[llvm] [AArch64][ISel] Recover ADDHN from OR comparison masks (PR #210943)

[David Green via llvm-commits]

================

----------------
davemgreen wrote:

This one isn't really a multiclass, and isn't saving too much over just having the patterns written separately. You could also, ideally, replace int_aarch64_neon_addhn with the new node (LowerINTRINSIC_WO_CHAIN) and use the new node in the place of it above instead (in SIMDNarrowThreeVectorBHS).

https://github.com/llvm/llvm-project/pull/210943