Author: Olaf Jenkner
Date: 10:20:26 10/02/02
Go up one level in this thread
>Of course you are right with the better algorithm. It was a kind of expanding my >awareness, after Steffan Westcott teached me about flood fill and Kogge-Stone. >And now and then i love this low level assembler stuff so much, specially with >these fascinating algos, which seem to be made for mmx-registers. > >There are always more important things to do, but it's fun, hobby and sometimes >obsession. > >But i think it is more than a few percent. The main feature of this routines is >IMHO, that they are able to simultaniously generate all attacks for multiple >pieces of one kind and color (even with 128 bit xmm for both colors in >parallel). With the same effort than determing all pieces which attack one >square, it is possible to get a set of all attacked squares by any piece of one >side. > >If you pass the result of this attackGetters (slightly modified by anding with >some mask) to a second call of this routine you get a set of all squares >reachable in two moves... > >There are so many applications, king move (castle) generation, in check move >generation, generating check moves, attacking heavy pieces, sorting moves, >eval... > >One nice side-effect is that there is no need for rotated bitboards, and no need >to access lookup tables. > >Regards, >Gerd Note that a speedup of 300 percent in one part of the program may result in a speedup of 2 percent in general. I know it's fun. By myself I wasted a lot of time in bringing assembler stuff to my program. The speedup was less then 20 percent. It is decreasing with every new processor generation. OJe
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.