as stated in this page, why std::atomic_thread_fence() with full fence does not hold for StoreLoad operations ?
I understand that it is combination of Acquire fence and Release fence, but isn’t there a scenario you would want to have a fence between a Store operation and a Load operation ?