Optimizing Binaries for Faster Performance?

Bolt was developed by facebook, and it's been used internally there for a while. It is used to optimize the code layout for binaries generated from g c c and clang. And they have some nice graps here where they're showing improvement that you get from bolt alone versus performance improvement from p g o. The closest corollary i can think of in my own work is when i've used profile guided optimization incorrectly.