[llvm] [VPlan] Introduce recipes for VP loads and stores. (PR #87816)

[Florian Hahn via llvm-commits]

================
@@ -2372,15 +2391,144 @@ class VPWidenMemoryInstructionRecipe : public VPRecipeBase {
   bool onlyFirstLaneUsed(const VPValue *Op) const override {
     assert(is_contained(operands(), Op) &&
            "Op must be an operand of the recipe");
-
     // Widened, consecutive memory operations only demand the first lane of
     // their address, unless the same operand is also stored. That latter can
     // happen with opaque pointers.
-    return Op == getAddr() && isConsecutive() &&
-           (!isStore() || Op != getStoredValue());
+    return Op == getAddr() && isConsecutive();
   }
+};
 
-  Instruction &getIngredient() const { return Ingredient; }
+/// A recipe for widening load operations with vector-predication intrinsics,
+/// using the address to load from, the explicit vector length and an optional
+/// mask.
+struct VPWidenVPLoadRecipe final : public VPWidenMemoryRecipe, public VPValue {
----------------
fhahn wrote:

Yes, any recipe must be fully supported; all existing code dealing with memory ops should already do this, either by directly dealing with the common base class or various may-read/may-write helpers.

https://github.com/llvm/llvm-project/pull/87816