Author: Uri Blass
Date: 15:48:52 06/04/02
Go up one level in this thread
On June 04, 2002 at 18:29:08, Robert Hyatt wrote: >On June 04, 2002 at 18:20:53, Gian-Carlo Pascutto wrote: > >>Odd, I get 39 secs for Crafty, 23.5 for Yace, which is a much bigger difference. >> >>(Crafty 18.12 vs Yace 0.99.68b) >> >>Dieter did some optimisations here or so? >> >>-- >>GCP > > >No idea. I didn't try to make this "fast" at all. It was designed to >be a debugging tool for the move generator (for me), not as a performance >test. Although it has been used for that. It is reasonable to compare >perft for crafty version X on machine Y vs perft for crafty version X >on machine Z, to compare the two machines. Compareing two different >programs' perft times doesn't mean a lot unless both programs have been >optimized to make it efficient. I simply wrote it to be simple. IE I >would turn off a _lot_ of Make/Unmake overhead to simply run perft as fast >as possible... I tried to make it fast in the first place but after I got something that is faster than everything that I know I decided to stop to work on improving the speed of it inspite of the fact that I could continue to work on making it faster. I will have to look at some older version of movei without hash table in order to find the fastest version that I have today that calculates perft. There is no reason to update the zobrist key after every move in order to calculate perft but movei today does it as part of the makemove function. There are also other calculations that are used today in the makemove or in generating moves and are not needed(for example I do some calculation for better order of moves in the move generator and I update some variables in my makemove). Movei is using pure C and no assembler so I am sure that it's speed can be also improved by assembler tricks. 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.