[TTI] Add a Mask to getShuffleCost

[dmgreen]

Summary

This adds an Mask ArrayRef to getShuffleCost, so that if an exact mask can be provided, a more accurate cost can be provided by the backend. For example VREV costs could be returned by the ARM backend. This should be an NFC until then, laying the groundwork for that to be added.

[lebedev.ri]

What is VREV?
How is that different from TargetTransformInfo::ShuffleKind::SK_Reverse?
Can a new shuffle kind be added instead of increasing problem surface?

[dmgreen]

Hello. A VREV is a little different from a SK_Reverse, as far as I understand. A vrev mask will reverse lanes in a block, not the whole register. So they have masks like <3, 2, 1, 0, 7, 6, 5, 4> or <1, 0, 3, 2, 5, 4, 7, 6>. In MVE there are vrev16.8, vrev32.8, vrev32.16, vrev64.8, vrev64.16 and vrev64.32. The way I think of them is that taking a 128 bit legal vector, they reverse the bits in the first digit, then reverse back in the second.

This seemed like the most general way to fix it to me, giving the backend the exact shuffle mask where available can help it provide better info. Adding a collection of SK_VREVXY might work, but I was hoping to add some other shuffles cheap for the Arm backend like VMOVN's, which would make further use of it eventually.

[RKSimon]

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

3482

Should isShuffle() create a shuffle mask to pass on to getShuffleCost?

[dmgreen]

Nice suggestion. Calculate the Mask in IsShuffle.

[dmgreen]

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

3470

I don't know the SLP vectorizer very well. Could these use E->ReuseShuffleIndices?

[ABataev]

llvm/include/llvm/Analysis/TargetTransformInfo.h

1047–1048

ArrayRef<int> Mask = llvm::None? How about make it a param with the default value?

llvm/include/llvm/CodeGen/BasicTTIImpl.h

1260

Better to use llvm::None here and in the other places

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

293

SmalVectorImpl<int> &?

312

UndefMaskElem

3470

I think yes.

3550

You can call inversePermutation to get a mask out of E->ReorderIndices

3776

Same, call inversePermutation to get a mask out of E->ReorderIndices

3793

Same, call inversePermutation to get a mask out of E->ReorderIndices

3861

The mask here is something like 1010101..., just in case

[dmgreen]

Thanks for the comments. None is a new one to me, that's nice that it works for ArrayRefs.
This now uses None, including as a default parameter value, and tries to use more accurate masks from the SLP vectorizer.

[ABataev]

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

3861

Mask(E->Scalars.size());

3862

for (unsigned I = 0, End = E->Scalars.size(); I < End; ++I) {

3865

Mask[I] = ....;

[dmgreen]

Updated SLP loop and added more accurate masks for a couple of VectorCombines.

[dmgreen]

Ping. Any other comments?

[RKSimon]

LGTM

[dmgreen]

Thanks.

llvm/include/llvm/Analysis/TargetTransformInfo.h

int getArithmeticInstrCost(
       OperandValueKind Opd1Info = OK_AnyValue,
       OperandValueKind Opd2Info = OK_AnyValue,
       OperandValueProperties Opd1PropInfo = OP_None,
       OperandValueProperties Opd2PropInfo = OP_None,
       ArrayRef<const Value *> Args = ArrayRef<const Value *>(),
       const Instruction *CxtI = nullptr) const;
 
   /// \return The cost of a shuffle instruction of kind Kind and of type Tp.
+  /// The exact mask may be passed as Mask, or else the array will be empty.
   /// The index and subtype parameters are used by the subvector insertion and
   /// extraction shuffle kinds to show the insert/extract point and the type of
   /// the subvector being inserted/extracted.
   /// NOTE: For subvector extractions Tp represents the source type.
-  int getShuffleCost(ShuffleKind Kind, VectorType *Tp, int Index = 0,
-                     VectorType *SubTp = nullptr) const;
+  int getShuffleCost(ShuffleKind Kind, VectorType *Tp, ArrayRef<int> Mask,
+                     int Index = 0, VectorType *SubTp = nullptr) const;
 
   /// Represents a hint about the context in which a cast is used.
   ///
   /// For zext/sext, the context of the cast is the operand, which must be a
   /// load of some kind. For trunc, the context is of the cast is the single
   /// user of the instruction, which must be a store of some kind.
   ///
   /// This enum allows the vectorizer to give getCastInstrCost an idea of the
public:
   virtual unsigned getMaxInterleaveFactor(unsigned VF) = 0;
   virtual unsigned getArithmeticInstrCost(
       unsigned Opcode, Type *Ty,
       TTI::TargetCostKind CostKind,
       OperandValueKind Opd1Info,
       OperandValueKind Opd2Info, OperandValueProperties Opd1PropInfo,
       OperandValueProperties Opd2PropInfo, ArrayRef<const Value *> Args,
       const Instruction *CxtI = nullptr) = 0;
-  virtual int getShuffleCost(ShuffleKind Kind, VectorType *Tp, int Index,
+  virtual int getShuffleCost(ShuffleKind Kind, VectorType *Tp,
+                             ArrayRef<int> Mask, int Index,
                              VectorType *SubTp) = 0;
   virtual int getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
                                CastContextHint CCH,
                                TTI::TargetCostKind CostKind,
                                const Instruction *I) = 0;
   virtual int getExtractWithExtendCost(unsigned Opcode, Type *Dst,
                                        VectorType *VecTy, unsigned Index) = 0;
   virtual int getCFInstrCost(unsigned Opcode,
unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty,
                                   OperandValueKind Opd2Info,
                                   OperandValueProperties Opd1PropInfo,
                                   OperandValueProperties Opd2PropInfo,
                                   ArrayRef<const Value *> Args,
                                   const Instruction *CxtI = nullptr) override {
     return Impl.getArithmeticInstrCost(Opcode, Ty, CostKind, Opd1Info, Opd2Info,
                                        Opd1PropInfo, Opd2PropInfo, Args, CxtI);
   }
-  int getShuffleCost(ShuffleKind Kind, VectorType *Tp, int Index,
-                     VectorType *SubTp) override {
-    return Impl.getShuffleCost(Kind, Tp, Index, SubTp);
+  int getShuffleCost(ShuffleKind Kind, VectorType *Tp, ArrayRef<int> Mask,
+                     int Index, VectorType *SubTp) override {
+    return Impl.getShuffleCost(Kind, Tp, Mask, Index, SubTp);
   }
   int getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
                        CastContextHint CCH, TTI::TargetCostKind CostKind,
                        const Instruction *I) override {
     return Impl.getCastInstrCost(Opcode, Dst, Src, CCH, CostKind, I);
   }
   int getExtractWithExtendCost(unsigned Opcode, Type *Dst, VectorType *VecTy,
                                unsigned Index) override {

llvm/include/llvm/Analysis/TargetTransformInfoImpl.h

unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty,
     case Instruction::UDiv:
     case Instruction::URem:
       // FIXME: Unlikely to be true for CodeSize.
       return TTI::TCC_Expensive;
     }
     return 1;
   }
 
-  unsigned getShuffleCost(TTI::ShuffleKind Kind, VectorType *Ty, int Index,
+  unsigned getShuffleCost(TTI::ShuffleKind Kind, VectorType *Ty,
+                          ArrayRef<int> Mask, int Index,
                           VectorType *SubTp) const {
     return 1;
   }
 
   unsigned getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
                             TTI::CastContextHint CCH,
                             TTI::TargetCostKind CostKind,
                             const Instruction *I) const {
case Instruction::ShuffleVector: {
         return TTI::TCC_Basic; // FIXME
       auto *VecTy = cast<VectorType>(U->getType());
       auto *VecSrcTy = cast<VectorType>(U->getOperand(0)->getType());
 
       // TODO: Identify and add costs for insert subvector, etc.
       int SubIndex;
       if (Shuffle->isExtractSubvectorMask(SubIndex))
         return TargetTTI->getShuffleCost(TTI::SK_ExtractSubvector, VecSrcTy,
-                                         SubIndex, VecTy);
+                                         Shuffle->getShuffleMask(), SubIndex,
+                                         VecTy);
       else if (Shuffle->changesLength())
         return CostKind == TTI::TCK_RecipThroughput ? -1 : 1;
       else if (Shuffle->isIdentity())
         return 0;
       else if (Shuffle->isReverse())
-        return TargetTTI->getShuffleCost(TTI::SK_Reverse, VecTy, 0, nullptr);
+        return TargetTTI->getShuffleCost(TTI::SK_Reverse, VecTy,
+                                         Shuffle->getShuffleMask(), 0, nullptr);
       else if (Shuffle->isSelect())
-        return TargetTTI->getShuffleCost(TTI::SK_Select, VecTy, 0, nullptr);
+        return TargetTTI->getShuffleCost(TTI::SK_Select, VecTy,
+                                         Shuffle->getShuffleMask(), 0, nullptr);
       else if (Shuffle->isTranspose())
-        return TargetTTI->getShuffleCost(TTI::SK_Transpose, VecTy, 0, nullptr);
+        return TargetTTI->getShuffleCost(TTI::SK_Transpose, VecTy,
+                                         Shuffle->getShuffleMask(), 0, nullptr);
       else if (Shuffle->isZeroEltSplat())
-        return TargetTTI->getShuffleCost(TTI::SK_Broadcast, VecTy, 0, nullptr);
+        return TargetTTI->getShuffleCost(TTI::SK_Broadcast, VecTy,
+                                         Shuffle->getShuffleMask(), 0, nullptr);
       else if (Shuffle->isSingleSource())
-        return TargetTTI->getShuffleCost(TTI::SK_PermuteSingleSrc, VecTy, 0,
-                                         nullptr);
+        return TargetTTI->getShuffleCost(TTI::SK_PermuteSingleSrc, VecTy,
+                                         Shuffle->getShuffleMask(), 0, nullptr);
 
-      return TargetTTI->getShuffleCost(TTI::SK_PermuteTwoSrc, VecTy, 0,
-                                       nullptr);
+      return TargetTTI->getShuffleCost(TTI::SK_PermuteTwoSrc, VecTy,
+                                       Shuffle->getShuffleMask(), 0, nullptr);
     }
     case Instruction::ExtractElement: {
       unsigned Idx = -1;
       auto *EEI = dyn_cast<ExtractElementInst>(U);
       if (!EEI)
         return TTI::TCC_Basic; // FIXME
 
       auto *CI = dyn_cast<ConstantInt>(EEI->getOperand(1));

llvm/include/llvm/CodeGen/BasicTTIImpl.h

if (auto *VTy = dyn_cast<VectorType>(Ty)) {
       SmallVector<Type *> Tys(Args.size(), Ty);
       return getScalarizationOverhead(VTy, Args, Tys) + Num * Cost;
     }
 
     // We don't know anything about this scalar instruction.
     return OpCost;
   }
 
-  unsigned getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, int Index,
-                          VectorType *SubTp) {
+  unsigned getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp,
+                          ArrayRef<int> Mask, int Index, VectorType *SubTp) {
 
     switch (Kind) {
     case TTI::SK_Broadcast:
       return getBroadcastShuffleOverhead(cast<FixedVectorType>(Tp));
     case TTI::SK_Select:
     case TTI::SK_Reverse:
     case TTI::SK_Transpose:
     case TTI::SK_PermuteSingleSrc:
case Intrinsic::masked_gather: {
                                              VarMask, Alignment, CostKind, I);
     }
     case Intrinsic::experimental_vector_extract: {
       // FIXME: Handle case where a scalable vector is extracted from a scalable
       // vector
       if (isa<ScalableVectorType>(RetTy))
         return BaseT::getIntrinsicInstrCost(ICA, CostKind);
       unsigned Index = cast<ConstantInt>(Args[1])->getZExtValue();
-      return thisT()->getShuffleCost(TTI::SK_ExtractSubvector,
-                                     cast<VectorType>(Args[0]->getType()),
-                                     Index, cast<VectorType>(RetTy));
+      return thisT()->getShuffleCost(
+          TTI::SK_ExtractSubvector, cast<VectorType>(Args[0]->getType()),
+          ArrayRef<int>(), Index, cast<VectorType>(RetTy));
     }
     case Intrinsic::experimental_vector_insert: {
       // FIXME: Handle case where a scalable vector is inserted into a scalable
       // vector
       if (isa<ScalableVectorType>(Args[1]->getType()))
         return BaseT::getIntrinsicInstrCost(ICA, CostKind);
       unsigned Index = cast<ConstantInt>(Args[2])->getZExtValue();
       return thisT()->getShuffleCost(
-          TTI::SK_InsertSubvector, cast<VectorType>(Args[0]->getType()), Index,
-          cast<VectorType>(Args[1]->getType()));
+          TTI::SK_InsertSubvector, cast<VectorType>(Args[0]->getType()),
+          ArrayRef<int>(), Index, cast<VectorType>(Args[1]->getType()));
     }
     case Intrinsic::experimental_vector_reverse: {
-      return thisT()->getShuffleCost(TTI::SK_Reverse,
-                                     cast<VectorType>(Args[0]->getType()), 0,
-                                     cast<VectorType>(RetTy));
+      return thisT()->getShuffleCost(
+          TTI::SK_Reverse, cast<VectorType>(Args[0]->getType()),
+          ArrayRef<int>(), 0, cast<VectorType>(RetTy));
     }
     case Intrinsic::vector_reduce_add:
     case Intrinsic::vector_reduce_mul:
     case Intrinsic::vector_reduce_and:
     case Intrinsic::vector_reduce_or:
     case Intrinsic::vector_reduce_xor:
     case Intrinsic::vector_reduce_smax:
     case Intrinsic::vector_reduce_smin:
std::pair<unsigned, MVT> LT =
         thisT()->getTLI()->getTypeLegalizationCost(DL, Ty);
     unsigned LongVectorCount = 0;
     unsigned MVTLen =
         LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
     while (NumVecElts > MVTLen) {
       NumVecElts /= 2;
       VectorType *SubTy = FixedVectorType::get(ScalarTy, NumVecElts);
       // Assume the pairwise shuffles add a cost.
-      ShuffleCost +=
-          (IsPairwise + 1) * thisT()->getShuffleCost(TTI::SK_ExtractSubvector,
-                                                     Ty, NumVecElts, SubTy);
+      ShuffleCost += (IsPairwise + 1) * thisT()->getShuffleCost(
+                                            TTI::SK_ExtractSubvector, Ty,
+                                            ArrayRef<int>(), NumVecElts, SubTy);
       ArithCost += thisT()->getArithmeticInstrCost(Opcode, SubTy, CostKind);
       Ty = SubTy;
       ++LongVectorCount;
     }
 
     NumReduxLevels -= LongVectorCount;
 
     // The minimal length of the vector is limited by the real length of vector
     // operations performed on the current platform. That's why several final
     // reduction operations are performed on the vectors with the same
     // architecture-dependent length.
 
     // Non pairwise reductions need one shuffle per reduction level. Pairwise
     // reductions need two shuffles on every level, but the last one. On that
     // level one of the shuffles is <0, u, u, ...> which is identity.
     unsigned NumShuffles = NumReduxLevels;
     if (IsPairwise && NumReduxLevels >= 1)
       NumShuffles += NumReduxLevels - 1;
-    ShuffleCost += NumShuffles *
-                   thisT()->getShuffleCost(TTI::SK_PermuteSingleSrc, Ty, 0, Ty);
+    ShuffleCost +=
+        NumShuffles * thisT()->getShuffleCost(TTI::SK_PermuteSingleSrc, Ty,
+                                              ArrayRef<int>(), 0, Ty);
     ArithCost += NumReduxLevels * thisT()->getArithmeticInstrCost(Opcode, Ty);
     return ShuffleCost + ArithCost +
            thisT()->getVectorInstrCost(Instruction::ExtractElement, Ty, 0);
   }
 
   /// Try to calculate op costs for min/max reduction operations.
   /// \param CondTy Conditional type for the Select instruction.
   unsigned getMinMaxReductionCost(VectorType *Ty, VectorType *CondTy,
unsigned getMinMaxReductionCost(VectorType *Ty, VectorType *CondTy,
     unsigned MVTLen =
         LT.second.isVector() ? LT.second.getVectorNumElements() : 1;
     while (NumVecElts > MVTLen) {
       NumVecElts /= 2;
       auto *SubTy = FixedVectorType::get(ScalarTy, NumVecElts);
       CondTy = FixedVectorType::get(ScalarCondTy, NumVecElts);
 
       // Assume the pairwise shuffles add a cost.
-      ShuffleCost +=
-          (IsPairwise + 1) * thisT()->getShuffleCost(TTI::SK_ExtractSubvector,
-                                                     Ty, NumVecElts, SubTy);
+      ShuffleCost += (IsPairwise + 1) * thisT()->getShuffleCost(
+                                            TTI::SK_ExtractSubvector, Ty,
+                                            ArrayRef<int>(), NumVecElts, SubTy);
       MinMaxCost +=
           thisT()->getCmpSelInstrCost(CmpOpcode, SubTy, CondTy,
                                       CmpInst::BAD_ICMP_PREDICATE, CostKind) +
           thisT()->getCmpSelInstrCost(Instruction::Select, SubTy, CondTy,
                                       CmpInst::BAD_ICMP_PREDICATE, CostKind);
       Ty = SubTy;
       ++LongVectorCount;
     }
 
     NumReduxLevels -= LongVectorCount;
 
     // The minimal length of the vector is limited by the real length of vector
     // operations performed on the current platform. That's why several final
     // reduction opertions are perfomed on the vectors with the same
     // architecture-dependent length.
 
     // Non pairwise reductions need one shuffle per reduction level. Pairwise
     // reductions need two shuffles on every level, but the last one. On that
     // level one of the shuffles is <0, u, u, ...> which is identity.
     unsigned NumShuffles = NumReduxLevels;
     if (IsPairwise && NumReduxLevels >= 1)
       NumShuffles += NumReduxLevels - 1;
-    ShuffleCost += NumShuffles *
-                   thisT()->getShuffleCost(TTI::SK_PermuteSingleSrc, Ty, 0, Ty);
+    ShuffleCost +=
+        NumShuffles * thisT()->getShuffleCost(TTI::SK_PermuteSingleSrc, Ty,
+                                              ArrayRef<int>(), 0, Ty);
     MinMaxCost +=
         NumReduxLevels *
         (thisT()->getCmpSelInstrCost(CmpOpcode, Ty, CondTy,
                                      CmpInst::BAD_ICMP_PREDICATE, CostKind) +
          thisT()->getCmpSelInstrCost(Instruction::Select, Ty, CondTy,
                                      CmpInst::BAD_ICMP_PREDICATE, CostKind));
     // The last min/max should be in vector registers and we counted it above.
     // So just need a single extractelement.

llvm/lib/Analysis/TargetTransformInfo.cpp

int TargetTransformInfo::getArithmeticInstrCost(
   int Cost = TTIImpl->getArithmeticInstrCost(
       Opcode, Ty, CostKind, Opd1Info, Opd2Info, Opd1PropInfo, Opd2PropInfo,
       Args, CxtI);
   assert(Cost >= 0 && "TTI should not produce negative costs!");
   return Cost;
 }
 
 int TargetTransformInfo::getShuffleCost(ShuffleKind Kind, VectorType *Ty,
-                                        int Index, VectorType *SubTp) const {
-  int Cost = TTIImpl->getShuffleCost(Kind, Ty, Index, SubTp);
+                                        ArrayRef<int> Mask, int Index,
+                                        VectorType *SubTp) const {
+  int Cost = TTIImpl->getShuffleCost(Kind, Ty, Mask, Index, SubTp);
   assert(Cost >= 0 && "TTI should not produce negative costs!");
   return Cost;
 }
 
 TTI::CastContextHint
 TargetTransformInfo::getCastContextHint(const Instruction *I) {
   if (!I)
     return CastContextHint::None;

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h

public:
 
   bool isLegalToVectorizeReduction(RecurrenceDescriptor RdxDesc,
                                    ElementCount VF) const;
 
   int getArithmeticReductionCost(unsigned Opcode, VectorType *Ty,
                                  bool IsPairwiseForm,
                                  TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput);
 
-  int getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, int Index,
-                     VectorType *SubTp);
+  int getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, ArrayRef<int> Mask,
+                     int Index, VectorType *SubTp);
   /// @}
 };
 
 } // end namespace llvm
 
 #endif // LLVM_LIB_TARGET_AARCH64_AARCH64TARGETTRANSFORMINFO_H

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

int AArch64TTIImpl::getArithmeticReductionCost(unsigned Opcode,
   if (const auto *Entry = CostTableLookup(CostTblNoPairwise, ISD, MTy))
     return LT.first * Entry->Cost;
 
   return BaseT::getArithmeticReductionCost(Opcode, ValTy, IsPairwiseForm,
                                            CostKind);
 }
 
 int AArch64TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp,
-                                   int Index, VectorType *SubTp) {
+                                   ArrayRef<int> Mask, int Index,
+                                   VectorType *SubTp) {
   if (Kind == TTI::SK_Broadcast || Kind == TTI::SK_Transpose ||
       Kind == TTI::SK_Select || Kind == TTI::SK_PermuteSingleSrc ||
       Kind == TTI::SK_Reverse) {
     static const CostTblEntry ShuffleTbl[] = {
       // Broadcast shuffle kinds can be performed with 'dup'.
       { TTI::SK_Broadcast, MVT::v8i8,  1 },
       { TTI::SK_Broadcast, MVT::v16i8, 1 },
       { TTI::SK_Broadcast, MVT::v4i16, 1 },
static const CostTblEntry ShuffleTbl[] = {
       { TTI::SK_Reverse, MVT::nxv4f32,  1 },
       { TTI::SK_Reverse, MVT::nxv2f64,  1 },
     };
     std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);
     if (const auto *Entry = CostTableLookup(ShuffleTbl, Kind, LT.second))
       return LT.first * Entry->Cost;
   }
 
-  return BaseT::getShuffleCost(Kind, Tp, Index, SubTp);
+  return BaseT::getShuffleCost(Kind, Tp, Mask, Index, SubTp);
 }

llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.h

public:
   Optional<Value *> simplifyDemandedVectorEltsIntrinsic(
       InstCombiner &IC, IntrinsicInst &II, APInt DemandedElts, APInt &UndefElts,
       APInt &UndefElts2, APInt &UndefElts3,
       std::function<void(Instruction *, unsigned, APInt, APInt &)>
           SimplifyAndSetOp) const;
 
   unsigned getVectorSplitCost() { return 0; }
 
-  unsigned getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, int Index,
-                          VectorType *SubTp);
+  unsigned getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp,
+                          ArrayRef<int> Mask, int Index, VectorType *SubTp);
 
   bool areInlineCompatible(const Function *Caller,
                            const Function *Callee) const;
 
   unsigned getInliningThresholdMultiplier() { return 11; }
   unsigned adjustInliningThreshold(const CallBase *CB) const;
 
   int getInlinerVectorBonusPercent() { return 0; }

llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp

return B.CreateIntrinsic(Intrinsic::ptrmask, {NewV->getType(), MaskTy},
                              {NewV, MaskOp});
   }
   default:
     return nullptr;
   }
 }
 
 unsigned GCNTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, VectorType *VT,
-                                    int Index, VectorType *SubTp) {
+                                    ArrayRef<int> Mask, int Index,
+                                    VectorType *SubTp) {
   if (ST->hasVOP3PInsts()) {
     if (cast<FixedVectorType>(VT)->getNumElements() == 2 &&
         DL.getTypeSizeInBits(VT->getElementType()) == 16) {
       // With op_sel VOP3P instructions freely can access the low half or high
       // half of a register, so any swizzle is free.
 
       switch (Kind) {
       case TTI::SK_Broadcast:
       case TTI::SK_Reverse:
       case TTI::SK_PermuteSingleSrc:
         return 0;
       default:
         break;
       }
     }
   }
 
-  return BaseT::getShuffleCost(Kind, VT, Index, SubTp);
+  return BaseT::getShuffleCost(Kind, VT, Mask, Index, SubTp);
 }
 
 bool GCNTTIImpl::areInlineCompatible(const Function *Caller,
                                      const Function *Callee) const {
   const TargetMachine &TM = getTLI()->getTargetMachine();
   const GCNSubtarget *CallerST
     = static_cast<const GCNSubtarget *>(TM.getSubtargetImpl(*Caller));
   const GCNSubtarget *CalleeST

llvm/lib/Target/ARM/ARMTargetTransformInfo.h

public:
   bool isLegalMaskedScatter(Type *Ty, Align Alignment) {
     return isLegalMaskedGather(Ty, Alignment);
   }
 
   int getMemcpyCost(const Instruction *I);
 
   int getNumMemOps(const IntrinsicInst *I) const;
 
-  int getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, int Index,
-                     VectorType *SubTp);
+  int getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, ArrayRef<int> Mask,
+                     int Index, VectorType *SubTp);
 
   bool preferInLoopReduction(unsigned Opcode, Type *Ty,
                              TTI::ReductionFlags Flags) const;
 
   bool preferPredicatedReductionSelect(unsigned Opcode, Type *Ty,
                                        TTI::ReductionFlags Flags) const;
 
   bool shouldExpandReduction(const IntrinsicInst *II) const { return false; }

llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp

int ARMTTIImpl::getMemcpyCost(const Instruction *I) {
   // To model the cost of a library call, we assume 1 for the call, and
   // 3 for the argument setup.
   if (NumOps == -1)
     return 4;
   return NumOps;
 }
 
 int ARMTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp,
-                               int Index, VectorType *SubTp) {
+                               ArrayRef<int> Mask, int Index,
+                               VectorType *SubTp) {
   if (ST->hasNEON()) {
     if (Kind == TTI::SK_Broadcast) {
       static const CostTblEntry NEONDupTbl[] = {
           // VDUP handles these cases.
           {ISD::VECTOR_SHUFFLE, MVT::v2i32, 1},
           {ISD::VECTOR_SHUFFLE, MVT::v2f32, 1},
           {ISD::VECTOR_SHUFFLE, MVT::v2i64, 1},
           {ISD::VECTOR_SHUFFLE, MVT::v2f64, 1},
if (Kind == TTI::SK_Broadcast) {
                                               LT.second))
         return LT.first * Entry->Cost *
                ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput);
     }
   }
   int BaseCost = ST->hasMVEIntegerOps() && Tp->isVectorTy()
                      ? ST->getMVEVectorCostFactor(TTI::TCK_RecipThroughput)
                      : 1;
-  return BaseCost * BaseT::getShuffleCost(Kind, Tp, Index, SubTp);
+  return BaseCost * BaseT::getShuffleCost(Kind, Tp, Mask, Index, SubTp);
 }
 
 int ARMTTIImpl::getArithmeticInstrCost(unsigned Opcode, Type *Ty,
                                        TTI::TargetCostKind CostKind,
                                        TTI::OperandValueKind Op1Info,
                                        TTI::OperandValueKind Op2Info,
                                        TTI::OperandValueProperties Opd1PropInfo,
                                        TTI::OperandValueProperties Opd2PropInfo,

llvm/lib/Target/Hexagon/HexagonTargetTransformInfo.h

public:
   unsigned getMemoryOpCost(unsigned Opcode, Type *Src, MaybeAlign Alignment,
                            unsigned AddressSpace,
                            TTI::TargetCostKind CostKind,
                            const Instruction *I = nullptr);
   unsigned
   getMaskedMemoryOpCost(unsigned Opcode, Type *Src, Align Alignment,
                         unsigned AddressSpace,
                         TTI::TargetCostKind CostKind = TTI::TCK_SizeAndLatency);
-  unsigned getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index,
-            Type *SubTp);
+  unsigned getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, ArrayRef<int> Mask,
+                          int Index, Type *SubTp);
   unsigned getGatherScatterOpCost(unsigned Opcode, Type *DataTy,
                                   const Value *Ptr, bool VariableMask,
                                   Align Alignment, TTI::TargetCostKind CostKind,
                                   const Instruction *I);
   unsigned getInterleavedMemoryOpCost(
       unsigned Opcode, Type *VecTy, unsigned Factor, ArrayRef<unsigned> Indices,
       Align Alignment, unsigned AddressSpace,
       TTI::TargetCostKind CostKind = TTI::TCK_SizeAndLatency,

llvm/lib/Target/Hexagon/HexagonTargetTransformInfo.cpp

unsigned HexagonTTIImpl::getMaskedMemoryOpCost(unsigned Opcode, Type *Src,
                                                Align Alignment,
                                                unsigned AddressSpace,
                                                TTI::TargetCostKind CostKind) {
   return BaseT::getMaskedMemoryOpCost(Opcode, Src, Alignment, AddressSpace,
                                       CostKind);
 }
 
 unsigned HexagonTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, Type *Tp,
-      int Index, Type *SubTp) {
+                                        ArrayRef<int> Mask, int Index,
+                                        Type *SubTp) {
   return 1;
 }
 
 unsigned HexagonTTIImpl::getGatherScatterOpCost(
     unsigned Opcode, Type *DataTy, const Value *Ptr, bool VariableMask,
     Align Alignment, TTI::TargetCostKind CostKind, const Instruction *I) {
   return BaseT::getGatherScatterOpCost(Opcode, DataTy, Ptr, VariableMask,
                                        Alignment, CostKind, I);

llvm/lib/Target/PowerPC/PPCTargetTransformInfo.h

int getArithmeticInstrCost(
       unsigned Opcode, Type *Ty,
       TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput,
       TTI::OperandValueKind Opd1Info = TTI::OK_AnyValue,
       TTI::OperandValueKind Opd2Info = TTI::OK_AnyValue,
       TTI::OperandValueProperties Opd1PropInfo = TTI::OP_None,
       TTI::OperandValueProperties Opd2PropInfo = TTI::OP_None,
       ArrayRef<const Value *> Args = ArrayRef<const Value *>(),
       const Instruction *CxtI = nullptr);
-  int getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index, Type *SubTp);
+  int getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, ArrayRef<int> Mask,
+                     int Index, Type *SubTp);
   int getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
                        TTI::CastContextHint CCH, TTI::TargetCostKind CostKind,
                        const Instruction *I = nullptr);
   int getCFInstrCost(unsigned Opcode, TTI::TargetCostKind CostKind);
   int getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy,
                          CmpInst::Predicate VecPred,
                          TTI::TargetCostKind CostKind,
                          const Instruction *I = nullptr);

llvm/lib/Target/PowerPC/PPCTargetTransformInfo.cpp

int PPCTTIImpl::getArithmeticInstrCost(unsigned Opcode, Type *Ty,
 
   // Fallback to the default implementation.
   int Cost = BaseT::getArithmeticInstrCost(Opcode, Ty, CostKind, Op1Info,
                                            Op2Info,
                                            Opd1PropInfo, Opd2PropInfo);
   return vectorCostAdjustment(Cost, Opcode, Ty, nullptr);
 }
 
-int PPCTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index,
-                               Type *SubTp) {
+int PPCTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, Type *Tp,
+                               ArrayRef<int> Mask, int Index, Type *SubTp) {
   // Legalize the type.
   std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);
 
   // PPC, for both Altivec/VSX, support cheap arbitrary permutations
   // (at least in the sense that there need only be one non-loop-invariant
   // instruction). We need one such shuffle instruction for each actual
   // register (this is not true for arbitrary shuffles, but is true for the
   // structured types of shuffles covered by TTI::ShuffleKind).

llvm/lib/Target/SystemZ/SystemZTargetTransformInfo.h

int getArithmeticInstrCost(
       unsigned Opcode, Type *Ty,
       TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput,
       TTI::OperandValueKind Opd1Info = TTI::OK_AnyValue,
       TTI::OperandValueKind Opd2Info = TTI::OK_AnyValue,
       TTI::OperandValueProperties Opd1PropInfo = TTI::OP_None,
       TTI::OperandValueProperties Opd2PropInfo = TTI::OP_None,
       ArrayRef<const Value *> Args = ArrayRef<const Value *>(),
       const Instruction *CxtI = nullptr);
-  int getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, int Index,
-                     VectorType *SubTp);
+  int getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, ArrayRef<int> Mask,
+                     int Index, VectorType *SubTp);
   unsigned getVectorTruncCost(Type *SrcTy, Type *DstTy);
   unsigned getVectorBitmaskConversionCost(Type *SrcTy, Type *DstTy);
   unsigned getBoolVecToIntConversionCost(unsigned Opcode, Type *Dst,
                                          const Instruction *I);
   int getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
                        TTI::CastContextHint CCH, TTI::TargetCostKind CostKind,
                        const Instruction *I = nullptr);
   int getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy,

llvm/lib/Target/SystemZ/SystemZTargetTransformInfo.cpp

int SystemZTTIImpl::getArithmeticInstrCost(
   }
 
   // Fallback to the default implementation.
   return BaseT::getArithmeticInstrCost(Opcode, Ty, CostKind, Op1Info, Op2Info,
                                        Opd1PropInfo, Opd2PropInfo, Args, CxtI);
 }
 
 int SystemZTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp,
-                                   int Index, VectorType *SubTp) {
+                                   ArrayRef<int> Mask, int Index,
+                                   VectorType *SubTp) {
   if (ST->hasVector()) {
     unsigned NumVectors = getNumVectorRegs(Tp);
 
     // TODO: Since fp32 is expanded, the shuffle cost should always be 0.
 
     // FP128 values are always in scalar registers, so there is no work
     // involved with a shuffle, except for broadcast. In that case register
     // moves are done with a single instruction per element.
if (ST->hasVector()) {
 
     default:
 
       // SystemZ supports single instruction permutation / replication.
       return NumVectors;
     }
   }
 
-  return BaseT::getShuffleCost(Kind, Tp, Index, SubTp);
+  return BaseT::getShuffleCost(Kind, Tp, Mask, Index, SubTp);
 }
 
 // Return the log2 difference of the element sizes of the two vector types.
 static unsigned getElSizeLog2Diff(Type *Ty0, Type *Ty1) {
   unsigned Bits0 = Ty0->getScalarSizeInBits();
   unsigned Bits1 = Ty1->getScalarSizeInBits();
 
   if (Bits1 >  Bits0)

llvm/lib/Target/X86/X86TargetTransformInfo.h

int getArithmeticInstrCost(
       unsigned Opcode, Type *Ty,
       TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput,
       TTI::OperandValueKind Opd1Info = TTI::OK_AnyValue,
       TTI::OperandValueKind Opd2Info = TTI::OK_AnyValue,
       TTI::OperandValueProperties Opd1PropInfo = TTI::OP_None,
       TTI::OperandValueProperties Opd2PropInfo = TTI::OP_None,
       ArrayRef<const Value *> Args = ArrayRef<const Value *>(),
       const Instruction *CxtI = nullptr);
-  int getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, int Index,
-                     VectorType *SubTp);
+  int getShuffleCost(TTI::ShuffleKind Kind, VectorType *Tp, ArrayRef<int> Mask,
+                     int Index, VectorType *SubTp);
   int getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
                        TTI::CastContextHint CCH, TTI::TargetCostKind CostKind,
                        const Instruction *I = nullptr);
   int getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy,
                          CmpInst::Predicate VecPred,
                          TTI::TargetCostKind CostKind,
                          const Instruction *I = nullptr);
   int getVectorInstrCost(unsigned Opcode, Type *Val, unsigned Index);

llvm/lib/Target/X86/X86TargetTransformInfo.cpp

if (LT.second.isVector() && (ISD == ISD::SDIV || ISD == ISD::SREM ||
     return 20 * LT.first * LT.second.getVectorNumElements() * ScalarCost;
   }
 
   // Fallback to the default implementation.
   return BaseT::getArithmeticInstrCost(Opcode, Ty, CostKind, Op1Info, Op2Info);
 }
 
 int X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, VectorType *BaseTp,
-                               int Index, VectorType *SubTp) {
+                               ArrayRef<int> Mask, int Index,
+                               VectorType *SubTp) {
   // 64-bit packed float vectors (v2f32) are widened to type v4f32.
   // 64-bit packed integer vectors (v2i32) are widened to type v4i32.
   std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, BaseTp);
 
   // Treat Transpose as 2-op shuffles - there's no difference in lowering.
   if (Kind == TTI::SK_Transpose)
     Kind = TTI::SK_PermuteTwoSrc;
 
if (SubLT.second.isVector()) {
         assert(NumElts >= NumSubElts && NumElts > OrigSubElts &&
                "Unexpected number of elements!");
         auto *VecTy = FixedVectorType::get(BaseTp->getElementType(),
                                            LT.second.getVectorNumElements());
         auto *SubTy = FixedVectorType::get(BaseTp->getElementType(),
                                            SubLT.second.getVectorNumElements());
         int ExtractIndex = alignDown((Index % NumElts), NumSubElts);
         int ExtractCost = getShuffleCost(TTI::SK_ExtractSubvector, VecTy,
-                                         ExtractIndex, SubTy);
+                                         ArrayRef<int>(), ExtractIndex, SubTy);
 
         // If the original size is 32-bits or more, we can use pshufd. Otherwise
         // if we have SSSE3 we can use pshufb.
         if (SubTp->getPrimitiveSizeInBits() >= 32 || ST->hasSSSE3())
           return ExtractCost + 1; // pshufd or pshufb
 
         assert(SubTp->getPrimitiveSizeInBits() == 16 &&
                "Unexpected vector size");
if (LegalVT.isVector() &&
       unsigned NumOfSrcs = (VecTySize + LegalVTSize - 1) / LegalVTSize;
       // Number of destination vectors after legalization:
       unsigned NumOfDests = LT.first;
 
       auto *SingleOpTy = FixedVectorType::get(BaseTp->getElementType(),
                                               LegalVT.getVectorNumElements());
 
       unsigned NumOfShuffles = (NumOfSrcs - 1) * NumOfDests;
-      return NumOfShuffles *
-             getShuffleCost(TTI::SK_PermuteTwoSrc, SingleOpTy, 0, nullptr);
+      return NumOfShuffles * getShuffleCost(TTI::SK_PermuteTwoSrc, SingleOpTy,
+                                            ArrayRef<int>(), 0, nullptr);
     }
 
-    return BaseT::getShuffleCost(Kind, BaseTp, Index, SubTp);
+    return BaseT::getShuffleCost(Kind, BaseTp, Mask, Index, SubTp);
   }
 
   // For 2-input shuffles, we must account for splitting the 2 inputs into many.
   if (Kind == TTI::SK_PermuteTwoSrc && LT.first != 1) {
     // We assume that source and destination have the same vector type.
     int NumOfDests = LT.first;
     int NumOfShufflesPerDest = LT.first * 2 - 1;
     LT.first = NumOfDests * NumOfShufflesPerDest;
static const CostTblEntry SSE1ShuffleTbl[] = {
     { TTI::SK_PermuteSingleSrc, MVT::v4f32, 1 }, // shufps
     { TTI::SK_PermuteTwoSrc,    MVT::v4f32, 2 }, // 2*shufps
   };
 
   if (ST->hasSSE1())
     if (const auto *Entry = CostTableLookup(SSE1ShuffleTbl, Kind, LT.second))
       return LT.first * Entry->Cost;
 
-  return BaseT::getShuffleCost(Kind, BaseTp, Index, SubTp);
+  return BaseT::getShuffleCost(Kind, BaseTp, Mask, Index, SubTp);
 }
 
 int X86TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
                                  TTI::CastContextHint CCH,
                                  TTI::TargetCostKind CostKind,
                                  const Instruction *I) {
   int ISD = TLI->InstructionOpcodeToISD(Opcode);
   assert(ISD && "Invalid opcode");
if (Index != -1U && (Opcode == Instruction::ExtractElement ||
     // If the vector type is already less than 128-bits then don't reduce it.
     // TODO: Under what circumstances should we shuffle using the full width?
     int ShuffleCost = 1;
     if (Opcode == Instruction::InsertElement) {
       auto *SubTy = cast<VectorType>(Val);
       EVT VT = TLI->getValueType(DL, Val);
       if (VT.getScalarType() != MScalarTy || VT.getSizeInBits() >= 128)
         SubTy = FixedVectorType::get(ScalarType, SubNumElts);
-      ShuffleCost = getShuffleCost(TTI::SK_PermuteTwoSrc, SubTy, 0, SubTy);
+      ShuffleCost = getShuffleCost(TTI::SK_PermuteTwoSrc, SubTy,
+                                   ArrayRef<int>(), 0, SubTy);
     }
     int IntOrFpCost = ScalarType->isFloatingPointTy() ? 0 : 1;
     return ShuffleCost + IntOrFpCost + RegisterFileMoveCost;
   }
 
   // Add to the base cost if we know that the extracted element of a vector is
   // destined to be moved to and used in the integer register file.
   if (Opcode == Instruction::ExtractElement && ScalarType->isPointerTy())
int X86TTIImpl::getMaskedMemoryOpCost(unsigned Opcode, Type *SrcTy,
 
   // Legalize the type.
   std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, SrcVTy);
   auto VT = TLI->getValueType(DL, SrcVTy);
   int Cost = 0;
   if (VT.isSimple() && LT.second != VT.getSimpleVT() &&
       LT.second.getVectorNumElements() == NumElem)
     // Promotion requires expand/truncate for data and a shuffle for mask.
-    Cost += getShuffleCost(TTI::SK_PermuteTwoSrc, SrcVTy, 0, nullptr) +
-            getShuffleCost(TTI::SK_PermuteTwoSrc, MaskTy, 0, nullptr);
+    Cost += getShuffleCost(TTI::SK_PermuteTwoSrc, SrcVTy, ArrayRef<int>(), 0,
+                           nullptr) +
+            getShuffleCost(TTI::SK_PermuteTwoSrc, MaskTy, ArrayRef<int>(), 0,
+                           nullptr);
 
   else if (LT.second.getVectorNumElements() > NumElem) {
     auto *NewMaskTy = FixedVectorType::get(MaskTy->getElementType(),
                                            LT.second.getVectorNumElements());
     // Expanding requires fill mask with zeroes
-    Cost += getShuffleCost(TTI::SK_InsertSubvector, NewMaskTy, 0, MaskTy);
+    Cost += getShuffleCost(TTI::SK_InsertSubvector, NewMaskTy, ArrayRef<int>(),
+                           0, MaskTy);
   }
 
   // Pre-AVX512 - each maskmov load costs 2 + store costs ~8.
   if (!ST->hasAVX512())
     return Cost + LT.first * (IsLoad ? 2 : 8);
 
   // AVX-512 masked load/store is cheapper
   return Cost + LT.first;
int X86TTIImpl::getArithmeticReductionCost(unsigned Opcode, VectorType *ValTy,
   // at each level.
   while (NumVecElts > 1) {
     // Determine the size of the remaining vector we need to reduce.
     unsigned Size = NumVecElts * ScalarSize;
     NumVecElts /= 2;
     // If we're reducing from 256/512 bits, use an extract_subvector.
     if (Size > 128) {
       auto *SubTy = FixedVectorType::get(ValVTy->getElementType(), NumVecElts);
-      ReductionCost +=
-          getShuffleCost(TTI::SK_ExtractSubvector, Ty, NumVecElts, SubTy);
+      ReductionCost += getShuffleCost(TTI::SK_ExtractSubvector, Ty,
+                                      ArrayRef<int>(), NumVecElts, SubTy);
       Ty = SubTy;
     } else if (Size == 128) {
       // Reducing from 128 bits is a permute of v2f64/v2i64.
       FixedVectorType *ShufTy;
       if (ValVTy->isFloatingPointTy())
         ShufTy =
             FixedVectorType::get(Type::getDoubleTy(ValVTy->getContext()), 2);
       else
         ShufTy =
             FixedVectorType::get(Type::getInt64Ty(ValVTy->getContext()), 2);
-      ReductionCost +=
-          getShuffleCost(TTI::SK_PermuteSingleSrc, ShufTy, 0, nullptr);
+      ReductionCost += getShuffleCost(TTI::SK_PermuteSingleSrc, ShufTy,
+                                      ArrayRef<int>(), 0, nullptr);
     } else if (Size == 64) {
       // Reducing from 64 bits is a shuffle of v4f32/v4i32.
       FixedVectorType *ShufTy;
       if (ValVTy->isFloatingPointTy())
         ShufTy =
             FixedVectorType::get(Type::getFloatTy(ValVTy->getContext()), 4);
       else
         ShufTy =
             FixedVectorType::get(Type::getInt32Ty(ValVTy->getContext()), 4);
-      ReductionCost +=
-          getShuffleCost(TTI::SK_PermuteSingleSrc, ShufTy, 0, nullptr);
+      ReductionCost += getShuffleCost(TTI::SK_PermuteSingleSrc, ShufTy,
+                                      ArrayRef<int>(), 0, nullptr);
     } else {
       // Reducing from smaller size is a shift by immediate.
       auto *ShiftTy = FixedVectorType::get(
           Type::getIntNTy(ValVTy->getContext(), Size), 128 / Size);
       ReductionCost += getArithmeticInstrCost(
           Instruction::LShr, ShiftTy, CostKind,
           TargetTransformInfo::OK_AnyValue,
           TargetTransformInfo::OK_UniformConstantValue,
int X86TTIImpl::getMinMaxReductionCost(VectorType *ValTy, VectorType *CondTy,
   // at each level.
   while (NumVecElts > 1) {
     // Determine the size of the remaining vector we need to reduce.
     unsigned Size = NumVecElts * ScalarSize;
     NumVecElts /= 2;
     // If we're reducing from 256/512 bits, use an extract_subvector.
     if (Size > 128) {
       auto *SubTy = FixedVectorType::get(ValVTy->getElementType(), NumVecElts);
-      MinMaxCost +=
-          getShuffleCost(TTI::SK_ExtractSubvector, Ty, NumVecElts, SubTy);
+      MinMaxCost += getShuffleCost(TTI::SK_ExtractSubvector, Ty,
+                                   ArrayRef<int>(), NumVecElts, SubTy);
       Ty = SubTy;
     } else if (Size == 128) {
       // Reducing from 128 bits is a permute of v2f64/v2i64.
       VectorType *ShufTy;
       if (ValTy->isFloatingPointTy())
         ShufTy =
             FixedVectorType::get(Type::getDoubleTy(ValTy->getContext()), 2);
       else
         ShufTy = FixedVectorType::get(Type::getInt64Ty(ValTy->getContext()), 2);
-      MinMaxCost +=
-          getShuffleCost(TTI::SK_PermuteSingleSrc, ShufTy, 0, nullptr);
+      MinMaxCost += getShuffleCost(TTI::SK_PermuteSingleSrc, ShufTy,
+                                   ArrayRef<int>(), 0, nullptr);
     } else if (Size == 64) {
       // Reducing from 64 bits is a shuffle of v4f32/v4i32.
       FixedVectorType *ShufTy;
       if (ValTy->isFloatingPointTy())
         ShufTy = FixedVectorType::get(Type::getFloatTy(ValTy->getContext()), 4);
       else
         ShufTy = FixedVectorType::get(Type::getInt32Ty(ValTy->getContext()), 4);
-      MinMaxCost +=
-          getShuffleCost(TTI::SK_PermuteSingleSrc, ShufTy, 0, nullptr);
+      MinMaxCost += getShuffleCost(TTI::SK_PermuteSingleSrc, ShufTy,
+                                   ArrayRef<int>(), 0, nullptr);
     } else {
       // Reducing from smaller size is a shift by immediate.
       auto *ShiftTy = FixedVectorType::get(
           Type::getIntNTy(ValTy->getContext(), Size), 128 / Size);
       MinMaxCost += getArithmeticInstrCost(
           Instruction::LShr, ShiftTy, TTI::TCK_RecipThroughput,
           TargetTransformInfo::OK_AnyValue,
           TargetTransformInfo::OK_UniformConstantValue,
if (Opcode == Instruction::Load) {
 
     // Kind of shuffle depends on number of loaded values.
     // If we load the entire data in one register, we can use a 1-src shuffle.
     // Otherwise, we'll merge 2 sources in each operation.
     TTI::ShuffleKind ShuffleKind =
         (NumOfMemOps > 1) ? TTI::SK_PermuteTwoSrc : TTI::SK_PermuteSingleSrc;
 
     unsigned ShuffleCost =
-        getShuffleCost(ShuffleKind, SingleMemOpTy, 0, nullptr);
+        getShuffleCost(ShuffleKind, SingleMemOpTy, ArrayRef<int>(), 0, nullptr);
 
     unsigned NumOfLoadsInInterleaveGrp =
         Indices.size() ? Indices.size() : Factor;
     auto *ResultTy = FixedVectorType::get(VecTy->getElementType(),
                                           VecTy->getNumElements() / Factor);
     unsigned NumOfResults =
         getTLI()->getTypeLegalizationCost(DL, ResultTy).first *
         NumOfLoadsInInterleaveGrp;
int X86TTIImpl::getInterleavedMemoryOpCostAVX512(
   if (const auto *Entry =
           CostTableLookup(AVX512InterleavedStoreTbl, Factor, VT))
     return NumOfMemOps * MemOpCost + Entry->Cost;
   //If an entry does not exist, fallback to the default implementation.
 
   // There is no strided stores meanwhile. And store can't be folded in
   // shuffle.
   unsigned NumOfSources = Factor; // The number of values to be merged.
-  unsigned ShuffleCost =
-      getShuffleCost(TTI::SK_PermuteTwoSrc, SingleMemOpTy, 0, nullptr);
+  unsigned ShuffleCost = getShuffleCost(TTI::SK_PermuteTwoSrc, SingleMemOpTy,
+                                        ArrayRef<int>(), 0, nullptr);
   unsigned NumOfShufflesPerStore = NumOfSources - 1;
 
   // The SK_MergeTwoSrc shuffle clobbers one of src operands.
   // We need additional instructions to keep sources.
   unsigned NumOfMoves = NumOfMemOps * NumOfShufflesPerStore / 2;
   int Cost = NumOfMemOps * (MemOpCost + NumOfShufflesPerStore * ShuffleCost) +
              NumOfMoves;
   return Cost;

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

if (Legal->isMaskRequired(I))
     Cost += TTI.getMaskedMemoryOpCost(I->getOpcode(), VectorTy, Alignment, AS,
                                       CostKind);
   else
     Cost += TTI.getMemoryOpCost(I->getOpcode(), VectorTy, Alignment, AS,
                                 CostKind, I);
 
   bool Reverse = ConsecutiveStride < 0;
   if (Reverse)
-    Cost += TTI.getShuffleCost(TargetTransformInfo::SK_Reverse, VectorTy, 0);
+    Cost += TTI.getShuffleCost(TargetTransformInfo::SK_Reverse, VectorTy,
+                               ArrayRef<int>(), 0);
   return Cost;
 }
 
 InstructionCost
 LoopVectorizationCostModel::getUniformMemOpCost(Instruction *I,
                                                 ElementCount VF) {
   assert(Legal->isUniformMemOp(*I));
 
   Type *ValTy = getMemInstValueType(I);
   auto *VectorTy = cast<VectorType>(ToVectorTy(ValTy, VF));
   const Align Alignment = getLoadStoreAlignment(I);
   unsigned AS = getLoadStoreAddressSpace(I);
   enum TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
   if (isa<LoadInst>(I)) {
     return TTI.getAddressComputationCost(ValTy) +
            TTI.getMemoryOpCost(Instruction::Load, ValTy, Alignment, AS,
                                CostKind) +
-           TTI.getShuffleCost(TargetTransformInfo::SK_Broadcast, VectorTy);
+           TTI.getShuffleCost(TargetTransformInfo::SK_Broadcast, VectorTy,
+                              ArrayRef<int>());
   }
   StoreInst *SI = cast<StoreInst>(I);
 
   bool isLoopInvariantStoreValue = Legal->isUniform(SI->getValueOperand());
   return TTI.getAddressComputationCost(ValTy) +
          TTI.getMemoryOpCost(Instruction::Store, ValTy, Alignment, AS,
                              CostKind) +
          (isLoopInvariantStoreValue
InstructionCost Cost = TTI.getInterleavedMemoryOpCost(
       I->getOpcode(), WideVecTy, Group->getFactor(), Indices, Group->getAlign(),
       AS, TTI::TCK_RecipThroughput, Legal->isMaskRequired(I), UseMaskForGaps);
 
   if (Group->isReverse()) {
     // TODO: Add support for reversed masked interleaved access.
     assert(!Legal->isMaskRequired(I) &&
            "Reverse masked interleaved access not supported.");
     Cost += Group->getNumMembers() *
-            TTI.getShuffleCost(TargetTransformInfo::SK_Reverse, VectorTy, 0);
+            TTI.getShuffleCost(TargetTransformInfo::SK_Reverse, VectorTy,
+                               ArrayRef<int>(), 0);
   }
   return Cost;
 }
 
 InstructionCost LoopVectorizationCostModel::getReductionPatternCost(
     Instruction *I, ElementCount VF, Type *Ty, TTI::TargetCostKind CostKind) {
   // Early exit for no inloop reductions
   if (InLoopReductionChains.empty() || VF.isScalar() || !isa<VectorType>(Ty))
case Instruction::Br: {
     // may decide in the future to call TTI for all branches.
   }
   case Instruction::PHI: {
     auto *Phi = cast<PHINode>(I);
 
     // First-order recurrences are replaced by vector shuffles inside the loop.
     // NOTE: Don't use ToVectorTy as SK_ExtractSubvector expects a vector type.
     if (VF.isVector() && Legal->isFirstOrderRecurrence(Phi))
-      return TTI.getShuffleCost(
-          TargetTransformInfo::SK_ExtractSubvector, cast<VectorType>(VectorTy),
-          VF.getKnownMinValue() - 1, FixedVectorType::get(RetTy, 1));
+      return TTI.getShuffleCost(TargetTransformInfo::SK_ExtractSubvector,
+                                cast<VectorType>(VectorTy), ArrayRef<int>(),
+                                VF.getKnownMinValue() - 1,
+                                FixedVectorType::get(RetTy, 1));
 
     // Phi nodes in non-header blocks (not inductions, reductions, etc.) are
     // converted into select instructions. We require N - 1 selects per phi
     // node, where N is the number of incoming values.
     if (VF.isVector() && Phi->getParent() != TheLoop->getHeader())
       return (Phi->getNumIncomingValues() - 1) *
              TTI.getCmpSelInstrCost(
                  Instruction::Select, ToVectorTy(Phi->getType(), VF),

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

 /// %ins3 = insertelement <4 x i8> %ins2, i8 %8, i32 2
 /// %9 = extractelement <4 x i8> %5, i32 3
 /// %ins4 = insertelement <4 x i8> %ins3, i8 %9, i32 3
 /// ret <4 x i8> %ins4
 /// InstCombiner transforms this into a shuffle and vector mul
 /// TODO: Can we split off and reuse the shuffle mask detection from
 /// TargetTransformInfo::getInstructionThroughput?
 static Optional<TargetTransformInfo::ShuffleKind>
 isShuffle(ArrayRef<Value *> VL) {
   auto *EI0 = cast<ExtractElementInst>(VL[0]);
   unsigned Size =
       cast<FixedVectorType>(EI0->getVectorOperandType())->getNumElements();
   Value *Vec1 = nullptr;
   Value *Vec2 = nullptr;
   enum ShuffleMode { Unknown, Select, Permute };
   ShuffleMode CommonShuffleMode = Unknown;
   for (unsigned I = 0, E = VL.size(); I < E; ++I) {
     auto *EI = cast<ExtractElementInst>(VL[I]);
     auto *Vec = EI->getVectorOperand();
     // All vector operands must have the same number of vector elements.
     if (cast<FixedVectorType>(Vec->getType())->getNumElements() != Size)
       return None;
     auto *Idx = dyn_cast<ConstantInt>(EI->getIndexOperand());
     if (!Idx)
       return None;
     // Undefined behavior if Idx is negative or >= Size.
     if (Idx->getValue().uge(Size))
       continue;
     unsigned IntIdx = Idx->getValue().getZExtValue();
     // We can extractelement from undef or poison vector.
     if (isa<UndefValue>(Vec))
       continue;
     // For correct shuffling we have to have at most 2 different vector operands
     // in all extractelement instructions.
     if (!Vec1 || Vec1 == Vec)
       Vec1 = Vec;
InstructionCost BoUpSLP::getEntryCost(TreeEntry *E) {
   if (MinBWs.count(VL[0]))
     VecTy = FixedVectorType::get(
         IntegerType::get(F->getContext(), MinBWs[VL[0]].first), VL.size());
 
   unsigned ReuseShuffleNumbers = E->ReuseShuffleIndices.size();
   bool NeedToShuffleReuses = !E->ReuseShuffleIndices.empty();
   InstructionCost ReuseShuffleCost = 0;
   if (NeedToShuffleReuses) {
-    ReuseShuffleCost =
-        TTI->getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, VecTy);
+    ReuseShuffleCost = TTI->getShuffleCost(
+        TargetTransformInfo::SK_PermuteSingleSrc, VecTy, ArrayRef<int>());
   }
   if (E->State == TreeEntry::NeedToGather) {
     if (allConstant(VL))
       return 0;
     if (isSplat(VL)) {
       return ReuseShuffleCost +
-             TTI->getShuffleCost(TargetTransformInfo::SK_Broadcast, VecTy, 0);
+             TTI->getShuffleCost(TargetTransformInfo::SK_Broadcast, VecTy,
+                                 ArrayRef<int>(), 0);
     }
     if (E->getOpcode() == Instruction::ExtractElement &&
         allSameType(VL) && allSameBlock(VL)) {
       Optional<TargetTransformInfo::ShuffleKind> ShuffleKind = isShuffle(VL);
       if (ShuffleKind.hasValue()) {
         InstructionCost Cost =
-            TTI->getShuffleCost(ShuffleKind.getValue(), VecTy);
+            TTI->getShuffleCost(ShuffleKind.getValue(), VecTy, ArrayRef<int>());
         for (auto *V : VL) {
           // If all users of instruction are going to be vectorized and this
           // instruction itself is not going to be vectorized, consider this
           // instruction as dead and remove its cost from the final cost of the
           // vectorized tree.
           if (areAllUsersVectorized(cast<Instruction>(V)) &&
               !ScalarToTreeEntry.count(V)) {
             auto *IO = cast<ConstantInt>(
case Instruction::ExtractElement: {
             --Idx;
           }
           ReuseShuffleCost +=
               TTI->getVectorInstrCost(Instruction::ExtractElement, VecTy, Idx);
         }
         CommonCost = ReuseShuffleCost;
       } else if (!E->ReorderIndices.empty()) {
         CommonCost = TTI->getShuffleCost(
-            TargetTransformInfo::SK_PermuteSingleSrc, VecTy);
+            TargetTransformInfo::SK_PermuteSingleSrc, VecTy, ArrayRef<int>());
       }
       for (unsigned I = 0, E = VL.size(); I < E; ++I) {
         Instruction *EI = cast<Instruction>(VL[I]);
         // If all users are going to be vectorized, instruction can be
         // considered as dead.
         // The same, if have only one user, it will be vectorized for sure.
         if (areAllUsersVectorized(EI)) {
           // Take credit for instruction that will become dead.
case Instruction::Load: {
       } else {
         assert(E->State == TreeEntry::ScatterVectorize && "Unknown EntryState");
         VecLdCost = TTI->getGatherScatterOpCost(
             Instruction::Load, VecTy, cast<LoadInst>(VL0)->getPointerOperand(),
             /*VariableMask=*/false, alignment, CostKind, VL0);
       }
       if (!NeedToShuffleReuses && !E->ReorderIndices.empty())
         VecLdCost += TTI->getShuffleCost(
-            TargetTransformInfo::SK_PermuteSingleSrc, VecTy);
+            TargetTransformInfo::SK_PermuteSingleSrc, VecTy, ArrayRef<int>());
       LLVM_DEBUG(dumpTreeCosts(E, ReuseShuffleCost, VecLdCost, ScalarLdCost));
       return ReuseShuffleCost + VecLdCost - ScalarLdCost;
     }
     case Instruction::Store: {
       // We know that we can merge the stores. Calculate the cost.
       bool IsReorder = !E->ReorderIndices.empty();
       auto *SI =
           cast<StoreInst>(IsReorder ? VL[E->ReorderIndices.front()] : VL0);
       Align Alignment = SI->getAlign();
       InstructionCost ScalarEltCost = TTI->getMemoryOpCost(
           Instruction::Store, ScalarTy, Alignment, 0, CostKind, VL0);
       InstructionCost ScalarStCost = VecTy->getNumElements() * ScalarEltCost;
       InstructionCost VecStCost = TTI->getMemoryOpCost(
           Instruction::Store, VecTy, Alignment, 0, CostKind, VL0);
       if (IsReorder)
         VecStCost += TTI->getShuffleCost(
-            TargetTransformInfo::SK_PermuteSingleSrc, VecTy);
+            TargetTransformInfo::SK_PermuteSingleSrc, VecTy, ArrayRef<int>());
       LLVM_DEBUG(dumpTreeCosts(E, ReuseShuffleCost, VecStCost, ScalarStCost));
       return VecStCost - ScalarStCost;
     }
     case Instruction::Call: {
       CallInst *CI = cast<CallInst>(VL0);
       Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
 
       // Calculate the cost of the scalar and vector calls.
case Instruction::ShuffleVector: {
         Type *Src1SclTy = E->getAltOp()->getOperand(0)->getType();
         auto *Src0Ty = FixedVectorType::get(Src0SclTy, VL.size());
         auto *Src1Ty = FixedVectorType::get(Src1SclTy, VL.size());
         VecCost = TTI->getCastInstrCost(E->getOpcode(), VecTy, Src0Ty,
                                         TTI::CastContextHint::None, CostKind);
         VecCost += TTI->getCastInstrCost(E->getAltOpcode(), VecTy, Src1Ty,
                                          TTI::CastContextHint::None, CostKind);
       }
-      VecCost += TTI->getShuffleCost(TargetTransformInfo::SK_Select, VecTy, 0);
+      VecCost += TTI->getShuffleCost(TargetTransformInfo::SK_Select, VecTy,
+                                     ArrayRef<int>(), 0);
       LLVM_DEBUG(dumpTreeCosts(E, ReuseShuffleCost, VecCost, ScalarCost));
       return ReuseShuffleCost + VecCost - ScalarCost;
     }
     default:
       llvm_unreachable("Unknown instruction");
   }
 }
 
 bool BoUpSLP::isFullyVectorizableTinyTree() const {
   LLVM_DEBUG(dbgs() << "SLP: Check whether the tree with height "
                     << VectorizableTree.size() << " is fully vectorizable .\n");
 
BoUpSLP::getGatherCost(FixedVectorType *Ty,
   APInt DemandedElts = APInt::getNullValue(NumElts);
   for (unsigned I = 0; I < NumElts; ++I)
     if (!ShuffledIndices.count(I))
       DemandedElts.setBit(I);
   InstructionCost Cost =
       TTI->getScalarizationOverhead(Ty, DemandedElts, /*Insert*/ true,
                                     /*Extract*/ false);
   if (!ShuffledIndices.empty())
-    Cost += TTI->getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, Ty);
+    Cost += TTI->getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, Ty,
+                                ArrayRef<int>());
   return Cost;
 }
 
 InstructionCost BoUpSLP::getGatherCost(ArrayRef<Value *> VL) const {
   // Find the type of the operands in VL.
   Type *ScalarTy = VL[0]->getType();
   if (StoreInst *SI = dyn_cast<StoreInst>(VL[0]))
     ScalarTy = SI->getValueOperand()->getType();

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

bool VectorCombine::vectorizeLoadInsert(Instruction &I) {
   OldCost += TTI.getScalarizationOverhead(MinVecTy, DemandedElts,
                                           /* Insert */ true, HasExtract);
 
   // New pattern: load VecPtr
   InstructionCost NewCost =
       TTI.getMemoryOpCost(Instruction::Load, MinVecTy, Alignment, AS);
   // Optionally, we are shuffling the loaded vector element(s) into place.
   if (OffsetEltIndex)
-    NewCost += TTI.getShuffleCost(TTI::SK_PermuteSingleSrc, MinVecTy);
+    NewCost +=
+        TTI.getShuffleCost(TTI::SK_PermuteSingleSrc, MinVecTy, ArrayRef<int>());
 
   // We can aggressively convert to the vector form because the backend can
   // invert this transform if it does not result in a performance win.
   if (OldCost < NewCost || !NewCost.isValid())
     return false;
 
   // It is safe and potentially profitable to load a vector directly:
   // inselt undef, load Scalar, 0 --> load VecPtr
if (ConvertToShuffle) {
 
     // If we are extracting from 2 different indexes, then one operand must be
     // shuffled before performing the vector operation. The shuffle mask is
     // undefined except for 1 lane that is being translated to the remaining
     // extraction lane. Therefore, it is a splat shuffle. Ex:
     // ShufMask = { undef, undef, 0, undef }
     // TODO: The cost model has an option for a "broadcast" shuffle
     //       (splat-from-element-0), but no option for a more general splat.
-    NewCost +=
-        TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, VecTy);
+    NewCost += TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc,
+                                  VecTy, ArrayRef<int>());
   }
 
   // Aggressively form a vector op if the cost is equal because the transform
   // may enable further optimization.
   // Codegen can reverse this transform (scalarize) if it was not profitable.
   return OldCost < NewCost;
 }
 
bool VectorCombine::foldBitcastShuf(Instruction &I) {
   // mask for scalable type is a splat or not.
   // 2) Disallow non-vector casts and length-changing shuffles.
   // TODO: We could allow any shuffle.
   auto *DestTy = dyn_cast<FixedVectorType>(I.getType());
   auto *SrcTy = dyn_cast<FixedVectorType>(V->getType());
   if (!SrcTy || !DestTy || I.getOperand(0)->getType() != SrcTy)
     return false;
 
-  // The new shuffle must not cost more than the old shuffle. The bitcast is
-  // moved ahead of the shuffle, so assume that it has the same cost as before.
-  InstructionCost DestCost =
-      TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, DestTy);
-  InstructionCost SrcCost =
-      TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, SrcTy);
-  if (DestCost > SrcCost || !DestCost.isValid())
-    return false;
-
   unsigned DestNumElts = DestTy->getNumElements();
   unsigned SrcNumElts = SrcTy->getNumElements();
   SmallVector<int, 16> NewMask;
   if (SrcNumElts <= DestNumElts) {
     // The bitcast is from wide to narrow/equal elements. The shuffle mask can
     // always be expanded to the equivalent form choosing narrower elements.
     assert(DestNumElts % SrcNumElts == 0 && "Unexpected shuffle mask");
     unsigned ScaleFactor = DestNumElts / SrcNumElts;
     narrowShuffleMaskElts(ScaleFactor, Mask, NewMask);
   } else {
     // The bitcast is from narrow elements to wide elements. The shuffle mask
     // must choose consecutive elements to allow casting first.
     assert(SrcNumElts % DestNumElts == 0 && "Unexpected shuffle mask");
     unsigned ScaleFactor = SrcNumElts / DestNumElts;
     if (!widenShuffleMaskElts(ScaleFactor, Mask, NewMask))
       return false;
   }
+
+  // The new shuffle must not cost more than the old shuffle. The bitcast is
+  // moved ahead of the shuffle, so assume that it has the same cost as before.
+  InstructionCost DestCost = TTI.getShuffleCost(
+      TargetTransformInfo::SK_PermuteSingleSrc, DestTy, NewMask);
+  InstructionCost SrcCost =
+      TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, SrcTy, Mask);
+  if (DestCost > SrcCost || !DestCost.isValid())
+    return false;
+
   // bitcast (shuf V, MaskC) --> shuf (bitcast V), MaskC'
   ++NumShufOfBitcast;
   Value *CastV = Builder.CreateBitCast(V, DestTy);
   Value *Shuf = Builder.CreateShuffleVector(CastV, NewMask);
   replaceValue(I, *Shuf);
   return true;
 }
 
bool VectorCombine::foldExtractedCmps(Instruction &I) {
 
   // The proposed vector pattern is:
   // vcmp = cmp Pred X, VecC
   // ext (binop vNi1 vcmp, (shuffle vcmp, Index1)), Index0
   int CheapIndex = ConvertToShuf == Ext0 ? Index1 : Index0;
   int ExpensiveIndex = ConvertToShuf == Ext0 ? Index0 : Index1;
   auto *CmpTy = cast<FixedVectorType>(CmpInst::makeCmpResultType(X->getType()));
   InstructionCost NewCost = TTI.getCmpSelInstrCost(CmpOpcode, X->getType());
-  NewCost +=
-      TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, CmpTy);
+  NewCost += TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, CmpTy,
+                                ArrayRef<int>());
   NewCost += TTI.getArithmeticInstrCost(I.getOpcode(), CmpTy);
   NewCost += TTI.getVectorInstrCost(Ext0->getOpcode(), CmpTy, CheapIndex);
 
   // Aggressively form vector ops if the cost is equal because the transform
   // may enable further optimization.
   // Codegen can reverse this transform (scalarize) if it was not profitable.
   if (OldCost < NewCost || !NewCost.isValid())
     return false;