Skip some operations in interior parents computation#4358
Skip some operations in interior parents computation#4358leonardo-mutti-aks wants to merge 3 commits intolibMesh:develfrom
Conversation
|
FYI @jwpeterson . |
So ... ex falso? It's not used as often as for side elements, but we set the I'd like to see a test case that this speeds up, though; maybe there's some other way to accomplish that. |
d4a36c8 to
1abd1ce
Compare
Thanks for pointing this out, I have updated some of the comments in my commits.
Here are a couple of screenshots of a model we are interested in (static structural). You can see that it contains many (0D) nodeelems because of all the RBE3 connections plotted in blue. I profiled the mesh stitching time (14 mins), of which 5:45 mins are spent in I also pushed one last commit to be able to completely disable the computation of interior parents, |
|
Job Coverage on 1abd1ce : invalidated by @jwpeterson We had to retarget this PR from master -> devel, I think that may have confused the Coverage job so I'm invalidating it. |



Given that an interior parent
pof an elementelmust satisfyp.dim() == el.dim() + 1,we can skip some computations in
MeshBase::detect_interior_parents():elsuch that there is nopwithp.dim() == el.dim() + 1,pif there is noelwithel.dim() == p.dim() - 1.In models with many elements, especially building the node-to-elem map can be a bottleneck,
so optimization 2 may help easing the computational effort.