chardan

I think it's at the very least been hinted at below (and there are other good points), but note that their implementation makes copies. If the copy ctor of type T is the least bit expensive, the STL (implementation in glibc) for this same benchmark is far faster than theirs. Try it. :)

A strategy to manage complexity by decomposing problems into hierarchical type relationships through the mechanisms of encapsulation (code and data together in an "object"), polymorphism (actions on abstract types trigger behaviors without knowledge of their specific implementation), and derivation (relationships between classes) such that specific behaviors increase toward the leaves and decrease toward the roots, leaving implementation and general behavior isolated at appropriate levels.