Rebase patch after resect changes.

Update pr35210.ll after recent upstream changes.

Thank you for the review!

Dropping this, since the issue is not resolvable in general AFAICT. Please see D51960.

Moving the MSSA after cloning earlier to avoid the ValueMap invalidation.
Adding another test.
And ping.

Final update before commit: changing a cast_or_null to dyn_cast_or_null for cloned instructions. LoopRotate can place Value in the map, as the clone of an instruction.

Nit: remove braces.

Address comment. Thank you for the review!

Adding third test-case for new failure.

Avoid duplicates (use set).

Add test for second failure.

Need to bypass geps when traversing users, use worklist.
Adding test in dependent patch.

Update test and format.

Enable tests. Add new tests to check analysis is preserved.

Thank you for the patience on this review! :)

Small update triggered by a test in rotate (patch to be uploaded soon).

Fix bug introduced by refactoring.

Missed an update.

Fix nit.

This LGTM.
FWIW, I think we're having an underlying problem and this and the previous PRs may be related.
In unswitchTrivialSwitch, at the very beginning we're removing switch cases and querying LoopInfo to get the outer loop for each switch successor without updating it. LoopInfo is updated only at the very end in hoistLoopToNewParent. Adding an artificial assert to check if LI should be updated after removing switch cases, triggers on 2/3 tests from D51375 and existing trivial-unswitch tests. Looking into it further.

Address comments. Undo const-ifies no longer needed.

Updated test, thank you for the comments!

Address comments.

I'm planning to add a few more unittests, but sending the updates now for additional comments.

Thanks a lot of the comments, working to address them.

Thank you for the review!

Thank you for the review!

Rebase and cleanup. Ready to review.

Dropping this, updater is movable, some mishap on my side.

Thank you for the review!
I'll submit this one as is, and the usages will follow shortly.

This is great, thank you for the cleanup!
LGTM.

Update to normal constructor.

Update to simple constructor.

Nope, don't need that.

Updated to allocate mssaupdater.

My bad, previous update won't work, updater needs to be allocated. Please comment on the use of make_unique (= operator in Optional does an std::move).

Address comment.

Address comments.