Author: Dan Newman
Date: 03:08:31 09/30/99
Go up one level in this thread
On September 29, 1999 at 21:30:17, Robert Hyatt wrote: >OK... you are doing a hybrid scheme... which is worth considering, although >there are other parts of your engine that need analysis. IE I assume you have >some sort of "SEE" evaluator? The question of interest is whether you need to >do more computation when you do the SEE stuff, to offset the lack of lookups >for sliding pieces. > >It is something I will add to my list to do however... Yeah, I'm doing the hybrid thing mainly to avoid the difficulty of working out the rotated bitboard stuff and also in the hopes that (at least on the Intel machines) it might be a little faster to do some of the sliding stuff the "old-fashioned" way. Already the code seems to be a little more compact than my usual. I tended to unroll things in the 0x88 program, but this one seems to lend itself to loops for some reason. Anyway, I'll find out after awhile whether this was a good way to do it or not. I hope I don't get to the last thing on my list and find it's impossible to implement in any reasonable fashion... I'm going to rip the SEE out of my other program and -see- if I can adapt it. I hope I can figure out something to speed it up. (It currently takes about 25% of the cpu time in the 0x88 engine--that is an engine with no appreciable eval yet though...so maybe the 25% isn't too bad.) I think the bitboards will help at least a little. If nothing else, they make it easy to calculate attacks in the proper order. (In my 0x88 engine promotions disturbed the order of the piece lists, so I can now remove a couple of sorts that I was doing. It was pretty ugly.) One thing that I don't do in my SEE is look at x-ray attacks like Crafty's does. I've always wondered if this hurts it much. I think I might find the x-ray stuff easier to do this time around though. -Dan.
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.