Morton hashing still requires while loop in some cases in Curvilinear Grid

[fluidnumerics-joe]

In discussion with @erikvansebille , there is desire to remove the while loop in the curvilinear grid cell search immediately following the spatial hash query. In the query method, I decided to relax the condition to match the particle's morton encoded position with an exact morton code from source grid as this strict matching caused a number of tests to fail. The thinking is that if a particle position has a "nearby" morton encoding then it will be spatially close to a cell centroid with a morton encoding in close proximity.

However, relaxing strict matching prevents us from safely removing the while loop in the curvilinear grid cell search. It sounds like we need to spend some time reviewing quantization and bit interleaving strategies to see if we can enforce exact morton code matching on particle positions.

[fluidnumerics-joe]

I'm looking into counting the number of iterations of the whole loop and how often it happens for some of the examples and tests already defined. If you have examples that demonstrate inexact matches from the spatial hash query, please provide minimal reproducers here.

In thinking on this a bit, do we really want to get rid of the while loop ? If someone provides a face ID for a particle, having this while loop helps execute the nearest neighbor search.

@erikvansebille

[erikvansebille]

I just uploaded a minimal example (i.e. a breaking case) in #2177. It's one of the unit tests: tests/v4/test_index_search.py::test_grid_indexing_fpoints