Allow normal function results of @yield_once coroutines

[asl]

This adds SIL-level support and LLVM codegen for normal results of a coroutine.

The main user of this will be autodiff as VJP of a coroutine must be a coroutine itself (in order to produce the yielded result) and return a pullback closure as a normal result.

The LLVM IR codegen depends on the following patch from LLVM mainline: swiftlang/llvm-project@51d5d7b (that is already a part of apple/llvm-project/next)

For now only direct results are supported, but this seems to be enough for autodiff purposes.

[asl]

Tagging @rxwei @BradLarson

[asl]

@rjmccall @hborla @slavapestov @xedin Note that we cannot run the tests as we need to have the corresponding LLVM change in stable branch first. However, then we'd also need to make a corresponding IRgen change (as LLVM API changes are not backward compatible), so we're having mutual interdependence between changes.

Let me know how this is usually resolved in swift repo.

[rxwei]

What's the corresponding LLVM branch this should be tested against? I think you can do cross-PR testing, for example:

apple/llvm-project#<PR number>
@ swift-ci please test

[asl]

What's the corresponding LLVM branch this should be tested against? I think you can do cross-PR testing, for example:

apple/llvm-project#<PR number>
@ swift-ci please test

Ah, ok. Let me try this! Thanks.

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[asl]

@rxwei @rjmccall @hborla @slavapestov @xedin
Tests passed (cross-repo), please check when you will have a chance

[asl]

@rxwei @rjmccall @hborla @slavapestov @xedin
ping

[hborla]

Hey @asl, apologies for the delay, many folks have been out of office the last few weeks. I'm not the right person to review this change (and I suspect Pavel isn't either; we were only automatically tagged because of the minor change in AST/Types.h). @rjmccall or @jckarter are probably the right reviewers, and they are both OOO this week.

[asl]

Hey @asl, apologies for the delay, many folks have been out of office the last few weeks. I'm not the right person to review this change (and I suspect Pavel isn't either; we were only automatically tagged because of the minor change in AST/Types.h). @rjmccall or @jckarter are probably the right reviewers, and they are both OOO this week.

Thanks! Yeah, waiting mostly for @rjmccall – we discussed this during the last US LLVM Dev Meeting.

[tbkka]

CC: @nate-chandler and @atrick

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test macos platform

[asl]

@rjmccall ping

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[rjmccall]

I'm not sure I get the LLVM IR pattern that we're emitting here. There's a coro.end.results instruction that may or may not be tied to the coro.end with the token? What do we do to register the direct results on the unwind path?

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[asl]

@rxwei @rjmccall @hborla @jckarter

Is there something that could be done to move this PR forward? It's in the review for ~2 months and is currently blocking submission of several other PRs. And it is required to be rebased each time module format is changed...

Thanks!

[asl]

@rjmccall Friendly weekly ping. Thanks!

[asl]

@rjmccall @rxwei @hborla Another ping, thanks!

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[rjmccall]

Alright, let's move forward here.

[asl]

Alright, let's move forward here.

Thanks a lot! Much appreciated

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test

[asl]

Please test with swiftlang/llvm-project#7797
@swift-ci please test