what I have seen so far with the SSDF tests, Chess Tiger for Palm could be
>>a little bit under 2100 SSDF elo.
>
>Thank you very much for your answers. My engine does not use bitboard either.
>You make me feel much more confident about how strong a non-bitboard engine
>could reach :)
>
>Another curious and important question (for me): do you use attack table?
>Currently, I don't use that table because it is very expensive for non-bitboard
>one and I am still wondering if it should be implemented (I have implemented
>once and then removed it immediately because of slowness). Your advice will help
>me much. Many thanks in advance :)
I do not know what you mean exactly by "attack tables".
If you mean a table computed after every move and that gives the list of
attacker for every piece (or the opposite) then I would advise you to not use
this approach.
I have done (a long time ago) versions of Chess Tiger with a very efficient
attack table update routine. It was incredibly fast in regard to the job it was
doing and worked by updating the tables instead of computing them from scratch
after every move.
And it was unfortunately very unefficient.
In computer chess, don't compute anything if you do not have the immediate need
for it. The reason is that 99% of the time an information computed in advance is
going to be useless because you get a cutoff (or some other event allowing you
to skip a lot of computations) before the computed information is used.
Design your data structures with this in mind.
Attack tables are exactly the kind of apparently good idea that does not work at
all.
(IMO, IIRC, etc, etc...)
Christophe
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.