I haven't profiled the code, but I don't think better predictability factors significantly in the computed goto approach. Because of the global history table pr

I haven't profiled the code, but I don't think better predictability factors significantly in the computed goto approach. Because of the global history table probably both sample implementations are likely perfectly predicted. The advantage is that CPUs have an hard limit on how many taken jumps per cycle they can execute, as it requires resteering the fetch unit even before the jump is decoded. For example typically intel cpus can only take one jump every other cycle. It is possible that the apple cpu in question has a tighter limit than the amd one.