[SLP]Try detect strided loads, if any pointer op require extraction.

If any pointer operand of the non-cosencutive loads is an instructions
with the user, which is not part of the current graph, and, thus,
requires emission of the extractelement instruction, better to try to
detect if the load sequence can be repsented as strided load and
extractelement instructions for pointers are not required.

Reviewers: preames, RKSimon, topperc

Reviewed By: RKSimon

Pull Request: #101668

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -4617,7 +4617,17 @@ BoUpSLP::LoadsState BoUpSLP::canVectorizeLoads(
       // 3. The loads are ordered, or number of unordered loads <=
       // MaxProfitableUnorderedLoads, or loads are in reversed order.
       // (this check is to avoid extra costs for very expensive shuffles).
-      if (IsPossibleStrided && (((Sz > MinProfitableStridedLoads ||
+      // 4. Any pointer operand is an instruction with the users outside of the
+      // current graph (for masked gathers extra extractelement instructions
+      // might be required).
+      auto IsAnyPointerUsedOutGraph =
+          IsPossibleStrided && any_of(PointerOps, [&](Value *V) {
+            return isa<Instruction>(V) && any_of(V->users(), [&](User *U) {
+                     return !getTreeEntry(U) && !MustGather.contains(U);
+                   });
+          });
+      if (IsPossibleStrided && (IsAnyPointerUsedOutGraph ||
+                                ((Sz > MinProfitableStridedLoads ||
                                   (static_cast<unsigned>(std::abs(*Diff)) <=
                                        MaxProfitableLoadStride * Sz &&
                                    isPowerOf2_32(std::abs(*Diff)))) &&