Author: Uri Blass
Date: 02:54:04 12/19/03
Go up one level in this thread
On December 19, 2003 at 03:32:28, Albert Bertilsson wrote: >I just forgot to mention one very important thing... > >Hashing is only used for a single perft(8) calculation, for the next problem the >hash is cleared, so collisions can only occur within the same perft(8) >calculation which is crazy unlikly with 128 bit keys (even very unlikely with >64bit keys). I can add that it is probably possible to use 128 bits by hashing only changes in the board in this case. There can be 64 bits for the squares that information was changed and 4 bits to define the type of change. If not more than 16 squares were changed from the intial position then 128 bits is enough and I guess that in most cases not more than 16 squares were changes. I could be sure in it in case that every move change only 2 squares but there are moves that change more than it enpassent capture and castling. Innspite of it for perft 8 cacluation the only relevant positions are positions that happen 7 plies after the root position so more than one castle need to take place in order to get impossible to hash position and I guess that these positions are relatively rare. Big part of the changes in the board are changes to empty squares and I believe that it is also possible to take advantage of it but I did not formulate a way to do it. 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.