Question about simple profiling implementation

[Andrew de Quincey]

On Tuesday 02 December 2003 04:13, Alexandre Julliard wrote:
> Andrew de Quincey <adq_dvb at lidskialf.net> writes:
> > However, if no one minds, I think I'll still implement the stuff I was
> > doing. I found being able to examine the call tree with ballpark figures
> > of how long was spent in each call was very invaluable.
>
> Note that the relay debugging adds a huge overhead, especially for
> functions that call other parts of Wine, so adding precise timings in
> there is pretty much useless. That may also explain why you get such
> strange results.

I've been having a look into oprofile, and it looks very useful. However, its 
not really suited to what I'm trying to do right now... its more useful for 
identifying performance problems on algorithms that can be run for a longer 
time.

At the moment I want to do a single GUI operation in an application, then view 
the call tree, annotated with some indication of how long each call took. Its 
not intended to be a very accurate profile; its simply meant to help identify 
which functions merit a closer look.

As you say, relay debugging adds a huge overhead... however, would you say 
that this overhead would be fairly constant for each particular function? 

I'm thinking of not outputting the raw values as they are quite misleading; 
instead percentages would probably be better.

What do you think?