mutrace and tiles3

[Boudewijn Rempt]

On Wed, 16 Sep 2009, Cyrille Berger wrote:

> On Wednesday 16 September 2009, Boudewijn Rempt wrote:
> > I did exactly the same thing as with tiles1. It was much slower -- and the
> > results are certainly puzzling, since the main contention doesn't seem to
> > come from the tile engine!

> I wouldn't be so sure, since 190168 is a mutex used by the event loop, if 
> tiles3 use more signals it can be the explanation. Might be worth to try by 
> disabling the glib loop event and see what happen.
> Use QT_NO_GLIB to disable glib event loop without recompiling Qt:
> export QT_NO_GLIB=1

That was plenty interesting. I tried this during my lunch break and found
that with QT_NO_GLIB=1, there is virtually no lock contentions _at all_
with tiles3.

Boudewijn