We all have to write loops that do something more interesting than just for_each.
In modern C++ it's often considered to be a good idea to write such loops as generic algorithms. On top of that we have a bunch of algorithms from the standard library that often do what you need.
This talk focuses on two related topics: benchmarking such generic algorithms and optimising after the benchmarks are written. It will show some typical problems/mistakes that occure when working on such algorithms, suggest approaches to benchmarking and share a few success stories from libc++ and other production code.