John Carmack wrote:
>
> There should probably be a set of standard functions that synthesize lines
> and points with quads, similar to the automatic quad-to-two-triangles
> routines.
>
> The matrox glx driver currently handles colored, depth buffered, and
> blended lines, but can't handle wide lines or textured lines directly.
>
> To fix that, we will need to build two triangles for each line and use the
> full featured triangle path. Obviously not high performance, but better
> than syncing with the hardware to do software fallback.
>
> We currently program a trapazoid directly for points, but pure triangle
> engine hardware would also need to do those as two tris.
>
> Does the 3dfx driver currently do this? This should be generally usefull
> functionality.
This is done inside glide, mesa doesn't see it. I was wondering about this for
the mga driver - do you have the magic incantations to build the displaced
vertices for the two triangles? I don't see any way to avoid taking 1/sqrt(area)
to get the increments. It's the square root I'd like to avoid...
On the other hand, syncing is much much slower again. It'd be good to do this
once, in Mesa, with assembly where possible.
> Related to that, but needing a bit more work, would be synthesizing pixel
> and bitmap operations by uploading into a driver work texture, then using
> texture mapped triangles to image.
>
> It would be nice for driver implementation to be able to just do textured
> triangles and have all the other primitives synthesized automatically until
> someone cares enough to pick out the functionality that can be more
> directly accelerated.
Are you talking about the read/write spans/pixels functions in dd.h?
Keith
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
