Hi, Until very recently, I was using OpenMP-enabled builds of trunk to work on Wesnoth and user-made content. My main focus all this time was performance and not CPU usage, since the latter doesn't usually cause a significant difference with my current laptop and kernel configuration (Linux 2.6.38 can automatically balance load to keep the system responsive); for this reason, I didn't even look at the CPU usage until I stumbled upon it by accident some nights ago.
You can see my surprise in the March 29th #wesnoth-dev IRC log [1] around 09:04:54 when I discovered that Wesnoth was using more than 100% CPU (according to top's process-based output) while idling in the gamemap view of a scenario with no more than 4 units visible at a time, and no complex standing animations. I later restricted my test case to HttT scenario 1 with the view centered on player 1's initial keep (thus, only two units visible), with all kinds of animations disabled in Preferences (unit/standing, unit/idle, animated terrains and halos). In all cases, Wesnoth would idle with one of its four threads (apparently the main one) using around 99% of CPU time according to a thread-based output. I am not sure how to collect further diagnostic information, but if I had to guess I'd say Wesnoth is wasting its time waiting for something during the idle loop. Naturally, this does not affect the default builds with OpenMP disabled. The system in question runs Debian GNU/Linux 6.0.1 x86_64, thus the default compiler is GCC 4.4.5 using libgomp (GCC OpenMP support lib) 4.4.5. [1]: http://wesnoth.org/irclogs/2011/03/%23wesnoth-dev.2011-03-29.log -- Regards Ignacio Riquelme Morelle <shadowmaster>
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ Wesnoth-dev mailing list Wesnoth-dev@gna.org https://mail.gna.org/listinfo/wesnoth-dev