[X86] Don't compare i128 through vector if construction not cheap (PR…

…41971)

Fix for https://bugs.llvm.org/show_bug.cgi?id=41971. Make the
combineVectorSizedSetCCEquality() transform more conservative by
checking that the bitcast to the vector type will be cheap/free
for both operands. I'm considering it cheap if it's a constant,
a load or already a vector. I've dropped the explicit check for
f128 because it should fall out naturally (in the cases where
it'd be detrimental).

Differential Revision: https://reviews.llvm.org/D62220

llvm-svn: 361352

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -41349,9 +41349,14 @@ static SDValue combineVectorSizedSetCCEquality(SDNode *SetCC, SelectionDAG &DAG,
   if (isNullConstant(Y) && !IsOrXorXorCCZero)
     return SDValue();
 
-  // Bail out if we know that this is not really just an oversized integer.
-  if (peekThroughBitcasts(X).getValueType() == MVT::f128 ||
-      peekThroughBitcasts(Y).getValueType() == MVT::f128)
+  // Don't perform this combine if constructing the vector will be expensive.
+  auto IsVectorBitCastCheap = [](SDValue X) {
+    X = peekThroughBitcasts(X);
+    return isa<ConstantSDNode>(X) || X.getValueType().isVector() ||
+           X.getOpcode() == ISD::LOAD;
+  };
+  if ((!IsVectorBitCastCheap(X) || !IsVectorBitCastCheap(Y)) &&
+      !IsOrXorXorCCZero)
     return SDValue();
 
   // TODO: Use PXOR + PTEST for SSE4.1 or later?