Optimizing Program Performance via Similarity, Using Feature-aware and Feature-agnostic Characterization Approaches