I changed update_components to use a fast union-find disjoint set implementation with wait-free and path-compression properties.
I benchmark a dataset with 5M vertices in texrecon at 30% less total runtime than with the old implementation.
I'm happy to work on issues/problems you might find.
I changed update_components to use a fast union-find disjoint set implementation with wait-free and path-compression properties. I benchmark a dataset with 5M vertices in texrecon at 30% less total runtime than with the old implementation. I'm happy to work on issues/problems you might find.