On Sun, 11 Dec 2005 23:51:24 +0900 Carsten writes:
> On Sun, 11 Dec 2005 02:00:55 -0500 Jose O Gonzalez
> <[EMAIL PROTECTED]> babbled:
>
> >
> >
> > On Sun, 11 Dec 2005 13:59:10 +0900 Carsten writes:
> > > On Sat, 10 Dec 2005 20:47:56 -0600 Brian Mattern
> > > <[EMAIL PROTECTED]>
> > > babbled:
> > >
> > > > Some recent commit (after the 2nd but before the 8th) broke
> evas
> > > shape
> > > > mask calculations. I'm going to go out on a limb and blame
> Jose's
> > > > gradient patch :)
> > >
> > > correct. found it. the destination alpha math definitely gets
> things
> > > wrong..
> > > like forgetting in 2's compliment binary math on single 32bit
> values
> > > that
> > > values can overflow... into the next, etc. unlike mmx/'sse where
> the
> > > simd ops
> > > protect the values form eachother... i ended up backign out all
> the
> > > blend
> > > changes as it just woudl take too long to debug it all.
> > >
> > > > For some reason regions that have transparent images blended
> onto
> > > opaque
> > > > ones get shaped out now.
> > > >
> > > > For an example, see: http://rephorm.com/files/shots/icky.png
> > > >
> > > > The AA parts of the text are shaped out, as is the window
> icon.
> > > > This isn't noticeable in the default theme, since it doesn't
> use
> > > shaping.
> > > >
> > > > I'll try to take a look, bu this isn't really my area of
> > > expertise, so I
> > > > wanted to let everyone else know about the issue (Raster and
> Jose
> > > > especially).
> > > >
> > > > Take care,
> > > > Brian
> > > >
> >
> > Ummmm.... It's likely that some of the blend routines
> > have bugs -- they were written very quickly, and went basically
> > unchecked and untested. But it's also likely that most of them
> > are fine.. they do need to be checked and tested. I'll see if I
> > can get to that later -- but if you know of particular routines
> > that are incorrect, let me know and I can look at them first :)
> >
> > Moving to pre-mul alpha is a good idea, but is *far*
> > more radical in several respects, especially in relation to
> > programs that have been getting/setting image data.
>
> well it has other problems i wont go into here... :) there arent
> that many
> programs yet... thank god.
>
> > Right now I was thinking of sending a small patch to add
> > "rounded rectangles" to evas, but we'll see..
>
> for now - let it cool for a bit. i just reverted these :)
>
They'll remain in deep freeze til you let me know otherwise :)
As far as the blend functions...
From that one screenshot, it's unlikely that the dst-alpha
blend functions are causing what is seen, since the areas where
the src (icon and text) images are being drawn to, seem to be opague,
and there the dst-alpha blend funcs just do non-dst-alpha blending..
Also, the corners aren't being touched by any src data, I presume,
and yet they are not transparent, etc.. But I can't say much more
from just that screenshot (I don't have the 'theme' or its images).
I thus took a bit of time to run some tests on the blend funcs.
Not exhaustive by any means, but a few... Blending to dst images
with alpha, with src images with/without alpha, and with alpha masks
plus colors with/without alpha, for both the c and mmx versions, and...
I sure don't see any problems.
But, you seem certain, and could well be right...
> > > correct. found it. the destination alpha math definitely gets
> things
> > > wrong..
> > > like forgetting in 2's compliment binary math on single 32bit
> values
> > > that
> > > values can overflow... into the next, etc. unlike mmx/'sse where
> the
> > > simd ops
> > > protect the values form eachother...
If you can tell me exactly which functions... :)
-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems? Stop! Download the new AJAX search engine that makes
searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
http://ads.osdn.com/?ad_id=7637&alloc_id=16865&op=click
_______________________________________________
enlightenment-devel mailing list
enlightenment-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
