[ODE] Ideas for threading ODE...

Bram Stolk bram at sara.nl
Sat Aug 20 23:44:03 MST 2005


Aras Pranckevicius wrote:

>
>Returning to the topic: for the near future, I think the most you can
>get out of multicore stuff is to run big clearly separate things on
>the different cores (i.e. one for physics, one for rendering, etc.).
>Trying to schedule&manage lots of small tasks (like OpenMP would do)
>probably won't help much.
>  
>

Sure you can do multithreaded physics.
Like someone mentioned earlier: islands of objects are
independent problems.

And the step before the solver can also be done
in parallel by separate threats: collision detection.
Each geom vs geom intersection test is independent
of all others, and highly suited for multiple threads,
as the thread don't even require write access to the
data, so no semaphores are needed. Collision detection
is a read-only problem that returns simple results.

OpenMP should not be req'd, as it is better to do roll
your own threads using pthread library.

  Bram



More information about the ODE mailing list