yade-dev team mailing list archive

[Václav Šmilauer <eudoxos@xxxxxxxx>]

> Monolithic
> 
> OMP_NUM_THREADS=1:
>                       Iter/sec
> r1712_thread1 83.4262
> r1716_thread1 83.9633
> 
> 
> OMP_NUM_THREADS=2:
>                       Iter/sec
> r1712_thread2 120.561
> r1716_thread2 112.497

Eh, and what did you test? ;-)

Anyway, the impact will be close to zero as both InteractionDispatchers
(and AABBMetaEngine as well) cache the functor so that they look up the
dispatch matrix only once per interaction/body.

If you used InteractionGeometryMetaEngine, InteractionPhysicsMetaEngine
and ConstitutiveLawDispatcher separately, then the impact would be
bigger.

But don't bother. Anyone who cares about performance has switched to
InteractionDispatchers long time ago.

V.