[llvm] [DAGCombiner] Share byte provider steps with AMDGPU target (PR #221959)

[Matt Arsenault via llvm-commits]

================
@@ -15184,30 +15185,16 @@ calculateByteProvider(const SDValue &Op, unsigned Index, unsigned Depth,
   if (Index > BitWidth / 8 - 1)
     return std::nullopt;
 
+  auto Recurse = [&](SDValue NextOp, unsigned NextIndex) {
+    return calculateByteProvider(NextOp, NextIndex, Depth + 1, StartingIndex);
----------------
arsenm wrote:

Avoid this lambda and just inline all the uses? 

https://github.com/llvm/llvm-project/pull/221959