[ODE] ODE-GIMPACT performance and advices

[Francisco Leon]

Hi Pierre.

At this moment we have integrated both engines (OPCODE
and GIMPACT) to the ODE engine. So I could make a more
transparent benchmarking to these collision engines.

Last days you've discussed about the importance of
have an entire destructible world, when all objects
can be deformable and moveable. Well, this is our most
desirable goal, and that's the reason why are we
working so hard. :)

About GIMPACT, I'll clarify that this engine asumes
that all objects are deformable. And it is designed in
the manner that when we have almost none objects
moving, it would run faster. If we have some objects
moving and interacting, it still runs pretty. But when
we have all objects moving in a huge scene, it runs
badly. It is obvious.

But concerning to Rigid trimeshes, I have to say that
you're right. I've noticed that OPCODE runs faster
that  GIMPACT does, and I already know the reason. So
OPCODE is the winner when we are dealing with rigid
bodies. 

I've mentioned that GIMPACT consumes more memory than
OPCODE, do you remember that? I'll explain what it
happens: With OPCODE, you'll have one model for many
instances. But with GIMPACT, you'll have one reply of
that model for each instance, and with rigid bodies
this is not desirable. 

But when we could simulate deformable bodies, we'll
meet a very different situation. So I could say that
with OPCODE, we need many replies of the original
model for representing each deformable body, and we'll
have replies of the AABB tree too. It would be higher
memory consuming. With GIMPACT, we have replies too,
but those replies would be most cheaper. I'm not sure
of this, so leave to the Time this decision. We will
test these libraries for deformable bodies, and it is
possible that another better algorithm will arises of
these experiments, very different from the current.

At this moment OPCODE has the advantage of
performance, but GIMPACT has the advantage of
multi-plattform and compatibility. That's the fact.

I didn't have a paper for explain the GIMPACT
algorithm. I'm currently working on that, so I have to
ask you to be patient. You have to understand that
I've just starting the documentation last week, and
GIMPACT is pretty new. 

And if I've just talked from OPCODE badly in the past,
it's because  I've integrated this engine in the wrong
way. At 2005, I've integrated it to the ODE v0.5  and
my Gammaleon engine hasn't been optimized yet. But
when I've started with GIMPACT I've made many
improvements to the Gammaleon Engine and also to the
ODE engine, so I had better results than the last
year.

Pierre, I want to talk with you even more about this.
You have a huge knowledge of game physics which you
could share with us.

Thanks for your time.




"Technocracy Rules with Supremacy"
Visit http://gimpact.sourceforge.net


 
____________________________________________________________________________________
Everyone is raving about the all-new Yahoo! Mail 
(http://advision.webevents.yahoo.com/mailbeta/)