[maemo-developers] [maemo-developers] Cairo benchmarking

From: Matthew Allum mallum at gmail.com
Date: Fri Oct 28 14:46:14 EEST 2005
Hi;

On 10/28/05, Clemens Eisserer <linuxhippy at gmail.com> wrote:
>
> I wonder why the API should make such a bit difference - I am quite
> sure that cpu/memory throughput is far more a problem. The ARM9 in tne
> 770 is quite a bit, well, turkey ... and to fill a 100x100 shape which
> is quite a bit more complex than a rectangle needs many million
> cylcles.

An FP API will make a difference if your calling it thousands of times
( i.e in an animation ). Of course this is only one factor as I
previously mentioned but believe me FP operations really hurt on
hardfloat ARM - a simple timed test of adding ints vs floats will
prove this.

> You can see this when benchmarking the X11 implementation of the
> device - quite nice but to fill a larger pixmap  with a color takes
> ages of course.
>

I depends on *how* you paint that pixmap of course. Using a 'core' X
fillrect should be pretty fast. Doing the same thing via cairo is
going to be slower - it offers much more feature wise that 'core' X
cannot do and as not yet fully optimised basic operations may not be
'fast pathed'.

  -- Matthew

More information about the maemo-developers mailing list