[llvm] [AArch64] Increase scatter overhead on Neoverse-V2 (PR #101296)

[David Green via llvm-commits]

================
@@ -0,0 +1,8 @@
+; RUN: opt -mtriple aarch64 -mcpu=neoverse-v2 -passes="print<cost-model>" -disable-output < %s 2>&1 | FileCheck %s
+; CHECK: Cost Model: Found an estimated cost of 52 for instruction: call void @llvm.masked.scatter.nxv4f32
----------------
davemgreen wrote:

Oh that's probably because it's scalable with no sve support. Can you add -mattr=+sve2 to make them comparable?

https://github.com/llvm/llvm-project/pull/101296