Re: Is GTK+ 3.x 2x slower than GTK+ 2.x?

[Clemens Eisserer <linuxhippy gmail com>]

I found the same slowdown during the GTK-1.2 -> GTK-2 transition, and
it hasn't become a lot faster during the 2.x releases - moore's law
took care of it. I expect the same to happen with GTK3 sooner or later
(wether I like it or not, is a different question).
There simply don't seem a lot of devs arround optimizing code for fun,
whereas paid devs are usually busy implementing feature requests and
repairing behaviour/crash/blocker-bugs.

- Clemens

2011/7/23 Jakub Misak <jmisak centrum cz>:
>
> Is GTK+ 3.x 2x slower than GTK+ 2.x?
>
> Hello,
>
> When I upgraded to GTK+ 3.0, the first thing that struck me was how
> sluggish it was compared to 2.24. The same dialogs in the same
> application appear immediately on the screen in the 2x version, while
> it takes time to paint them on the screen with the 3.x version.
> Browsing through menus or switching between tabs consumed 2x or 3x more
> CPU cycles in the 3.x version. The file open/save dialog is much slower
> (slower to show up, slower to change directories). Or hovering over the
> toolbuttons in 3.x-based Glade consumed 80% CPU cycles and the
> highlighting could barely keep up with the mouse cursor. Application
> startup is a bit slower with 3.x, too.
>
> So I made a (very) simple and stupid benchmark that tests a couple of
> basic things in a loop, to see how the two GTK+ versions compare.
> If you're interested, you can download it here:
>
> http://www.fileupyours.com/files/310686/perftest.tar.gz
>
> (A simple "make" command should build both versions.)
>
> I am testing it in xfwm4 with compositor turned off. If you test it in
> a compositing WM with window effects (open/close), especially with
> test #3 which opens/closes a dialog 50 times, it may measure something
> else than GTK+ speed...
>
> The default GTK+ theme (Raleigh) was used for both 2.24 and 3.0.11.
> Both GTK+ 2.x and 3.x software was already running, so both library
> versions were loaded in memory (so it was probably not an issue for the
> startup speed).
>
> The results from my 3GHz dual CPU desktop machine with the open-source
> radeon driver (I ran each test 7 times and took the best result for each
> test):
>
> Startup, GTK+ 2: 0.11081 s, GTK+ 3: 0.14468 s
> Test 1, GTK+ 2:  4.34362 s, GTK+ 3: 8.83771 s
> Test 2, GTK+ 2:  1.72010 s, GTK+ 3: 2.33266 s
> Test 3, GTK+ 2:  2.27047 s, GTK+ 3: 4.79270 s
>
> It's not just my computer, as I tried it on an HP 625 laptop, again
> using the radeon driver:
>
> Startup, GTK+ 2: 0.09357 s, GTK+ 3: 0.13600 s
> Test 1, GTK+ 2:  3.27795 s, GTK+ 3: 5.20183 s
> Test 2, GTK+ 2:  1.47700 s, GTK+ 3: 1.46733 s
> Test 3, GTK+ 2:  1.35302 s, GTK+ 3: 5.14810 s
>
> It's not the radeon driver either, as I tried it with the proprietary
> fglrx driver, too:
>
> Startup, GTK+ 2: 0.09058 s, GTK+ 3: 0.13770 s
> Test 1, GTK+ 2:  2.42151 s, GTK+ 3: 6.67091 s
> Test 2, GTK+ 2:  1.15294 s, GTK+ 3: 1.44557 s
> Test 3, GTK+ 2:  1.31779 s, GTK+ 3: 4.64865 s
>
> It's probably not the default theme either. First, the default themes
> look very similar in both versions. Second, test 2 does not seem to be
> influenced by the GTK+ version too much. Third, when I use slow fancy
> themes in both GTK+ versions (Victory, Adwaita, New Wave...), I still
> get the same pattern again - even though a fancy theme makes both
> versions (sometimes even 6x) slower, 3.x is still 50-150% slower than
> 2.x (except for test 2, which seems to be mainly influenced by the
> theme itself). And fourth, even a slow fancy GTK+ 2.x theme is faster
> than the plain default 3.x theme.
>
> Finally, it's not my Linux distribution either. I ran the tests in 3
> different distros (Arch Linux, LMDE and Fedora 15, with Xfce in all of
> them), and always got similar results.
>
> So my question is - is there something in GTK+ 3.0 that could possibly
> cause such a dramatic slowdown compared to 2.24?
> _______________________________________________
> gtk-list mailing list
> gtk-list gnome org
> http://mail.gnome.org/mailman/listinfo/gtk-list
>