User-agent: Mozilla Thunderbird 0.8 (Macintosh/20040913)
Hi Y'all,
I have a tex env combine hack that goes something like this:
Unit 0:
RGB: replace from tex 0
alpha: subtract (tex fragment alpha - primary color alpha), then scale by 4x
Unit 1:
RGB: blend previous tex vs. tex 1, controlled by previous alpha
alpha: set to constant color (1)
The goal is to use the vertex color (primary color) and the texture 0
alpha almost like a "choose" function.
If the texture fragment is less than the primary color alpha (failing to
meet the bar), we get 0 for alpha and end up with all of tex 0.
If the the difference is less than 0.25, then after multiplying by 4.0
we get at least 1.0 alpha and we get all of the second color.
The problem is with clamping...if I read the combine spec right, the
resulting alpha from tex unit 0 should be clamped to [0, 1]. So I
should always get tex 0, tex 1, or something in between.
But on my friend's G5 with an ATI Radeon card, he is seeing strange blue
artifacts...upon close investigation, it _looks_ like we're getting an
alpha blend factor > 1, causing the RGB colors to "clip" as they are
overdriven...as they clip in different places we have a hue shift and
our texure goes from a nice green and brown turf to a strange very
brigiht cyan radioactive pile of goo. (At least that's how he describes
it.)
- Am I interpretting the spec wrong, or is there a requirement to
clamp after each TU operation on all four channels?
- Anyone else seen this?
- Is this even worth reporting as a bug, or is this the kind of thing
that's going to be so low down in the ATI driver implementation that
it's not worth fixing for one user's special case app?
Our workaround is to not do the multiply by 4, which prevents color
distortion, but gives us a lot less of the 'choose'-ness that we
want...we always have a blend across all alpha values in our texture,
not just from 0.0 to 0.25, with clamping for the rest.
*cheers*
Ben
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Mac-opengl mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/mac-opengl/email@hidden
