[BasicBlockUtils] Check for unreachable preds before updating LI in UpdateAnalysisInformation

[anna]

Summary

We are incorrectly updating the LI when loop-simplify generates
dedicated exit blocks for a loop. The issue is that there's an implicit
assumption that the Preds passed into UpdateAnalysisInformation are
reachable. However, this is not true and breaks LI by incorrectly
updating the header of a loop.

One such case is when we generate dedicated exits when the exit block is
a landing pad (through SplitLandingPadPredecessors). There maybe other
cases as well, since we do not guarantee that Preds passed in are
reachable basic blocks.

The added test case shows how loop-simplify breaks LI for the outer loop (and DT in turn)
after we try to generate the LoopSimplifyForm.

[davide]

I think this is on the right track, but, can you please simplify the test case so I can read it better :) ?

[anna]

Simplified the testcase to make the issue clearer.

[anna]

In D41519#962880, @davide wrote:

I think this is on the right track, but, can you please simplify the test case so I can read it better :) ?

Yes, absolutely agree. Actually, the testcase was a *manually simplified* version of bugpoint's output - bugpoint's simplified output had 230 lines in the testcase!
Clearly, the manual simplification (75 lines) could still be improved :)

As an aside I think bugpoint needs improvement around invokes and landing blocks (unless there's some specific option to use here other than -compile-custom).

[anna]

ping

[davide]

LGTM, thanks Anna. (sorry for the delay, I was in Europe for the end of the year).

[anna]

In D41519#965788, @davide wrote:

LGTM, thanks Anna. (sorry for the delay, I was in Europe for the end of the year).

Thanks Davide! Happy new year :)

lib/Transforms/Utils/BasicBlockUtils.cpp

static void UpdateAnalysisInformation(BasicBlock *OldBB, BasicBlock *NewBB,
 
   Loop *L = LI->getLoopFor(OldBB);
 
   // If we need to preserve loop analyses, collect some information about how
   // this split will affect loops.
   bool IsLoopEntry = !!L;
   bool SplitMakesNewLoopHeader = false;
   for (BasicBlock *Pred : Preds) {
+    // Preds that are not reachable from entry should not be used to identify if
+    // OldBB is a loop entry or if SplitMakesNewLoopHeader. Blocks that do not
+    // have any predecessors are not within any loops, so we incorrectly mark
+    // SplitMakesNewLoopHeader as true and make the NewBB the header of some
+    // loop. This breaks LI.
+    if (!DT->isReachableFromEntry(Pred))
+      continue;
     // If we need to preserve LCSSA, determine if any of the preds is a loop
     // exit.
     if (PreserveLCSSA)
       if (Loop *PL = LI->getLoopFor(Pred))
         if (!PL->contains(OldBB))
           HasLoopExit = true;
 
     // If we need to preserve LoopInfo, note whether any of the preds crosses

test/Transforms/LoopSimplify/unreachable-loop-pred.ll

header:                                       ; preds = %if.end206, %cond.end66, %if.end23
   br label %while.body115
 
 while.body115:                                    ; preds = %9, %if.end192, %if.end101
   br i1 undef, label %header, label %while.body115
 
 foo:
   br label %while.body115
 }
+
+; When loopsimplify generates dedicated exit blocks for blocks that are landing
+; pads, we should not incorrectly mark that exit block as the header for the outer
+; loop. This happens when it gets confused with the unreachable pred to the inner
+; loop exit block
+define align 8 void @baz() personality i32* ()* @wobble {
+bb:
+  br label %outerH
+
+outerH:
+  %tmp30 = invoke align 8 i8 * undef()
+          to label %outerLExiting1 unwind label %unwindblock
+
+unwindblock:
+  %tmp34 = landingpad { i8*, i32 }
+          cleanup
+  ret void
+
+outerLExiting1:
+  br i1 undef, label %outerLExiting2, label %exit
+
+exit:
+  ret void
+
+outerLExiting2:
+  invoke void @foo() 
+          to label %innerPreheader unwind label %innerLExit
+
+innerPreheader:                                             ; preds = %outerLExiting2
+  br label %innerH
+
+innerH:                                             ; preds = %innerLatch, %innerPreheader
+  %tmp50 = invoke i8 * undef()
+          to label %innerExiting unwind label %innerLExit
+
+innerExiting:                                             ; preds = %bb57
+  invoke align 8 void @foo() 
+          to label %innerLatch unwind label %innerLExit
+
+innerLatch:                                             ; preds = %innerExiting
+  br label %innerH
+
+; This is a predecessor for inner loop exit block which is a landing pad.
+unreachableB:                                             ; No predecessors!
+  %tmp62 = invoke i8 * undef()
+          to label %unwindblock2 unwind label %innerLExit
+
+unwindblock2:                                             ; preds = %unreachableB
+  ret void
+
+innerLExit:
+  %tmp65 = landingpad { i8*, i32 }
+          cleanup
+  invoke void @foo() 
+          to label %outerLatch unwind label %bb66
+
+bb66:                                             ; preds = %innerLExit
+  %tmp67 = landingpad { i8*, i32 }
+          cleanup
+  ret void
+
+
+outerLatch:                                             ; preds = %bb70
+  br label %outerH
+}
+
+; Function Attrs: nounwind
+declare i32* @wobble()
+
+; Function Attrs: uwtable
+declare void @foo()