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

[via llvm-commits]

================
@@ -1385,6 +1385,24 @@ static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
                 VPValue *NewMask = GetNewMask(Red->getCondOp());
                 return new VPReductionEVLRecipe(*Red, EVL, NewMask);
               })
+              .Case<VPWidenIntrinsicRecipe>(
+                  [&](VPWidenIntrinsicRecipe *CInst) -> VPRecipeBase * {
+                    auto *CI = cast<CallInst>(CInst->getUnderlyingInstr());
+                    SmallVector<VPValue *> Ops(CInst->operands());
+                    Ops.push_back(&EVL);
+                    Intrinsic::ID VPID = VPIntrinsic::getForIntrinsic(
+                        CI->getCalledFunction()->getIntrinsicID());
+                    if (VPID == Intrinsic::not_intrinsic)
+                      return nullptr;
+                    // FIXME: In fact, can we really not pass the
+                    // underlyingInstr? In this case, how to set the Flag and
+                    // add metadata in execute?
+                    return new VPWidenIntrinsicRecipe(
+                        VPID, Ops, TypeInfo.inferScalarType(CInst), false,
+                        false, false);
----------------
LiqinWeng wrote:

It seems that the addition of this metadata still needs to be done through getUnderlyingValue. Do we really not need to pass getUnderlyingValue？
`   State.addMetadata(VPInst,
                      dyn_cast_or_null<Instruction>(getUnderlyingValue()));`

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