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

[Artem Belevich via llvm-commits]

================
@@ -1577,7 +1584,7 @@ bool NVPTXDAGToDAGISel::tryLDGLDU(SDNode *N) {
       NumElts /= 2;
     } else if (OrigType == MVT::v4i8) {
       EltVT = OrigType;
-      NumElts = 1;
+      NumElts /= 4;
----------------
Artem-B wrote:

It should also have an assert that NumElts is a multiple of 4.

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