local-waker (perf): Avoid Waker::clone on 2nd registration

[nurmohammed840]

No description provided.

[JohnTitor]

Do you have any benchmark results against this? It'd be great if we could have cofidence on this change.

[JohnTitor]

Safety note should be added.

[nurmohammed840]

Do you have any benchmark results against this? It'd be great if we could have cofidence on this change.

It's hard to measure actual impact.

However Waker::clone is expensive, takio and async-task implementation require Atomic instructions. For example:
ArcWake require Arc::clone for each Waker::clone call.

timeout, join!, select! trigger repeated re-registration. and if they’re inside a loop. Multiple that for each iteration.

Waker instance very unlikely to change (In practice, I would say never).
That's a lot of unnecessary Atomic instructions.