Thanks Chien!
Shawn, You are welcome to share that info. with the interest alias. BTW, you might not have to wait till j3d1.3 to have this fixed. As always, performance ranks highly on our tasklist. Thanks, Chien Yang Java 3D Team. > Date: Tue, 31 Oct 2000 15:55:24 -0500 > From: "Shawn Kendall" <[EMAIL PROTECTED]> > X-Accept-Language: en > MIME-Version: 1.0 > To: Joe Kiniry <[EMAIL PROTECTED]>, Chien Yang <[EMAIL PROTECTED]> > Subject: Re: [JAVA3D] Rendering scalability withBranchGroup-centricscenegraphs. > > That someone was actually me! :-) > Classification is done in the Transform3D whenever the tranform is changed. It > is a rather expensive test that sets the matrix classification - > identity,scale,rigid,affine etc. (see Transform3D API docs) > Hwever, classification does not seem to be the problem because it would be a > constant cost. > The real problem with Transforms is due to the child graph wide LocalToVworld > updating that come with a Transform change. > In the current implementation, everytime a Transform is updated, ALL of it's > child Transforms LocalToVworld's are also updated. > > This creates a nearly exponential cost for non-flat scene graphs. > Here is a snippet I got on this. > > Date: > Thu, 12 Oct 2000 11:08:29 -0700 (PDT) > > From Chien Yang: > > You're right. Based on the above observation, in j3d1.2, we did TG message > compaction when multiple TG change are send to the same TG node. > To improve the performance of scene with S3D under multiple TG nodes, in > j3d1.2.1beta1, we only recompute vwcBounds of affected S3D once per frame > regardless of the number of TG change send to it. Yes, we can also apply the > same technique to recompute affected local to vword matrics, but this > enhancement require more effort than we can commit for 1.2.1beta1. Hence, > we've to defer it for later release. Thanks for the reminder. > > > Chien Yang > Java 3D Team. > > BTW, Chien, if you wish I can/will post our emails on this topic, but since > they were in private, I will let you make the call! > Thanks. > > Joe Kiniry wrote: > > > Thanks for the email Shawn. > > > > I saw some of that thread just as I joined the list. It seems that Sun > > finally fixed their problems with the web-archive service (finally!), so > > I'll do some poking about. > > > > I was under the impression that someone did some serious looking at this > > problem (which came down to automatic matrix classification during > > setTransform) and it turned out not to be a problem (for them) (e.g. > > <http://archives.java.sun.com/cgi-bin/wa?A2=ind0010&L=java3d-interest&P=R13 > > 581>. Perhaps I'm unclear when classification takes place. > > > > It seems to me this is a very serious problem with Java3D if there is no > > reasonable work-around. I'd be quite happy tagging matrices (since we know > > what characteristics they have because we created them) and letting Java3D > > use that hinting for significantly increased performance. > > > > Best, > > Joe Kiniry > > > > --On Tuesday, October 31, 2000 14:49:15 -0500 Shawn Kendall > > <[EMAIL PROTECTED]> wrote: > > > > > We found nearly the same result when we started to implement a basic > > > character animation. > > > Search the archive for : > > > > > > serious performance problems when setting Transforms > > > > > > or just > > > > > > setting Transforms > > > > > > This is a known problem that is not planned on being fixed until Java3D > > > 1.3, which has no published release schedule. > > > This is equally a problem for us as you, because we are trying to do > > > commerical quality game demo's and can not do many types because of the > > > hit we take with Transforms. > > -- > ___________________________________________________________ > > Shawn Kendall Full Sail Real World Education > Course Director 3300 University BLVD > Real Time 3D for Gaming Winter Park FL 32792 > [EMAIL PROTECTED] http://www.fullsail.com > ___________________________________________________________ >
