!fixup address latest comments, thanks

Author: fhahn

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -7696,7 +7696,7 @@ DenseMap<const SCEV *, Value *> LoopVectorizationPlanner::executePlan(
       ((VectorizingEpilogue && ExpandedSCEVs) ||
        (!VectorizingEpilogue && !ExpandedSCEVs)) &&
       "expanded SCEVs to reuse can only be used during epilogue vectorization");
-  VPlanTransforms::materializeBroadcasts(BestVPlan);
+  VPlanTransforms::materializeLiveInBroadcasts(BestVPlan);
   // TODO: Move to VPlan transform stage once the transition to the VPlan-based
   // cost model is complete for better cost estimates.
   VPlanTransforms::runPass(VPlanTransforms::unrollByUF, BestVPlan, BestUF,

llvm/lib/Transforms/Vectorize/VPlan.h

@@ -1520,7 +1520,7 @@ class VPWidenGEPRecipe : public VPRecipeWithIRFlags {
   bool onlyFirstLaneUsed(const VPValue *Op) const override {
     assert(is_contained(operands(), Op) &&
            "Op must be an operand of the recipe");
-    return Op == getOperand(0);
+    return Op == getOperand(0) && isPointerLoopInvariant();
   }
 };
 
@@ -3705,7 +3705,14 @@ class VPlan {
   VPValue *getLiveIn(Value *V) const { return Value2VPValue.lookup(V); }
 
   /// Return the list of live-in VPValues available in the VPlan.
-  ArrayRef<VPValue *> getLiveIns() const { return VPLiveIns; }
+  ArrayRef<VPValue *> getLiveIns() const {
+    assert(all_of(Value2VPValue,
+                  [this](const auto &P) {
+                    return is_contained(VPLiveIns, P.second);
+                  }) &&
+           "all VPValues in Value2VPValue must also be in VPLiveIns");
+    return VPLiveIns;
+  }
 
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
   /// Print the live-ins of this VPlan to \p O.

llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp

@@ -92,7 +92,7 @@ Type *VPTypeAnalysis::inferScalarTypeForRecipe(const VPInstruction *R) {
     return IntegerType::get(Ctx, 1);
   case VPInstruction::Broadcast:
   case VPInstruction::PtrAdd:
-    // Return the type based on the pointer argument (i.e. first operand).
+    // Return the type based on first operand.
     return inferScalarType(R->getOperand(0));
   case VPInstruction::BranchOnCond:
   case VPInstruction::BranchOnCount:

llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp

@@ -2122,11 +2122,13 @@ void VPlanTransforms::handleUncountableEarlyExit(
   LatchExitingBranch->eraseFromParent();
 }
 
-void VPlanTransforms::materializeBroadcasts(VPlan &Plan) {
+void VPlanTransforms::materializeLiveInBroadcasts(VPlan &Plan) {
   VPDominatorTree VPDT;
   VPDT.recalculate(Plan);
   auto *VectorPreheader = Plan.getVectorPreheader();
   VPBuilder Builder(VectorPreheader);
+  if (Plan.hasScalarVFOnly())
+    return;
   for (VPValue *LiveIn : Plan.getLiveIns()) {
     if (all_of(LiveIn->users(),
                [LiveIn](VPUser *U) {

llvm/lib/Transforms/Vectorize/VPlanTransforms.h

@@ -175,7 +175,7 @@ struct VPlanTransforms {
                              DenseMap<VPValue *, VPValue *> &EndValues);
 
   /// Add explicit broadcasts for live-ins used as vectors.
-  static void materializeBroadcasts(VPlan &Plan);
+  static void materializeLiveInBroadcasts(VPlan &Plan);
 };
 
 } // namespace llvm