This doesn't seem to be an issue for x86, possibly due to the way Clang is setup for AArch64, possibly just an artefact of llvm's code gen. Either way, rather than rely on llvm eliminating these jumps ...