Author: Ernst A. Heinz
Date: 13:11:44 12/03/97
Go up one level in this thread
On December 03, 1997 at 15:54:41, Don Dailey wrote: > >[...] The only possible problem with Cilkchess is the extra state >I will have to carry around. We basically encapsulate the position and >supporting data structures for searching, in a structure we can neatly >pass up the search tree and across processors. This turned out to be >much more efficient than I thought it would be and is quite workable >even >for a serial program. We never unmake() a move and do not have a >"master" >board. We do try to minimize the total amount of state, the rotated bit >boards will add somewhat to this state but it may be worth it. Yes, copying is efficient on the Alpha -- in "DarkThought" we can switch between a global board with make()/unmake() move and copying with make() move only. With board data structures <= 256 bytes both settings perform about equal on the Alpha but *not* on most x86 CPUs where a global board with make()/unmake() is much faster. With our brand new board data structure that comprises only 128 bytes copying may well be faster on the Alpha ... > One possibility now that we have the superior bitmap data structure is > to measure mobility by counting squares not attacked by certain pieces, > probably just pawns. This is exactly how "DarkThought" has always done it and we are quite satisfied with it. =Ernst=
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.