[ODE] test problems?

Russ Smith russ at q12.org
Mon Nov 19 20:13:02 MST 2001


> > DynaMechs is O(N), the rest are O(N^3) (except i don't know about
> > Darwin2K).
> 
> When you say O(N) I suppose you mean Nlog(N)?

errr ... nope, i'm pretty sure that dynamechs is O(N), where N is the
number of bodies (which also corresponds to the number of constraints
for a tree system).

> On your homepage you say that Havok is O(N^3) and medium speed, while
> Vortex/Karma (CM-Labs/MathEngine) is O(N^3) and high speed.

i took a look at havok a long time ago and concluded that it was
slower that MathEngine per step, and also because it seems to
use an acceleration constraint / baumgarte stabilization method
which was a bit less stable. but i'm not knocking havok, because
i think they've done a great job, especially with the 3DS max plugins
(where speed doesn't matter). anyway, their implementation could well
be different / faster now.

> The relations between CML/MathEngine (Vortex & Karma) and ODE are
> indeed very, very puzzling...

to make things clear: ODE's API and feature set looks like
Vortex & Karma because i wrote the original core of MathEngine's
Karma, which was subsequently rebranded as Vortex by CML. so when i
did ODE, i did it kind of the same way (different code, but similar
techniques). things that Vortex & Karma have that ODE does not:
  - more joint types
  - more collision types
  - PS2 (playstation) and intel SSE optimizations
  - other ... ?

russ.

--
Russell Smith
http://www.q12.org



More information about the ODE mailing list