Re: Floating point in pango

[Richard Purdie <richard openedhand com>]

On Thu, 2006-07-13 at 18:26 +0300, Jorn Baayen wrote:
> I have uploaded oprofile profiles of the test program I posted earlier
> to:
> 
> http://www.o-hand.com/~jorn/pango-benchmarks/
> 
> There are profiles against both GTK+ 2.6.8 as well as 2.8.18. For both
> pango 1.10.0 and glib 2.10.3 were used.
> 
> Some things that jumped to my eye:
> 
>  o Using GTK+ 2.8.18 the testcase runs about 30% faster.
> 
>  o pango_renderer_draw_layout_line() jumped from 2.3% to 8.7% in the
>    pango profile between 2.6.8 and 2.8.18.
> 
>  o Why 7% of the time spent in GTK+ spent in
>    _gtk_style_peek_property_value() using 2.8 ?

Also of note is that the time spent in no-vmlinux increased
significantly (11%). These tests were done against a hardfloat image so
floating point instructions cause exceptions and get handled in kernel
space. That jump is extremely likely to be the consequence of a
significant increase in floating instruction usage.

I realise you could avoid some of the overhead of using such
instructions using softfloat but that's missing the point as it would be
preferable to avoid the instructions entirely if possible.

Regards,

Richard

-- 
Richard Purdie
Openedhand Ltd.