[llvm] [InstCombine] Optimize x * !x to 0 for vector #84608 (PR #84792)

[Craig Topper via llvm-commits]

================
@@ -198,6 +198,22 @@ Instruction *InstCombinerImpl::visitMul(BinaryOperator &I) {
   if (SimplifyAssociativeOrCommutative(I))
     return &I;
 
+  // mul (sext (icmp eq x, 0)), x -> 0
+  Value *SExtOp, *V0;
+  if ((match(Op0, m_SExt(m_Value(SExtOp))) && match(Op1, m_Value(V0))) ||
+      (match(Op1, m_SExt(m_Value(SExtOp))) && match(Op0, m_Value(V0)))) {
+    Constant *CV;
+    ICmpInst::Predicate Pred;
+    Value *V1;
+    const APInt *IV;
+    if ((match(SExtOp, m_ICmp(Pred, m_Value(V1), m_Constant(CV))) ||
+         match(SExtOp, m_ICmp(Pred, m_Constant(CV), m_Value(V1)))) &&
----------------
topperc wrote:

Constants will get canonicalized to the RHS side. Do we need this check?

https://github.com/llvm/llvm-project/pull/84792