From: broehl@sunee.uwaterloo.ca (Bernie Roehl)
Subject: Re: Optimizing projections
Organization: University of Waterloo
Lines: 31

In article <1sua3tINNqs2@no-names.nerdc.ufl.edu> LIONESS@ufcc.ufl.edu writes:
>My three-d library does a lot of projections ( duh ), but currently it
>is projecting an object's vertices on a _per triangle basis_.  This is
>grossly inefficient for 99% of the objects displayed ( which can
>be optimized by doing projections ONE time, once for each vertex ), but
>objects whose Z-extents intersect the hither plane can't benefit from
>this because new vertices must be created during Z-clipping.

>Anyone have any better ideas?

Yes.  Here's what you should do.

Keep the vertices in an array, and have the polygons (triangles are okay,
but n-sided polygons are slightly more efficient) store the indices into
the array of the vertices that comprise them.  You set a flag for each
vertex when you transform it, so you don't have to transform any vertex
more than once; you also do backface elimination before processing the
polygon, so that vertices that belong only to bacfacing polys don't have
to be transformed at all.

Whenever you transform a vertex, check if it's on the far side of the hither
plane; if it is, you can project it right away and store the result.

Then do your Z clip; any vertices that get produced will have to have their
projection done at that stage.

-- 
	Bernie Roehl
        University of Waterloo Dept of Electrical and Computer Engineering
	Mail: broehl@sunee.UWaterloo.ca
	Voice:  (519) 885-1211 x 2607 [work]