[llvm] [AArch64] Optimize vector slide shuffles with zeros to use shift instructions (PR #185170)

[David Green via llvm-commits]

================
@@ -1509,6 +1509,13 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
       }
     }
 
+    if (Subtarget->hasF16F32DOT()) {
----------------
davemgreen wrote:

Remove this, it looks like it comes from elsewhere.

https://github.com/llvm/llvm-project/pull/185170