Author: Peter Fendrich
Date: 13:48:00 10/23/02
Go up one level in this thread
On October 23, 2002 at 10:29:27, Gerd Isenberg wrote: >On October 23, 2002 at 03:36:58, Peter Fendrich wrote: > >>On October 22, 2002 at 19:16:42, Gerd Isenberg wrote: >> >>>On October 22, 2002 at 19:01:43, Peter Fendrich wrote: >>> >>>>On October 22, 2002 at 18:27:11, Gerd Isenberg wrote: >>>> >>>>Hi Gerd, >>>>I'm not sure I understand this: >>>> >>>>- snip - >>>>>Due to the >>>>>simultanious feature this algorithms seem a favorably alternative to rotated >>>>>bitboards, even without assembler. >>>>- snip - >>>> >>>>Do you really mean that the c-code in the previous post in this thread is faster >>>>than rotated bitboards? >>>> >>>>Peter >>> >>>Hi Peter, >>> >>>I believe yes on 64-bit processors in conjunktion with using this routines with >>>sets of multiple pieces. >>> >>>I use it currently with mmx-registers in Leiden. >>> >>>http://www.talkchess.com/forums/1/message.html?259095 >>> >>>See also the reply of Steffan using move target set for each direction! >>> >>>Regards, >>>Gerd >> >>Yes, I saw your mmx code but didn't think that c-code could match that. Maybe >>64-bits processors would make a difference. > >Yes, with hammers sixteen general purpuse registers even c-code may be faster >then. > >>BTW your link is outdated! >>Peter > >oups, only valid for one week. If you like, i may send it by mail Yes, please! >- or hopefully it goes to the archives soon, It will take some time IIRC. >specially the replies of Steffan. >His point was that i lose information, if i combine the directions to early into >the final attack set register. His Idea is to hold eight disjoint direction sets >for all sliding pieces, to gain really from generating attack sets for multiple >pieces. These sets anded with ~ownPieces may be usefull as a kind of movelist >entry with implicite bookholding. > >I stated that i use the mmx-stuff now in IsiChess instead of rotated bitboards, >without changing function interfaces (so far). >Due to valid move generation the winner is currently a mmx-routine, that gets >all attacks by all pieces of one (opposite) side. So i get all tabu squares for >the king to move. Instead of calling multiple getAttackedBy for possible king >move (castle) target squares, i call getAllAttacks once. >I made also dedicated mmx-routines for determing pinned pieces and covered >checkers (remove checker?). > >Gerd You, Steffen et al have posted some interesting ideas about this! I'm using rotated bitboards as described by Bob Hyatt (but a somewhat different implementation). I have to be very convinced before I rewrite all that stuff! Peter
This page took 0.01 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.