Benchmark Design for Robust Profile-Directed Optimization

This work proposes an evaluation methodology appropriate for profile-guided optimization based on cross-validation, a methodology from machine learning that takes input sensitivity into account, and provides a measure of the generalizability of results.