SGTM.

Oh, wow! Might I ask you add the same for fmul? We may get a revision like that next time :).
Thank you!

LGTM.

LGTM.

Thank you!

Thanks you for taking this on!

Minor updates.

Nit: remove F.

Add comment explaining invalidate rational.
GlobalsAA is stateless and is preserved, unless it's explicitly invalidated.

Address comment.

Not a bug fix, but a prerequisite for a change I'm working on for the new pass manager, requiring all passes to have an invalidate method.
I sent it for review to get another pair of eyes on the invalidate implementation in case I missed something.

Regarding handling the branching example, the solution fhahn proposes sounds reasonable to me.
I was thinking something similar, along the lines of: do the checks bottom-up from both second and third stores during the main algorithm, both find the dead store but do not postdominate it; don't abandon them, but keep this info (include the checks for no intervening reads) for after the main search. The data structure could be a Hashmap<PotentialDeadStore, ListOfDefsOrBlocksWhoFoundThisPotentialDeadStore>. If all paths are covered in the List for a PotentialDeadStore, then it's truly dead.

Just a general comment here: DSE shouldn't necessarily make all optimizations possible. This one looks like it should be caught by DCE. If it's not, perhaps it would be good to improve DCE. Alternatively, if it's cheap enough, add a comment that it's still worth adding it to DSE.

Thanks, this looks useful!

Thank you for the other patch! I added a few comments on the bottom up approach. I think overall it's a better high-level approach.

Some comments inline.

Sorry, I forgot about this one, thank you for the reminder!

lgtm, thank you!

This revision appears to be the root-cause of https://bugs.llvm.org/show_bug.cgi?id=44231.

Filed https://bugs.llvm.org/show_bug.cgi?id=44231 with a small reproducer.

This revision came back as the root-cause for an internal iOS crash.
It's hitting the following assertion:
assert.h assertion failed at llvm/lib/Target/ARM/ARMFrameLowering.cpp:468 in virtual void llvm::ARMFrameLowering::emitPrologue(llvm::MachineFunction &, llvm::MachineBasicBlock &) const: getMaxFPOffset(MF.getFunction(), *AFI) <= FPOffset && "Max FP estimation is wrong"

I have no expertise here, but this fixes PR44222 (many AArch64 internal crashes caused by D70673), so I'm going to go ahead and approve this to get the fix in tree.

Thank you for this fix. LGTM.

Combine 3 methods.

This looks like a great improvement, IMO.

Remove missed comment.

I updated the patch to be more generic, while still addressing this usecase.
The condition is: if there are any internal methods whose address is taken but for which we do not conclude that they do not modify globals, then cannot conclude lower than ModRef for any other (LocalLinkageCall, GlobalValue).
Tests included.

Set KnowNothing.

I think this is a great example of a unit test actually.
Getting to the scenario of forcing ScEv to cache the Load, so we can verify it is no longer properly cached after the Load is hoisted is harder to do in a lit test such that it's so easy to understand, like it is in this unit test.

Thank you for the patch! This LGTM as is.

Since I don't really understand the CodeExtractor, I'm relying on the author or another reviewer to confirm the correctness here.
This patch by itself looks good, but please wait to see if there are additional comments for the other reviewers before checking in.

Some comments, but it looks reasonable overall.

Update iterator var names.

Nit: rename optional argument.
I'd like to check this in, in an effort to clear my patch queue. Please let me know if you have any comments.