On Thu, Jul 24, 2008 at 10:33 PM, Nicolas Grunbaum <[EMAIL PROTECTED]>wrote:
> I have been profiling Wesnoth with Shark and Instruments on OS 10.5, > the results which still need a bit of mining and statistical analysis > show that Wesnoth spends a majority of its time ( ~ 85% ) in the > graphics and audio routines, a large amount of time blocking in the > main loop ( between ~3% and ~7% ), and lastly about ( ~ 3% ) combined > allocating std::string, copying std::strings and comparing > std::strings in maps. Most other operations weigh in under 0.1%. These > numbers, as I said need refining, and as I only have had time for a > sample size of three runs ( they take quite some time ) they may change. > > Before I do anymore runs, do these numbers line up roughly with > expectations and previous analysis? > first and formost, thanks a lot for the job I wanted to do something like that but real time always comes in the way. If you can post your raw data somewhere that would be great (I only use gprof, I don't know if your tool provides a "raw output" that we could use) I am very interested in the time spent in the graphic routines, since i'm responsible for the high level ones... I have a question for you though : how did you test (what kind of run, did you have an AI, was it network play...) that's very important for us to analyze your result (obviously) and will help us provide with other test cases for further analysis. These results match our "gut feelings" but little has been done so far WRT basic CPU performance, and there is probably a lot to gain there... I really hope to hear some more from you, possibly with some patches submitted to correct the worth offenders... Cheers Boucman
_______________________________________________ Wesnoth-dev mailing list [email protected] https://mail.gna.org/listinfo/wesnoth-dev
