Author: Carlos del Cacho
Date: 15:58:07 07/02/01
Go up one level in this thread
On July 02, 2001 at 00:44:30, Robert Hyatt wrote: >On July 01, 2001 at 15:05:44, Carlos del Cacho wrote: > >>On July 01, 2001 at 13:21:04, Uri Blass wrote: >> >>>On July 01, 2001 at 12:58:40, Bruce Moreland wrote: >>> >>>>On July 01, 2001 at 06:44:23, Uri Blass wrote: >>>> >>>>>On July 01, 2001 at 05:09:45, stefan wrote: >>>>> >>>>>> >>>>>>see also >>>>>> >>>>>>http://members.tripod.com/~RyanMack/hypertech.htm >>>>> >>>>>If it is truth than it seems that we are going to see a progress of more than >>>>>200 elo in comp-comp games only because of better software for the PIII >>>>>hardware. >>>>> >>>>>I have not enough knowledge to understand if he is right >>>>> >>>>>Uri >>>> >>>>If the move generator in my own program took zero time it would increase in Elo >>>>points by maybe 20 or 30, and that's probably high. >>>> >>>>bruce >>> >>>You are right that only move generation is not enough but the point is that I >>>understand that the data structure helps to do everything faster. >>> >>>He suggests in the last 3 lines when you click on the link that the program can >>>see 10,000,000 nodes per second with the evaluation function >> >>I can't see how you can extract useful attack information from a reversed >>bitboard (without being costly) so the evaluation function is going to be quite >>stupid or it's going to rely on preprocessing. >> >>The method he describes is faster for move generation than rotated bitboards for >>sure, but I don't think it's going to be faster than mailbox or 0x88 (just my >>opinion). I may be wrong but we won't see an Hyperbola engine going to 10 Mnps >>this year, he'll probably have to wait for a long time before release :-) >> >>Carlos >>\\\ > > > > > >I didn't see any evidence that it is faster. The "compact attacks" way it is >done in crafty is very cache-friendly. And it produces the attack bitmaps >on a complete rank or file, in one operation. The approach given takes 4 >steps for rooks/bishops, and 8 for queens. I was talking for myslef. It sounds as if it were faster than what I'm doing (I followed an excellent introduction to bitboards by James Swafford). I'll take a look at your compact attacks scheme, though I think I didn't got to understand how it worked when browsing through Crafty's code. Maybe you could enlighten my narrow mind on the subject. Anyway there are dubious claims in that document, I don't see how you can make a move generator 30 times faster than crafty's, so I doubt if he ever got to implement it, and if he did what was the HW difference he used to measure it. This person sounds like a day dreamer to me. >I don't believe it is faster at all. It avoids updating two bitboards which >means that the table lookup move generation approach won't work. That does >save time. But how does this approach gain over the direct simplicity of real >rotated bitmaps when actually generating moves? > >I think someone wrote this description without knowing exactly what I am >doing, using Crafty for my example... > > > >>> >>>If you rememeber that nodes is only legal move because he talked about legal >>>move generator then the result is more impressive. >>> >>>We need to wait and see if he is right. >>> >>>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.