[llvm] [Thumb] Resolve FIXME: We can trust live-in information from successor blocks now (PR #84111)

[Craig Topper via llvm-commits]

topperc wrote:

> > > > The PR21403 mentioned in the comment still seems to reproduce. It says R6 is missing from the liveins lists. R6 in those cases is reserved and being used as a base pointer. Reserved registers probably aren't required to be in the liveins list. Not sure if R6 is relevant to the code being changed here.
> > > 
> > > 
> > > Can you tell me the case this happens in? I cannot find it.
> > 
> > 
> > Sorry 21403 is the PR number in github. The number in the comment is the bugzilla number. #21403
> 
> So I used the .ll file used to repro the issue, converted it to modern synax via only addressing the errors llc gave me which oonly consisted of removing the metadata keywords themselves (but not the actual metadata) and the getelementptr/load syntax.
> 
> I compared the output before and after the patch. No difference.

Agreed that test would not be affected by this patch.

The title of this patch is "We can trust live-in information from successor blocks now" implying something is different today than it was when that FIXME was written. The test file from https://github.com/llvm/llvm-project/issues/21403 still has the same live-in behavior today as it did when the bug was filed.

Was that livein issue misinterpreted and was never really an issue?

https://github.com/llvm/llvm-project/pull/84111