On 25.06.2006, Dave Hooper wrote:

> I'm secretly sorta surprised that using two framebuffers (one
> for the background and one for the foreground/text) would make
> a significant difference to just blitting the background and
> rewriting the text (as we currently do). If it does make a
> significant difference doesn't it just mean our text blit code
> is in need of optimising?

Yes, that's necessary on the colour targets. While text blitting
is already significantly faster than it was in the beginning,
there's still a lot to be desired. Unfortunately ¡'ve got no
clever idea how to generalise some optimisation ideas of mine
(mostly working on pixel blocks at a time, utilising burst mode
(known on coldfire, unknown on portalplayer), and using IRAM as
a cache. Maybe both methods won't help much on portalplayer,
since PP has got data cache.

Scrolling the framebuffer instead of redrawing is a bad idea for
two reasons:

1) (already mentioned) It interacts badly with bakdrop.

2) While pixel-wise scrolling is reasonably fast and simple on
>=8bpp targets, it's more complex on 1bpp and 2bpp. So if we
   do that for all, it means for b&w & greyscale targets: More
   complexity, nothing gained. Otherwise it means lots of
   ifdefing...

Regards, Jens
Received on Sun Jun 25 22:19:22 2006