Author: Paul J. Messmer
Date: 14:11:12 08/05/99
Go up one level in this thread
Okay, time to de-lurk I guess. I've been working on a chess engine with the idea of seeing how fast one can make it, and utilizing x86 assembly (MMX technology in particular) where possible. Not only does the MSVC profiler slow you down drastically, but I have found (at least on my machine) that it does not produce correct results. Perhaps for some things it is great, but for the kind of data I was look for it was crap. I would suggest trying Intel's VTune product; it is very useable, easy to get started, and gives good (and correct) profiling information. The way it samples does not slow down your program; I do think it requires a real Intel processor though. It's a little pricy but there is a 14 day eval you can download for free from Intel. I don't work for Intel, BTW... :) Check the eval out... --- Paul J. Messmer http://www.spootopia.com messmer@netcom.com On August 05, 1999 at 16:01:28, Ren Wu wrote: >On August 05, 1999 at 15:01:40, James Robertson wrote: > >>I finally figured out how to make the profiler work.... unfortunately, every >>time I profile my program, I get different results. For instance, with three >>identical profiles, my SEE code took 30.5%, 10.8%, and 18.8%, respectively. My >>UnMakeMove routine took 2.6%, 14.4%, and 0.0% with the three profiles. > >You need many positions, and run many times. This is life. > >>Does anyone else have this problem? There is no hard disk swapping during the >>profiles, and I let my program think for 60 seconds after which it exits >>automatically. > >60 seconds is way too short, MS's profiler slow down your program by maybe 50x, >so you need run it with 50x more time, or 3000 seconds. > >You can try to run it over night on a set of positions. > >Ren.
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.