[llvm] [AArch64][GlobalISel] Lower G_BUILD_VECTOR to G_INSERT_VECTOR_ELT (PR #105686)

[Thorsten Schütt via llvm-commits]

================
@@ -1048,6 +1048,45 @@ void applyLowerVectorFCMP(MachineInstr &MI, MachineRegisterInfo &MRI,
   MI.eraseFromParent();
 }
 
+// Intend to match the last part of
+// AArch64TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG)
+bool matchLowerBuildToInsertVecElt(MachineInstr &MI, MachineRegisterInfo &MRI) {
+  assert(MI.getOpcode() == TargetOpcode::G_BUILD_VECTOR &&
+         "Expected G_BUILD_VECTOR instruction");
+  bool isConstant = true;
+
+  // Check if the values are the same
+  for (unsigned i = 1; i < MI.getNumOperands(); i++) {
+    auto ConstVal =
+        getAnyConstantVRegValWithLookThrough(MI.getOperand(i).getReg(), MRI);
+    if (!ConstVal.has_value()) {
+      isConstant = false;
+    }
----------------
tschuett wrote:

`break;`

https://github.com/llvm/llvm-project/pull/105686