[llvm] [AArch64] Enable maximize vector bandwidth for Neoverse-V1, V2 and N1 (PR #166748)

[Sander de Smalen via llvm-commits]

================
@@ -371,8 +371,15 @@ AArch64TTIImpl::getInlineCallPenalty(const Function *F, const CallBase &Call,
 bool AArch64TTIImpl::shouldMaximizeVectorBandwidth(
     TargetTransformInfo::RegisterKind K) const {
   assert(K != TargetTransformInfo::RGK_Scalar);
-  return (K == TargetTransformInfo::RGK_FixedWidthVector &&
-          ST->isNeonAvailable());
+  switch (ST->getProcFamily()) {
+  case AArch64Subtarget::NeoverseV2:
+  case AArch64Subtarget::NeoverseV1:
+  case AArch64Subtarget::NeoverseN1:
+    return true;
----------------
sdesmalen-arm wrote:

Two things:
* This is overriding the default for fixed-length vectors, is that intentional?
* For scalable vectors, I'd rather we make it the default to enable this unless we have found an instance where maximising bandwidth is not profitable.

https://github.com/llvm/llvm-project/pull/166748