Link to the (great) CppCon talk: https://www.youtube.com/watch?v=bSkpMdDe4g4

The one thing I would say is: as you dig more into C++ perf, you can develop an intuition for the kinds of things the compiler can and can't optimize. For instance, the C++ example from the article uses `std::unordered_map`, which is just an absolute mess from a cache locality perspective—the best compilers today (or of the foreseeable future) can't do a thing to fix that. Improving the programmer's choice of data structure is just not on the radar. :(