wxWidget ( wxGtk) profiling
truecool
I am trying for the profiling of wxGtk. I tried enabling profiling
while configure of wxGtk but when I run gprof for minimal sample,
given in source code , it give all time data as 0.0.( I also build
sample application with -pg option) . I am not very good in Linux.
Could anyone please suggest some good profiling tool or existing set-
up which can help me in improving performance of wxGtk( wxWidgets) or
any related documentation or correct approach .
On comparing sample window creation time of wxGtk with gtk
applications , there is good time difference ( 3 times) which I want
to improve some how.
Regards,
Gaurav Verma
Vadim Zeitlin
t> I am trying for the profiling of wxGtk. I tried enabling profiling
t> while configure of wxGtk but when I run gprof for minimal sample,
t> given in source code , it give all time data as 0.0.( I also build
t> sample application with -pg option) . I am not very good in Linux.
I don't know what could explain it, I did use gprof with wx in the past
although it was mostly for non-GUI code.
t> Could anyone please suggest some good profiling tool or existing set-
t> up which can help me in improving performance of wxGtk( wxWidgets) or
t> any related documentation or correct approach .
I use kcachegrind for profiling myself. Maybe there are better choices, I
didn't look at the state of Linux profiling tools since many years.
t> On comparing sample window creation time of wxGtk with gtk
t> applications , there is good time difference ( 3 times) which I want
t> to improve some how.
This would be very welcome, of course. Good luck!
VZ
tonda...@post.cz
I succesfully profiled wxWindets application by Oprofile. But I focused to the application not the wxWidgets...
I added flags "-g -pg" to the compiler and linker, to enable debugging symbols and allow profiling on source line accuracy.
I don't know well the "valgrind --tool=callgrind" way, but the raw comparison is that the oprofile slow down the application only little but the valgrind takes ages.
Also you can start/stop oprofile independently on the application (profile only part of the session).
On the other hand oprofile is statistical tool, so accuracy is smaller.
Other experience:
- sometimes is problem to measure inline function, the time is added to calling function
- profile very depends on optimization settings, in my case the "debug" cause a lot of time wasted in STL classes
Have a nice day,
TonyMi