Re: [glob2-devel] To Erik: My Profiling Information

[Erik Søe Sørensen]

Kieran P skrev:

> > Running valgrind today paid off, I think - one significant (i.e. 
> memory-corrupting) bug and one memory leak (0.4MB leaked in a 
> relatively short run) found & fixed :-)
>
> Wow. Very nice job. Was that what you commited?

Thanks :-)
Yes, that's changesets eb0f02742864 and 
<http://hg.globulation2.org/glob2/rev/eb0f02742864>59f2c9ee8128 
<http://hg.globulation2.org/glob2/rev/59f2c9ee8128>.

> > Well, thank you for the effort, but I'm not sure these virtualized 
> timings say much. They won't correspond to running the game on a 
> machine's 'real' OS... the virtualization software influences the 
> measurements.
>
> Very well. I'll run it on my Linux computer (non virtualized) and a 
> gaming system I should point out..................

Of course it is - these days I use it for playing the profiling game :)
(Oh, and that nifty 'Valgrind' game. It even has a boss key.)

> Here you go: http://dl.sv.nongnu.org/releases/glob2/profiling.tar.bz2
>
> Those are the new results. Much different. Perlin and clouds seem to 
> dominate?

Yes, this is interesting!
I haven't used the '-h' switch myself until now.

The busy callpath is
 Game::drawAll -> Dyn.Clouds::compute -> PerlinNoise::Noise -> 
PerlinNoise::Noise3d
Is there any chance that the noise can be computed using integers, or 
cached?
It seems it's all multiplying and adding, so I'm guessing it might.

A visualization of the results are available here:
 http://www.daimi.au.dk/~eriksoe/misc/glob2-prof2.png
or in detailed version (200k, 10000x1200) here:
 http://www.daimi.au.dk/~eriksoe/misc/glob2-prof.png

/Erik