[llvm] [NVPTX] Make i16x2 a native type and add supported vec instructions (PR #65432)

[Artem Belevich via llvm-commits]

================
@@ -5198,6 +5296,7 @@ static void ReplaceLoadVector(SDNode *N, SelectionDAG &DAG,
   case MVT::v4f16:
   case MVT::v4f32:
   case MVT::v8f16: // <4 x f16x2>
+  case MVT::v8i16: // <4 x i16x2>
----------------
Artem-B wrote:

Looks like we're missing bf16 here.

https://github.com/llvm/llvm-project/pull/65432