Mark Allan wrote:
>
> Mark Allan wrote:
> >
> > Brian Paul wrote:
> > >
> > > Pasi Kärkkäinen wrote:
> > > >
> > > > Hello!
> > > >
> > > > Are there known blending bugs in the mga-driver (g400)? When I render many
> > > > additive polygons on the top of each other with small alpha-value
> > > > (something like 0.05) the result is something it shouldn't be with g400.
> > > > With software-mesa it's correct as well as with nvidia's drivers.
> > > >
> > > > If this is not a known issue, I might do little app that demonstrates
> > > > this..
> > >
> > > I've found that the accuracy of blending isn't always very good on
> > > some hardware. Not much can be done about that.
> > >
> > > -Brian
> >
> > I doubt it's a problem with the hardware itself. I've done comparisons
> > with DRI, Utah, XIG, and Win2K G400 drivers - both Utah and DRI produce
> > poor results with additive blending, whereas the XIG and Windows drivers
> > look normal. I'll try and assemble some screenshots that demonstrate the
> > discrepancies.
>
> I've created a test case and put up some screenshots that show the
> blending problems with the G400:
> http://www.reptilelabour.com/TEMP/G400_DRI_blend/
>
> G400 DRI and Utah are the only drivers that display the problem. WinNT,
> Win2K and XIG G400 drivers all display expected blending results.
OK, I fixed this. The problem was in the alpha blend register setup.
We were effectively modulating the post-textured fragment alpha by
itself if the texture env mode was GL_MODULATE. I added a comment to
the code to explain the situation.
I've checked this into the trunk and will put it into the 3.5 branch
to be sure it's not repeated.
Thanks for the test program, Mark!
The point I made earlier about the accuracy of blending still stands.
The Glean blendFunc test prints the error (in bits) of the various
blending modes compared to an expected result. Some hardware is
indeed more accurate than others.
-Brian
_______________________________________________
Dri-devel mailing list
[EMAIL PROTECTED]
http://lists.sourceforge.net/lists/listinfo/dri-devel
