[llvm] [AArch64] Optimize vector multiplications by certain constants for v2i64 (PR #183827)

[David Green via llvm-commits]

================
@@ -5919,6 +5919,44 @@ static unsigned selectUmullSmull(SDValue &N0, SDValue &N1, SelectionDAG &DAG,
   return 0;
 }
 
+// Transform mul<v2i64, splat(2^n +-1)> into a SHL and ADD/SUB
+// this transormation is much faster when vector mul is not supported
+static SDValue convertMulToShlAdd(SDNode *N, SelectionDAG &DAG) {
+  const SDNode *Operand = N->getOperand(1).getNode();
+  APInt SplatValue;
+  ISD::isConstantSplatVector(Operand, SplatValue);
+
+  // Not a constant splat so should just stay as a mulitplcation operation
+  if (!SplatValue.getBoolValue())
----------------
davemgreen wrote:

`if (!ISD::isConstantSplatVector(Operand, SplatValue) || !SplatValue.getBoolValue())`

https://github.com/llvm/llvm-project/pull/183827