[llvm] [AMDGPU]Refactor `lowerWaveReduce` for maintainability (PR #189223)

[Matt Arsenault via llvm-commits]

================
@@ -5689,6 +5760,28 @@ static MachineBasicBlock *lowerWaveReduce(MachineInstr &MI,
   unsigned Stratergy = static_cast<unsigned>(MI.getOperand(2).getImm());
   enum WAVE_REDUCE_STRATEGY : unsigned { DEFAULT = 0, ITERATIVE = 1, DPP = 2 };
   MachineBasicBlock *RetBB = nullptr;
+  auto ExtractSubRegs = [&](MachineInstr &MI, MachineOperand &Op,
+                            const TargetRegisterClass *SrcRC)
+      -> std::tuple<MachineOperand, MachineOperand> {
----------------
arsenm wrote:

Avoid value Machineoperands. These also probably shouldn't' be lambdas 

https://github.com/llvm/llvm-project/pull/189223