Author: Robert Hyatt
Date: 21:00:40 02/27/03
Go up one level in this thread
On February 25, 2003 at 00:48:09, Christophe Theron wrote: >On February 24, 2003 at 13:40:59, Andrew Dados wrote: > >>On February 24, 2003 at 11:41:03, Albert Bertilsson wrote: >> >>>Hello! >>> >>>I've done some testing and thinking and I ask the following question: >>>Why have an UnMakeMove or UndoMove function? >>> >>>When testing with Sharper I found that actually copying the board and then >>>calling DoMove on the new board was faster than using DoMove() / UndoMove() on >>>the same board. >>> >>>My Board size is 376 bytes, and I figure that will cause a copy function to >>>require some 94 copy instruction (on a 32 bit machine), this will of course also >>>require some sort of condition of when to stop copying and so, but I figure this >>>must be a much easier task than doing branches and testing if it was a capture >>>etc. and updating piece lists. >>> >>>Is this a unique case for me? Have anybody else tried it? What is your board >>>size in bytes? Maybe this only works for me because my board is small (although >>>I certainly don't think it is small). >>> >>>Regards Albert Bertilsson >> >>Once your program reaches decent depth you'll find out that 376 extra bytes per >>depth will eventually blow up cache and slow things down. Probably somewhere >>around 10-12th ply. >> >>-Andrew- > > > >Certainly not, even if you reach 100 plies. > >Only the LAST plies are going to be accessed the most often, and only these will >reside in the cache. > >When you are at ply depth 100, only the boards of ply depth 97-100 are going to >be copied "often" for example. > >The cache is MRU, don't forget it. > >So overflowing the cache is not a problem with his approach, unless your cache >is very small. > > > > Christophe Also not _all_ caches are set associative, some are direct-mapped, which means that most recently used data will overwrite data used earlier, killing overall performance significantly...
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.