[IndVars] Speculative fix for an assertion failure seen in bots

I don't have an IR sample which is actually failing, but the issue described in the comment is theoretically possible, and should be guarded against even if there's a different root cause for the bot failures.

llvm-svn: 366241

1 files changed, 6 insertions, 1 deletions

diff --git a/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp b/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
index 70508bf7525..f9fc698a4a9 100644
--- a/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
+++ b/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
@@ -2810,7 +2810,12 @@ bool IndVarSimplify::run(Loop *L) {
       if (isa<SCEVCouldNotCompute>(ExitCount))
         continue;
 
-      assert(!ExitCount->isZero() && "Should have been folded above");
+      // This was handled above, but as we form SCEVs, we can sometimes refine
+      // existing ones; this allows exit counts to be folded to zero which
+      // weren't when optimizeLoopExits saw them.  Arguably, we should iterate
+      // until stable to handle cases like this better.
+      if (ExitCount->isZero())
+        continue;
       
       PHINode *IndVar = FindLoopCounter(L, ExitingBB, ExitCount, SE, DT);
       if (!IndVar)