Performance change from X in Fedora Core 4 to Fedora Core 5

[Matthew Allum]

Hi;

On 7/9/06, Keith Packard <keithp at keithp.com> wrote:
> On Sun, 2006-07-09 at 16:27 +0100, Ross Burton wrote:
>
> > I should point out that at GUADEC we have people with benchmarks proving
> > that GTK+ 2.8 is slower than 2.6, the GTK+ developers, and the Cairo
> > developers in the same room.  At the moment profiling is being done,
> > Cairo is being optimised, and any non-essential and proven slow use of
> > Cairo in GTK is being removed.
>
> While cairo is certainly in need of performance improvements, the usage
> of cairo by Gtk+ 2.8 is very limited, to the point where cairo
> performance should have very little impact on Gtk+ application
> performance.
>
> As far as I understand it, the only Gtk+ widget using cairo is the color
> selector,

In varying degrees widgets that use cairo are;

        GtkCalendar
        GtkCellRenderer
        GtkCellRendererPixbuf
        GtkCellRendererProgress
        GtkCellRendererText
        GtkCellView
        GtkColorSel
        GtkDnD
        GtkEntry
        GtkHRuler
        GtkHSV
        GtkIconView
        GtkTextView
        GtkTreeView
        GtkVRuler

And the Default theme engine has;

        draw_arrow
        gtk_default_draw_check
        gtk_default_draw_option
        gtk_default_draw_focus
        gtk_default_draw_expander

Making cairo calls ( these can be avoided easily by using a non-cairo
theme engine. )

> and the only other significant usage is pango, which hits
> essentially the same Render paths through cairo as it did through Xft.
>

This seems to perform about the same on x86 ( if a little better ) but
is a lot slower on ARM ( a lot fo FP calls in play ? ).

> We need to profile Gtk+ 2.6 and Gtk+ 2.8 environments before attempting
> to diagnose performance problems; uninformed speculation is not helpful.
>

Agree. Working on some stuff here. Are you on GNOME perfromance list ?

  -- Matthew