Recent AIGLX Breakage?

[Antonio Vargas]

On 6/5/06, Adam Jackson <ajackson at redhat.com> wrote:
> Antonio Vargas wrote:
> > The usual real-time way is to render the original to a texture,
> > generate mipmaps for the texture and then use texture-bias so that you
> > paint into the screen a lower-resolution texture before overlaying the
> > translucent item. Notice that this does not need any pixel shader,
> > just a fast mipmap generator (depends on the driver support), or even
> > if this is not provided, then use multitexturing to implement a
> > box-filter yourself: you can get 4 samples/pixel with each texture
> > source by sampling the original pixels with a (+0.5,+0.5) difference
> > in (U,V) coordinates
>
> Cute trick!  I think right now we're not doing hardware mipmap
> generation, though it wouldn't be hard to add in principle.  The mipmap
> control code in Mesa could use a little work anyway, if someone's
> looking for a relatively easy project to take on.
>
> Even if we were, though, almost all of our textures are
> non-power-of-two, and the intersection of the sets of (hardware without
> pixel shaders) and (hardware that can do mipmapped rectangular textures)
> is pretty small, so it might be difficult to get this to work acceptably
> with just plain mipmapping.  However, pretty much all hardware that can
> do EXT_texture_rectangle can also multitexture so that'd be an option
> too, the only exception I know of is the Intel i740 which we don't have
> a driver for anyway.

The original Riva TNT 1 did have 2 texture sources per pixel (TNT
meant twin texturing ;) and of course the ever popular 3dfx did also
have 2 sources/pixel, so I think any interesting hardware we are
likely to try AIGLX on will support both rectangles and multitexturing
just fine.


-- 
Greetz, Antonio Vargas aka winden of network