3D Rtree

[fdrmrc]

So far it was hardcoded for the 2D case.

[fdrmrc]

• 3D hexahedral piston, 470 agglomerates.

• Here, some selected agglomerates:

@luca-heltai I'm going to add some instrumentations to measure how it behaves in comparison with METIS.

[luca-heltai]

Nice!

[fdrmrc]

Just to give some very prelimiary numbers for the piston case:

Size of tria: 365712 hexas

N agglomerates = 90
R-tree agglomerates built in 0.0920437 seconds [Wall Clock]
METIS agglomerates built in 0.7282 seconds [Wall Clock]

N agglomerates = 715
R-tree agglomerates built in 0.0859208 seconds [Wall Clock]
METIS agglomerates built in 2.49369 seconds [Wall Clock]

N agglomerates = 5715
R-tree agglomerates built in 0.0832403 seconds [Wall Clock]
METIS agglomerates built in 13.9057 seconds [Wall Clock]

Release mode. For the tree, I measure the time to:

• pack it
• visit it
• assign subdomain ids to relevant cells (i.e. the true time to create the agglomerates).

With METIS, I measure the time of

 GridTools::partition_triangulation(n_subdomains,
                                         tria,
                                         SparsityTools::Partitioner::metis);

@luca-heltai Do you have some other specific tests for what concerns the timings in mind?

[fdrmrc]

I've added the 3D meshes and the relative benchmark file. I tested the convergence on a 3D cube agglomerated with the tree, so I'll merge.