Re: Vertex submission (was Re: [Plib-devel] Re: 3DNow! acceleration)

[Steve B. <sjb...@ai...>]

Sam Stickland wrote:

> To get the best speed out of CVA's won't it need some adjustment to the
> scene graph traversal?  Say we had this structure:
> 
>              Transform Node                    Transform Node
>                            |                                            |
>                            +------- CVA Node ---------+
> 
> Then you'd want to lock the arrarys, draw the CVA Node in both positions and
> then unlock it - rather than locking it twice.

Well, only if the two references to that node were at the same position
in the scene.
 
> I guess this might start to get a little to complex, particulary if you
> start to factor in different states.

Yep. 


  The CVA will only be re-usable within a frame if it's drawn
more than once at exactly the same place in the scene.  Hence, you
couldn't use it to draw twenty identical buildings in a scene - but
you could use it to draw the building with one texture - and then
again at the precise same location with alpha blending and a second
texture.

Hence, we would want that ability to simulate more complex rendering
than the underlying OpenGL can manage. In a word: Multipass.

I think this would be best implemented by creating a new kind of
ssgState that contained multiple internal ssgState nodes (much
like the current ssgStateSelector node) - but with the semantics
being that the ssgLeaf that uses it would render multiple times -
each using a different sub-state.  For regular ssgVtxList nodes,
that wouldn't be too efficient - but for CVA nodes, it would be
fast.

This would look really nice to the user - just create a 'multistate',
to describe whatever 'look' you want your geometry to have - and
whatever you connect it to get's rendered using multipass automatically.

-- 
Steve Baker                  http://web2.airmail.net/sjbaker1
sjb...@ai... (home)  http://www.woodsoup.org/~sbaker
sj...@ht...      (work)