[llvm] [LV][EVL] Support call instruction with EVL-vectorization (PR #110412)

[Shih-Po Hung via llvm-commits]

================
@@ -994,6 +994,16 @@ void VPWidenIntrinsicRecipe::execute(VPTransformState &State) {
     Args.push_back(Arg);
   }
 
+  if (VPIntrinsic::isVPIntrinsic(VectorIntrinsicID) &&
+      VectorIntrinsicID != Intrinsic::vp_select) {
+    Value *Mask =
+        State.Builder.CreateVectorSplat(State.VF, State.Builder.getTrue());
+    Value *EVL = Args.back();
+    Args.pop_back();
+    Args.push_back(Mask);
+    Args.push_back(EVL);
----------------
arcbbb wrote:

@fhahn The VPValue `BlockInMask` currently represents an all-one mask as a no-mask (`nullptr`). It seems we have two options: either model the all-one mask directly in the plan, which may require adding a new recipe, or convert the nullptr (no-mask) to an all-one mask specifically for VP intrinsics in `VPWidenIntrinsic`. Which approach would you recommend? 

https://github.com/llvm/llvm-project/pull/110412