Author: Reinhard Scharnagl
Date: 12:51:02 02/22/06
Go up one level in this thread
Hi Alessandro, On February 22, 2006 at 15:15:00, Alessandro Scotti wrote: >On February 22, 2006 at 13:00:22, Reinhard Scharnagl wrote: >>how has it been calculated? > >Hi Reinhard, >I'm quite sure Fabio isn't using caching. It's just Delfi's (legal) move >generator that's lightning fast! well, that is really fast! Are the computed moves also fully informed, that is: bearing information on being check threat or mate? Smirf puts a lot of time to support additional move flags either with zero, check threat direction, double check code or mating code. E.g. because of that timing without built statistics hardly could be compared. >>There are of course faster methods. Smirf also has one (with caching). > >I've tried caching too, and if I enable it I usually get the result in half the >required time... not bad! (But I haven't done it for speed, just for testing the >hash table, something you suggested a while ago IIRC.) That is of course the desired purpose! I modified several types of caching and replacement schemes until I got the current lay-out. And you could see, that the caching effect is increasing every level now. But more important is, that the results of the countings are identic. So early hard hash collisions would become obvious, and thus a wrong hash-key generating could be detected in time. >>And do not forget, it costs time to analyse moves and to fill the statistic >>using 64 Bit counters. > >I know... just tried the next depth and got a negative node count! :-) Nevertheless Delfi's generator is remarkable fast. Reinhard.
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.