Allow empty IGs when allocating unwind info for a fragment

[kunalspathak]

The empty IGs are sometimes created before the loop that we are trying to align to ensure the loop size calculation is correct.

Fixes: #62250

[kunalspathak]

@dotnet/jit-contrib

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT
See info in area-owners.md if you want to be subscribed.

Issue Details

---

The empty IGs are sometimes created before the loop that we are trying to align to ensure the loop size calculation is correct.

Fixes: #62250

Author:	kunalspathak
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[BruceForstall]

I apologize, since I suggested to you to loosen the assert, but after looking at the test case for a while, I think this is the wrong fix. I think there's a bug in the split algorithm (emitSplit) for zero sized IGs. Do you mind if I take the issue?

[kunalspathak]

Now that I was able to debug the issue with your minimum repro, I think the problem is that we should not report the candidate if it has size zero. So, may be something like this? I verified and it fixes the issue. Were you thinking similar fix?

@@ -2764,6 +2764,10 @@ void emitter::emitSplit(emitLocation*         startLoc,
         {
             // We can't update the candidate
         }
+        else if (ig->igSize == 0)
+        {
+            // We can't update the candiate because it is empty.
+        }
         else
         {
             igLastCandidate = ig;

[BruceForstall]

@kunalspathak Almost; here's my proposal: #62931

[kunalspathak]

#62931 will take care of this.