Welcome! Log In Create A New Profile

Advanced
Re: T-Junctions re-visited
March 20, 2013 06:42AM
Hi Travis,

I'm using a related technique: I have a single array of vertices, but they are sorted by a "lexicographal" comparison function (that is, sort by X, if X is equal sort by Y, if Y is equal, sort by Z, etc.).

This gives you similar performance times as the balanced binary tree - NlogN to get the index built (due to sort time) and log-time lookups via a binary search.

The big thing you lose with the array is cheap insertion of "more stuff." But in the case of my code, I know the total vertex count before I even start working (because I have the unsmoothed directive and can count up the triangles and quads and lines) so this is acceptable.

The other thing I discovered is that, given a vertex in the array, it is a bit faster to find the range of equal vertices by linear search (E.g. just walk in both directions in the array) than to do a binary search of the whole sorted array. I chalk this up to locality - more cache hits if the search stays close by. I've been meaning to do a 'step' search - that is, start in the array, walk in units of N vertices, then walk back by N/2, etc. If we start N at a small number like 16 or so, we can find groups of equal value vertices in only a few steps. But...performance profiles indicate that it's a small optimization at this point; the algorithm is currently dominated by the 'big sort' at the beginning.

I'm currently sorting the _entire_ part together -- lines and tris and quads. When I had the lines separate it was faster (it's quicker to sort two small things than one big thing) but I want to have one 'pile' of vertices for snapping things together, so I think one big pile it is.

I think the big win is going to be recursion optimizations, e.g. smooth the stud, recognize in the smoothing process that the stud is "sealed" from other parts and doesn't need to be resmoothed when it is included, then smooth the parent part without the studs. I base this on the fact that (1) my code can detect that the stud is sealed and (2) small parts optimize almost instantly and (3) the big baseplates minus studs have similar vertex counts to small parts. It's a big refactor to actually get the code working this way though; I'm going to finish T junction repair first.

Cheers
Ben
SubjectAuthorViewsPosted
T-Junctions re-visited Travis Cobbs841March 11, 2013 10:38PM
Re: T-Junctions re-visited Tim Gould441March 11, 2013 11:31PM
Re: T-Junctions re-visited Roland Melkert396March 12, 2013 11:17AM
Re: T-Junctions re-visited Ben Supnik413March 12, 2013 11:17AM
Re: T-Junctions re-visited Roland Melkert383March 12, 2013 11:22AM
Re: T-Junctions re-visited Ben Supnik368March 12, 2013 11:32AM
Re: T-Junctions re-visited Roland Melkert361March 12, 2013 11:41AM
Re: T-Junctions re-visited Ben Supnik369March 12, 2013 11:55AM
Re: T-Junctions re-visited Roland Melkert371March 13, 2013 10:08AM
Re: T-Junctions re-visited Travis Cobbs391March 13, 2013 11:27AM
Re: T-Junctions re-visited Allen Smith402March 12, 2013 02:01PM
Re: T-Junctions re-visited Michael Heidemann386March 13, 2013 09:27AM
Re: T-Junctions re-visited Ben Supnik370March 13, 2013 10:37AM
Re: T-Junctions re-visited Tim Gould380March 13, 2013 02:31PM
Re: T-Junctions re-visited Michael Heidemann360March 13, 2013 03:42PM
Re: T-Junctions re-visited Michael Heidemann392March 17, 2013 04:35AM
Re: T-Junctions re-visited Ben Supnik333March 17, 2013 08:43AM
Re: T-Junctions re-visited Michael Heidemann330March 17, 2013 08:59AM
Re: T-Junctions re-visited Ben Supnik321March 17, 2013 09:02AM
Re: T-Junctions re-visited Magnus Forsberg344March 17, 2013 09:15AM
Re: T-Junctions re-visited Travis Cobbs401March 17, 2013 10:09PM
Re: T-Junctions re-visited Ben Supnik367March 18, 2013 07:30AM
Re: T-Junctions re-visited Travis Cobbs361March 18, 2013 10:57AM
Re: T-Junctions re-visited Ben Supnik379March 18, 2013 11:27AM
Re: T-Junctions re-visited Travis Cobbs357March 18, 2013 12:23PM
Re: T-Junctions re-visited Ben Supnik406March 18, 2013 01:41PM
Re: T-Junctions re-visited Roland Melkert424March 18, 2013 02:51PM
Re: T-Junctions re-visited Travis Cobbs395March 18, 2013 02:59PM
Re: T-Junctions re-visited Roland Melkert458March 18, 2013 03:41PM
Re: T-Junctions re-visited Ben Supnik440March 18, 2013 06:09PM
Re: T-Junctions re-visited Roland Melkert393March 19, 2013 11:38AM
Re: T-Junctions re-visited Ben Supnik407March 19, 2013 11:43AM
Re: T-Junctions re-visited Roland Melkert433March 19, 2013 07:07PM
Re: T-Junctions re-visited Travis Cobbs402March 19, 2013 10:13PM
Re: T-Junctions re-visited Tim Gould409March 20, 2013 03:56AM
Re: T-Junctions re-visited Travis Cobbs373March 20, 2013 10:13AM
Re: T-Junctions re-visited Ben Supnik368March 20, 2013 10:23AM
Re: T-Junctions re-visited Travis Cobbs400March 20, 2013 10:29AM
Re: T-Junctions re-visited Travis Cobbs377March 20, 2013 10:26AM
Re: T-Junctions re-visited Ben Supnik358March 20, 2013 11:18AM
Re: T-Junctions re-visited Ben Supnik382March 20, 2013 06:42AM
Re: T-Junctions re-visited Roland Melkert358March 20, 2013 11:36AM
Re: T-Junctions re-visited Philippe Hurbain391March 20, 2013 06:36AM
Re: T-Junctions re-visited Ben Supnik425March 20, 2013 06:58AM
Re: T-Junctions re-visited Allen Smith419March 20, 2013 09:08AM
Re: T-Junctions re-visited Roland Melkert466March 20, 2013 11:35AM
Re: T-Junctions re-visited Ben Supnik892March 20, 2013 11:50AM
Re: T-Junctions re-visited Michael Heidemann430March 18, 2013 03:54PM
Re: T-Junctions re-visited Roland Melkert370March 13, 2013 10:03AM
Re: T-Junctions re-visited Tim Gould337March 13, 2013 03:56PM
Re: T-Junctions re-visited Sergio Reano330March 13, 2013 03:51PM
Re: T-Junctions re-visited Roland Melkert406March 13, 2013 04:36PM
Re: T-Junctions re-visited Tim Gould357March 13, 2013 05:10PM
Re: T-Junctions re-visited Ben Supnik399March 14, 2013 08:17PM
Re: T-Junctions re-visited Tim Gould448March 14, 2013 08:41PM
Re: T-Junctions re-visited Ben Supnik425March 15, 2013 10:56AM
Re: T-Junctions re-visited Roland Melkert376March 15, 2013 11:36AM
Re: T-Junctions re-visited Ben Supnik326March 13, 2013 06:58PM
Re: T-Junctions re-visited Roland Melkert340March 13, 2013 07:27PM
Re: T-Junctions re-visited Ben Supnik330March 13, 2013 07:37PM
Re: T-Junctions re-visited Roland Melkert366March 13, 2013 07:59PM
Re: T-Junctions re-visited Ben Supnik371March 13, 2013 08:18PM
Re: T-Junctions re-visited Travis Cobbs437March 14, 2013 03:23PM
Re: T-Junctions re-visited Roland Melkert448March 14, 2013 03:34PM
Re: T-Junctions re-visited Tim Gould438March 14, 2013 03:41PM
Re: T-Junctions re-visited Roland Melkert425March 15, 2013 11:33AM
Re: T-Junctions re-visited Tim Gould370March 13, 2013 07:43PM
Re: T-Junctions re-visited Roland Melkert352March 13, 2013 08:08PM
Re: T-Junctions re-visited Tim Gould367March 13, 2013 08:18PM
Re: T-Junctions re-visited Ben Supnik441March 14, 2013 08:15PM
Re: T-Junctions re-visited Roland Melkert422March 15, 2013 11:29AM
Re: T-Junctions re-visited Travis Cobbs398March 15, 2013 12:27PM
Re: T-Junctions re-visited Ben Supnik388March 15, 2013 05:08PM
Re: T-Junctions re-visited Travis Cobbs403March 16, 2013 11:29PM
Re: T-Junctions re-visited Ben Supnik318March 17, 2013 08:58AM
Re: T-Junctions re-visited Roland Melkert319March 17, 2013 12:35PM
Re: T-Junctions re-visited Travis Cobbs362March 18, 2013 12:28PM
Re: T-Junctions re-visited Travis Cobbs439March 18, 2013 12:32PM
Re: T-Junctions re-visited Roland Melkert397March 18, 2013 05:00PM
Re: T-Junctions re-visited Travis Cobbs424March 18, 2013 11:05PM
Re: T-Junctions re-visited Roland Melkert480March 19, 2013 11:38AM
Re: T-Junctions re-visited Sergio Reano428March 26, 2013 02:46PM
Re: T-Junctions re-visited Roland Melkert358March 27, 2013 11:19AM
Re: T-Junctions re-visited Sergio Reano453March 27, 2013 01:48PM



Sorry, only registered users may post in this forum.

Click here to login