On Fri, Dec 17, 1999 at 07:37:03PM +0000, Neal Tringham wrote:
> By the way, could you comment at all on the problems we both found with
> the 3.1 Windows / FX builds before - the crashes on Voodoo 1 boards and
> the mysterious red to blue colour swapping effects which might have had
> something to do with Glide? Are these fixed? (Apologies for not
> managing to explore these further - I've been swamped with stuff to do
> lately:-))
I don't know about the V1 problems, but I can comment on some of the other
recent changes.
There was a bug in texture validation that caused textures not to reset
their clamping values. This caused problems with Q3A bounce pads for
example.
There was a bug with color blending where it would attempt to read
destination alpha which doesn't exist on Voodoo hardware. This caused
lighting bugs in Q3A.
There was the Red/Blue swapping bug. This was caused by us not being
careful about which boards did which color ordering. This was agrivated
by Glide having an option to select byte ordering, but not all the
boards being able to do it. Setting up the ordering is now part of the
board/context initialization.
I reworked some of the texture memory management code. It just serves to
simplify the code base somewhat.
There's also a current bug with a missing triangle strip in a vertex
buffer. This shows up in performer if you try to view the blob.nff
object.
There's still some texture management problems that I'm trying to track
down. The most obvious bug is that the hell fog in Q3A flickers.
I've currently got two optimizations of the driver turned off. One is
that textures were defaulting to being LODblended. This caused some very
obvious texture map problems in Q3A.
The second is that we're forcing all non-multitexture textures into
TMU0. It should decide if TMU0 is full and then place them in TMU1, but
that leads to more texture corruption in Q3A. It appears that when
texture 0 is in TMU1 and texture 1 is in TMU0 it fails. This case is
supossed to work.
I highly suspect the last three issues are all the same bug. I've done
some investigation, but I haven't nailed the issue yet. I'm also
inclined to rework the texture unit setup code. It seems it has a more
redundant code and data than it should, but I don't see that happening
until next year. The new V4/V5 boards have more texture capabilities, so
that may be a good opportunity to improve the code.
- |Daryll
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
