[InstCombine] limit icmp transform to ConstantInt (PR28011)

In r271810 ( http://reviews.llvm.org/rL271810 ), I loosened the check
above this to work for any Constant rather than ConstantInt. AFAICT, 
that part makes sense if we can determine that the shrunken/extended 
constant remained equal. But it doesn't make sense for this later 
transform where we assume that the constant DID change. 

This could assert for a ConstantExpr:
https://llvm.org/bugs/show_bug.cgi?id=28011

And it could be wrong for a vector as shown in the added regression test.

llvm-svn: 271908

1 files changed, 5 insertions, 3 deletions

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index cb9c7e5b893..b3727e4bbe5 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -2459,12 +2459,14 @@ Instruction *InstCombiner::visitICmpInstWithCastAndCast(ICmpInst &ICmp) {
     return new ICmpInst(ICmp.getUnsignedPredicate(), LHSCIOp, Res1);
   }
 
-  // The re-extended constant changed so the constant cannot be represented
-  // in the shorter type. Consequently, we cannot emit a simple comparison.
+  // The re-extended constant changed, partly changed (in the case of a vector),
+  // or could not be determined to be equal (in the case of a constant
+  // expression), so the constant cannot be represented in the shorter type.
+  // Consequently, we cannot emit a simple comparison.
   // All the cases that fold to true or false will have already been handled
   // by SimplifyICmpInst, so only deal with the tricky case.
 
-  if (isSignedCmp || !isSignedExt)
+  if (isSignedCmp || !isSignedExt || !isa<ConstantInt>(C))
     return nullptr;
 
   // Evaluate the comparison for LT (we invert for GT below). LE and GE cases