Presently, Thread::SpinRelease uses OrderAccess::fence() followed by an ordinary store. It also has a bunch of commentary stating the desired semantics are those of OrderAccess::release_store. Using a full fence is overkill on most platforms.