[llvm] [LLVM][CodeGen] Add lowering for scalable vector bfloat operations. (PR #109803)

[Graham Hunter via llvm-commits]

================
@@ -11917,3 +11923,38 @@ bool TargetLowering::LegalizeSetCCCondCode(SelectionDAG &DAG, EVT VT,
   }
   return false;
 }
+
+SDValue TargetLowering::expandVectorNaryOpBySplitting(SDNode *Node,
+                                                      SelectionDAG &DAG) const {
+  EVT VT = Node->getValueType(0);
+  // Despite its documentation, GetSplitDestVTs will assert if VT cannot be
+  // split into two equal parts.
+  if (!VT.isVector() || !VT.getVectorElementCount().isKnownMultipleOf(2))
+    return SDValue();
+
+  EVT LoVT, HiVT;
+  std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(VT);
----------------
huntergr-arm wrote:

Not a request, more of an style option to match the use of auto elsewhere in the patch for you to consider.
```suggestion
  auto [LoVT, HiVT] = DAG.GetSplitDestVTs(VT);
```
That lets you get rid of the separate declaration line.

https://github.com/llvm/llvm-project/pull/109803