Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: More info. on the VC++ profiler

Author: Peter Fendrich

Date: 03:29:06 02/02/00

Go up one level in this thread


On February 01, 2000 at 20:24:59, Tom Kerrigan wrote:

>Many people have given me tips about using the VC++ profiler, and this afternoon
>I actually got it to work!
>
>If other people are brave enough to try this, here are the two main tricks:
>1. Modify your program so it doesn't use stdin/stdout

I've never had any problems with this.

>2. Set the program's working directory in Project/Settings
>
>Here are my comments.
>
>First, when profiling, my program ran __72 times__ slower than normal. I was
>only doing function timing! I think this is absurd.
>
>Second, I did two consecutive profiling runs, and the resulting data was TOTALLY
>DIFFERENT. This isn't an issue of one function being a little higher than
>another. Example: in the first run, my move generator wasn't even in the top ten
>functions. In the second one, it was the most expensive function by a long shot!

This is exactly what I've expeienced to. The method used by Profile is to
interupt now and then doing its meassurements. I think that a chess program is
quite cyclic in its nature and it is a big risk for biased Profile results
because of this. If you run the tests for a longer time I think the results will
be somewhat better but not much.

>None of the other data matched either. And I was running my program for more
>than 2 minutes per run!
>
>So, the upshot is that I would prefer hitting my toes with a hammer to using
>this stupid profiler POS again.
>
>-Tom



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.