[llvm] [NVPTX] Generalize and extend upsizing when lowering 8/16-bit-element vector loads/stores (PR #119622)

[Drew Kersnar via llvm-commits]

================
@@ -3162,122 +3234,80 @@ NVPTXTargetLowering::LowerSTOREVector(SDValue Op, SelectionDAG &DAG) const {
   SDLoc DL(N);
   EVT ValVT = Val.getValueType();
 
-  if (ValVT.isVector()) {
-    // We only handle "native" vector sizes for now, e.g. <4 x double> is not
-    // legal.  We can (and should) split that into 2 stores of <2 x double> here
-    // but I'm leaving that as a TODO for now.
-    if (!ValVT.isSimple())
-      return SDValue();
-    switch (ValVT.getSimpleVT().SimpleTy) {
-    default:
-      return SDValue();
-    case MVT::v2i8:
-    case MVT::v2i16:
-    case MVT::v2i32:
-    case MVT::v2i64:
-    case MVT::v2f16:
-    case MVT::v2bf16:
-    case MVT::v2f32:
-    case MVT::v2f64:
-    case MVT::v4i8:
-    case MVT::v4i16:
-    case MVT::v4i32:
-    case MVT::v4f16:
-    case MVT::v4bf16:
-    case MVT::v4f32:
-    case MVT::v8f16: // <4 x f16x2>
-    case MVT::v8bf16: // <4 x bf16x2>
-    case MVT::v8i16:  // <4 x i16x2>
-      // This is a "native" vector type
-      break;
-    }
+  auto VectorLoweringParams = tryGetVectorLoweringParams(ValVT);
----------------
dakersnar wrote:

Keeping the current method, I think with only two elements std::pair is fine

https://github.com/llvm/llvm-project/pull/119622