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

[Mel Chen via llvm-commits]

================
@@ -2126,8 +2126,13 @@ void VPVectorPointerRecipe::execute(VPTransformState &State) {
   Value *Ptr = State.get(getOperand(0), VPLane(0));
 
   Value *Increment = createStepForVF(Builder, IndexTy, State.VF, CurrentPart);
+  // TODO: Support non-unit-reverse strided accesses.
+  Value *Index =
+      Strided
+          ? Builder.CreateMul(Increment, ConstantInt::getSigned(IndexTy, -1))
----------------
Mel-Chen wrote:

This is the initial patch for strided memory accesses, currently only supporting cases where the stride is -1. A follow-up patch will focus on stride analysis to support more strided memory accesses.

However, I believe it would be beneficial to discuss and decide how VPlan should handle stride in this initial patch. My current plan is to extend the existing `VPWidenLoadRecipe/VPWidenStoreRecipe` and `VPVectorPointerRecipe` by explicitly passing the stride into the recipe. For example:
```
VPWidenLoadRecipe(LoadInst &Load, VPValue *Addr, VPValue *Mask, 
                  VPValue *Stride,  
                  bool Consecutive, bool Reverse, DebugLoc DL)
```
A memory access will only be considered strided when `Stride` is explicitly provided (i.e., not nullptr).

@fhahn @ayalz What do you think about this approach?

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