Correction: those numbers are correct, but the PNG that I linked before is for approximately 1/4 of that data. Try it on your phone.
Ok, that explains things :). Please compare it with q-hull and s-hull, and perhaps with whatever CGAL
offers.
Looks like nobody wants to look at the benchmarks: http://svn.boost.org/svn/boost/sandbox/gtl/doc/voronoi_benchmark.htm I'm not hiding anything; I posted most of the code. But this test was not
constructed as a benchmark, so don't read too much into it.
I suspect that the main difference between this test system and yours is
the cache size.
Yes, you should be right. I just wanted to clarify for the users that Voronoi is indeed fast. Have a look at how Boost.Graph does "bundled properties":
Thanks, I've read it and I like it a lot. However the main problem is that mapping between input geometries and output cells is not one-to-one for Voronoi of segments, because each input segment is treated as three input sites (segment itself and both its endpoints). Nevertheless I am going to investigate "bundled properties" a bit more. Regards, Andrii