Author: Gerd Isenberg
Date: 00:19:07 10/02/04
Go up one level in this thread
On October 02, 2004 at 02:14:32, Tony Werten wrote: >On October 01, 2004 at 19:41:24, Gerd Isenberg wrote: > >>On September 28, 2004 at 17:49:21, martin fierz wrote: >> >>>On September 27, 2004 at 17:27:45, Gerd Isenberg wrote: >>> >>>>On September 25, 2004 at 06:21:13, martin fierz wrote: >>>> >>Yes, Kogge-Stone and even more dumb7fill do often a lot of "stupid" work. >>But 4 for 1 compensates that a bit. Depending on the design, one may profit from >>the other parallel nature of fill-routines - to work setwise (not to mention >>bitscanless). >> >>That becomes even more interesting if you feed up a set of safe target squares >>of a sliding piece to get a set of progessive mobility - move targets in two >>moves. >> > >I was thinking about that. > >Generate safe squares for a piece, feed the resulting BB back to the generate >squares and if the resulting BB does not differ very much from the first, the >piece is locked up ? Yes, and this is a case, where one profits from disjoint directions as well, even if it is most likely not so performance critical for a hashed evaluation and considering alpha/beta bounds (lazy eval). E.g. for rooks, if you have disjoint vertical and horicontal attacks, it is only neccessary to feed up vertical attack sets to horicontal filler and vice versa. Of course if safe capture targets are involved, one may ignore it or use one additional fill in the same direction as the capture move. I vote for "ignore". Safe captures are handled by (q)search anyway, a too "pessimistic" evaluation of the capturing piece is probably not so important. Gerd > >Tony
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.