[llvm] [AMDGPU] Add custom lowering of llvm.convert.from.arbitrary.fp for FP8 (PR #194144)

[Matt Arsenault via llvm-commits]

================
@@ -1430,6 +1430,10 @@ void VectorLegalizer::Expand(SDNode *Node, SmallVectorImpl<SDValue> &Results) {
       Results.push_back(DAG.getPOISON(Node->getValueType(0)));
     return;
   case ISD::CONVERT_FROM_ARBITRARY_FP:
+    if (SDValue Expanded = TLI.expandVectorNaryOpBySplitting(Node, DAG)) {
+      Results.push_back(Expanded);
+      return;
+    }
----------------
arsenm wrote:

You shouldn't need to do this. I do however see this pattern is increasing in frequency 

https://github.com/llvm/llvm-project/pull/194144