[VPlan] Introduce new VPWidenCallRecipe (NFC).

[fhahn]

Summary

This patch moves calls to their own recipe, to simplify the transition
to VPUser for operands of VPWidenRecipe, as discussed in D76992.

Subsequently additional information can be added to the recipe rather
than computing it during the execute step.

[fhahn]

Update comment.

[gilr]

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

4234

Instruction::Call now belongs here with the other special-widening instructions to explicitly denote it's being handled by another recipe.

llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h

110

Missing a comment (and a newline).

llvm/lib/Transforms/Vectorize/VPlan.cpp

717

"WIDEN-CALL"

717

Since it's a single-ingredient recipe, better make this a single-line print as in VPWidenPHIRecipe (actually also the case for VPWidenRecipe now).

llvm/lib/Transforms/Vectorize/VPlan.h

787

Missing a comment.

[fhahn]

Address comments, thanks!

[fhahn]

llvm/lib/Transforms/Vectorize/VPlan.cpp

717

Right, I wasn't entirely sure about the DOT syntax. I think the DOT syntax might be a bit verbose for just printing it in -debug. Maybe we should add a way to print a VPlan without DOT syntax (for regular debug output) and an option to toggle DOT/non-DOT output. WDYT?

actually also the case for VPWidenRecipe now).

If the style here is fine I'll push a similar fix for VPWidenRecipe.

[Ayal]

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

7074–7075

add here

(Recipe = tryToWidenCall(Instr, Range)) ||

promoting VPWidenCallRecipe to join the "specific widening recipes", keeping tryToWiden() called below as kitchen-sink? It's somewhat confusing to have tryToWiden() call tryToWidenCall(). Similar to the promotion of inlined "tryToWidenGEP" in D69067.

Also augmenting above "with memory operations, [calls,] inductions and Phi nodes" comment.

llvm/lib/Transforms/Vectorize/VPlan.h

618

Lexical order.

[fhahn]

Address Ayal's comments, thanks!

[Ayal]

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

4364

Note that the *decision* not to widen calls to debug intrinsics should be made when building VPlan, in tryToWidenCall() or perhaps added to DeadInstructions(?), instead of here during code-gen.
Can be taken care of here, or in a subsequent patch along with migrating NeedToScalarize/UseVectorIntrinsics decisions "how" to widen calls.

[fhahn]

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

4364

I noticed the same thing but I didn't want to turn this into an assertion in this patch, but rather as a follow-up. There might be something better we can do for debug info intrinsics (e.g. update the values they reference to the vector values), but as long as we just drop them, we should drop them early. I think best addressed in subsequent patches. WDYT?

[Ayal]

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

4364

Sure; perhaps worth leaving a TODO behind.

[fhahn]

Add todo about handling of debug intrinsics.

[gilr]

LGTM, thanks!

llvm/lib/Transforms/Vectorize/VPlan.cpp

717

Sure, would be good to also have a text-only dump; AFAIK DOTs are indeed usually dumped in response to some -dot-xxx flag.

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

public:
 
   /// Create a new empty loop. Unlink the old loop and connect the new one.
   /// Return the pre-header block of the new loop.
   BasicBlock *createVectorizedLoopSkeleton();
 
   /// Widen a single instruction within the innermost loop.
   void widenInstruction(Instruction &I);
 
+  /// Widen a single call instruction within the innermost loop.
+  void widenCallInstruction(CallInst &I);
+
   /// Fix the vectorized code, taking care of header phi's, live-outs, and more.
   void fixVectorizedLoop();
 
   // Return true if any runtime check is added.
   bool areSafetyChecksAdded() { return AddedSafetyChecks; }
 
   /// A type for vectorized values in the new loop. Each value from the
   /// original loop, when vectorized, is represented by UF vector values in the
assert((I.getOpcode() == Instruction::UDiv ||
          "Unexpected instruction");
   Value *Divisor = I.getOperand(1);
   auto *CInt = dyn_cast<ConstantInt>(Divisor);
   return !CInt || CInt->isZero();
 }
 
 void InnerLoopVectorizer::widenInstruction(Instruction &I) {
   switch (I.getOpcode()) {
+  case Instruction::Call:
   case Instruction::Br:
   case Instruction::PHI:
   case Instruction::GetElementPtr:
     llvm_unreachable("This instruction is handled by a different recipe.");
   case Instruction::UDiv:
   case Instruction::SDiv:
   case Instruction::SRem:
   case Instruction::URem:
   case Instruction::Add:
   case Instruction::FAdd:
   case Instruction::Sub:
   case Instruction::FSub:
case Instruction::BitCast: {
     for (unsigned Part = 0; Part < UF; ++Part) {
       Value *A = getOrCreateVectorValue(CI->getOperand(0), Part);
       Value *Cast = Builder.CreateCast(CI->getOpcode(), A, DestTy);
       VectorLoopValueMap.setVectorValue(&I, Part, Cast);
       addMetadata(Cast, &I);
     }
     break;
   }
+  default:
+    // This instruction is not vectorized by simple widening.
+    LLVM_DEBUG(dbgs() << "LV: Found an unhandled instruction: " << I);
+    llvm_unreachable("Unhandled instruction!");
+  } // end of switch.
+}
 
-  case Instruction::Call: {
-    // Ignore dbg intrinsics.
-    if (isa<DbgInfoIntrinsic>(I))
-      break;
-    setDebugLocFromInst(Builder, &I);
+void InnerLoopVectorizer::widenCallInstruction(CallInst &I) {
+  // TODO: Debug intrinsics should be skipped/handled during VPlan construction
+  // rather than dropping them here.
+    return;
 
-    Module *M = I.getParent()->getParent()->getParent();
-    auto *CI = cast<CallInst>(&I);
 
-    SmallVector<Type *, 4> Tys;
-    for (Value *ArgOperand : CI->arg_operands())
-      Tys.push_back(ToVectorTy(ArgOperand->getType(), VF));
 
-    Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
 
-    // The flag shows whether we use Intrinsic or a usual Call for vectorized
-    // version of the instruction.
-    // Is it beneficial to perform intrinsic call compared to lib call?
-    bool NeedToScalarize = false;
-    unsigned CallCost = Cost->getVectorCallCost(CI, VF, NeedToScalarize);
-    bool UseVectorIntrinsic =
-        ID && Cost->getVectorIntrinsicCost(CI, VF) <= CallCost;
-    assert((UseVectorIntrinsic || !NeedToScalarize) &&
-           "Instruction should be scalarized elsewhere.");
 
-    for (unsigned Part = 0; Part < UF; ++Part) {
-      SmallVector<Value *, 4> Args;
-      for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) {
-        Value *Arg = CI->getArgOperand(i);
-        // Some intrinsics have a scalar argument - don't replace it with a
-        // vector.
-        if (!UseVectorIntrinsic || !hasVectorInstrinsicScalarOpd(ID, i))
-          Arg = getOrCreateVectorValue(CI->getArgOperand(i), Part);
-        Args.push_back(Arg);
-      }
 
-      Function *VectorF;
-      if (UseVectorIntrinsic) {
-        // Use vector version of the intrinsic.
-        Type *TysForDecl[] = {CI->getType()};
-        if (VF > 1)
-          TysForDecl[0] = VectorType::get(CI->getType()->getScalarType(), VF);
-        VectorF = Intrinsic::getDeclaration(M, ID, TysForDecl);
-      } else {
-        // Use vector version of the function call.
-        const VFShape Shape =
-            VFShape::get(*CI, {VF, false} /*EC*/, false /*HasGlobalPred*/);
 #ifndef NDEBUG
         const SmallVector<VFInfo, 8> Infos = VFDatabase::getMappings(*CI);
         assert(std::find_if(Infos.begin(), Infos.end(),
                             [&Shape](const VFInfo &Info) {
                               return Info.Shape == Shape;
                             }) != Infos.end() &&
                "Vector function shape is missing from the database.");
 #endif
         VectorF = VFDatabase(*CI).getVectorizedFunction(Shape);
-      }
       assert(VectorF && "Can't create vector function.");
 
       SmallVector<OperandBundleDef, 1> OpBundles;
       CI->getOperandBundlesAsDefs(OpBundles);
       CallInst *V = Builder.CreateCall(VectorF, Args, OpBundles);
 
       if (isa<FPMathOperator>(V))
         V->copyFastMathFlags(CI);
 
       VectorLoopValueMap.setVectorValue(&I, Part, V);
       addMetadata(V, &I);
-    }
-
-    break;
-  }
-
-  default:
-    // This instruction is not vectorized by simple widening.
-    LLVM_DEBUG(dbgs() << "LV: Found an unhandled instruction: " << I);
-    llvm_unreachable("Unhandled instruction!");
-  } // end of switch.
 }
 
 void LoopVectorizationCostModel::collectLoopScalars(unsigned VF) {
   // We should not collect Scalars more than once per VF. Right now, this
   // function is called from collectUniformsAndScalars(), which already does
   // this check. Collecting Scalars for VF=1 does not make any sense.
   assert(VF >= 2 && Scalars.find(VF) == Scalars.end() &&
          "This function should not be visited twice for the same VF");
for (unsigned In = 0; In < NumIncoming; In++) {
     assert((EdgeMask || NumIncoming == 1) &&
            "Multiple predecessors with one having a full mask");
     if (EdgeMask)
       Masks.push_back(EdgeMask);
   }
   return new VPBlendRecipe(Phi, Masks);
 }
 
-VPWidenRecipe *VPRecipeBuilder::tryToWiden(Instruction *I, VFRange &Range) {
+VPWidenCallRecipe *VPRecipeBuilder::tryToWidenCall(Instruction *I,
+                                                   VFRange &Range) {
 
   bool IsPredicated = LoopVectorizationPlanner::getDecisionAndClampRange(
       [&](unsigned VF) { return CM.isScalarWithPredication(I, VF); }, Range);
 
+  CallInst *CI = dyn_cast<CallInst>(I);
+  if (IsPredicated || !CI)
+    return nullptr;
+
+  Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
+  if (ID && (ID == Intrinsic::assume || ID == Intrinsic::lifetime_end ||
+             ID == Intrinsic::lifetime_start || ID == Intrinsic::sideeffect))
+    return nullptr;
+
+  auto willWiden = [&](unsigned VF) -> bool {
+    Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
+    // The following case may be scalarized depending on the VF.
+    // The flag shows whether we use Intrinsic or a usual Call for vectorized
+    // version of the instruction.
+    // Is it beneficial to perform intrinsic call compared to lib call?
+    bool NeedToScalarize;
+    unsigned CallCost = CM.getVectorCallCost(CI, VF, NeedToScalarize);
+    bool UseVectorIntrinsic =
+        ID && CM.getVectorIntrinsicCost(CI, VF) <= CallCost;
+    return UseVectorIntrinsic || !NeedToScalarize;
+  };
+
+  if (!LoopVectorizationPlanner::getDecisionAndClampRange(willWiden, Range))
+    return nullptr;
+
+  // Success: widen this call.
+  return new VPWidenCallRecipe(*CI);
+}
+
+VPWidenRecipe *VPRecipeBuilder::tryToWiden(Instruction *I, VFRange &Range) {
+  bool IsPredicated = LoopVectorizationPlanner::getDecisionAndClampRange(
+      [&](unsigned VF) { return CM.isScalarWithPredication(I, VF); }, Range);
+
   if (IsPredicated)
     return nullptr;
 
   auto IsVectorizableOpcode = [](unsigned Opcode) {
     switch (Opcode) {
     case Instruction::Add:
     case Instruction::And:
     case Instruction::AShr:
     case Instruction::BitCast:
     case Instruction::Br:
-    case Instruction::Call:
     case Instruction::FAdd:
     case Instruction::FCmp:
     case Instruction::FDiv:
     case Instruction::FMul:
     case Instruction::FNeg:
     case Instruction::FPExt:
     case Instruction::FPToSI:
     case Instruction::FPToUI:
case Instruction::ZExt:
       return true;
     }
     return false;
   };
 
   if (!IsVectorizableOpcode(I->getOpcode()))
     return nullptr;
 
-  if (CallInst *CI = dyn_cast<CallInst>(I)) {
-    Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
-    if (ID && (ID == Intrinsic::assume || ID == Intrinsic::lifetime_end ||
-               ID == Intrinsic::lifetime_start || ID == Intrinsic::sideeffect))
-      return nullptr;
-  }
-
   auto willWiden = [&](unsigned VF) -> bool {
     if (!isa<PHINode>(I) && (CM.isScalarAfterVectorization(I, VF) ||
                              CM.isProfitableToScalarize(I, VF)))
       return false;
-    if (CallInst *CI = dyn_cast<CallInst>(I)) {
-      Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
-      // The following case may be scalarized depending on the VF.
-      // The flag shows whether we use Intrinsic or a usual Call for vectorized
-      // version of the instruction.
-      // Is it beneficial to perform intrinsic call compared to lib call?
-      bool NeedToScalarize;
-      unsigned CallCost = CM.getVectorCallCost(CI, VF, NeedToScalarize);
-      bool UseVectorIntrinsic =
-          ID && CM.getVectorIntrinsicCost(CI, VF) <= CallCost;
-      return UseVectorIntrinsic || !NeedToScalarize;
-    }
     if (isa<LoadInst>(I) || isa<StoreInst>(I)) {
       assert(CM.getWideningDecision(I, VF) ==
                  LoopVectorizationCostModel::CM_Scalarize &&
              "Memory widening decisions should have been taken care by now");
       return false;
     }
     return true;
   };
VPRegionBlock *VPRecipeBuilder::createReplicateRegion(Instruction *Instr,
 
   return Region;
 }
 
 bool VPRecipeBuilder::tryToCreateRecipe(Instruction *Instr, VFRange &Range,
                                         VPlanPtr &Plan, VPBasicBlock *VPBB) {
   VPRecipeBase *Recipe = nullptr;
 
-  // First, check for specific widening recipes that deal with memory
+  // First, check for specific widening recipes that deal with calls, memory
   // operations, inductions and Phi nodes.
-  if ((Recipe = tryToWidenMemory(Instr, Range, Plan)) ||
+  if ((Recipe = tryToWidenCall(Instr, Range)) ||
+      (Recipe = tryToWidenMemory(Instr, Range, Plan)) ||
       (Recipe = tryToOptimizeInduction(Instr, Range)) ||
       (Recipe = tryToBlend(Instr, Plan)) ||
       (isa<PHINode>(Instr) &&
        (Recipe = new VPWidenPHIRecipe(cast<PHINode>(Instr))))) {
     setRecipe(Instr, Recipe);
     VPBB->appendRecipe(Recipe);
     return true;
   }
void VPInterleaveRecipe::print(raw_ostream &O, const Twine &Indent,
   }
   O << "\\l\"";
   for (unsigned i = 0; i < IG->getFactor(); ++i)
     if (Instruction *I = IG->getMember(i))
       O << " +\n"
         << Indent << "\"  " << VPlanIngredient(I) << " " << i << "\\l\"";
 }
 
+void VPWidenCallRecipe::execute(VPTransformState &State) {
+  State.ILV->widenCallInstruction(Ingredient);
+}
+
 void VPWidenRecipe::execute(VPTransformState &State) {
   State.ILV->widenInstruction(Ingredient);
 }
 
 void VPWidenGEPRecipe::execute(VPTransformState &State) {
   State.ILV->widenGEP(GEP, State.UF, State.VF, IsPtrLoopInvariant,
                       IsIndexLoopInvariant);
 }

llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h

public:
   VPWidenIntOrFpInductionRecipe *tryToOptimizeInduction(Instruction *I,
                                                         VFRange &Range);
 
   /// Handle non-loop phi nodes. Currently all such phi nodes are turned into
   /// a sequence of select instructions as the vectorizer currently performs
   /// full if-conversion.
   VPBlendRecipe *tryToBlend(Instruction *I, VPlanPtr &Plan);
 
+  /// Handle call instruction. If \p I is a call that can be widened for \p
+  /// Range.Start, return a new VPWidenCallRecipe. Range.End may be decreased to
+  /// ensure same decision from \p Range.Start to \p Range.End.
+  VPWidenCallRecipe *tryToWidenCall(Instruction *I, VFRange &Range);
+
   /// Check if \p I can be widened within the given VF \p Range. If \p I can be
   /// widened for \p Range.Start, build a new VPWidenRecipe and return it.
   /// Range.End may be decreased to ensure same decision from \p Range.Start to
   /// \p Range.End.
   VPWidenRecipe *tryToWiden(Instruction *I, VFRange &Range);
 
   /// Create a replicating region for instruction \p I that requires
   /// predication. \p PredRecipe is a VPReplicateRecipe holding \p I.

llvm/lib/Transforms/Vectorize/VPlan.h

public:
   /// type identification.
   using VPRecipeTy = enum {
     VPBlendSC,
     VPBranchOnMaskSC,
     VPInstructionSC,
     VPInterleaveSC,
     VPPredInstPHISC,
     VPReplicateSC,
+    VPWidenCallSC,
     VPWidenGEPSC,
     VPWidenIntOrFpInductionSC,
     VPWidenMemoryInstructionSC,
     VPWidenPHISC,
     VPWidenSC,
   };
 
   VPRecipeBase(const unsigned char SC) : SubclassID(SC) {}
   virtual ~VPRecipeBase() = default;
 
   /// \return an ID for the concrete type of this object.
   /// This is used to implement the classof checks. This should not be used
   /// for any other purpose, as the values may change as LLVM evolves.
   unsigned getVPRecipeID() const { return SubclassID; }
public:
   /// Produce widened copies of all Ingredients.
   void execute(VPTransformState &State) override;
 
   /// Print the recipe.
   void print(raw_ostream &O, const Twine &Indent,
              VPSlotTracker &SlotTracker) const override;
 };
 
+/// A recipe for widening Call instructions.
+class VPWidenCallRecipe : public VPRecipeBase {
+private:
+  /// Hold the call to be widened.
+  CallInst &Ingredient;
+
+public:
+  VPWidenCallRecipe(CallInst &I) : VPRecipeBase(VPWidenCallSC), Ingredient(I) {}
+
+  ~VPWidenCallRecipe() override = default;
+
+  /// Method to support type inquiry through isa, cast, and dyn_cast.
+  static inline bool classof(const VPRecipeBase *V) {
+    return V->getVPRecipeID() == VPRecipeBase::VPWidenCallSC;
+  }
+
+  /// Produce a widened version of the call instruction.
+  void execute(VPTransformState &State) override;
+
+  /// Print the recipe.
+  void print(raw_ostream &O, const Twine &Indent,
+             VPSlotTracker &SlotTracker) const override;
+};
+
 /// A recipe for handling GEP instructions.
 class VPWidenGEPRecipe : public VPRecipeBase {
 private:
   GetElementPtrInst *GEP;
   bool IsPtrLoopInvariant;
   SmallBitVector IsIndexLoopInvariant;
 
 public:

llvm/lib/Transforms/Vectorize/VPlan.cpp

if (E > 0) {
         Inst->getOperand(I)->printAsOperand(RSO << ", ", false);
     }
   } else // !Inst
     V->printAsOperand(RSO, false);
   RSO.flush();
   O << DOT::EscapeString(IngredientString);
 }
 
+void VPWidenCallRecipe::print(raw_ostream &O, const Twine &Indent,
+                              VPSlotTracker &SlotTracker) const {
+  O << " +\n"
+    << Indent << "\"WIDEN-CALL " << VPlanIngredient(&Ingredient) << "\\l\"";
+}
+
 void VPWidenRecipe::print(raw_ostream &O, const Twine &Indent,
                           VPSlotTracker &SlotTracker) const {
   O << " +\n" << Indent << "\"WIDEN\\l\"";
   O << "\"  " << VPlanIngredient(&Ingredient) << "\\l\"";
 }
 
 void VPWidenIntOrFpInductionRecipe::print(raw_ostream &O, const Twine &Indent,
                                           VPSlotTracker &SlotTracker) const {