Author: Sune Fischer
Date: 17:22:45 06/27/03
Go up one level in this thread
On June 27, 2003 at 20:04:29, Uri Blass wrote: >>If you are determained to get the fastest possible make move, then that excludes >>the use of attack tables, this will come back to haunt you in the search and the >>eval. > >And this is also counter productive for perft because if you have fast >make move then you cannot use attack tables for faster move generation. > >I believe that this is the reason that Movei was faster than Sharp in >calculating perft. Everything seems counter productive for perfting. Perft has a really bizare profile, hotspots appear where usually there aren't any. Perft is simply painting the wrong picture, IMO. > IMO it's very clear that it makes little sense to optimize a >>makemove+movegen without considering the remaining program. > >I agree but I think that the difference between the things is not big. Yes I would not be surprised if many roads lead to Rome. :) >>If I wanted a fast perft I'd only have to optimize my makemove, if I could move >>the big load into genmove I'd go a lot faster. > > >No >It is better for you not to make the last ply but only generate legal moves. Yes, sure for perft. But I don't see a reason to build a long list of moves at the last ply, where I usually won't be needing them. I think there are more effective ways of extracting any information one might be interested in. > However in a *real* search you >>don't get to make all the moves you generate, so that means the genmoves would >>profile-wise be just as important as makemove, perhaps. > >In real perft I also do not make most of the moves that I generate so >you see that for me it is similiar. > >Uri Yes, but you do what you do _not_ because of an optimization in the makemove+movegen, but because you need this information in the search. I can see why it would be interesting to have the entire list of moves before searching the first move. Such a list could certainly help one to decide when to nullmove and what extensions to use. -S.
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.