[concurrency] SILGen: emit hop_to_executor instructions

[eeckstein]

Emit hop_to_executor instruction in the prolog of actor-isolated async functions and after all async calls in such functions.

rdar://problem/70299168

[slavapestov]

I think we also need to emit hop_to_executor after an apply instruction, don't we? I was thinking it would be better to do this in a mandatory SIL pass, since SILGen emits applies in various places that would all need this handling, such as protocol witness thunks. The only bit of plumbing required then would be storing the actor isolation kind in the SILFunction instead of pulling it out of the Decl.

[slavapestov]

If you're going to keep doing this in SILGen, how about passing this down as a parameter to emitProlog()? The caller knows the function being emitted, and I'd like to remove FunctionDC eventually since it's used for various hacks that should go away.

[eeckstein]

Can I leave this up to you? emitProlog is called from multiple places (which I'm not familiar with) and I don't want to introduce a bug with this refactoring.

[eeckstein]

I think we also need to emit hop_to_executor after an apply instruction, don't we? I was thinking it would be better to do this in a mandatory SIL pass, since SILGen emits applies in various places that would all need this handling, such as protocol witness thunks.

That's a good point. Yes, we need hop_to_executor after each async apply (just async applies) - and only in functions, which actually need to be actor isolated, e.g. because they access an actor property. I assume that this is not the case for all the thunks we generate.
Anyway, it's easy to emit the after-apply hop_to_executors in a mandatory pass. The main thing of this PR is to set up the prolog (which would be hard outside SILGen).

[eeckstein]

@swift-ci test

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 523840a834303189feb1e8a8a99b7365af7d0255

[eeckstein]

@swift-ci smoke test