yade-dev team mailing list archive
-
yade-dev team
-
Mailing list archive
-
Message #01447
Re: strided collider speedups
Václav Šmilauer a écrit :
I think this sounds just awesome.
This really changes the priorities in terms of "where to optimize" and
paralellisation, doesn't it?
In what way?
If the collider takes 6% with 50k spheres, then optimisation should
focus on interactions handling (but perhaps this really optimised already)
I actually don't think so. At this point, this
"optimization" is not even reliable (bodies may get out of their BB,
which is not acceptable).
Well, fundamentally, there is no difference, bodies get out of BBs at
each time step already. With detectionFactor = 1, you can miss contacts
created DURING the timestep, if spheres are x-aligned.
Just imagine that activating the collider each 10 steps is like setting
dt 10 times larger...
I am looking at
http://www3.interscience.wiley.com/journal/112140923/abstract now, but
I don't think I will have time for proper implementation.
Just read the abstract and forget it :
"The algorithm is applicable to systems comprising bodies of a similar
sizes."
Bruno
--
_______________
Chareyre Bruno
Maître de Conferences
Grenoble INP
Laboratoire 3SR - bureau E145
BP 53 - 38041, Grenoble cedex 9 - France
Tél : 33 4 56 52 86 21
Fax : 33 4 76 82 70 43
________________
Follow ups
References