Interesting, I was lacking this context. Could you provide me with more information about this? I only saw atomic_shared_ptr come up in discussions about bugs up to now.
The target of this optimization is low-latency code. Rendezvous will not work for that.