Thee optimization in question here was not obvious at all. It's a bs clang codegen quirk