Author: scott farrell
Date: 05:51:31 01/06/03
Go up one level in this thread
On January 05, 2003 at 17:46:03, Vincent Diepeveen wrote: >On January 05, 2003 at 17:07:41, Bruce Moreland wrote: > >>How much stronger would your program be if move generation and make/unmake took >>zero cycles? > >also in the evaluation in order to calculate mobility and other things >with zero cycles? > >Move generation in diep ==> 0.6% system time >makemove/unmakemove ==> 2.x% system time (doing incremental attacktables > there). > >So i get faster like 3.x% > >no check detection needed. I just check it in the attacktable whether >i am in check. SEE goes a lot easier too if you already got all the >attacks in the attacktable. > >>My program spends 43% of its time on this. That is move generation, >>make/unmake, static exchange, and check detection. > >>So if on some hypothetical processor it got 100K nps, it would now get 175K nps >>if it got everything for free. That seems pretty substantial, but remember that >>this essentially removes the guts of the program and replaces them with nothing. > >single cpu K7 1.6Ghz 70k nps - 80k nps for DIEP. That seems extremely slow. Crafty does about 600Knps on about a PIII-1000. My little chompster under java does about 70-80Knps on a PIII-1000, it is bit board also (dare I mention that?). I am guessing that you are using compiled c/c++ and maybe some assembler. Do you have a heavier eval than crafty, more SEE type evals ? I ahve seen many other amateur program do arounf the 80-100Knps speed, and many commericials are more like crafty, in the 500-1000Knps. Scott > >>If I could get 50% speedup in these areas of my program, say that I get rid of >>22% of the execution time. > >>This program would go from 100K nps to 128K nps. That seems a little less >>important. A 60 second solution would now take 47 seconds. > >>I think that it would be extremely difficult or impossible to make all of this >>stuff go twice as fast, so why even try? >>bruce > >all the loops in the expensive chess code in eval matters a lot. If i do it >with bitboards i get slower 4 times or so. > >Best regards, >Vincent
This page took 0.03 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.