Tuning LLVM's SLP Vectorizer Cost Model / hacker news