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

[Luke Lau via llvm-commits]

================
@@ -1571,12 +1571,15 @@ class VPVectorPointerRecipe : public VPRecipeWithIRFlags,
                               public VPUnrollPartAccessor<1> {
   Type *IndexedTy;
 
+  /// Indicate whether to compute the pointer for strided memory accesses.
+  bool Strided;
+
 public:
-  VPVectorPointerRecipe(VPValue *Ptr, Type *IndexedTy, GEPNoWrapFlags GEPFlags,
-                        DebugLoc DL)
+  VPVectorPointerRecipe(VPValue *Ptr, Type *IndexedTy, bool Strided,
----------------
lukel97 wrote:

> To me, VPReverseVectorPointerRecipe behaves like extracting the last pointer of each Part for reverse accesses.

I think this is a good description, maybe it would be more clear if it was called something like `VPVectorEndPointerRecipe`? Not for this PR though, I might submit an NFC for this and see what people think

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