02.03.2011 11:55, Václav Šmilauer пишет:
Cause is a non-parallelised InsertionSortCollider, who need about 80%
time with 4 threads.
The cause is not (only) that it is not parallelized, but that the
algorithm scales worse than O(n). You see that for 1 thread collider's
percentage gets up with increasing n. Parallelizing the collider will
only have very limited effect, and I doubt it is worth the work that
would be necessary for that; FYI the trivial "let's sort along the 3
axes in parallel" does not make it any faster. v.
Agree. But scaling O(NlogN) is not so bad.. And perfomance can be
significantly optimized now by playing with verletDist, I think. But it
need some tries.
But what about
40% of collider time is spent in the "find(id1,id2)" ???
Because, if more verletDist => more interactions per body => yet more time
for collider?!
--
Best regards,
Sergei D.
_______________________________________________
Mailing list: https://launchpad.net/~yade-dev
Post to : yade-dev@lists.launchpad.net
Unsubscribe : https://launchpad.net/~yade-dev
More help : https://help.launchpad.net/ListHelp