https://github.com/aengelke commented: The generated code could be better. I still would like to see some actual (real) benchmark results as well as some data on the performance impact of only-slow-case/mixed-with-many-branch-misses. https://github.com/llvm/llvm-project/pull/148343