Faster uncontended atomics should improve uncontended lock performance too, at least if you’re using pthread mutexes (they’re implemented via atomic operations

Faster uncontended atomics should improve uncontended lock performance too, at least if you’re using pthread mutexes (they’re implemented via atomic operations with an os_unfair_lock fallback for the contended case on Apple platforms).