[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-core-devel
Subject:    Re: Alpha blending
From:       Stephan Kulow <coolo () kde ! org>
Date:       2000-10-10 9:54:58
[Download RAW message or body]

Rik Hemsley wrote:
> 
> I hacked KIconLoader to allow me to get a QImage instead of a QPixmap,
> which was quite easy, because it uses QImage internally and only converts
> to QPixmap just before returning a processed pic.
> 
> I also wrote a (very) fast blend function that takes 2 QImages and blends
> the first over the second, honouring the alpha channel. I couldn't find
> anything in Qt or KDE to do this, so I had to hand-roll it. In tests on my
> Celeron 600 it can blend 65 million pixels (about 1/2 of which have alpha
> != 0) in 5 seconds. That's about 5ms for a 256x256 image, so approximately
> 0.17ms for our largest icons (48x48).
> 
> Fast functions aside, of course this would be configurable and probably off
> by default. I asked someone to try my function on a K6/233 and the speed
> goes from ~0.17ms to ~1ms, which is clearly not going to be fast enough
> for even slower machines (thanks to that person on IRC !)
> 
> Now, my implementation could be improved somewhat. I had a chat with Dirk
> and it seems likely we can use KPixmap in some way - it seems to have access
> to the XImage held by QPixmap. I'm not clear on how we would do this yet,
> but I'm sure I'll figure it out.
> 
> Using blending for e.g. toolbar icons would actually be comparatively easy
> and source compatible, assuming that developers use KAction when writing
> their apps (KAction takes a string as a parameter for the icon - you don't
> mess about with QPixmaps IIRC.)
> 
> Anyway, enough talk. I'm tired and I'm sure a screenshot is in order ;)
> 
> http://www.geoid.clara.net/kicker_alpha_test.png
> 
> Apologies for the hideous background tile, but I needed something to
> illustrate the point.
> 
> Rik - wondering if tackat will a) kill me when I suggest we make new versions
> of all hicolour icons to support transparency or b) be so happy that this is
> possible that the amount of work is somehow irrelevant ;)

I love it though - and I bet I'm already on Tackat's list anyway :)
Then we at least have a nice feature that people should have for 2.1.

Greetings, Stephan

-- 
... but you ain't had mine

[prev in list] [next in list] [prev in thread] [next in thread]