[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [E-devel] evas premul changes



On Sat, 16 Sep 2006 07:21:09 GMT "jose_ogp@juno.com" <jose_ogp@juno.com>
babbled:

> 
> > > Much ado about gradients, but hopefully all this will make
> > > it easier to work with grads in current edje and otherwise.
> > > 
> > >	Comments...?
> > 
> > sounds good - one thing i did notice - gradient fills are a lot
> > slower than images...  :)  i noticed when cross-fading wallpapers
> > @ 1600x1200 with gradient bg's  :) 
> > 
> 	I'll get back to you on the rest later... I've a fairly
> busy wekend ahead and need to get some sleeeeeep :)  Nearly done
> with things - just a tiiiny bit more to do, and then merely check
> every evas src file to see if something stands out.....
> 
> 	But as far as grad rendering speed... They'll never be
> as fast as image rendering, ever. Not with any kind of software
> implementation.
> 	The best you can do is to match it in certain special cases.
> Every dst pixel needs to be composited with some src pixel, and
> with grads the src pixel needs to be generated on the fly via
> possibly several interp steps.. at least one step more than images.

wel grads do have an advantage on modern cpu's - much less memory bandwidth- no
need to read the src image pixels - as you can generate on the fly. :) they can
be fast - it's just a matter of making the calculation as fast as possible :)

> 	You can match it by special casing (aka more code) or by
> caching results as images for redraws.. but every special case
> that one can do for grads (eg. horz/vert grads etc), one can have
> the same for images. On the 'plus' side, grads require a lot less
> input data to generate large regions with varied geometrical shapes,
> etc.

sure :) mind you - i am comparing grads vs. images when being scaled and
interpolated - the non-scaled image is smooth. the scaled image i think
visually was a bit smoother than grads. anyway - it was just an observation for
now :)

> 	Nevertheless, yes grads can definately benefit from some
> optimizations :)

there we go - that's all i want to mention :) they aren't "close to
optimal".... YET :)

-- 
------------- Codito, ergo sum - "I code, therefore I am" --------------
The Rasterman (Carsten Haitzler)    raster@rasterman.com
裸好多
Tokyo, Japan (東京 日本)