VAR and shared vertices

Hi,

I’m implementing a terrain renderer using a quad tree. Until now some vertices of neighbouring nodes are inserted into the vertex array twice. I did that because I thought I read that GeForce had a post T&L vertex cache of 10 or something. If these nodes really shared some vertices they wouldn’t be in the cache anyway when the recursive rendering algorithm reached there. Am I corrcet about that ?

How are you rendering you quads ? I break each quad down into an NxN grid made up from Tri Strips. So a 4x4 column grid looks like

X x x x X
x x x x x
x x x x x
x x x x x
X x x x X
Which converts to 4 strips of 8 tri’s. Because I draw the tri-strips within each quad, it doesn’t matter if I don’t share edge vertices between quads. As you say, they wont be in the cache anyway. If you are using the same texture across the quads then you can reuse shared vertices to save space.

If your using LOD on your Quads you can also share vertices between parent and child quads as well as adjacent quads.

But again if you need a unique texture per quad you have to duplicate.

OK, that’s what I thought.

I’m using continuous lod algorithm by Thatcher Ulrich http://www.gamasutra.com/features/20000228/ulrich_01.htm

Nodes are rendered recursively but I’m experimenting with assigning a vertex array to parts of the tree to optimize for VAR extension.

Sharing vertices between parent and child nodes is possible and I thing I’ll implement it but I’ll habe to test wether sharing between neighbours will actually increase performance or not.

[This message has been edited by muhkuh (edited 05-09-2002).]

[This message has been edited by muhkuh (edited 05-09-2002).]