Two questions

[Kurt Akeley]

In article <9006060619.aa20941 at VGR.BRL.MIL>, XBR2D96D at DDATHD21.BITNET
(Knobi der Rechnerschrat) writes:

> 2) This is a more fundamental question. The molecular surfaces used by
>    MOLCAD are constructed from triangles, which are generated from a set
>    of arbitrary - non-analytically defined - dots, which are numbered from
>    0-n. For each triangle we know the numbers (and from that the 3-D
>    coordinates, of cource) of the three vertices (dots). The triangles are
>    read from a file and then analyzed "for that we can display them as meshes
>    to please the lords of the pipeline". We have two sources of dots and
>    two algorithms to generate the triangles. The first combination generates
>    a good (in terms of tmeshes) ratio between new-meshes, continuation-
>    vertices and swap-vertices. Within the 256-vertex limit of the gl, we
>    get a pretty good performance. Unfortunately the second combination
>    of dots and algorithm (which is really fast in generating the triangles)
>    does generate very poor meshes. Over 90% of the triangles are singles, and
>    the remaining meshes are rather short. This still gives excellent graphics
>    performance on the GT and GTX, but we would like to get the best possible
>    results of course (having in mind the VGX, which is told to be tuned for
>    meshes). After this rather long introduction: does anybody know of
>    algorithm that can be used to rearrange a given set of more or less single
>    triangles into meshes? The algorithm has to be fast, as we want to use it
>    on surfaces with several 100.000 triangles. Any code examples, any
>    published material?

i have code that generates reasonable meshes from independent triangle data.
further optimized code can be had from Paul Haeberli.  my attempt to include
the code in a reply to the newsgroup crashed xrn, so i'l simply publish my
mail address and let those interested get in touch with me.

    kurt at cashew.asd.sgi.com

-- kurt