[llvm] [AArch64] Select ADDHN instead of SHRN when possible (PR #225094)

[via llvm-commits]

================
@@ -1351,3 +1351,84 @@ define <4 x i32> @neg_narrow_i32(<4 x i64> %a) {
   %vshrn_n = trunc nuw <4 x i64> %s to <4 x i32>
   ret <4 x i32> %vshrn_n
 }
+
+define <8 x i8> @addhn_shift_i8(ptr %a) nounwind {
+; CHECK-LABEL: addhn_shift_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr q0, [x0]
+; CHECK-NEXT:    addhn v0.8b, v0.8h, v0.8h
+; CHECK-NEXT:    ret
+  %1 = load <8 x i16>, ptr %a
----------------
KRM7 wrote:

Sure, I've updated the tests

https://github.com/llvm/llvm-project/pull/225094