[llvm] [LV] Support strided memory accesses with a stride of -1 (PR #128718)

[Alexey Bataev via llvm-commits]

================
@@ -1312,6 +1313,22 @@ class LoopVectorizationCostModel {
            (SI && TTI.isLegalMaskedScatter(Ty, Align));
   }
 
+  /// Returns true if the target machine can represent \p V as a strided load
+  /// or store operation.
+  bool isLegalStridedLoadStore(Value *V, ElementCount VF) const {
----------------
alexey-bataev wrote:

I'm suggesting to use the prototype of isLegalMaskedLoad here, most of the function do the analysis based on type/alignment info

https://github.com/llvm/llvm-project/pull/128718