[maemo-developers] [maemo-developers] GtkPerf data for GTK+ 2.6 vs GTK+ 2.10 vs Maemo-GTK+ 2.6 (on a Nokia 770)
From: Kalle Vahlman kalle.vahlman at gmail.comDate: Fri Jul 21 00:26:29 EEST 2006
- Previous message: [maemo-developers] Missing libgmodule-2.0.la
- Next message: [maemo-developers] [maemo-users] [Fwd: PERSONAL TECHNOLOGY: Nokia's Tiny Computer Is Crisp but Slow]
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
So while the GtkPerf benchmarking may or may not give accurate data on actual user experience, I went ahead and ran the thing at 1000 iterations on the fabeled ARM platform we all know (to miss a FPU unit) and love (to hate for that). The results are somewhat expected, with a couple of funny things. First the setups (relevant libs with their versions, everything else from the Maemo 2.0 developer rootfs): 1. GTK+ 2.6 atk-1.9.0 glib-2.6.6 gtk+-2.6.10 pango-1.8.2 2. GTK+ 2.10 atk-1.10.3 cairo-1.2.0 glib-2.12.0 gtk+-2.10.0 pango-1.13.3 3. Maemo-GTK+ 2.6 atk-1.10.3-0osso1 glib-2.8.6-1osso5 gtk+-2.6.10-1.osso2 pango-1.8.1-1osso10 I ran GtkPerf 0.4 first a couple of times with low iteration count for each setup to get the libs to memory[1] and then ran it for 1000 iterations (should be enough for resolution, at least it took ages to finish :). All tests used the default builtin theme of GTK+. Attached the gnumeric sheet with data and a png of the graph that shows the relative performance of the setups (GTK+ 2.6 being the one compared to). There are some peculiar results there, of which I have nothing to really say yet as I haven't checked the GtkPerf code or have more detailed data on the hotspots (other than pango rendering perhaps, but that's already being discussed right?). But what the heck was wrong with the ComboBoxEntry in 2.6?-) Also noteworthy is that even though the Maemo team has worked hard on the performance (and succeeding overall), there is some regressions there too when compared to regular 2.6. [1] I had shut down some services to free up memory, but it's possible that memory exhaustion could have kicked in in some of the tests that use a lot of it (the textview test). After all, I still had the maemo-gtk and other unrelated libs in memory all the time for practical reasons... -- Kalle Vahlman, zuh at iki.fi Powered by http://movial.fi Interesting stuff at http://syslog.movial.fi -------------- next part -------------- A non-text attachment was scrubbed... Name: GTK+ performance comparison.png Type: image/png Size: 27903 bytes Desc: not available Url : http://lists.maemo.org/pipermail/maemo-developers/attachments/20060721/40c65ce8/attachment.png -------------- next part -------------- A non-text attachment was scrubbed... Name: GTK+ performance comparison.gnumeric Type: application/octet-stream Size: 3865 bytes Desc: not available Url : http://lists.maemo.org/pipermail/maemo-developers/attachments/20060721/40c65ce8/attachment.obj
- Previous message: [maemo-developers] Missing libgmodule-2.0.la
- Next message: [maemo-developers] [maemo-users] [Fwd: PERSONAL TECHNOLOGY: Nokia's Tiny Computer Is Crisp but Slow]
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]