Adding vertexbuffer support to ReplicantBody shouldnt be that much work.
Wouldnt that make more sense so that the whole osg community would make use
of a better character animation toolkit?

Although, I dont think it will affect performance that much, the same amount
of data have to be submitted, its fewer OpenGL calls, but my experience of
glBegin() glEnd() is that its the glEnd() that consumes time, not glVertex()
glNormal() calls, as I guess the data is transmitted after the glEnd().
The only really good way to go to get better performance is to do hw
skinning...

/Anders


On 4/18/07, David Guthrie <[EMAIL PROTECTED]> wrote:

The problem with rbody is that it's pushing all of the vertices to the GPU
every frame.  Using a display list stops the animation because it's
precompiling the vertices onto the GPU, so it doesn't change.  Using display
lists would only help for a stationary character.  You can write the
vertices straight to a vertex buffer and see some performance benefit, but I
don't think rbody supports that.  We are currently removing rbody from our
engine (Delta3D) to use cal3d straight.  Partly so that we can control the
rendering.  If you want to see more characters and you use a vertex buffer,
you can do things like only updating the animations on timer somewhat less
than the frame-rate, which will allow a vertex buffer to not be written to
the card every frame.  Another option is to do hardware skinning where you
only use cal3d to update the bone positions and have a vertex shader use the
bone positions to move the vertices of static pieces of geometry in the
GPU.
On Apr 18, 2007, at 3:06 PM, Zach Deedler wrote:

*PROBLEM*
I want more people in my scenes rendering at 60 Hz.

*INVESTIGATION*

   1. Well I spent all this time putting in LODs, but I didn't get
   spectacular results. :(
   I render a man model, and looked at the stats.

   High level LOD  3497 vertices  0.5 Draw time
   Low level  LOD  805 vertices   1.0 Draw time

   The draw time is the draw time of the actual model (I cancelled out
   the 0.5 draw time of drawing nothing)
   2. Stopping his animation doesn't improve the stats.
   3. Removing the textures doesn't improve the stats.
   4. Rendering a non-cal3d model of 4837 vertices gives me 0.25 draw
   time.
   5. If I tell the geometry of the rbody Submesh to
   setSupportsDisplayList(true), the man no longer animates, but the draw time
   goes to 0.2.

*QUESTIONS*

   1. Where is all the draw time being spent for cal3d/rbody?
   2. What can I do to reduce the draw time?
   3. Any suggestions on how to find the performance bottleneck?

For now, I'll continue to hack away.

Thank you.

Zach

-----Original Message-----
From: [EMAIL PROTECTED] [
mailto:[EMAIL PROTECTED]<[EMAIL PROTECTED]>]
On Behalf Of Jan Ciger
Sent: Wednesday, April 18, 2007 14:25
To: osg users
Subject: Re: [osg-users] cal3d lod

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Jeremy L. Moles wrote:
> Palle Raabjerg and I rewrote most of the Cal3D exporter for Blender
> (he as a Google SOC project and myself just out of curiosity) that is
> currently in Cal3D SVN (I also put up the new funky website!). I
> didn't write any of the LOD code however, and I can confirm it is
> certainly not in a working state at the moment; I do plan on revisiting
it eventually.
> There isn't a lot of activity lately unfortunately, but I don't
> believe the project is dead...
>

That's a good news - that exporter could use some love. I have also a
similar waiting project there - to finish a BVH skeleton and animation
loader for Blender. There are some really nasty and non-trivial issues there
so I cannot blame you for the Cal3D one - I have seen the code when I needed
a converter for our in-house animation format to Cal3D and it is certainly
not a piece of cake.

Jan
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (GNU/Linux)
Comment: Using GnuPG with Mandriva - http://enigmail.mozdev.org

iD8DBQFGJmJ0n11XseNj94gRAtSbAJ0dc6s93SQSeDCOuno/nje55etdngCeNa62
Vl+RKgDdS6TKwBXzBgb0a0U=
=xU0R
-----END PGP SIGNATURE-----
_______________________________________________
osg-users mailing list
[email protected]
http://openscenegraph.net/mailman/listinfo/osg-users
http://www.openscenegraph.org/
_______________________________________________
osg-users mailing list
[email protected]
http://openscenegraph.net/mailman/listinfo/osg-users
http://www.openscenegraph.org/



_______________________________________________
osg-users mailing list
[email protected]
http://openscenegraph.net/mailman/listinfo/osg-users
http://www.openscenegraph.org/




--


________________________________________________________________
Anders Backman               Email:    [EMAIL PROTECTED]
HPC2N/VRlab                  Phone:    +46 (0)90-786 9936
Umea university              Cellular: +46 (0)70-392 64 67
S-901 87 UMEA SWEDEN         Fax:      +46 90-786 6126
                              http://www.cs.umu.se/~andersb
_______________________________________________
osg-users mailing list
[email protected]
http://openscenegraph.net/mailman/listinfo/osg-users
http://www.openscenegraph.org/

Reply via email to