arm64 added a load acquire instruction which I think it is fast enough on actual hardware that might not be worth bothering with consume. If it isn't, then load

arm64 added a load acquire instruction which I think it is fast enough on actual hardware that might not be worth bothering with consume. If it isn't, then load-relaxed plus atomic_signal_fence might be your best bet. Good luck!