Re: [AD] small bug in 15 and 16 bit asm color converters

[Elias Pschernig <elias@xxxxxxxxxx>]

On Sat, 2004-10-23 at 08:27 -0700, Chris wrote:
> On Sat, 23 Oct 2004 16:23:44 +0200, Elias Pschernig <allefant@xxxxxxxxxx> wrote:
> > And doing some testing, I agree to this. I set this monitor from the
> > standard 10% brightness 10% contrast to blinding 100% 100%, and still,
> > color 7/3/7 is not to be distinguished from 0/0/0.
> 
> That won't quite work. While the difference there may be negligible,
> once you start adding transluceny affects, any difference from 0 could
> make all the difference (multiply blending, for example).
> 

True, that's another problem. But depending on which blending you use,
the same applies currently. What if I have a blender which first does a
color inversion and then a multiply? It will fail right now as well. So,
this argument probably doesn't count much - the patch makes it neither
worse nor better in the general case. I can always find a blender for
which one is good and the other is bad.

So the only concern should be normal display, without increasing the
effect by reading back from the screen. If I run a filter over my
testcase, of course I will notice where there is a difference :) With,
and without the patch.

But just run the testcase, and say if you notice the discontinuity more
in the upper or the lower half of the screen. If we find a single case
for who it is the upper - the patch probably shouldn't be applied. Or
even if for the majority both are about the same, it shouldn't be
applied if there's a performance penalty (but I guess the difference
with that POR instruction will be in the area of 1%.. maybe even much
smaller - Bob hopefully can tell).

-- 
Elias Pschernig