Making profile-guided optimization reproducible

This blog post discusses the challenges of making profile-guided optimization (PGO) reproducible in package managers like Nix and Guix. The author explains their investigation into using Rewind, a deterministic virtual machine, to achieve reproducibility during the training run of PGO. The post details the practical implementation of this method, including code snippets and a breakdown of the results, highlighting how it allows optimizations to be made without compromising reproducibility.