[llvm] [NVPTX] Promote v2i8 to v2i16 (PR #111189)

[Artem Belevich via llvm-commits]

================
@@ -325,10 +325,17 @@ define <3 x i8> @byte3() {
   ret <3 x i8> zeroinitializer
 }
 
-; FIXME: This test causes a crash. 
-; define <2 x i8> @byte2() {
-;   ret <2 x i8> zeroinitializer
-; }
+define <2 x i8> @byte2() {
+; CHECK-LABEL: byte2(
+; CHECK:       {
+; CHECK-NEXT:    .reg .b32 %r<2>;
+; CHECK-EMPTY:
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    mov.b32 %r1, 0;
+; CHECK-NEXT:    st.param.b32 [func_retval0+0], %r1;
+; CHECK-NEXT:    ret;
+  ret <2 x i8> zeroinitializer
----------------
Artem-B wrote:

It may be useful to use a non-zero value, so we can verify that we've stored the vector elements in correct order.
That, actually applies to all the tests here, so it may be worth fixing that in a separate patch.

https://github.com/llvm/llvm-project/pull/111189