[llvm] [AArch64] Correct the throughput of reductions on Neoverse V3ae (PR #221151)

[Asher Dobrescu via llvm-commits]

================
@@ -2038,9 +2038,9 @@
 # CHECK-NEXT:  1      2     0.50                        saddlt	z0.h, z1.b, z2.b
 # CHECK-NEXT:  1      2     0.50                        saddlt	z29.s, z30.h, z31.h
 # CHECK-NEXT:  1      2     0.50                        saddlt	z31.d, z31.s, z31.s
-# CHECK-NEXT:  6      9     4.00                        saddv	d0, p7, z31.b
-# CHECK-NEXT:  4      8     2.00                        saddv	d0, p7, z31.h
-# CHECK-NEXT:  4      6     2.00                        saddv	d0, p7, z31.s
+# CHECK-NEXT:  4      9     2.00                        saddv	d0, p7, z31.b
+# CHECK-NEXT:  3      8     1.50                        saddv	d0, p7, z31.h
+# CHECK-NEXT:  2      6     1.00                        saddv	d0, p7, z31.s
----------------
Asher8118 wrote:

I see, I was unaware of this. In that case I'm happy to approve.

https://github.com/llvm/llvm-project/pull/221151