[llvm] [SelectionDAG] Fold (icmp eq/ne (shift X, C), 0) -> (icmp eq/ne X, 0) (PR #88801)

[Björn Pettersson via llvm-commits]

================
@@ -4282,6 +4282,38 @@ SelectionDAG::computeOverflowForSignedMul(SDValue N0, SDValue N1) const {
   return OFK_Sometime;
 }
 
+SDNodeFlags SelectionDAG::computeShiftFlags(SDValue Op) const {
+  assert((Op.getOpcode() == ISD::SRL ||  Op.getOpcode() == ISD::SRA ||
+          Op.getOpcode() == ISD::SHL) &&
+         "Expecting an SHL/SRL/SRA operation.");
+  SDValue Op0 = Op.getOperand(0);
+  SDNodeFlags Flags = Op->getFlags();
+
----------------
bjope wrote:

I've tried to guard the most expensive computations below by checking if the flags already are present. So not quite sure what you propose.

https://github.com/llvm/llvm-project/pull/88801