[llvm] [AArch64] Use ADDP tree for v16i8 to i16 bitmask extraction (PR #192974)

[Conor Kotwasinski via llvm-commits]

================
@@ -2,302 +2,131 @@
 ; RUN: opt -passes=expand-memcmp -mtriple=aarch64-none-linux-gnu -mattr=+neon -S < %s | llc -mtriple=aarch64-none-linux-gnu -mattr=+neon | FileCheck %s
 
 define i1 @combine_setcc_eq_vecreduce_or_v8i1(<8 x i8> %a) {
----------------
conorKotwasinski wrote:

You're right. The test runs opt -passes=expand-memcmp | llc but the script just invokes llc, so the regen was checking codegen for IR that hadn't been through expand-memcmp. Reverted and updated just the four _add_nsw checks that go through the v16i8 bitmask path.

https://github.com/llvm/llvm-project/pull/192974