Closed Alex-Vasile closed 5 months ago
Just to clarify, the optimization was to replace the cross product operation implemented using np.cross() with inline/algebraic formulation, removing the overhead of a function call.
The overhead and the extra cost needed to obtained the generality afforded by np.cross.
Both
_calculate_facet_normals
and_calculate_facet_sizes
usednp.cross
extensively. The overhead of calling the general function was very high, this is 5x faster which makes a big difference for large meshes.Will need to investigate further how to rework the functions as they take up a very large percentage of the total mesh conversion time.