Updating based on comments, forgot to upload this diff several months ago

Makes sense, I don't think this will cause any issues then.

I could write a test case that would become a problem if this was not implemented when https://reviews.llvm.org/D139338 is merged on top of it. But I can't write a test case that would fail without this check for the current version of the outliner.

In D139336#3975102, @paquette wrote:

Is this one of the compile time patches we discussed offline? If so, could you mention that + an overview of the improvements in the commit message?

exclude parent patch code

Fix diff order

LGTM with small nit.

There shouldn't be a case where there is a ret void and a ret <Constant Int> in a function. I am pretty sure that there will only more than one return value when the function has more than one exit block. In that case, there are multiple return values have an integer value. So the cases you laid out should be correct.

Yes, this was written to fix that crash. Should have noted it in the commit message.

Updating case when there are no predecessors to the outlined region form outside the region.

I don't know that much about this analysis pass to say if it is correct to skip dbg intrinsics like this. (And should for example lifetime instrinsics be skipped in the same way?)

In D122207#3455485, @bjope wrote:

Seen assertion failures downstream with this patch. Here is a reproducer:

Reduced reproducer here: https://godbolt.org/z/4jW3KsEns

In D121627#3396435, @hkmatsumoto wrote:

Since I don't have commit access, can you land this patch for me?

LGTM

https://github.com/llvm/llvm-project/issues/54212

Reducing test case

Updating with context

Updating with nits