The state sorting structure that we use does treat ordered groups
differently because of their extra semantics, but you should not see
this much of a drop. Especially if the only thing you are changing
is a transform. This sounds like a bug in our stuff.
As for transparency/opaqueness changing the framerate, we are more at
the mercy of the graphics card for that one. We should "mostly" match
their speed when switching modes.
Doug Twilleager
Sun Microsystems
>MIME-Version: 1.0
>Content-Transfer-Encoding: 7bit
>X-Priority: 3
>X-MSMail-Priority: Normal
>X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2615.200
>Subject: Re: [JAVA3D] Ordered Groups and Optimizations
>To: [EMAIL PROTECTED]
>
>I ran a little experiment. I put 1000 flowers into the scene (thats 2
>shapes each) using links and shared groups. With say, 500 visible at one
>time. The first time I put them in an ordered group and the fps was 5. The
>second time I jumbled them into a single branchgroup. The fps was then 17.
>So putting them in an ordered group seems to cost a lot more. I suspect
>that you give up the appearance sorting when you do that. I then made the
>plants non-transparent.. it will interest you to know there was no change in
>fps.
>
>As an aside, it is interesting that 4000 vertices can have that big an
>impact on the scene. Its really all the same geometry, the only difference
>being the translation. Technically in opengl you should be able to set the
>appearance, set the display list and just set the transform and render the
>shape. I am a little suprised at the impact. The next thing to try is to
>build a "batch" of plants all in the same shape and see how that renders.
>
>Dave Yazel
>
>----- Original Message -----
>From: David <[EMAIL PROTECTED]>
>To: <[EMAIL PROTECTED]>
>Sent: Monday, March 05, 2001 12:57 PM
>Subject: [JAVA3D] Ordered Groups and Optimizations
>
>
>Hi all:
>
>I am trying to get a handle on managing the sorting of transparent
>vegetation. My first stab at it used an ordered group, with each shape
>being in a seperate branch group. After the user moves a short distance I
>check to see if the nodes are out of order (based on distance from eye to
>center of sub-scene. If they are out of order I detach the BG parent of the
>ordered group (because the following detaches are very expensive if they are
>live), and detach all the nodes from the ordered group and then place the
>children back into the ordered group in the proper sorted order. This is
>extremely expensive when the sorted sub-graphs reach past 100 or so.
>
>1. Does the renderer still optimize at all, if the sub-graphs have the same
>material, appearance, etc? Say I have 100 sub-graphs which are branch
>groups with links to the exact same geometry. Technically the only thing
>different is the transform. Is the ordered group causing java3d to
>repeatedly send all the appearance info for each ordered group?
>
>
>2. I would like to consider an alternate approach for a specific case. The
>case is that I have, say 4 batches of 100 sub-graphs. Each batch of 100 has
>exactly the same sub-graph, but with different transformations. Would it be
>more efficient to just sort the trsnaformations? In other words don't sort
>the groups each pass, just sort the transforms, swapping transforms as I
>sort?
>
>3. One problem I have been having is with intersecting alpha blended shapes.
>Like a simple bush which has 2 or 3 intersecting planes with alpha blended
>textures. The problem is that ordering the entire bush with other such
>bushes is fine as far as its relation to each bush goes, but the ordering of
>the planes within a single bush can (and almost always do) cause them to
>step on each other. So sometimes you look at it from one angle and you see
>that the back to front pane must have been drawn before the left to right
>pane, causing you to see behind the bush on the left, missing the left hand
>part of one pane. I could make those 4 seperate planes, so there is no
>intersection and sort them.
>
> A. This is more sorting, would increase the objects being sorted 400
>percent.
> B. Could I place all four planes in one shape node? Would I need to
>sort the geometries then?
> C. If I have to sort the geometry arrays then I can no longer use links
>and shared groups right?
>
>Any suggestions would be helpful,
>
>I would love to hear how the Sun engineers would code this if they were
>doing it in Java3d.
>
>Sincerely,
>
>Dave Yazel
>Cosm Development Team
>
>===========================================================================
>To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
>of the message "signoff JAVA3D-INTEREST". For general help, send email to
>[EMAIL PROTECTED] and include in the body of the message "help".
>
>===========================================================================
>To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
>of the message "signoff JAVA3D-INTEREST". For general help, send email to
>[EMAIL PROTECTED] and include in the body of the message "help".
===========================================================================
To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
of the message "signoff JAVA3D-INTEREST". For general help, send email to
[EMAIL PROTECTED] and include in the body of the message "help".
