[llvm] [LV] Autovectorization for the all-in-one histogram intrinsic (PR #91458)

[Graham Hunter via llvm-commits]

================
@@ -1939,6 +1942,7 @@ class TargetTransformInfo::Concept {
                                               unsigned *Fast) = 0;
   virtual PopcntSupportKind getPopcntSupport(unsigned IntTyWidthInBit) = 0;
   virtual bool haveFastSqrt(Type *Ty) = 0;
+  virtual InstructionCost getHistogramCost(Type *Ty) = 0;
----------------
huntergr-arm wrote:

Committed in #97397

We may need a separate method at some point anyway to determine likely addressing modes, but AArch64 currently doesn't do that for gathers and scatters anyway. Something to do after the cost model refactoring to generate the cost from the plan.

https://github.com/llvm/llvm-project/pull/91458