- //===- polly/ScopInfo.h -----------------------------------------*- C++ -*-===// 
- // 
- // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 
- // See https://llvm.org/LICENSE.txt for license information. 
- // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 
- // 
- //===----------------------------------------------------------------------===// 
- // 
- // Store the polyhedral model representation of a static control flow region, 
- // also called SCoP (Static Control Part). 
- // 
- // This representation is shared among several tools in the polyhedral 
- // community, which are e.g. CLooG, Pluto, Loopo, Graphite. 
- // 
- //===----------------------------------------------------------------------===// 
-   
- #ifndef POLLY_SCOPINFO_H 
- #define POLLY_SCOPINFO_H 
-   
- #include "polly/ScopDetection.h" 
- #include "polly/Support/SCEVAffinator.h" 
- #include "polly/Support/ScopHelper.h" 
- #include "llvm/ADT/ArrayRef.h" 
- #include "llvm/ADT/MapVector.h" 
- #include "llvm/ADT/SetVector.h" 
- #include "llvm/Analysis/RegionPass.h" 
- #include "llvm/IR/DebugLoc.h" 
- #include "llvm/IR/Instruction.h" 
- #include "llvm/IR/Instructions.h" 
- #include "llvm/IR/PassManager.h" 
- #include "llvm/IR/ValueHandle.h" 
- #include "llvm/Pass.h" 
- #include "isl/isl-noexceptions.h" 
- #include <cassert> 
- #include <cstddef> 
- #include <forward_list> 
- #include <optional> 
-   
- namespace polly { 
- using llvm::AnalysisInfoMixin; 
- using llvm::ArrayRef; 
- using llvm::AssertingVH; 
- using llvm::AssumptionCache; 
- using llvm::cast; 
- using llvm::DataLayout; 
- using llvm::DenseMap; 
- using llvm::DenseSet; 
- using llvm::function_ref; 
- using llvm::isa; 
- using llvm::iterator_range; 
- using llvm::LoadInst; 
- using llvm::make_range; 
- using llvm::MapVector; 
- using llvm::MemIntrinsic; 
- using llvm::PassInfoMixin; 
- using llvm::PHINode; 
- using llvm::RegionNode; 
- using llvm::RegionPass; 
- using llvm::RGPassManager; 
- using llvm::SetVector; 
- using llvm::SmallPtrSetImpl; 
- using llvm::SmallVector; 
- using llvm::SmallVectorImpl; 
- using llvm::StringMap; 
- using llvm::Type; 
- using llvm::Use; 
- using llvm::Value; 
- using llvm::ValueToValueMap; 
-   
- class MemoryAccess; 
-   
- //===---------------------------------------------------------------------===// 
-   
- extern bool UseInstructionNames; 
-   
- // The maximal number of basic sets we allow during domain construction to 
- // be created. More complex scops will result in very high compile time and 
- // are also unlikely to result in good code. 
- extern unsigned const MaxDisjunctsInDomain; 
-   
- /// The different memory kinds used in Polly. 
- /// 
- /// We distinguish between arrays and various scalar memory objects. We use 
- /// the term ``array'' to describe memory objects that consist of a set of 
- /// individual data elements arranged in a multi-dimensional grid. A scalar 
- /// memory object describes an individual data element and is used to model 
- /// the definition and uses of llvm::Values. 
- /// 
- /// The polyhedral model does traditionally not reason about SSA values. To 
- /// reason about llvm::Values we model them "as if" they were zero-dimensional 
- /// memory objects, even though they were not actually allocated in (main) 
- /// memory.  Memory for such objects is only alloca[ed] at CodeGeneration 
- /// time. To relate the memory slots used during code generation with the 
- /// llvm::Values they belong to the new names for these corresponding stack 
- /// slots are derived by appending suffixes (currently ".s2a" and ".phiops") 
- /// to the name of the original llvm::Value. To describe how def/uses are 
- /// modeled exactly we use these suffixes here as well. 
- /// 
- /// There are currently four different kinds of memory objects: 
- enum class MemoryKind { 
-   /// MemoryKind::Array: Models a one or multi-dimensional array 
-   /// 
-   /// A memory object that can be described by a multi-dimensional array. 
-   /// Memory objects of this type are used to model actual multi-dimensional 
-   /// arrays as they exist in LLVM-IR, but they are also used to describe 
-   /// other objects: 
-   ///   - A single data element allocated on the stack using 'alloca' is 
-   ///     modeled as a one-dimensional, single-element array. 
-   ///   - A single data element allocated as a global variable is modeled as 
-   ///     one-dimensional, single-element array. 
-   ///   - Certain multi-dimensional arrays with variable size, which in 
-   ///     LLVM-IR are commonly expressed as a single-dimensional access with a 
-   ///     complicated access function, are modeled as multi-dimensional 
-   ///     memory objects (grep for "delinearization"). 
-   Array, 
-   
-   /// MemoryKind::Value: Models an llvm::Value 
-   /// 
-   /// Memory objects of type MemoryKind::Value are used to model the data flow 
-   /// induced by llvm::Values. For each llvm::Value that is used across 
-   /// BasicBlocks, one ScopArrayInfo object is created. A single memory WRITE 
-   /// stores the llvm::Value at its definition into the memory object and at 
-   /// each use of the llvm::Value (ignoring trivial intra-block uses) a 
-   /// corresponding READ is added. For instance, the use/def chain of a 
-   /// llvm::Value %V depicted below 
-   ///              ______________________ 
-   ///              |DefBB:              | 
-   ///              |  %V = float op ... | 
-   ///              ---------------------- 
-   ///               |                  | 
-   /// _________________               _________________ 
-   /// |UseBB1:        |               |UseBB2:        | 
-   /// |  use float %V |               |  use float %V | 
-   /// -----------------               ----------------- 
-   /// 
-   /// is modeled as if the following memory accesses occurred: 
-   /// 
-   ///                        __________________________ 
-   ///                        |entry:                  | 
-   ///                        |  %V.s2a = alloca float | 
-   ///                        -------------------------- 
-   ///                                     | 
-   ///                    ___________________________________ 
-   ///                    |DefBB:                           | 
-   ///                    |  store %float %V, float* %V.s2a | 
-   ///                    ----------------------------------- 
-   ///                           |                   | 
-   /// ____________________________________ ___________________________________ 
-   /// |UseBB1:                           | |UseBB2:                          | 
-   /// |  %V.reload1 = load float* %V.s2a | |  %V.reload2 = load float* %V.s2a| 
-   /// |  use float %V.reload1            | |  use float %V.reload2           | 
-   /// ------------------------------------ ----------------------------------- 
-   /// 
-   Value, 
-   
-   /// MemoryKind::PHI: Models PHI nodes within the SCoP 
-   /// 
-   /// Besides the MemoryKind::Value memory object used to model the normal 
-   /// llvm::Value dependences described above, PHI nodes require an additional 
-   /// memory object of type MemoryKind::PHI to describe the forwarding of values 
-   /// to 
-   /// the PHI node. 
-   /// 
-   /// As an example, a PHIInst instructions 
-   /// 
-   /// %PHI = phi float [ %Val1, %IncomingBlock1 ], [ %Val2, %IncomingBlock2 ] 
-   /// 
-   /// is modeled as if the accesses occurred this way: 
-   /// 
-   ///                    _______________________________ 
-   ///                    |entry:                       | 
-   ///                    |  %PHI.phiops = alloca float | 
-   ///                    ------------------------------- 
-   ///                           |              | 
-   /// __________________________________  __________________________________ 
-   /// |IncomingBlock1:                 |  |IncomingBlock2:                 | 
-   /// |  ...                           |  |  ...                           | 
-   /// |  store float %Val1 %PHI.phiops |  |  store float %Val2 %PHI.phiops | 
-   /// |  br label % JoinBlock          |  |  br label %JoinBlock           | 
-   /// ----------------------------------  ---------------------------------- 
-   ///                             \            / 
-   ///                              \          / 
-   ///               _________________________________________ 
-   ///               |JoinBlock:                             | 
-   ///               |  %PHI = load float, float* PHI.phiops | 
-   ///               ----------------------------------------- 
-   /// 
-   /// Note that there can also be a scalar write access for %PHI if used in a 
-   /// different BasicBlock, i.e. there can be a memory object %PHI.phiops as 
-   /// well as a memory object %PHI.s2a. 
-   PHI, 
-   
-   /// MemoryKind::ExitPHI: Models PHI nodes in the SCoP's exit block 
-   /// 
-   /// For PHI nodes in the Scop's exit block a special memory object kind is 
-   /// used. The modeling used is identical to MemoryKind::PHI, with the 
-   /// exception 
-   /// that there are no READs from these memory objects. The PHINode's 
-   /// llvm::Value is treated as a value escaping the SCoP. WRITE accesses 
-   /// write directly to the escaping value's ".s2a" alloca. 
-   ExitPHI 
- }; 
-   
- /// Maps from a loop to the affine function expressing its backedge taken count. 
- /// The backedge taken count already enough to express iteration domain as we 
- /// only allow loops with canonical induction variable. 
- /// A canonical induction variable is: 
- /// an integer recurrence that starts at 0 and increments by one each time 
- /// through the loop. 
- using LoopBoundMapType = std::map<const Loop *, const SCEV *>; 
-   
- using AccFuncVector = std::vector<std::unique_ptr<MemoryAccess>>; 
-   
- /// A class to store information about arrays in the SCoP. 
- /// 
- /// Objects are accessible via the ScoP, MemoryAccess or the id associated with 
- /// the MemoryAccess access function. 
- /// 
- class ScopArrayInfo final { 
- public: 
-   /// Construct a ScopArrayInfo object. 
-   /// 
-   /// @param BasePtr        The array base pointer. 
-   /// @param ElementType    The type of the elements stored in the array. 
-   /// @param IslCtx         The isl context used to create the base pointer id. 
-   /// @param DimensionSizes A vector containing the size of each dimension. 
-   /// @param Kind           The kind of the array object. 
-   /// @param DL             The data layout of the module. 
-   /// @param S              The scop this array object belongs to. 
-   /// @param BaseName       The optional name of this memory reference. 
-   ScopArrayInfo(Value *BasePtr, Type *ElementType, isl::ctx IslCtx, 
-                 ArrayRef<const SCEV *> DimensionSizes, MemoryKind Kind, 
-                 const DataLayout &DL, Scop *S, const char *BaseName = nullptr); 
-   
-   /// Destructor to free the isl id of the base pointer. 
-   ~ScopArrayInfo(); 
-   
-   ///  Update the element type of the ScopArrayInfo object. 
-   /// 
-   ///  Memory accesses referencing this ScopArrayInfo object may use 
-   ///  different element sizes. This function ensures the canonical element type 
-   ///  stored is small enough to model accesses to the current element type as 
-   ///  well as to @p NewElementType. 
-   /// 
-   ///  @param NewElementType An element type that is used to access this array. 
-   void updateElementType(Type *NewElementType); 
-   
-   ///  Update the sizes of the ScopArrayInfo object. 
-   /// 
-   ///  A ScopArrayInfo object may be created without all outer dimensions being 
-   ///  available. This function is called when new memory accesses are added for 
-   ///  this ScopArrayInfo object. It verifies that sizes are compatible and adds 
-   ///  additional outer array dimensions, if needed. 
-   /// 
-   ///  @param Sizes       A vector of array sizes where the rightmost array 
-   ///                     sizes need to match the innermost array sizes already 
-   ///                     defined in SAI. 
-   ///  @param CheckConsistency Update sizes, even if new sizes are inconsistent 
-   ///                          with old sizes 
-   bool updateSizes(ArrayRef<const SCEV *> Sizes, bool CheckConsistency = true); 
-   
-   /// Set the base pointer to @p BP. 
-   void setBasePtr(Value *BP) { BasePtr = BP; } 
-   
-   /// Return the base pointer. 
-   Value *getBasePtr() const { return BasePtr; } 
-   
-   // Set IsOnHeap to the value in parameter. 
-   void setIsOnHeap(bool value) { IsOnHeap = value; } 
-   
-   /// For indirect accesses return the origin SAI of the BP, else null. 
-   const ScopArrayInfo *getBasePtrOriginSAI() const { return BasePtrOriginSAI; } 
-   
-   /// The set of derived indirect SAIs for this origin SAI. 
-   const SmallSetVector<ScopArrayInfo *, 2> &getDerivedSAIs() const { 
-     return DerivedSAIs; 
-   } 
-   
-   /// Return the number of dimensions. 
-   unsigned getNumberOfDimensions() const { 
-     if (Kind == MemoryKind::PHI || Kind == MemoryKind::ExitPHI || 
-         Kind == MemoryKind::Value) 
-       return 0; 
-     return DimensionSizes.size(); 
-   } 
-   
-   /// Return the size of dimension @p dim as SCEV*. 
-   // 
-   //  Scalars do not have array dimensions and the first dimension of 
-   //  a (possibly multi-dimensional) array also does not carry any size 
-   //  information, in case the array is not newly created. 
-   const SCEV *getDimensionSize(unsigned Dim) const { 
-     assert(Dim < getNumberOfDimensions() && "Invalid dimension"); 
-     return DimensionSizes[Dim]; 
-   } 
-   
-   /// Return the size of dimension @p dim as isl::pw_aff. 
-   // 
-   //  Scalars do not have array dimensions and the first dimension of 
-   //  a (possibly multi-dimensional) array also does not carry any size 
-   //  information, in case the array is not newly created. 
-   isl::pw_aff getDimensionSizePw(unsigned Dim) const { 
-     assert(Dim < getNumberOfDimensions() && "Invalid dimension"); 
-     return DimensionSizesPw[Dim]; 
-   } 
-   
-   /// Get the canonical element type of this array. 
-   /// 
-   /// @returns The canonical element type of this array. 
-   Type *getElementType() const { return ElementType; } 
-   
-   /// Get element size in bytes. 
-   int getElemSizeInBytes() const; 
-   
-   /// Get the name of this memory reference. 
-   std::string getName() const; 
-   
-   /// Return the isl id for the base pointer. 
-   isl::id getBasePtrId() const; 
-   
-   /// Return what kind of memory this represents. 
-   MemoryKind getKind() const { return Kind; } 
-   
-   /// Is this array info modeling an llvm::Value? 
-   bool isValueKind() const { return Kind == MemoryKind::Value; } 
-   
-   /// Is this array info modeling special PHI node memory? 
-   /// 
-   /// During code generation of PHI nodes, there is a need for two kinds of 
-   /// virtual storage. The normal one as it is used for all scalar dependences, 
-   /// where the result of the PHI node is stored and later loaded from as well 
-   /// as a second one where the incoming values of the PHI nodes are stored 
-   /// into and reloaded when the PHI is executed. As both memories use the 
-   /// original PHI node as virtual base pointer, we have this additional 
-   /// attribute to distinguish the PHI node specific array modeling from the 
-   /// normal scalar array modeling. 
-   bool isPHIKind() const { return Kind == MemoryKind::PHI; } 
-   
-   /// Is this array info modeling an MemoryKind::ExitPHI? 
-   bool isExitPHIKind() const { return Kind == MemoryKind::ExitPHI; } 
-   
-   /// Is this array info modeling an array? 
-   bool isArrayKind() const { return Kind == MemoryKind::Array; } 
-   
-   /// Is this array allocated on heap 
-   /// 
-   /// This property is only relevant if the array is allocated by Polly instead 
-   /// of pre-existing. If false, it is allocated using alloca instead malloca. 
-   bool isOnHeap() const { return IsOnHeap; } 
-   
- #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) 
-   /// Dump a readable representation to stderr. 
-   void dump() const; 
- #endif 
-   
-   /// Print a readable representation to @p OS. 
-   /// 
-   /// @param SizeAsPwAff Print the size as isl::pw_aff 
-   void print(raw_ostream &OS, bool SizeAsPwAff = false) const; 
-   
-   /// Access the ScopArrayInfo associated with an access function. 
-   static const ScopArrayInfo *getFromAccessFunction(isl::pw_multi_aff PMA); 
-   
-   /// Access the ScopArrayInfo associated with an isl Id. 
-   static const ScopArrayInfo *getFromId(isl::id Id); 
-   
-   /// Get the space of this array access. 
-   isl::space getSpace() const; 
-   
-   /// If the array is read only 
-   bool isReadOnly(); 
-   
-   /// Verify that @p Array is compatible to this ScopArrayInfo. 
-   /// 
-   /// Two arrays are compatible if their dimensionality, the sizes of their 
-   /// dimensions, and their element sizes match. 
-   /// 
-   /// @param Array The array to compare against. 
-   /// 
-   /// @returns True, if the arrays are compatible, False otherwise. 
-   bool isCompatibleWith(const ScopArrayInfo *Array) const; 
-   
- private: 
-   void addDerivedSAI(ScopArrayInfo *DerivedSAI) { 
-     DerivedSAIs.insert(DerivedSAI); 
-   } 
-   
-   /// For indirect accesses this is the SAI of the BP origin. 
-   const ScopArrayInfo *BasePtrOriginSAI; 
-   
-   /// For origin SAIs the set of derived indirect SAIs. 
-   SmallSetVector<ScopArrayInfo *, 2> DerivedSAIs; 
-   
-   /// The base pointer. 
-   AssertingVH<Value> BasePtr; 
-   
-   /// The canonical element type of this array. 
-   /// 
-   /// The canonical element type describes the minimal accessible element in 
-   /// this array. Not all elements accessed, need to be of the very same type, 
-   /// but the allocation size of the type of the elements loaded/stored from/to 
-   /// this array needs to be a multiple of the allocation size of the canonical 
-   /// type. 
-   Type *ElementType; 
-   
-   /// The isl id for the base pointer. 
-   isl::id Id; 
-   
-   /// True if the newly allocated array is on heap. 
-   bool IsOnHeap = false; 
-   
-   /// The sizes of each dimension as SCEV*. 
-   SmallVector<const SCEV *, 4> DimensionSizes; 
-   
-   /// The sizes of each dimension as isl::pw_aff. 
-   SmallVector<isl::pw_aff, 4> DimensionSizesPw; 
-   
-   /// The type of this scop array info object. 
-   /// 
-   /// We distinguish between SCALAR, PHI and ARRAY objects. 
-   MemoryKind Kind; 
-   
-   /// The data layout of the module. 
-   const DataLayout &DL; 
-   
-   /// The scop this SAI object belongs to. 
-   Scop &S; 
- }; 
-   
- /// Represent memory accesses in statements. 
- class MemoryAccess final { 
-   friend class Scop; 
-   friend class ScopStmt; 
-   friend class ScopBuilder; 
-   
- public: 
-   /// The access type of a memory access 
-   /// 
-   /// There are three kind of access types: 
-   /// 
-   /// * A read access 
-   /// 
-   /// A certain set of memory locations are read and may be used for internal 
-   /// calculations. 
-   /// 
-   /// * A must-write access 
-   /// 
-   /// A certain set of memory locations is definitely written. The old value is 
-   /// replaced by a newly calculated value. The old value is not read or used at 
-   /// all. 
-   /// 
-   /// * A may-write access 
-   /// 
-   /// A certain set of memory locations may be written. The memory location may 
-   /// contain a new value if there is actually a write or the old value may 
-   /// remain, if no write happens. 
-   enum AccessType { 
-     READ = 0x1, 
-     MUST_WRITE = 0x2, 
-     MAY_WRITE = 0x3, 
-   }; 
-   
-   /// Reduction access type 
-   /// 
-   /// Commutative and associative binary operations suitable for reductions 
-   enum ReductionType { 
-     RT_NONE, ///< Indicate no reduction at all 
-     RT_ADD,  ///< Addition 
-     RT_MUL,  ///< Multiplication 
-     RT_BOR,  ///< Bitwise Or 
-     RT_BXOR, ///< Bitwise XOr 
-     RT_BAND, ///< Bitwise And 
-   }; 
-   
-   using SubscriptsTy = SmallVector<const SCEV *, 4>; 
-   
- private: 
-   /// A unique identifier for this memory access. 
-   /// 
-   /// The identifier is unique between all memory accesses belonging to the same 
-   /// scop statement. 
-   isl::id Id; 
-   
-   /// What is modeled by this MemoryAccess. 
-   /// @see MemoryKind 
-   MemoryKind Kind; 
-   
-   /// Whether it a reading or writing access, and if writing, whether it 
-   /// is conditional (MAY_WRITE). 
-   enum AccessType AccType; 
-   
-   /// Reduction type for reduction like accesses, RT_NONE otherwise 
-   /// 
-   /// An access is reduction like if it is part of a load-store chain in which 
-   /// both access the same memory location (use the same LLVM-IR value 
-   /// as pointer reference). Furthermore, between the load and the store there 
-   /// is exactly one binary operator which is known to be associative and 
-   /// commutative. 
-   /// 
-   /// TODO: 
-   /// 
-   /// We can later lift the constraint that the same LLVM-IR value defines the 
-   /// memory location to handle scops such as the following: 
-   /// 
-   ///    for i 
-   ///      for j 
-   ///        sum[i+j] = sum[i] + 3; 
-   /// 
-   /// Here not all iterations access the same memory location, but iterations 
-   /// for which j = 0 holds do. After lifting the equality check in ScopBuilder, 
-   /// subsequent transformations do not only need check if a statement is 
-   /// reduction like, but they also need to verify that that the reduction 
-   /// property is only exploited for statement instances that load from and 
-   /// store to the same data location. Doing so at dependence analysis time 
-   /// could allow us to handle the above example. 
-   ReductionType RedType = RT_NONE; 
-   
-   /// Parent ScopStmt of this access. 
-   ScopStmt *Statement; 
-   
-   /// The domain under which this access is not modeled precisely. 
-   /// 
-   /// The invalid domain for an access describes all parameter combinations 
-   /// under which the statement looks to be executed but is in fact not because 
-   /// some assumption/restriction makes the access invalid. 
-   isl::set InvalidDomain; 
-   
-   // Properties describing the accessed array. 
-   // TODO: It might be possible to move them to ScopArrayInfo. 
-   // @{ 
-   
-   /// The base address (e.g., A for A[i+j]). 
-   /// 
-   /// The #BaseAddr of a memory access of kind MemoryKind::Array is the base 
-   /// pointer of the memory access. 
-   /// The #BaseAddr of a memory access of kind MemoryKind::PHI or 
-   /// MemoryKind::ExitPHI is the PHI node itself. 
-   /// The #BaseAddr of a memory access of kind MemoryKind::Value is the 
-   /// instruction defining the value. 
-   AssertingVH<Value> BaseAddr; 
-   
-   /// Type a single array element wrt. this access. 
-   Type *ElementType; 
-   
-   /// Size of each dimension of the accessed array. 
-   SmallVector<const SCEV *, 4> Sizes; 
-   // @} 
-   
-   // Properties describing the accessed element. 
-   // @{ 
-   
-   /// The access instruction of this memory access. 
-   /// 
-   /// For memory accesses of kind MemoryKind::Array the access instruction is 
-   /// the Load or Store instruction performing the access. 
-   /// 
-   /// For memory accesses of kind MemoryKind::PHI or MemoryKind::ExitPHI the 
-   /// access instruction of a load access is the PHI instruction. The access 
-   /// instruction of a PHI-store is the incoming's block's terminator 
-   /// instruction. 
-   /// 
-   /// For memory accesses of kind MemoryKind::Value the access instruction of a 
-   /// load access is nullptr because generally there can be multiple 
-   /// instructions in the statement using the same llvm::Value. The access 
-   /// instruction of a write access is the instruction that defines the 
-   /// llvm::Value. 
-   Instruction *AccessInstruction = nullptr; 
-   
-   /// Incoming block and value of a PHINode. 
-   SmallVector<std::pair<BasicBlock *, Value *>, 4> Incoming; 
-   
-   /// The value associated with this memory access. 
-   /// 
-   ///  - For array memory accesses (MemoryKind::Array) it is the loaded result 
-   ///    or the stored value. If the access instruction is a memory intrinsic it 
-   ///    the access value is also the memory intrinsic. 
-   ///  - For accesses of kind MemoryKind::Value it is the access instruction 
-   ///    itself. 
-   ///  - For accesses of kind MemoryKind::PHI or MemoryKind::ExitPHI it is the 
-   ///    PHI node itself (for both, READ and WRITE accesses). 
-   /// 
-   AssertingVH<Value> AccessValue; 
-   
-   /// Are all the subscripts affine expression? 
-   bool IsAffine = true; 
-   
-   /// Subscript expression for each dimension. 
-   SubscriptsTy Subscripts; 
-   
-   /// Relation from statement instances to the accessed array elements. 
-   /// 
-   /// In the common case this relation is a function that maps a set of loop 
-   /// indices to the memory address from which a value is loaded/stored: 
-   /// 
-   ///      for i 
-   ///        for j 
-   ///    S:     A[i + 3 j] = ... 
-   /// 
-   ///    => { S[i,j] -> A[i + 3j] } 
-   /// 
-   /// In case the exact access function is not known, the access relation may 
-   /// also be a one to all mapping { S[i,j] -> A[o] } describing that any 
-   /// element accessible through A might be accessed. 
-   /// 
-   /// In case of an access to a larger element belonging to an array that also 
-   /// contains smaller elements, the access relation models the larger access 
-   /// with multiple smaller accesses of the size of the minimal array element 
-   /// type: 
-   /// 
-   ///      short *A; 
-   /// 
-   ///      for i 
-   ///    S:     A[i] = *((double*)&A[4 * i]); 
-   /// 
-   ///    => { S[i] -> A[i]; S[i] -> A[o] : 4i <= o <= 4i + 3 } 
-   isl::map AccessRelation; 
-   
-   /// Updated access relation read from JSCOP file. 
-   isl::map NewAccessRelation; 
-   // @} 
-   
-   isl::basic_map createBasicAccessMap(ScopStmt *Statement); 
-   
-   isl::set assumeNoOutOfBound(); 
-   
-   /// Compute bounds on an over approximated  access relation. 
-   /// 
-   /// @param ElementSize The size of one element accessed. 
-   void computeBoundsOnAccessRelation(unsigned ElementSize); 
-   
-   /// Get the original access function as read from IR. 
-   isl::map getOriginalAccessRelation() const; 
-   
-   /// Return the space in which the access relation lives in. 
-   isl::space getOriginalAccessRelationSpace() const; 
-   
-   /// Get the new access function imported or set by a pass 
-   isl::map getNewAccessRelation() const; 
-   
-   /// Fold the memory access to consider parametric offsets 
-   /// 
-   /// To recover memory accesses with array size parameters in the subscript 
-   /// expression we post-process the delinearization results. 
-   /// 
-   /// We would normally recover from an access A[exp0(i) * N + exp1(i)] into an 
-   /// array A[][N] the 2D access A[exp0(i)][exp1(i)]. However, another valid 
-   /// delinearization is A[exp0(i) - 1][exp1(i) + N] which - depending on the 
-   /// range of exp1(i) - may be preferable. Specifically, for cases where we 
-   /// know exp1(i) is negative, we want to choose the latter expression. 
-   /// 
-   /// As we commonly do not have any information about the range of exp1(i), 
-   /// we do not choose one of the two options, but instead create a piecewise 
-   /// access function that adds the (-1, N) offsets as soon as exp1(i) becomes 
-   /// negative. For a 2D array such an access function is created by applying 
-   /// the piecewise map: 
-   /// 
-   /// [i,j] -> [i, j] :      j >= 0 
-   /// [i,j] -> [i-1, j+N] :  j <  0 
-   /// 
-   /// We can generalize this mapping to arbitrary dimensions by applying this 
-   /// piecewise mapping pairwise from the rightmost to the leftmost access 
-   /// dimension. It would also be possible to cover a wider range by introducing 
-   /// more cases and adding multiple of Ns to these cases. However, this has 
-   /// not yet been necessary. 
-   /// The introduction of different cases necessarily complicates the memory 
-   /// access function, but cases that can be statically proven to not happen 
-   /// will be eliminated later on. 
-   void foldAccessRelation(); 
-   
-   /// Create the access relation for the underlying memory intrinsic. 
-   void buildMemIntrinsicAccessRelation(); 
-   
-   /// Assemble the access relation from all available information. 
-   /// 
-   /// In particular, used the information passes in the constructor and the 
-   /// parent ScopStmt set by setStatment(). 
-   /// 
-   /// @param SAI Info object for the accessed array. 
-   void buildAccessRelation(const ScopArrayInfo *SAI); 
-   
-   /// Carry index overflows of dimensions with constant size to the next higher 
-   /// dimension. 
-   /// 
-   /// For dimensions that have constant size, modulo the index by the size and 
-   /// add up the carry (floored division) to the next higher dimension. This is 
-   /// how overflow is defined in row-major order. 
-   /// It happens e.g. when ScalarEvolution computes the offset to the base 
-   /// pointer and would algebraically sum up all lower dimensions' indices of 
-   /// constant size. 
-   /// 
-   /// Example: 
-   ///   float (*A)[4]; 
-   ///   A[1][6] -> A[2][2] 
-   void wrapConstantDimensions(); 
-   
- public: 
-   /// Create a new MemoryAccess. 
-   /// 
-   /// @param Stmt       The parent statement. 
-   /// @param AccessInst The instruction doing the access. 
-   /// @param BaseAddr   The accessed array's address. 
-   /// @param ElemType   The type of the accessed array elements. 
-   /// @param AccType    Whether read or write access. 
-   /// @param IsAffine   Whether the subscripts are affine expressions. 
-   /// @param Kind       The kind of memory accessed. 
-   /// @param Subscripts Subscript expressions 
-   /// @param Sizes      Dimension lengths of the accessed array. 
-   MemoryAccess(ScopStmt *Stmt, Instruction *AccessInst, AccessType AccType, 
-                Value *BaseAddress, Type *ElemType, bool Affine, 
-                ArrayRef<const SCEV *> Subscripts, ArrayRef<const SCEV *> Sizes, 
-                Value *AccessValue, MemoryKind Kind); 
-   
-   /// Create a new MemoryAccess that corresponds to @p AccRel. 
-   /// 
-   /// Along with @p Stmt and @p AccType it uses information about dimension 
-   /// lengths of the accessed array, the type of the accessed array elements, 
-   /// the name of the accessed array that is derived from the object accessible 
-   /// via @p AccRel. 
-   /// 
-   /// @param Stmt       The parent statement. 
-   /// @param AccType    Whether read or write access. 
-   /// @param AccRel     The access relation that describes the memory access. 
-   MemoryAccess(ScopStmt *Stmt, AccessType AccType, isl::map AccRel); 
-   
-   MemoryAccess(const MemoryAccess &) = delete; 
-   MemoryAccess &operator=(const MemoryAccess &) = delete; 
-   ~MemoryAccess(); 
-   
-   /// Add a new incoming block/value pairs for this PHI/ExitPHI access. 
-   /// 
-   /// @param IncomingBlock The PHI's incoming block. 
-   /// @param IncomingValue The value when reaching the PHI from the @p 
-   ///                      IncomingBlock. 
-   void addIncoming(BasicBlock *IncomingBlock, Value *IncomingValue) { 
-     assert(!isRead()); 
-     assert(isAnyPHIKind()); 
-     Incoming.emplace_back(std::make_pair(IncomingBlock, IncomingValue)); 
-   } 
-   
-   /// Return the list of possible PHI/ExitPHI values. 
-   /// 
-   /// After code generation moves some PHIs around during region simplification, 
-   /// we cannot reliably locate the original PHI node and its incoming values 
-   /// anymore. For this reason we remember these explicitly for all PHI-kind 
-   /// accesses. 
-   ArrayRef<std::pair<BasicBlock *, Value *>> getIncoming() const { 
-     assert(isAnyPHIKind()); 
-     return Incoming; 
-   } 
-   
-   /// Get the type of a memory access. 
-   enum AccessType getType() { return AccType; } 
-   
-   /// Is this a reduction like access? 
-   bool isReductionLike() const { return RedType != RT_NONE; } 
-   
-   /// Is this a read memory access? 
-   bool isRead() const { return AccType == MemoryAccess::READ; } 
-   
-   /// Is this a must-write memory access? 
-   bool isMustWrite() const { return AccType == MemoryAccess::MUST_WRITE; } 
-   
-   /// Is this a may-write memory access? 
-   bool isMayWrite() const { return AccType == MemoryAccess::MAY_WRITE; } 
-   
-   /// Is this a write memory access? 
-   bool isWrite() const { return isMustWrite() || isMayWrite(); } 
-   
-   /// Is this a memory intrinsic access (memcpy, memset, memmove)? 
-   bool isMemoryIntrinsic() const { 
-     return isa<MemIntrinsic>(getAccessInstruction()); 
-   } 
-   
-   /// Check if a new access relation was imported or set by a pass. 
-   bool hasNewAccessRelation() const { return !NewAccessRelation.is_null(); } 
-   
-   /// Return the newest access relation of this access. 
-   /// 
-   /// There are two possibilities: 
-   ///   1) The original access relation read from the LLVM-IR. 
-   ///   2) A new access relation imported from a json file or set by another 
-   ///      pass (e.g., for privatization). 
-   /// 
-   /// As 2) is by construction "newer" than 1) we return the new access 
-   /// relation if present. 
-   /// 
-   isl::map getLatestAccessRelation() const { 
-     return hasNewAccessRelation() ? getNewAccessRelation() 
-                                   : getOriginalAccessRelation(); 
-   } 
-   
-   /// Old name of getLatestAccessRelation(). 
-   isl::map getAccessRelation() const { return getLatestAccessRelation(); } 
-   
-   /// Get an isl map describing the memory address accessed. 
-   /// 
-   /// In most cases the memory address accessed is well described by the access 
-   /// relation obtained with getAccessRelation. However, in case of arrays 
-   /// accessed with types of different size the access relation maps one access 
-   /// to multiple smaller address locations. This method returns an isl map that 
-   /// relates each dynamic statement instance to the unique memory location 
-   /// that is loaded from / stored to. 
-   /// 
-   /// For an access relation { S[i] -> A[o] : 4i <= o <= 4i + 3 } this method 
-   /// will return the address function { S[i] -> A[4i] }. 
-   /// 
-   /// @returns The address function for this memory access. 
-   isl::map getAddressFunction() const; 
-   
-   /// Return the access relation after the schedule was applied. 
-   isl::pw_multi_aff 
-   applyScheduleToAccessRelation(isl::union_map Schedule) const; 
-   
-   /// Get an isl string representing the access function read from IR. 
-   std::string getOriginalAccessRelationStr() const; 
-   
-   /// Get an isl string representing a new access function, if available. 
-   std::string getNewAccessRelationStr() const; 
-   
-   /// Get an isl string representing the latest access relation. 
-   std::string getAccessRelationStr() const; 
-   
-   /// Get the original base address of this access (e.g. A for A[i+j]) when 
-   /// detected. 
-   /// 
-   /// This address may differ from the base address referenced by the original 
-   /// ScopArrayInfo to which this array belongs, as this memory access may 
-   /// have been canonicalized to a ScopArrayInfo which has a different but 
-   /// identically-valued base pointer in case invariant load hoisting is 
-   /// enabled. 
-   Value *getOriginalBaseAddr() const { return BaseAddr; } 
-   
-   /// Get the detection-time base array isl::id for this access. 
-   isl::id getOriginalArrayId() const; 
-   
-   /// Get the base array isl::id for this access, modifiable through 
-   /// setNewAccessRelation(). 
-   isl::id getLatestArrayId() const; 
-   
-   /// Old name of getOriginalArrayId(). 
-   isl::id getArrayId() const { return getOriginalArrayId(); } 
-   
-   /// Get the detection-time ScopArrayInfo object for the base address. 
-   const ScopArrayInfo *getOriginalScopArrayInfo() const; 
-   
-   /// Get the ScopArrayInfo object for the base address, or the one set 
-   /// by setNewAccessRelation(). 
-   const ScopArrayInfo *getLatestScopArrayInfo() const; 
-   
-   /// Legacy name of getOriginalScopArrayInfo(). 
-   const ScopArrayInfo *getScopArrayInfo() const { 
-     return getOriginalScopArrayInfo(); 
-   } 
-   
-   /// Return a string representation of the access's reduction type. 
-   const std::string getReductionOperatorStr() const; 
-   
-   /// Return a string representation of the reduction type @p RT. 
-   static const std::string getReductionOperatorStr(ReductionType RT); 
-   
-   /// Return the element type of the accessed array wrt. this access. 
-   Type *getElementType() const { return ElementType; } 
-   
-   /// Return the access value of this memory access. 
-   Value *getAccessValue() const { return AccessValue; } 
-   
-   /// Return llvm::Value that is stored by this access, if available. 
-   /// 
-   /// PHI nodes may not have a unique value available that is stored, as in 
-   /// case of region statements one out of possibly several llvm::Values 
-   /// might be stored. In this case nullptr is returned. 
-   Value *tryGetValueStored() { 
-     assert(isWrite() && "Only write statement store values"); 
-     if (isAnyPHIKind()) { 
-       if (Incoming.size() == 1) 
-         return Incoming[0].second; 
-       return nullptr; 
-     } 
-     return AccessValue; 
-   } 
-   
-   /// Return the access instruction of this memory access. 
-   Instruction *getAccessInstruction() const { return AccessInstruction; } 
-   
-   ///  Return an iterator range containing the subscripts. 
-   iterator_range<SubscriptsTy::const_iterator> subscripts() const { 
-     return make_range(Subscripts.begin(), Subscripts.end()); 
-   } 
-   
-   /// Return the number of access function subscript. 
-   unsigned getNumSubscripts() const { return Subscripts.size(); } 
-   
-   /// Return the access function subscript in the dimension @p Dim. 
-   const SCEV *getSubscript(unsigned Dim) const { return Subscripts[Dim]; } 
-   
-   /// Compute the isl representation for the SCEV @p E wrt. this access. 
-   /// 
-   /// Note that this function will also adjust the invalid context accordingly. 
-   isl::pw_aff getPwAff(const SCEV *E); 
-   
-   /// Get the invalid domain for this access. 
-   isl::set getInvalidDomain() const { return InvalidDomain; } 
-   
-   /// Get the invalid context for this access. 
-   isl::set getInvalidContext() const { return getInvalidDomain().params(); } 
-   
-   /// Get the stride of this memory access in the specified Schedule. Schedule 
-   /// is a map from the statement to a schedule where the innermost dimension is 
-   /// the dimension of the innermost loop containing the statement. 
-   isl::set getStride(isl::map Schedule) const; 
-   
-   /// Is the stride of the access equal to a certain width? Schedule is a map 
-   /// from the statement to a schedule where the innermost dimension is the 
-   /// dimension of the innermost loop containing the statement. 
-   bool isStrideX(isl::map Schedule, int StrideWidth) const; 
-   
-   /// Is consecutive memory accessed for a given statement instance set? 
-   /// Schedule is a map from the statement to a schedule where the innermost 
-   /// dimension is the dimension of the innermost loop containing the 
-   /// statement. 
-   bool isStrideOne(isl::map Schedule) const; 
-   
-   /// Is always the same memory accessed for a given statement instance set? 
-   /// Schedule is a map from the statement to a schedule where the innermost 
-   /// dimension is the dimension of the innermost loop containing the 
-   /// statement. 
-   bool isStrideZero(isl::map Schedule) const; 
-   
-   /// Return the kind when this access was first detected. 
-   MemoryKind getOriginalKind() const { 
-     assert(!getOriginalScopArrayInfo() /* not yet initialized */ || 
-            getOriginalScopArrayInfo()->getKind() == Kind); 
-     return Kind; 
-   } 
-   
-   /// Return the kind considering a potential setNewAccessRelation. 
-   MemoryKind getLatestKind() const { 
-     return getLatestScopArrayInfo()->getKind(); 
-   } 
-   
-   /// Whether this is an access of an explicit load or store in the IR. 
-   bool isOriginalArrayKind() const { 
-     return getOriginalKind() == MemoryKind::Array; 
-   } 
-   
-   /// Whether storage memory is either an custom .s2a/.phiops alloca 
-   /// (false) or an existing pointer into an array (true). 
-   bool isLatestArrayKind() const { 
-     return getLatestKind() == MemoryKind::Array; 
-   } 
-   
-   /// Old name of isOriginalArrayKind. 
-   bool isArrayKind() const { return isOriginalArrayKind(); } 
-   
-   /// Whether this access is an array to a scalar memory object, without 
-   /// considering changes by setNewAccessRelation. 
-   /// 
-   /// Scalar accesses are accesses to MemoryKind::Value, MemoryKind::PHI or 
-   /// MemoryKind::ExitPHI. 
-   bool isOriginalScalarKind() const { 
-     return getOriginalKind() != MemoryKind::Array; 
-   } 
-   
-   /// Whether this access is an array to a scalar memory object, also 
-   /// considering changes by setNewAccessRelation. 
-   bool isLatestScalarKind() const { 
-     return getLatestKind() != MemoryKind::Array; 
-   } 
-   
-   /// Old name of isOriginalScalarKind. 
-   bool isScalarKind() const { return isOriginalScalarKind(); } 
-   
-   /// Was this MemoryAccess detected as a scalar dependences? 
-   bool isOriginalValueKind() const { 
-     return getOriginalKind() == MemoryKind::Value; 
-   } 
-   
-   /// Is this MemoryAccess currently modeling scalar dependences? 
-   bool isLatestValueKind() const { 
-     return getLatestKind() == MemoryKind::Value; 
-   } 
-   
-   /// Old name of isOriginalValueKind(). 
-   bool isValueKind() const { return isOriginalValueKind(); } 
-   
-   /// Was this MemoryAccess detected as a special PHI node access? 
-   bool isOriginalPHIKind() const { 
-     return getOriginalKind() == MemoryKind::PHI; 
-   } 
-   
-   /// Is this MemoryAccess modeling special PHI node accesses, also 
-   /// considering a potential change by setNewAccessRelation? 
-   bool isLatestPHIKind() const { return getLatestKind() == MemoryKind::PHI; } 
-   
-   /// Old name of isOriginalPHIKind. 
-   bool isPHIKind() const { return isOriginalPHIKind(); } 
-   
-   /// Was this MemoryAccess detected as the accesses of a PHI node in the 
-   /// SCoP's exit block? 
-   bool isOriginalExitPHIKind() const { 
-     return getOriginalKind() == MemoryKind::ExitPHI; 
-   } 
-   
-   /// Is this MemoryAccess modeling the accesses of a PHI node in the 
-   /// SCoP's exit block? Can be changed to an array access using 
-   /// setNewAccessRelation(). 
-   bool isLatestExitPHIKind() const { 
-     return getLatestKind() == MemoryKind::ExitPHI; 
-   } 
-   
-   /// Old name of isOriginalExitPHIKind(). 
-   bool isExitPHIKind() const { return isOriginalExitPHIKind(); } 
-   
-   /// Was this access detected as one of the two PHI types? 
-   bool isOriginalAnyPHIKind() const { 
-     return isOriginalPHIKind() || isOriginalExitPHIKind(); 
-   } 
-   
-   /// Does this access originate from one of the two PHI types? Can be 
-   /// changed to an array access using setNewAccessRelation(). 
-   bool isLatestAnyPHIKind() const { 
-     return isLatestPHIKind() || isLatestExitPHIKind(); 
-   } 
-   
-   /// Old name of isOriginalAnyPHIKind(). 
-   bool isAnyPHIKind() const { return isOriginalAnyPHIKind(); } 
-   
-   /// Get the statement that contains this memory access. 
-   ScopStmt *getStatement() const { return Statement; } 
-   
-   /// Get the reduction type of this access 
-   ReductionType getReductionType() const { return RedType; } 
-   
-   /// Update the original access relation. 
-   /// 
-   /// We need to update the original access relation during scop construction, 
-   /// when unifying the memory accesses that access the same scop array info 
-   /// object. After the scop has been constructed, the original access relation 
-   /// should not be changed any more. Instead setNewAccessRelation should 
-   /// be called. 
-   void setAccessRelation(isl::map AccessRelation); 
-   
-   /// Set the updated access relation read from JSCOP file. 
-   void setNewAccessRelation(isl::map NewAccessRelation); 
-   
-   /// Return whether the MemoryyAccess is a partial access. That is, the access 
-   /// is not executed in some instances of the parent statement's domain. 
-   bool isLatestPartialAccess() const; 
-   
-   /// Mark this a reduction like access 
-   void markAsReductionLike(ReductionType RT) { RedType = RT; } 
-   
-   /// Align the parameters in the access relation to the scop context 
-   void realignParams(); 
-   
-   /// Update the dimensionality of the memory access. 
-   /// 
-   /// During scop construction some memory accesses may not be constructed with 
-   /// their full dimensionality, but outer dimensions may have been omitted if 
-   /// they took the value 'zero'. By updating the dimensionality of the 
-   /// statement we add additional zero-valued dimensions to match the 
-   /// dimensionality of the ScopArrayInfo object that belongs to this memory 
-   /// access. 
-   void updateDimensionality(); 
-   
-   /// Get identifier for the memory access. 
-   /// 
-   /// This identifier is unique for all accesses that belong to the same scop 
-   /// statement. 
-   isl::id getId() const; 
-   
-   /// Print the MemoryAccess. 
-   /// 
-   /// @param OS The output stream the MemoryAccess is printed to. 
-   void print(raw_ostream &OS) const; 
-   
- #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) 
-   /// Print the MemoryAccess to stderr. 
-   void dump() const; 
- #endif 
-   
-   /// Is the memory access affine? 
-   bool isAffine() const { return IsAffine; } 
- }; 
-   
- raw_ostream &operator<<(raw_ostream &OS, MemoryAccess::ReductionType RT); 
-   
- /// Ordered list type to hold accesses. 
- using MemoryAccessList = std::forward_list<MemoryAccess *>; 
-   
- /// Helper structure for invariant memory accesses. 
- struct InvariantAccess { 
-   /// The memory access that is (partially) invariant. 
-   MemoryAccess *MA; 
-   
-   /// The context under which the access is not invariant. 
-   isl::set NonHoistableCtx; 
- }; 
-   
- /// Ordered container type to hold invariant accesses. 
- using InvariantAccessesTy = SmallVector<InvariantAccess, 8>; 
-   
- /// Type for equivalent invariant accesses and their domain context. 
- struct InvariantEquivClassTy { 
-   /// The pointer that identifies this equivalence class 
-   const SCEV *IdentifyingPointer; 
-   
-   /// Memory accesses now treated invariant 
-   /// 
-   /// These memory accesses access the pointer location that identifies 
-   /// this equivalence class. They are treated as invariant and hoisted during 
-   /// code generation. 
-   MemoryAccessList InvariantAccesses; 
-   
-   /// The execution context under which the memory location is accessed 
-   /// 
-   /// It is the union of the execution domains of the memory accesses in the 
-   /// InvariantAccesses list. 
-   isl::set ExecutionContext; 
-   
-   /// The type of the invariant access 
-   /// 
-   /// It is used to differentiate between differently typed invariant loads from 
-   /// the same location. 
-   Type *AccessType; 
- }; 
-   
- /// Type for invariant accesses equivalence classes. 
- using InvariantEquivClassesTy = SmallVector<InvariantEquivClassTy, 8>; 
-   
- /// Statement of the Scop 
- /// 
- /// A Scop statement represents an instruction in the Scop. 
- /// 
- /// It is further described by its iteration domain, its schedule and its data 
- /// accesses. 
- /// At the moment every statement represents a single basic block of LLVM-IR. 
- class ScopStmt final { 
-   friend class ScopBuilder; 
-   
- public: 
-   /// Create the ScopStmt from a BasicBlock. 
-   ScopStmt(Scop &parent, BasicBlock &bb, StringRef Name, Loop *SurroundingLoop, 
-            std::vector<Instruction *> Instructions); 
-   
-   /// Create an overapproximating ScopStmt for the region @p R. 
-   /// 
-   /// @param EntryBlockInstructions The list of instructions that belong to the 
-   ///                               entry block of the region statement. 
-   ///                               Instructions are only tracked for entry 
-   ///                               blocks for now. We currently do not allow 
-   ///                               to modify the instructions of blocks later 
-   ///                               in the region statement. 
-   ScopStmt(Scop &parent, Region &R, StringRef Name, Loop *SurroundingLoop, 
-            std::vector<Instruction *> EntryBlockInstructions); 
-   
-   /// Create a copy statement. 
-   /// 
-   /// @param Stmt       The parent statement. 
-   /// @param SourceRel  The source location. 
-   /// @param TargetRel  The target location. 
-   /// @param Domain     The original domain under which the copy statement would 
-   ///                   be executed. 
-   ScopStmt(Scop &parent, isl::map SourceRel, isl::map TargetRel, 
-            isl::set Domain); 
-   
-   ScopStmt(const ScopStmt &) = delete; 
-   const ScopStmt &operator=(const ScopStmt &) = delete; 
-   ~ScopStmt(); 
-   
- private: 
-   /// Polyhedral description 
-   //@{ 
-   
-   /// The Scop containing this ScopStmt. 
-   Scop &Parent; 
-   
-   /// The domain under which this statement is not modeled precisely. 
-   /// 
-   /// The invalid domain for a statement describes all parameter combinations 
-   /// under which the statement looks to be executed but is in fact not because 
-   /// some assumption/restriction makes the statement/scop invalid. 
-   isl::set InvalidDomain; 
-   
-   /// The iteration domain describes the set of iterations for which this 
-   /// statement is executed. 
-   /// 
-   /// Example: 
-   ///     for (i = 0; i < 100 + b; ++i) 
-   ///       for (j = 0; j < i; ++j) 
-   ///         S(i,j); 
-   /// 
-   /// 'S' is executed for different values of i and j. A vector of all 
-   /// induction variables around S (i, j) is called iteration vector. 
-   /// The domain describes the set of possible iteration vectors. 
-   /// 
-   /// In this case it is: 
-   /// 
-   ///     Domain: 0 <= i <= 100 + b 
-   ///             0 <= j <= i 
-   /// 
-   /// A pair of statement and iteration vector (S, (5,3)) is called statement 
-   /// instance. 
-   isl::set Domain; 
-   
-   /// The memory accesses of this statement. 
-   /// 
-   /// The only side effects of a statement are its memory accesses. 
-   using MemoryAccessVec = llvm::SmallVector<MemoryAccess *, 8>; 
-   MemoryAccessVec MemAccs; 
-   
-   /// Mapping from instructions to (scalar) memory accesses. 
-   DenseMap<const Instruction *, MemoryAccessList> InstructionToAccess; 
-   
-   /// The set of values defined elsewhere required in this ScopStmt and 
-   ///        their MemoryKind::Value READ MemoryAccesses. 
-   DenseMap<Value *, MemoryAccess *> ValueReads; 
-   
-   /// The set of values defined in this ScopStmt that are required 
-   ///        elsewhere, mapped to their MemoryKind::Value WRITE MemoryAccesses. 
-   DenseMap<Instruction *, MemoryAccess *> ValueWrites; 
-   
-   /// Map from PHI nodes to its incoming value when coming from this 
-   ///        statement. 
-   /// 
-   /// Non-affine subregions can have multiple exiting blocks that are incoming 
-   /// blocks of the PHI nodes. This map ensures that there is only one write 
-   /// operation for the complete subregion. A PHI selecting the relevant value 
-   /// will be inserted. 
-   DenseMap<PHINode *, MemoryAccess *> PHIWrites; 
-   
-   /// Map from PHI nodes to its read access in this statement. 
-   DenseMap<PHINode *, MemoryAccess *> PHIReads; 
-   
-   //@} 
-   
-   /// A SCoP statement represents either a basic block (affine/precise case) or 
-   /// a whole region (non-affine case). 
-   /// 
-   /// Only one of the following two members will therefore be set and indicate 
-   /// which kind of statement this is. 
-   /// 
-   ///{ 
-   
-   /// The BasicBlock represented by this statement (in the affine case). 
-   BasicBlock *BB = nullptr; 
-   
-   /// The region represented by this statement (in the non-affine case). 
-   Region *R = nullptr; 
-   
-   ///} 
-   
-   /// The isl AST build for the new generated AST. 
-   isl::ast_build Build; 
-   
-   SmallVector<Loop *, 4> NestLoops; 
-   
-   std::string BaseName; 
-   
-   /// The closest loop that contains this statement. 
-   Loop *SurroundingLoop; 
-   
-   /// Vector for Instructions in this statement. 
-   std::vector<Instruction *> Instructions; 
-   
-   /// Remove @p MA from dictionaries pointing to them. 
-   void removeAccessData(MemoryAccess *MA); 
-   
- public: 
-   /// Get an isl_ctx pointer. 
-   isl::ctx getIslCtx() const; 
-   
-   /// Get the iteration domain of this ScopStmt. 
-   /// 
-   /// @return The iteration domain of this ScopStmt. 
-   isl::set getDomain() const; 
-   
-   /// Get the space of the iteration domain 
-   /// 
-   /// @return The space of the iteration domain 
-   isl::space getDomainSpace() const; 
-   
-   /// Get the id of the iteration domain space 
-   /// 
-   /// @return The id of the iteration domain space 
-   isl::id getDomainId() const; 
-   
-   /// Get an isl string representing this domain. 
-   std::string getDomainStr() const; 
-   
-   /// Get the schedule function of this ScopStmt. 
-   /// 
-   /// @return The schedule function of this ScopStmt, if it does not contain 
-   /// extension nodes, and nullptr, otherwise. 
-   isl::map getSchedule() const; 
-   
-   /// Get an isl string representing this schedule. 
-   /// 
-   /// @return An isl string representing this schedule, if it does not contain 
-   /// extension nodes, and an empty string, otherwise. 
-   std::string getScheduleStr() const; 
-   
-   /// Get the invalid domain for this statement. 
-   isl::set getInvalidDomain() const { return InvalidDomain; } 
-   
-   /// Get the invalid context for this statement. 
-   isl::set getInvalidContext() const { return getInvalidDomain().params(); } 
-   
-   /// Set the invalid context for this statement to @p ID. 
-   void setInvalidDomain(isl::set ID); 
-   
-   /// Get the BasicBlock represented by this ScopStmt (if any). 
-   /// 
-   /// @return The BasicBlock represented by this ScopStmt, or null if the 
-   ///         statement represents a region. 
-   BasicBlock *getBasicBlock() const { return BB; } 
-   
-   /// Return true if this statement represents a single basic block. 
-   bool isBlockStmt() const { return BB != nullptr; } 
-   
-   /// Return true if this is a copy statement. 
-   bool isCopyStmt() const { return BB == nullptr && R == nullptr; } 
-   
-   /// Get the region represented by this ScopStmt (if any). 
-   /// 
-   /// @return The region represented by this ScopStmt, or null if the statement 
-   ///         represents a basic block. 
-   Region *getRegion() const { return R; } 
-   
-   /// Return true if this statement represents a whole region. 
-   bool isRegionStmt() const { return R != nullptr; } 
-   
-   /// Return a BasicBlock from this statement. 
-   /// 
-   /// For block statements, it returns the BasicBlock itself. For subregion 
-   /// statements, return its entry block. 
-   BasicBlock *getEntryBlock() const; 
-   
-   /// Return whether @p L is boxed within this statement. 
-   bool contains(const Loop *L) const { 
-     // Block statements never contain loops. 
-     if (isBlockStmt()) 
-       return false; 
-   
-     return getRegion()->contains(L); 
-   } 
-   
-   /// Return whether this statement represents @p BB. 
-   bool represents(BasicBlock *BB) const { 
-     if (isCopyStmt()) 
-       return false; 
-     if (isBlockStmt()) 
-       return BB == getBasicBlock(); 
-     return getRegion()->contains(BB); 
-   } 
-   
-   /// Return whether this statement contains @p Inst. 
-   bool contains(Instruction *Inst) const { 
-     if (!Inst) 
-       return false; 
-     if (isBlockStmt()) 
-       return llvm::is_contained(Instructions, Inst); 
-     return represents(Inst->getParent()); 
-   } 
-   
-   /// Return the closest innermost loop that contains this statement, but is not 
-   /// contained in it. 
-   /// 
-   /// For block statement, this is just the loop that contains the block. Region 
-   /// statements can contain boxed loops, so getting the loop of one of the 
-   /// region's BBs might return such an inner loop. For instance, the region's 
-   /// entry could be a header of a loop, but the region might extend to BBs 
-   /// after the loop exit. Similarly, the region might only contain parts of the 
-   /// loop body and still include the loop header. 
-   /// 
-   /// Most of the time the surrounding loop is the top element of #NestLoops, 
-   /// except when it is empty. In that case it return the loop that the whole 
-   /// SCoP is contained in. That can be nullptr if there is no such loop. 
-   Loop *getSurroundingLoop() const { 
-     assert(!isCopyStmt() && 
-            "No surrounding loop for artificially created statements"); 
-     return SurroundingLoop; 
-   } 
-   
-   /// Return true if this statement does not contain any accesses. 
-   bool isEmpty() const { return MemAccs.empty(); } 
-   
-   /// Find all array accesses for @p Inst. 
-   /// 
-   /// @param Inst The instruction accessing an array. 
-   /// 
-   /// @return A list of array accesses (MemoryKind::Array) accessed by @p Inst. 
-   ///         If there is no such access, it returns nullptr. 
-   const MemoryAccessList * 
-   lookupArrayAccessesFor(const Instruction *Inst) const { 
-     auto It = InstructionToAccess.find(Inst); 
-     if (It == InstructionToAccess.end()) 
-       return nullptr; 
-     if (It->second.empty()) 
-       return nullptr; 
-     return &It->second; 
-   } 
-   
-   /// Return the only array access for @p Inst, if existing. 
-   /// 
-   /// @param Inst The instruction for which to look up the access. 
-   /// @returns The unique array memory access related to Inst or nullptr if 
-   ///          no array access exists 
-   MemoryAccess *getArrayAccessOrNULLFor(const Instruction *Inst) const { 
-     auto It = InstructionToAccess.find(Inst); 
-     if (It == InstructionToAccess.end()) 
-       return nullptr; 
-   
-     MemoryAccess *ArrayAccess = nullptr; 
-   
-     for (auto Access : It->getSecond()) { 
-       if (!Access->isArrayKind()) 
-         continue; 
-   
-       assert(!ArrayAccess && "More then one array access for instruction"); 
-   
-       ArrayAccess = Access; 
-     } 
-   
-     return ArrayAccess; 
-   } 
-   
-   /// Return the only array access for @p Inst. 
-   /// 
-   /// @param Inst The instruction for which to look up the access. 
-   /// @returns The unique array memory access related to Inst. 
-   MemoryAccess &getArrayAccessFor(const Instruction *Inst) const { 
-     MemoryAccess *ArrayAccess = getArrayAccessOrNULLFor(Inst); 
-   
-     assert(ArrayAccess && "No array access found for instruction!"); 
-     return *ArrayAccess; 
-   } 
-   
-   /// Return the MemoryAccess that writes the value of an instruction 
-   ///        defined in this statement, or nullptr if not existing, respectively 
-   ///        not yet added. 
-   MemoryAccess *lookupValueWriteOf(Instruction *Inst) const { 
-     assert((isRegionStmt() && R->contains(Inst)) || 
-            (!isRegionStmt() && Inst->getParent() == BB)); 
-     return ValueWrites.lookup(Inst); 
-   } 
-   
-   /// Return the MemoryAccess that reloads a value, or nullptr if not 
-   ///        existing, respectively not yet added. 
-   MemoryAccess *lookupValueReadOf(Value *Inst) const { 
-     return ValueReads.lookup(Inst); 
-   } 
-   
-   /// Return the MemoryAccess that loads a PHINode value, or nullptr if not 
-   /// existing, respectively not yet added. 
-   MemoryAccess *lookupPHIReadOf(PHINode *PHI) const { 
-     return PHIReads.lookup(PHI); 
-   } 
-   
-   /// Return the PHI write MemoryAccess for the incoming values from any 
-   ///        basic block in this ScopStmt, or nullptr if not existing, 
-   ///        respectively not yet added. 
-   MemoryAccess *lookupPHIWriteOf(PHINode *PHI) const { 
-     assert(isBlockStmt() || R->getExit() == PHI->getParent()); 
-     return PHIWrites.lookup(PHI); 
-   } 
-   
-   /// Return the input access of the value, or null if no such MemoryAccess 
-   /// exists. 
-   /// 
-   /// The input access is the MemoryAccess that makes an inter-statement value 
-   /// available in this statement by reading it at the start of this statement. 
-   /// This can be a MemoryKind::Value if defined in another statement or a 
-   /// MemoryKind::PHI if the value is a PHINode in this statement. 
-   MemoryAccess *lookupInputAccessOf(Value *Val) const { 
-     if (isa<PHINode>(Val)) 
-       if (auto InputMA = lookupPHIReadOf(cast<PHINode>(Val))) { 
-         assert(!lookupValueReadOf(Val) && "input accesses must be unique; a " 
-                                           "statement cannot read a .s2a and " 
-                                           ".phiops simultaneously"); 
-         return InputMA; 
-       } 
-   
-     if (auto *InputMA = lookupValueReadOf(Val)) 
-       return InputMA; 
-   
-     return nullptr; 
-   } 
-   
-   /// Add @p Access to this statement's list of accesses. 
-   /// 
-   /// @param Access  The access to add. 
-   /// @param Prepend If true, will add @p Access before all other instructions 
-   ///                (instead of appending it). 
-   void addAccess(MemoryAccess *Access, bool Preprend = false); 
-   
-   /// Remove a MemoryAccess from this statement. 
-   /// 
-   /// Note that scalar accesses that are caused by MA will 
-   /// be eliminated too. 
-   void removeMemoryAccess(MemoryAccess *MA); 
-   
-   /// Remove @p MA from this statement. 
-   /// 
-   /// In contrast to removeMemoryAccess(), no other access will be eliminated. 
-   /// 
-   /// @param MA            The MemoryAccess to be removed. 
-   /// @param AfterHoisting If true, also remove from data access lists. 
-   ///                      These lists are filled during 
-   ///                      ScopBuilder::buildAccessRelations. Therefore, if this 
-   ///                      method is called before buildAccessRelations, false 
-   ///                      must be passed. 
-   void removeSingleMemoryAccess(MemoryAccess *MA, bool AfterHoisting = true); 
-   
-   using iterator = MemoryAccessVec::iterator; 
-   using const_iterator = MemoryAccessVec::const_iterator; 
-   
-   iterator begin() { return MemAccs.begin(); } 
-   iterator end() { return MemAccs.end(); } 
-   const_iterator begin() const { return MemAccs.begin(); } 
-   const_iterator end() const { return MemAccs.end(); } 
-   size_t size() const { return MemAccs.size(); } 
-   
-   unsigned getNumIterators() const; 
-   
-   Scop *getParent() { return &Parent; } 
-   const Scop *getParent() const { return &Parent; } 
-   
-   const std::vector<Instruction *> &getInstructions() const { 
-     return Instructions; 
-   } 
-   
-   /// Set the list of instructions for this statement. It replaces the current 
-   /// list. 
-   void setInstructions(ArrayRef<Instruction *> Range) { 
-     Instructions.assign(Range.begin(), Range.end()); 
-   } 
-   
-   std::vector<Instruction *>::const_iterator insts_begin() const { 
-     return Instructions.begin(); 
-   } 
-   
-   std::vector<Instruction *>::const_iterator insts_end() const { 
-     return Instructions.end(); 
-   } 
-   
-   /// The range of instructions in this statement. 
-   iterator_range<std::vector<Instruction *>::const_iterator> insts() const { 
-     return {insts_begin(), insts_end()}; 
-   } 
-   
-   /// Insert an instruction before all other instructions in this statement. 
-   void prependInstruction(Instruction *Inst) { 
-     Instructions.insert(Instructions.begin(), Inst); 
-   } 
-   
-   const char *getBaseName() const; 
-   
-   /// Set the isl AST build. 
-   void setAstBuild(isl::ast_build B) { Build = B; } 
-   
-   /// Get the isl AST build. 
-   isl::ast_build getAstBuild() const { return Build; } 
-   
-   /// Restrict the domain of the statement. 
-   /// 
-   /// @param NewDomain The new statement domain. 
-   void restrictDomain(isl::set NewDomain); 
-   
-   /// Get the loop for a dimension. 
-   /// 
-   /// @param Dimension The dimension of the induction variable 
-   /// @return The loop at a certain dimension. 
-   Loop *getLoopForDimension(unsigned Dimension) const; 
-   
-   /// Align the parameters in the statement to the scop context 
-   void realignParams(); 
-   
-   /// Print the ScopStmt. 
-   /// 
-   /// @param OS                The output stream the ScopStmt is printed to. 
-   /// @param PrintInstructions Whether to print the statement's instructions as 
-   ///                          well. 
-   void print(raw_ostream &OS, bool PrintInstructions) const; 
-   
-   /// Print the instructions in ScopStmt. 
-   /// 
-   void printInstructions(raw_ostream &OS) const; 
-   
-   /// Check whether there is a value read access for @p V in this statement, and 
-   /// if not, create one. 
-   /// 
-   /// This allows to add MemoryAccesses after the initial creation of the Scop 
-   /// by ScopBuilder. 
-   /// 
-   /// @return The already existing or newly created MemoryKind::Value READ 
-   /// MemoryAccess. 
-   /// 
-   /// @see ScopBuilder::ensureValueRead(Value*,ScopStmt*) 
-   MemoryAccess *ensureValueRead(Value *V); 
-   
- #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) 
-   /// Print the ScopStmt to stderr. 
-   void dump() const; 
- #endif 
- }; 
-   
- /// Print ScopStmt S to raw_ostream OS. 
- raw_ostream &operator<<(raw_ostream &OS, const ScopStmt &S); 
-   
- /// Static Control Part 
- /// 
- /// A Scop is the polyhedral representation of a control flow region detected 
- /// by the Scop detection. It is generated by translating the LLVM-IR and 
- /// abstracting its effects. 
- /// 
- /// A Scop consists of a set of: 
- /// 
- ///   * A set of statements executed in the Scop. 
- /// 
- ///   * A set of global parameters 
- ///   Those parameters are scalar integer values, which are constant during 
- ///   execution. 
- /// 
- ///   * A context 
- ///   This context contains information about the values the parameters 
- ///   can take and relations between different parameters. 
- class Scop final { 
- public: 
-   /// Type to represent a pair of minimal/maximal access to an array. 
-   using MinMaxAccessTy = std::pair<isl::pw_multi_aff, isl::pw_multi_aff>; 
-   
-   /// Vector of minimal/maximal accesses to different arrays. 
-   using MinMaxVectorTy = SmallVector<MinMaxAccessTy, 4>; 
-   
-   /// Pair of minimal/maximal access vectors representing 
-   /// read write and read only accesses 
-   using MinMaxVectorPairTy = std::pair<MinMaxVectorTy, MinMaxVectorTy>; 
-   
-   /// Vector of pair of minimal/maximal access vectors representing 
-   /// non read only and read only accesses for each alias group. 
-   using MinMaxVectorPairVectorTy = SmallVector<MinMaxVectorPairTy, 4>; 
-   
- private: 
-   friend class ScopBuilder; 
-   
-   /// Isl context. 
-   /// 
-   /// We need a shared_ptr with reference counter to delete the context when all 
-   /// isl objects are deleted. We will distribute the shared_ptr to all objects 
-   /// that use the context to create isl objects, and increase the reference 
-   /// counter. By doing this, we guarantee that the context is deleted when we 
-   /// delete the last object that creates isl objects with the context. This 
-   /// declaration needs to be the first in class to gracefully destroy all isl 
-   /// objects before the context. 
-   std::shared_ptr<isl_ctx> IslCtx; 
-   
-   ScalarEvolution *SE; 
-   DominatorTree *DT; 
-   
-   /// The underlying Region. 
-   Region &R; 
-   
-   /// The name of the SCoP (identical to the regions name) 
-   std::optional<std::string> name; 
-   
-   // Access functions of the SCoP. 
-   // 
-   // This owns all the MemoryAccess objects of the Scop created in this pass. 
-   AccFuncVector AccessFunctions; 
-   
-   /// Flag to indicate that the scheduler actually optimized the SCoP. 
-   bool IsOptimized = false; 
-   
-   /// True if the underlying region has a single exiting block. 
-   bool HasSingleExitEdge; 
-   
-   /// Flag to remember if the SCoP contained an error block or not. 
-   bool HasErrorBlock = false; 
-   
-   /// Max loop depth. 
-   unsigned MaxLoopDepth = 0; 
-   
-   /// Number of copy statements. 
-   unsigned CopyStmtsNum = 0; 
-   
-   /// Flag to indicate if the Scop is to be skipped. 
-   bool SkipScop = false; 
-   
-   using StmtSet = std::list<ScopStmt>; 
-   
-   /// The statements in this Scop. 
-   StmtSet Stmts; 
-   
-   /// Parameters of this Scop 
-   ParameterSetTy Parameters; 
-   
-   /// Mapping from parameters to their ids. 
-   DenseMap<const SCEV *, isl::id> ParameterIds; 
-   
-   /// The context of the SCoP created during SCoP detection. 
-   ScopDetection::DetectionContext &DC; 
-   
-   /// OptimizationRemarkEmitter object for displaying diagnostic remarks 
-   OptimizationRemarkEmitter &ORE; 
-   
-   /// A map from basic blocks to vector of SCoP statements. Currently this 
-   /// vector comprises only of a single statement. 
-   DenseMap<BasicBlock *, std::vector<ScopStmt *>> StmtMap; 
-   
-   /// A map from instructions to SCoP statements. 
-   DenseMap<Instruction *, ScopStmt *> InstStmtMap; 
-   
-   /// A map from basic blocks to their domains. 
-   DenseMap<BasicBlock *, isl::set> DomainMap; 
-   
-   /// Constraints on parameters. 
-   isl::set Context; 
-   
-   /// The affinator used to translate SCEVs to isl expressions. 
-   SCEVAffinator Affinator; 
-   
-   using ArrayInfoMapTy = 
-       std::map<std::pair<AssertingVH<const Value>, MemoryKind>, 
-                std::unique_ptr<ScopArrayInfo>>; 
-   
-   using ArrayNameMapTy = StringMap<std::unique_ptr<ScopArrayInfo>>; 
-   
-   using ArrayInfoSetTy = SetVector<ScopArrayInfo *>; 
-   
-   /// A map to remember ScopArrayInfo objects for all base pointers. 
-   /// 
-   /// As PHI nodes may have two array info objects associated, we add a flag 
-   /// that distinguishes between the PHI node specific ArrayInfo object 
-   /// and the normal one. 
-   ArrayInfoMapTy ScopArrayInfoMap; 
-   
-   /// A map to remember ScopArrayInfo objects for all names of memory 
-   ///        references. 
-   ArrayNameMapTy ScopArrayNameMap; 
-   
-   /// A set to remember ScopArrayInfo objects. 
-   /// @see Scop::ScopArrayInfoMap 
-   ArrayInfoSetTy ScopArrayInfoSet; 
-   
-   /// The assumptions under which this scop was built. 
-   /// 
-   /// When constructing a scop sometimes the exact representation of a statement 
-   /// or condition would be very complex, but there is a common case which is a 
-   /// lot simpler, but which is only valid under certain assumptions. The 
-   /// assumed context records the assumptions taken during the construction of 
-   /// this scop and that need to be code generated as a run-time test. 
-   isl::set AssumedContext; 
-   
-   /// The restrictions under which this SCoP was built. 
-   /// 
-   /// The invalid context is similar to the assumed context as it contains 
-   /// constraints over the parameters. However, while we need the constraints 
-   /// in the assumed context to be "true" the constraints in the invalid context 
-   /// need to be "false". Otherwise they behave the same. 
-   isl::set InvalidContext; 
-   
-   /// The context under which the SCoP must have defined behavior. Optimizer and 
-   /// code generator can assume that the SCoP will only be executed with 
-   /// parameter values within this context. This might be either because we can 
-   /// prove that other values are impossible or explicitly have undefined 
-   /// behavior, such as due to no-wrap flags. If this becomes too complex, can 
-   /// also be nullptr. 
-   /// 
-   /// In contrast to Scop::AssumedContext and Scop::InvalidContext, these do not 
-   /// need to be checked at runtime. 
-   /// 
-   /// Scop::Context on the other side is an overapproximation and does not 
-   /// include all requirements, but is always defined. However, there is still 
-   /// no guarantee that there is no undefined behavior in 
-   /// DefinedBehaviorContext. 
-   isl::set DefinedBehaviorContext; 
-   
-   /// The schedule of the SCoP 
-   /// 
-   /// The schedule of the SCoP describes the execution order of the statements 
-   /// in the scop by assigning each statement instance a possibly 
-   /// multi-dimensional execution time. The schedule is stored as a tree of 
-   /// schedule nodes. 
-   /// 
-   /// The most common nodes in a schedule tree are so-called band nodes. Band 
-   /// nodes map statement instances into a multi dimensional schedule space. 
-   /// This space can be seen as a multi-dimensional clock. 
-   /// 
-   /// Example: 
-   /// 
-   /// <S,(5,4)>  may be mapped to (5,4) by this schedule: 
-   /// 
-   /// s0 = i (Year of execution) 
-   /// s1 = j (Day of execution) 
-   /// 
-   /// or to (9, 20) by this schedule: 
-   /// 
-   /// s0 = i + j (Year of execution) 
-   /// s1 = 20 (Day of execution) 
-   /// 
-   /// The order statement instances are executed is defined by the 
-   /// schedule vectors they are mapped to. A statement instance 
-   /// <A, (i, j, ..)> is executed before a statement instance <B, (i', ..)>, if 
-   /// the schedule vector of A is lexicographic smaller than the schedule 
-   /// vector of B. 
-   /// 
-   /// Besides band nodes, schedule trees contain additional nodes that specify 
-   /// a textual ordering between two subtrees or filter nodes that filter the 
-   /// set of statement instances that will be scheduled in a subtree. There 
-   /// are also several other nodes. A full description of the different nodes 
-   /// in a schedule tree is given in the isl manual. 
-   isl::schedule Schedule; 
-   
-   /// Is this Scop marked as not to be transformed by an optimization heuristic? 
-   bool HasDisableHeuristicsHint = false; 
-   
-   /// Whether the schedule has been modified after derived from the CFG by 
-   /// ScopBuilder. 
-   bool ScheduleModified = false; 
-   
-   /// The set of minimal/maximal accesses for each alias group. 
-   /// 
-   /// When building runtime alias checks we look at all memory instructions and 
-   /// build so called alias groups. Each group contains a set of accesses to 
-   /// different base arrays which might alias with each other. However, between 
-   /// alias groups there is no aliasing possible. 
-   /// 
-   /// In a program with int and float pointers annotated with tbaa information 
-   /// we would probably generate two alias groups, one for the int pointers and 
-   /// one for the float pointers. 
-   /// 
-   /// During code generation we will create a runtime alias check for each alias 
-   /// group to ensure the SCoP is executed in an alias free environment. 
-   MinMaxVectorPairVectorTy MinMaxAliasGroups; 
-   
-   /// Mapping from invariant loads to the representing invariant load of 
-   ///        their equivalence class. 
-   ValueToValueMap InvEquivClassVMap; 
-   
-   /// List of invariant accesses. 
-   InvariantEquivClassesTy InvariantEquivClasses; 
-   
-   /// The smallest array index not yet assigned. 
-   long ArrayIdx = 0; 
-   
-   /// The smallest statement index not yet assigned. 
-   long StmtIdx = 0; 
-   
-   /// A number that uniquely represents a Scop within its function 
-   const int ID; 
-   
-   /// Map of values to the MemoryAccess that writes its definition. 
-   /// 
-   /// There must be at most one definition per llvm::Instruction in a SCoP. 
-   DenseMap<Value *, MemoryAccess *> ValueDefAccs; 
-   
-   /// Map of values to the MemoryAccess that reads a PHI. 
-   DenseMap<PHINode *, MemoryAccess *> PHIReadAccs; 
-   
-   /// List of all uses (i.e. read MemoryAccesses) for a MemoryKind::Value 
-   /// scalar. 
-   DenseMap<const ScopArrayInfo *, SmallVector<MemoryAccess *, 4>> ValueUseAccs; 
-   
-   /// List of all incoming values (write MemoryAccess) of a MemoryKind::PHI or 
-   /// MemoryKind::ExitPHI scalar. 
-   DenseMap<const ScopArrayInfo *, SmallVector<MemoryAccess *, 4>> 
-       PHIIncomingAccs; 
-   
-   /// Scop constructor; invoked from ScopBuilder::buildScop. 
-   Scop(Region &R, ScalarEvolution &SE, LoopInfo &LI, DominatorTree &DT, 
-        ScopDetection::DetectionContext &DC, OptimizationRemarkEmitter &ORE, 
-        int ID); 
-   
-   //@} 
-   
-   /// Return the access for the base ptr of @p MA if any. 
-   MemoryAccess *lookupBasePtrAccess(MemoryAccess *MA); 
-   
-   /// Create an id for @p Param and store it in the ParameterIds map. 
-   void createParameterId(const SCEV *Param); 
-   
-   /// Build the Context of the Scop. 
-   void buildContext(); 
-   
-   /// Add the bounds of the parameters to the context. 
-   void addParameterBounds(); 
-   
-   /// Simplify the assumed and invalid context. 
-   void simplifyContexts(); 
-   
-   /// Create a new SCoP statement for @p BB. 
-   /// 
-   /// A new statement for @p BB will be created and added to the statement 
-   /// vector 
-   /// and map. 
-   /// 
-   /// @param BB              The basic block we build the statement for. 
-   /// @param Name            The name of the new statement. 
-   /// @param SurroundingLoop The loop the created statement is contained in. 
-   /// @param Instructions    The instructions in the statement. 
-   void addScopStmt(BasicBlock *BB, StringRef Name, Loop *SurroundingLoop, 
-                    std::vector<Instruction *> Instructions); 
-   
-   /// Create a new SCoP statement for @p R. 
-   /// 
-   /// A new statement for @p R will be created and added to the statement vector 
-   /// and map. 
-   /// 
-   /// @param R                      The region we build the statement for. 
-   /// @param Name                   The name of the new statement. 
-   /// @param SurroundingLoop        The loop the created statement is contained 
-   ///                               in. 
-   /// @param EntryBlockInstructions The (interesting) instructions in the 
-   ///                               entry block of the region statement. 
-   void addScopStmt(Region *R, StringRef Name, Loop *SurroundingLoop, 
-                    std::vector<Instruction *> EntryBlockInstructions); 
-   
-   /// Removes @p Stmt from the StmtMap. 
-   void removeFromStmtMap(ScopStmt &Stmt); 
-   
-   /// Removes all statements where the entry block of the statement does not 
-   /// have a corresponding domain in the domain map (or it is empty). 
-   void removeStmtNotInDomainMap(); 
-   
-   /// Collect all memory access relations of a given type. 
-   /// 
-   /// @param Predicate A predicate function that returns true if an access is 
-   ///                  of a given type. 
-   /// 
-   /// @returns The set of memory accesses in the scop that match the predicate. 
-   isl::union_map 
-   getAccessesOfType(std::function<bool(MemoryAccess &)> Predicate); 
-   
-   /// @name Helper functions for printing the Scop. 
-   /// 
-   //@{ 
-   void printContext(raw_ostream &OS) const; 
-   void printArrayInfo(raw_ostream &OS) const; 
-   void printStatements(raw_ostream &OS, bool PrintInstructions) const; 
-   void printAliasAssumptions(raw_ostream &OS) const; 
-   //@} 
-   
- public: 
-   Scop(const Scop &) = delete; 
-   Scop &operator=(const Scop &) = delete; 
-   ~Scop(); 
-   
-   /// Increment actual number of aliasing assumptions taken 
-   /// 
-   /// @param Step    Number of new aliasing assumptions which should be added to 
-   /// the number of already taken assumptions. 
-   static void incrementNumberOfAliasingAssumptions(unsigned Step); 
-   
-   /// Get the count of copy statements added to this Scop. 
-   /// 
-   /// @return The count of copy statements added to this Scop. 
-   unsigned getCopyStmtsNum() { return CopyStmtsNum; } 
-   
-   /// Create a new copy statement. 
-   /// 
-   /// A new statement will be created and added to the statement vector. 
-   /// 
-   /// @param SourceRel  The source location. 
-   /// @param TargetRel  The target location. 
-   /// @param Domain     The original domain under which the copy statement would 
-   ///                   be executed. 
-   ScopStmt *addScopStmt(isl::map SourceRel, isl::map TargetRel, 
-                         isl::set Domain); 
-   
-   /// Add the access function to all MemoryAccess objects of the Scop 
-   ///        created in this pass. 
-   void addAccessFunction(MemoryAccess *Access) { 
-     AccessFunctions.emplace_back(Access); 
-   
-     // Register value definitions. 
-     if (Access->isWrite() && Access->isOriginalValueKind()) { 
-       assert(!ValueDefAccs.count(Access->getAccessValue()) && 
-              "there can be just one definition per value"); 
-       ValueDefAccs[Access->getAccessValue()] = Access; 
-     } else if (Access->isRead() && Access->isOriginalPHIKind()) { 
-       PHINode *PHI = cast<PHINode>(Access->getAccessInstruction()); 
-       assert(!PHIReadAccs.count(PHI) && 
-              "there can be just one PHI read per PHINode"); 
-       PHIReadAccs[PHI] = Access; 
-     } 
-   } 
-   
-   /// Add metadata for @p Access. 
-   void addAccessData(MemoryAccess *Access); 
-   
-   /// Add new invariant access equivalence class 
-   void 
-   addInvariantEquivClass(const InvariantEquivClassTy &InvariantEquivClass) { 
-     InvariantEquivClasses.emplace_back(InvariantEquivClass); 
-   } 
-   
-   /// Add mapping from invariant loads to the representing invariant load of 
-   ///        their equivalence class. 
-   void addInvariantLoadMapping(const Value *LoadInst, Value *ClassRep) { 
-     InvEquivClassVMap[LoadInst] = ClassRep; 
-   } 
-   
-   /// Remove the metadata stored for @p Access. 
-   void removeAccessData(MemoryAccess *Access); 
-   
-   /// Return the scalar evolution. 
-   ScalarEvolution *getSE() const; 
-   
-   /// Return the dominator tree. 
-   DominatorTree *getDT() const { return DT; } 
-   
-   /// Return the LoopInfo used for this Scop. 
-   LoopInfo *getLI() const { return Affinator.getLI(); } 
-   
-   /// Get the count of parameters used in this Scop. 
-   /// 
-   /// @return The count of parameters used in this Scop. 
-   size_t getNumParams() const { return Parameters.size(); } 
-   
-   /// Return whether given SCEV is used as the parameter in this Scop. 
-   bool isParam(const SCEV *Param) const { return Parameters.count(Param); } 
-   
-   /// Take a list of parameters and add the new ones to the scop. 
-   void addParams(const ParameterSetTy &NewParameters); 
-   
-   /// Return an iterator range containing the scop parameters. 
-   iterator_range<ParameterSetTy::iterator> parameters() const { 
-     return make_range(Parameters.begin(), Parameters.end()); 
-   } 
-   
-   /// Return an iterator range containing invariant accesses. 
-   iterator_range<InvariantEquivClassesTy::iterator> invariantEquivClasses() { 
-     return make_range(InvariantEquivClasses.begin(), 
-                       InvariantEquivClasses.end()); 
-   } 
-   
-   /// Return an iterator range containing all the MemoryAccess objects of the 
-   /// Scop. 
-   iterator_range<AccFuncVector::iterator> access_functions() { 
-     return make_range(AccessFunctions.begin(), AccessFunctions.end()); 
-   } 
-   
-   /// Return whether this scop is empty, i.e. contains no statements that 
-   /// could be executed. 
-   bool isEmpty() const { return Stmts.empty(); } 
-   
-   StringRef getName() { 
-     if (!name) 
-       name = R.getNameStr(); 
-     return *name; 
-   } 
-   
-   using array_iterator = ArrayInfoSetTy::iterator; 
-   using const_array_iterator = ArrayInfoSetTy::const_iterator; 
-   using array_range = iterator_range<ArrayInfoSetTy::iterator>; 
-   using const_array_range = iterator_range<ArrayInfoSetTy::const_iterator>; 
-   
-   inline array_iterator array_begin() { return ScopArrayInfoSet.begin(); } 
-   
-   inline array_iterator array_end() { return ScopArrayInfoSet.end(); } 
-   
-   inline const_array_iterator array_begin() const { 
-     return ScopArrayInfoSet.begin(); 
-   } 
-   
-   inline const_array_iterator array_end() const { 
-     return ScopArrayInfoSet.end(); 
-   } 
-   
-   inline array_range arrays() { 
-     return array_range(array_begin(), array_end()); 
-   } 
-   
-   inline const_array_range arrays() const { 
-     return const_array_range(array_begin(), array_end()); 
-   } 
-   
-   /// Return the isl_id that represents a certain parameter. 
-   /// 
-   /// @param Parameter A SCEV that was recognized as a Parameter. 
-   /// 
-   /// @return The corresponding isl_id or NULL otherwise. 
-   isl::id getIdForParam(const SCEV *Parameter) const; 
-   
-   /// Get the maximum region of this static control part. 
-   /// 
-   /// @return The maximum region of this static control part. 
-   inline const Region &getRegion() const { return R; } 
-   inline Region &getRegion() { return R; } 
-   
-   /// Return the function this SCoP is in. 
-   Function &getFunction() const { return *R.getEntry()->getParent(); } 
-   
-   /// Check if @p L is contained in the SCoP. 
-   bool contains(const Loop *L) const { return R.contains(L); } 
-   
-   /// Check if @p BB is contained in the SCoP. 
-   bool contains(const BasicBlock *BB) const { return R.contains(BB); } 
-   
-   /// Check if @p I is contained in the SCoP. 
-   bool contains(const Instruction *I) const { return R.contains(I); } 
-   
-   /// Return the unique exit block of the SCoP. 
-   BasicBlock *getExit() const { return R.getExit(); } 
-   
-   /// Return the unique exiting block of the SCoP if any. 
-   BasicBlock *getExitingBlock() const { return R.getExitingBlock(); } 
-   
-   /// Return the unique entry block of the SCoP. 
-   BasicBlock *getEntry() const { return R.getEntry(); } 
-   
-   /// Return the unique entering block of the SCoP if any. 
-   BasicBlock *getEnteringBlock() const { return R.getEnteringBlock(); } 
-   
-   /// Return true if @p BB is the exit block of the SCoP. 
-   bool isExit(BasicBlock *BB) const { return getExit() == BB; } 
-   
-   /// Return a range of all basic blocks in the SCoP. 
-   Region::block_range blocks() const { return R.blocks(); } 
-   
-   /// Return true if and only if @p BB dominates the SCoP. 
-   bool isDominatedBy(const DominatorTree &DT, BasicBlock *BB) const; 
-   
-   /// Get the maximum depth of the loop. 
-   /// 
-   /// @return The maximum depth of the loop. 
-   inline unsigned getMaxLoopDepth() const { return MaxLoopDepth; } 
-   
-   /// Return the invariant equivalence class for @p Val if any. 
-   InvariantEquivClassTy *lookupInvariantEquivClass(Value *Val); 
-   
-   /// Return the set of invariant accesses. 
-   InvariantEquivClassesTy &getInvariantAccesses() { 
-     return InvariantEquivClasses; 
-   } 
-   
-   /// Check if the scop has any invariant access. 
-   bool hasInvariantAccesses() { return !InvariantEquivClasses.empty(); } 
-   
-   /// Mark the SCoP as optimized by the scheduler. 
-   void markAsOptimized() { IsOptimized = true; } 
-   
-   /// Check if the SCoP has been optimized by the scheduler. 
-   bool isOptimized() const { return IsOptimized; } 
-   
-   /// Mark the SCoP to be skipped by ScopPass passes. 
-   void markAsToBeSkipped() { SkipScop = true; } 
-   
-   /// Check if the SCoP is to be skipped by ScopPass passes. 
-   bool isToBeSkipped() const { return SkipScop; } 
-   
-   /// Return the ID of the Scop 
-   int getID() const { return ID; } 
-   
-   /// Get the name of the entry and exit blocks of this Scop. 
-   /// 
-   /// These along with the function name can uniquely identify a Scop. 
-   /// 
-   /// @return std::pair whose first element is the entry name & second element 
-   ///         is the exit name. 
-   std::pair<std::string, std::string> getEntryExitStr() const; 
-   
-   /// Get the name of this Scop. 
-   std::string getNameStr() const; 
-   
-   /// Get the constraint on parameter of this Scop. 
-   /// 
-   /// @return The constraint on parameter of this Scop. 
-   isl::set getContext() const; 
-   
-   /// Return the context where execution behavior is defined. Might return 
-   /// nullptr. 
-   isl::set getDefinedBehaviorContext() const { return DefinedBehaviorContext; } 
-   
-   /// Return the define behavior context, or if not available, its approximation 
-   /// from all other contexts. 
-   isl::set getBestKnownDefinedBehaviorContext() const { 
-     if (!DefinedBehaviorContext.is_null()) 
-       return DefinedBehaviorContext; 
-   
-     return Context.intersect_params(AssumedContext).subtract(InvalidContext); 
-   } 
-   
-   /// Return space of isl context parameters. 
-   /// 
-   /// Returns the set of context parameters that are currently constrained. In 
-   /// case the full set of parameters is needed, see @getFullParamSpace. 
-   isl::space getParamSpace() const; 
-   
-   /// Return the full space of parameters. 
-   /// 
-   /// getParamSpace will only return the parameters of the context that are 
-   /// actually constrained, whereas getFullParamSpace will return all 
-   //  parameters. This is useful in cases, where we need to ensure all 
-   //  parameters are available, as certain isl functions will abort if this is 
-   //  not the case. 
-   isl::space getFullParamSpace() const; 
-   
-   /// Get the assumed context for this Scop. 
-   /// 
-   /// @return The assumed context of this Scop. 
-   isl::set getAssumedContext() const; 
-   
-   /// Return true if the optimized SCoP can be executed. 
-   /// 
-   /// In addition to the runtime check context this will also utilize the domain 
-   /// constraints to decide it the optimized version can actually be executed. 
-   /// 
-   /// @returns True if the optimized SCoP can be executed. 
-   bool hasFeasibleRuntimeContext() const; 
-   
-   /// Check if the assumption in @p Set is trivial or not. 
-   /// 
-   /// @param Set  The relations between parameters that are assumed to hold. 
-   /// @param Sign Enum to indicate if the assumptions in @p Set are positive 
-   ///             (needed/assumptions) or negative (invalid/restrictions). 
-   /// 
-   /// @returns True if the assumption @p Set is not trivial. 
-   bool isEffectiveAssumption(isl::set Set, AssumptionSign Sign); 
-   
-   /// Track and report an assumption. 
-   /// 
-   /// Use 'clang -Rpass-analysis=polly-scops' or 'opt 
-   /// -pass-remarks-analysis=polly-scops' to output the assumptions. 
-   /// 
-   /// @param Kind The assumption kind describing the underlying cause. 
-   /// @param Set  The relations between parameters that are assumed to hold. 
-   /// @param Loc  The location in the source that caused this assumption. 
-   /// @param Sign Enum to indicate if the assumptions in @p Set are positive 
-   ///             (needed/assumptions) or negative (invalid/restrictions). 
-   /// @param BB   The block in which this assumption was taken. Used to 
-   ///             calculate hotness when emitting remark. 
-   /// 
-   /// @returns True if the assumption is not trivial. 
-   bool trackAssumption(AssumptionKind Kind, isl::set Set, DebugLoc Loc, 
-                        AssumptionSign Sign, BasicBlock *BB); 
-   
-   /// Add the conditions from @p Set (or subtract them if @p Sign is 
-   /// AS_RESTRICTION) to the defined behaviour context. 
-   void intersectDefinedBehavior(isl::set Set, AssumptionSign Sign); 
-   
-   /// Add assumptions to assumed context. 
-   /// 
-   /// The assumptions added will be assumed to hold during the execution of the 
-   /// scop. However, as they are generally not statically provable, at code 
-   /// generation time run-time checks will be generated that ensure the 
-   /// assumptions hold. 
-   /// 
-   /// WARNING: We currently exploit in simplifyAssumedContext the knowledge 
-   ///          that assumptions do not change the set of statement instances 
-   ///          executed. 
-   /// 
-   /// @param Kind The assumption kind describing the underlying cause. 
-   /// @param Set  The relations between parameters that are assumed to hold. 
-   /// @param Loc  The location in the source that caused this assumption. 
-   /// @param Sign Enum to indicate if the assumptions in @p Set are positive 
-   ///             (needed/assumptions) or negative (invalid/restrictions). 
-   /// @param BB   The block in which this assumption was taken. Used to 
-   ///             calculate hotness when emitting remark. 
-   /// @param RTC  Does the assumption require a runtime check? 
-   void addAssumption(AssumptionKind Kind, isl::set Set, DebugLoc Loc, 
-                      AssumptionSign Sign, BasicBlock *BB, bool RTC = true); 
-   
-   /// Mark the scop as invalid. 
-   /// 
-   /// This method adds an assumption to the scop that is always invalid. As a 
-   /// result, the scop will not be optimized later on. This function is commonly 
-   /// called when a condition makes it impossible (or too compile time 
-   /// expensive) to process this scop any further. 
-   /// 
-   /// @param Kind The assumption kind describing the underlying cause. 
-   /// @param Loc  The location in the source that triggered . 
-   /// @param BB   The BasicBlock where it was triggered. 
-   void invalidate(AssumptionKind Kind, DebugLoc Loc, BasicBlock *BB = nullptr); 
-   
-   /// Get the invalid context for this Scop. 
-   /// 
-   /// @return The invalid context of this Scop. 
-   isl::set getInvalidContext() const; 
-   
-   /// Return true if and only if the InvalidContext is trivial (=empty). 
-   bool hasTrivialInvalidContext() const { return InvalidContext.is_empty(); } 
-   
-   /// Return all alias groups for this SCoP. 
-   const MinMaxVectorPairVectorTy &getAliasGroups() const { 
-     return MinMaxAliasGroups; 
-   } 
-   
-   void addAliasGroup(MinMaxVectorTy &MinMaxAccessesReadWrite, 
-                      MinMaxVectorTy &MinMaxAccessesReadOnly) { 
-     MinMaxAliasGroups.emplace_back(); 
-     MinMaxAliasGroups.back().first = MinMaxAccessesReadWrite; 
-     MinMaxAliasGroups.back().second = MinMaxAccessesReadOnly; 
-   } 
-   
-   /// Remove statements from the list of scop statements. 
-   /// 
-   /// @param ShouldDelete  A function that returns true if the statement passed 
-   ///                      to it should be deleted. 
-   /// @param AfterHoisting If true, also remove from data access lists. 
-   ///                      These lists are filled during 
-   ///                      ScopBuilder::buildAccessRelations. Therefore, if this 
-   ///                      method is called before buildAccessRelations, false 
-   ///                      must be passed. 
-   void removeStmts(function_ref<bool(ScopStmt &)> ShouldDelete, 
-                    bool AfterHoisting = true); 
-   
-   /// Get an isl string representing the context. 
-   std::string getContextStr() const; 
-   
-   /// Get an isl string representing the assumed context. 
-   std::string getAssumedContextStr() const; 
-   
-   /// Get an isl string representing the invalid context. 
-   std::string getInvalidContextStr() const; 
-   
-   /// Return the list of ScopStmts that represent the given @p BB. 
-   ArrayRef<ScopStmt *> getStmtListFor(BasicBlock *BB) const; 
-   
-   /// Get the statement to put a PHI WRITE into. 
-   /// 
-   /// @param U The operand of a PHINode. 
-   ScopStmt *getIncomingStmtFor(const Use &U) const; 
-   
-   /// Return the last statement representing @p BB. 
-   /// 
-   /// Of the sequence of statements that represent a @p BB, this is the last one 
-   /// to be executed. It is typically used to determine which instruction to add 
-   /// a MemoryKind::PHI WRITE to. For this purpose, it is not strictly required 
-   /// to be executed last, only that the incoming value is available in it. 
-   ScopStmt *getLastStmtFor(BasicBlock *BB) const; 
-   
-   /// Return the ScopStmts that represents the Region @p R, or nullptr if 
-   ///        it is not represented by any statement in this Scop. 
-   ArrayRef<ScopStmt *> getStmtListFor(Region *R) const; 
-   
-   /// Return the ScopStmts that represents @p RN; can return nullptr if 
-   ///        the RegionNode is not within the SCoP or has been removed due to 
-   ///        simplifications. 
-   ArrayRef<ScopStmt *> getStmtListFor(RegionNode *RN) const; 
-   
-   /// Return the ScopStmt an instruction belongs to, or nullptr if it 
-   ///        does not belong to any statement in this Scop. 
-   ScopStmt *getStmtFor(Instruction *Inst) const { 
-     return InstStmtMap.lookup(Inst); 
-   } 
-   
-   /// Return the number of statements in the SCoP. 
-   size_t getSize() const { return Stmts.size(); } 
-   
-   /// @name Statements Iterators 
-   /// 
-   /// These iterators iterate over all statements of this Scop. 
-   //@{ 
-   using iterator = StmtSet::iterator; 
-   using const_iterator = StmtSet::const_iterator; 
-   
-   iterator begin() { return Stmts.begin(); } 
-   iterator end() { return Stmts.end(); } 
-   const_iterator begin() const { return Stmts.begin(); } 
-   const_iterator end() const { return Stmts.end(); } 
-   
-   using reverse_iterator = StmtSet::reverse_iterator; 
-   using const_reverse_iterator = StmtSet::const_reverse_iterator; 
-   
-   reverse_iterator rbegin() { return Stmts.rbegin(); } 
-   reverse_iterator rend() { return Stmts.rend(); } 
-   const_reverse_iterator rbegin() const { return Stmts.rbegin(); } 
-   const_reverse_iterator rend() const { return Stmts.rend(); } 
-   //@} 
-   
-   /// Return the set of required invariant loads. 
-   const InvariantLoadsSetTy &getRequiredInvariantLoads() const { 
-     return DC.RequiredILS; 
-   } 
-   
-   /// Add @p LI to the set of required invariant loads. 
-   void addRequiredInvariantLoad(LoadInst *LI) { DC.RequiredILS.insert(LI); } 
-   
-   /// Return the set of boxed (thus overapproximated) loops. 
-   const BoxedLoopsSetTy &getBoxedLoops() const { return DC.BoxedLoopsSet; } 
-   
-   /// Return true if and only if @p R is a non-affine subregion. 
-   bool isNonAffineSubRegion(const Region *R) { 
-     return DC.NonAffineSubRegionSet.count(R); 
-   } 
-   
-   const MapInsnToMemAcc &getInsnToMemAccMap() const { return DC.InsnToMemAcc; } 
-   
-   /// Return the (possibly new) ScopArrayInfo object for @p Access. 
-   /// 
-   /// @param ElementType The type of the elements stored in this array. 
-   /// @param Kind        The kind of the array info object. 
-   /// @param BaseName    The optional name of this memory reference. 
-   ScopArrayInfo *getOrCreateScopArrayInfo(Value *BasePtr, Type *ElementType, 
-                                           ArrayRef<const SCEV *> Sizes, 
-                                           MemoryKind Kind, 
-                                           const char *BaseName = nullptr); 
-   
-   /// Create an array and return the corresponding ScopArrayInfo object. 
-   /// 
-   /// @param ElementType The type of the elements stored in this array. 
-   /// @param BaseName    The name of this memory reference. 
-   /// @param Sizes       The sizes of dimensions. 
-   ScopArrayInfo *createScopArrayInfo(Type *ElementType, 
-                                      const std::string &BaseName, 
-                                      const std::vector<unsigned> &Sizes); 
-   
-   /// Return the cached ScopArrayInfo object for @p BasePtr. 
-   /// 
-   /// @param BasePtr   The base pointer the object has been stored for. 
-   /// @param Kind      The kind of array info object. 
-   /// 
-   /// @returns The ScopArrayInfo pointer or NULL if no such pointer is 
-   ///          available. 
-   ScopArrayInfo *getScopArrayInfoOrNull(Value *BasePtr, MemoryKind Kind); 
-   
-   /// Return the cached ScopArrayInfo object for @p BasePtr. 
-   /// 
-   /// @param BasePtr   The base pointer the object has been stored for. 
-   /// @param Kind      The kind of array info object. 
-   /// 
-   /// @returns The ScopArrayInfo pointer (may assert if no such pointer is 
-   ///          available). 
-   ScopArrayInfo *getScopArrayInfo(Value *BasePtr, MemoryKind Kind); 
-   
-   /// Invalidate ScopArrayInfo object for base address. 
-   /// 
-   /// @param BasePtr The base pointer of the ScopArrayInfo object to invalidate. 
-   /// @param Kind    The Kind of the ScopArrayInfo object. 
-   void invalidateScopArrayInfo(Value *BasePtr, MemoryKind Kind) { 
-     auto It = ScopArrayInfoMap.find(std::make_pair(BasePtr, Kind)); 
-     if (It == ScopArrayInfoMap.end()) 
-       return; 
-     ScopArrayInfoSet.remove(It->second.get()); 
-     ScopArrayInfoMap.erase(It); 
-   } 
-   
-   /// Set new isl context. 
-   void setContext(isl::set NewContext); 
-   
-   /// Update maximal loop depth. If @p Depth is smaller than current value, 
-   /// then maximal loop depth is not updated. 
-   void updateMaxLoopDepth(unsigned Depth) { 
-     MaxLoopDepth = std::max(MaxLoopDepth, Depth); 
-   } 
-   
-   /// Align the parameters in the statement to the scop context 
-   void realignParams(); 
-   
-   /// Return true if this SCoP can be profitably optimized. 
-   /// 
-   /// @param ScalarsAreUnprofitable Never consider statements with scalar writes 
-   ///                               as profitably optimizable. 
-   /// 
-   /// @return Whether this SCoP can be profitably optimized. 
-   bool isProfitable(bool ScalarsAreUnprofitable) const; 
-   
-   /// Return true if the SCoP contained at least one error block. 
-   bool hasErrorBlock() const { return HasErrorBlock; } 
-   
-   /// Notify SCoP that it contains an error block 
-   void notifyErrorBlock() { HasErrorBlock = true; } 
-   
-   /// Return true if the underlying region has a single exiting block. 
-   bool hasSingleExitEdge() const { return HasSingleExitEdge; } 
-   
-   /// Print the static control part. 
-   /// 
-   /// @param OS The output stream the static control part is printed to. 
-   /// @param PrintInstructions Whether to print the statement's instructions as 
-   ///                          well. 
-   void print(raw_ostream &OS, bool PrintInstructions) const; 
-   
- #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) 
-   /// Print the ScopStmt to stderr. 
-   void dump() const; 
- #endif 
-   
-   /// Get the isl context of this static control part. 
-   /// 
-   /// @return The isl context of this static control part. 
-   isl::ctx getIslCtx() const; 
-   
-   /// Directly return the shared_ptr of the context. 
-   const std::shared_ptr<isl_ctx> &getSharedIslCtx() const { return IslCtx; } 
-   
-   /// Compute the isl representation for the SCEV @p E 
-   /// 
-   /// @param E  The SCEV that should be translated. 
-   /// @param BB An (optional) basic block in which the isl_pw_aff is computed. 
-   ///           SCEVs known to not reference any loops in the SCoP can be 
-   ///           passed without a @p BB. 
-   /// @param NonNegative Flag to indicate the @p E has to be non-negative. 
-   /// 
-   /// Note that this function will always return a valid isl_pw_aff. However, if 
-   /// the translation of @p E was deemed to complex the SCoP is invalidated and 
-   /// a dummy value of appropriate dimension is returned. This allows to bail 
-   /// for complex cases without "error handling code" needed on the users side. 
-   PWACtx getPwAff(const SCEV *E, BasicBlock *BB = nullptr, 
-                   bool NonNegative = false, 
-                   RecordedAssumptionsTy *RecordedAssumptions = nullptr); 
-   
-   /// Compute the isl representation for the SCEV @p E 
-   /// 
-   /// This function is like @see Scop::getPwAff() but strips away the invalid 
-   /// domain part associated with the piecewise affine function. 
-   isl::pw_aff 
-   getPwAffOnly(const SCEV *E, BasicBlock *BB = nullptr, 
-                RecordedAssumptionsTy *RecordedAssumptions = nullptr); 
-   
-   /// Check if an <nsw> AddRec for the loop L is cached. 
-   bool hasNSWAddRecForLoop(Loop *L) { return Affinator.hasNSWAddRecForLoop(L); } 
-   
-   /// Return the domain of @p Stmt. 
-   /// 
-   /// @param Stmt The statement for which the conditions should be returned. 
-   isl::set getDomainConditions(const ScopStmt *Stmt) const; 
-   
-   /// Return the domain of @p BB. 
-   /// 
-   /// @param BB The block for which the conditions should be returned. 
-   isl::set getDomainConditions(BasicBlock *BB) const; 
-   
-   /// Return the domain of @p BB. If it does not exist, create an empty one. 
-   isl::set &getOrInitEmptyDomain(BasicBlock *BB) { return DomainMap[BB]; } 
-   
-   /// Check if domain is determined for @p BB. 
-   bool isDomainDefined(BasicBlock *BB) const { return DomainMap.count(BB) > 0; } 
-   
-   /// Set domain for @p BB. 
-   void setDomain(BasicBlock *BB, isl::set &Domain) { DomainMap[BB] = Domain; } 
-   
-   /// Get a union set containing the iteration domains of all statements. 
-   isl::union_set getDomains() const; 
-   
-   /// Get a union map of all may-writes performed in the SCoP. 
-   isl::union_map getMayWrites(); 
-   
-   /// Get a union map of all must-writes performed in the SCoP. 
-   isl::union_map getMustWrites(); 
-   
-   /// Get a union map of all writes performed in the SCoP. 
-   isl::union_map getWrites(); 
-   
-   /// Get a union map of all reads performed in the SCoP. 
-   isl::union_map getReads(); 
-   
-   /// Get a union map of all memory accesses performed in the SCoP. 
-   isl::union_map getAccesses(); 
-   
-   /// Get a union map of all memory accesses performed in the SCoP. 
-   /// 
-   /// @param Array The array to which the accesses should belong. 
-   isl::union_map getAccesses(ScopArrayInfo *Array); 
-   
-   /// Get the schedule of all the statements in the SCoP. 
-   /// 
-   /// @return The schedule of all the statements in the SCoP, if the schedule of 
-   /// the Scop does not contain extension nodes, and nullptr, otherwise. 
-   isl::union_map getSchedule() const; 
-   
-   /// Get a schedule tree describing the schedule of all statements. 
-   isl::schedule getScheduleTree() const; 
-   
-   /// Update the current schedule 
-   /// 
-   /// NewSchedule The new schedule (given as a flat union-map). 
-   void setSchedule(isl::union_map NewSchedule); 
-   
-   /// Update the current schedule 
-   /// 
-   /// NewSchedule The new schedule (given as schedule tree). 
-   void setScheduleTree(isl::schedule NewSchedule); 
-   
-   /// Whether the schedule is the original schedule as derived from the CFG by 
-   /// ScopBuilder. 
-   bool isOriginalSchedule() const { return !ScheduleModified; } 
-   
-   /// Intersects the domains of all statements in the SCoP. 
-   /// 
-   /// @return true if a change was made 
-   bool restrictDomains(isl::union_set Domain); 
-   
-   /// Get the depth of a loop relative to the outermost loop in the Scop. 
-   /// 
-   /// This will return 
-   ///    0 if @p L is an outermost loop in the SCoP 
-   ///   >0 for other loops in the SCoP 
-   ///   -1 if @p L is nullptr or there is no outermost loop in the SCoP 
-   int getRelativeLoopDepth(const Loop *L) const; 
-   
-   /// Find the ScopArrayInfo associated with an isl Id 
-   ///        that has name @p Name. 
-   ScopArrayInfo *getArrayInfoByName(const std::string BaseName); 
-   
-   /// Simplify the SCoP representation. 
-   /// 
-   /// @param AfterHoisting Whether it is called after invariant load hoisting. 
-   ///                      When true, also removes statements without 
-   ///                      side-effects. 
-   void simplifySCoP(bool AfterHoisting); 
-   
-   /// Get the next free array index. 
-   /// 
-   /// This function returns a unique index which can be used to identify an 
-   /// array. 
-   long getNextArrayIdx() { return ArrayIdx++; } 
-   
-   /// Get the next free statement index. 
-   /// 
-   /// This function returns a unique index which can be used to identify a 
-   /// statement. 
-   long getNextStmtIdx() { return StmtIdx++; } 
-   
-   /// Get the representing SCEV for @p S if applicable, otherwise @p S. 
-   /// 
-   /// Invariant loads of the same location are put in an equivalence class and 
-   /// only one of them is chosen as a representing element that will be 
-   /// modeled as a parameter. The others have to be normalized, i.e., 
-   /// replaced by the representing element of their equivalence class, in order 
-   /// to get the correct parameter value, e.g., in the SCEVAffinator. 
-   /// 
-   /// @param S The SCEV to normalize. 
-   /// 
-   /// @return The representing SCEV for invariant loads or @p S if none. 
-   const SCEV *getRepresentingInvariantLoadSCEV(const SCEV *S) const; 
-   
-   /// Return the MemoryAccess that writes an llvm::Value, represented by a 
-   /// ScopArrayInfo. 
-   /// 
-   /// There can be at most one such MemoryAccess per llvm::Value in the SCoP. 
-   /// Zero is possible for read-only values. 
-   MemoryAccess *getValueDef(const ScopArrayInfo *SAI) const; 
-   
-   /// Return all MemoryAccesses that us an llvm::Value, represented by a 
-   /// ScopArrayInfo. 
-   ArrayRef<MemoryAccess *> getValueUses(const ScopArrayInfo *SAI) const; 
-   
-   /// Return the MemoryAccess that represents an llvm::PHINode. 
-   /// 
-   /// ExitPHIs's PHINode is not within the SCoPs. This function returns nullptr 
-   /// for them. 
-   MemoryAccess *getPHIRead(const ScopArrayInfo *SAI) const; 
-   
-   /// Return all MemoryAccesses for all incoming statements of a PHINode, 
-   /// represented by a ScopArrayInfo. 
-   ArrayRef<MemoryAccess *> getPHIIncomings(const ScopArrayInfo *SAI) const; 
-   
-   /// Return whether @p Inst has a use outside of this SCoP. 
-   bool isEscaping(Instruction *Inst); 
-   
-   struct ScopStatistics { 
-     int NumAffineLoops = 0; 
-     int NumBoxedLoops = 0; 
-   
-     int NumValueWrites = 0; 
-     int NumValueWritesInLoops = 0; 
-     int NumPHIWrites = 0; 
-     int NumPHIWritesInLoops = 0; 
-     int NumSingletonWrites = 0; 
-     int NumSingletonWritesInLoops = 0; 
-   }; 
-   
-   /// Collect statistic about this SCoP. 
-   /// 
-   /// These are most commonly used for LLVM's static counters (Statistic.h) in 
-   /// various places. If statistics are disabled, only zeros are returned to 
-   /// avoid the overhead. 
-   ScopStatistics getStatistics() const; 
-   
-   /// Is this Scop marked as not to be transformed by an optimization heuristic? 
-   /// In this case, only user-directed transformations are allowed. 
-   bool hasDisableHeuristicsHint() const { return HasDisableHeuristicsHint; } 
-   
-   /// Mark this Scop to not apply an optimization heuristic. 
-   void markDisableHeuristics() { HasDisableHeuristicsHint = true; } 
- }; 
-   
- /// Print Scop scop to raw_ostream OS. 
- raw_ostream &operator<<(raw_ostream &OS, const Scop &scop); 
-   
- /// The legacy pass manager's analysis pass to compute scop information 
- ///        for a region. 
- class ScopInfoRegionPass final : public RegionPass { 
-   /// The Scop pointer which is used to construct a Scop. 
-   std::unique_ptr<Scop> S; 
-   
- public: 
-   static char ID; // Pass identification, replacement for typeid 
-   
-   ScopInfoRegionPass() : RegionPass(ID) {} 
-   ~ScopInfoRegionPass() override = default; 
-   
-   /// Build Scop object, the Polly IR of static control 
-   ///        part for the current SESE-Region. 
-   /// 
-   /// @return If the current region is a valid for a static control part, 
-   ///         return the Polly IR representing this static control part, 
-   ///         return null otherwise. 
-   Scop *getScop() { return S.get(); } 
-   const Scop *getScop() const { return S.get(); } 
-   
-   /// Calculate the polyhedral scop information for a given Region. 
-   bool runOnRegion(Region *R, RGPassManager &RGM) override; 
-   
-   void releaseMemory() override { S.reset(); } 
-   
-   void print(raw_ostream &O, const Module *M = nullptr) const override; 
-   
-   void getAnalysisUsage(AnalysisUsage &AU) const override; 
- }; 
-   
- llvm::Pass *createScopInfoPrinterLegacyRegionPass(raw_ostream &OS); 
-   
- class ScopInfo { 
- public: 
-   using RegionToScopMapTy = MapVector<Region *, std::unique_ptr<Scop>>; 
-   using reverse_iterator = RegionToScopMapTy::reverse_iterator; 
-   using const_reverse_iterator = RegionToScopMapTy::const_reverse_iterator; 
-   using iterator = RegionToScopMapTy::iterator; 
-   using const_iterator = RegionToScopMapTy::const_iterator; 
-   
- private: 
-   /// A map of Region to its Scop object containing 
-   ///        Polly IR of static control part. 
-   RegionToScopMapTy RegionToScopMap; 
-   const DataLayout &DL; 
-   ScopDetection &SD; 
-   ScalarEvolution &SE; 
-   LoopInfo &LI; 
-   AAResults &AA; 
-   DominatorTree &DT; 
-   AssumptionCache &AC; 
-   OptimizationRemarkEmitter &ORE; 
-   
- public: 
-   ScopInfo(const DataLayout &DL, ScopDetection &SD, ScalarEvolution &SE, 
-            LoopInfo &LI, AAResults &AA, DominatorTree &DT, AssumptionCache &AC, 
-            OptimizationRemarkEmitter &ORE); 
-   
-   /// Get the Scop object for the given Region. 
-   /// 
-   /// @return If the given region is the maximal region within a scop, return 
-   ///         the scop object. If the given region is a subregion, return a 
-   ///         nullptr. Top level region containing the entry block of a function 
-   ///         is not considered in the scop creation. 
-   Scop *getScop(Region *R) const { 
-     auto MapIt = RegionToScopMap.find(R); 
-     if (MapIt != RegionToScopMap.end()) 
-       return MapIt->second.get(); 
-     return nullptr; 
-   } 
-   
-   /// Recompute the Scop-Information for a function. 
-   /// 
-   /// This invalidates any iterators. 
-   void recompute(); 
-   
-   /// Handle invalidation explicitly 
-   bool invalidate(Function &F, const PreservedAnalyses &PA, 
-                   FunctionAnalysisManager::Invalidator &Inv); 
-   
-   iterator begin() { return RegionToScopMap.begin(); } 
-   iterator end() { return RegionToScopMap.end(); } 
-   const_iterator begin() const { return RegionToScopMap.begin(); } 
-   const_iterator end() const { return RegionToScopMap.end(); } 
-   reverse_iterator rbegin() { return RegionToScopMap.rbegin(); } 
-   reverse_iterator rend() { return RegionToScopMap.rend(); } 
-   const_reverse_iterator rbegin() const { return RegionToScopMap.rbegin(); } 
-   const_reverse_iterator rend() const { return RegionToScopMap.rend(); } 
-   bool empty() const { return RegionToScopMap.empty(); } 
- }; 
-   
- struct ScopInfoAnalysis : AnalysisInfoMixin<ScopInfoAnalysis> { 
-   static AnalysisKey Key; 
-   
-   using Result = ScopInfo; 
-   
-   Result run(Function &, FunctionAnalysisManager &); 
- }; 
-   
- struct ScopInfoPrinterPass final : PassInfoMixin<ScopInfoPrinterPass> { 
-   ScopInfoPrinterPass(raw_ostream &OS) : Stream(OS) {} 
-   
-   PreservedAnalyses run(Function &, FunctionAnalysisManager &); 
-   
-   raw_ostream &Stream; 
- }; 
-   
- //===----------------------------------------------------------------------===// 
- /// The legacy pass manager's analysis pass to compute scop information 
- ///        for the whole function. 
- /// 
- /// This pass will maintain a map of the maximal region within a scop to its 
- /// scop object for all the feasible scops present in a function. 
- /// This pass is an alternative to the ScopInfoRegionPass in order to avoid a 
- /// region pass manager. 
- class ScopInfoWrapperPass final : public FunctionPass { 
-   std::unique_ptr<ScopInfo> Result; 
-   
- public: 
-   ScopInfoWrapperPass() : FunctionPass(ID) {} 
-   ~ScopInfoWrapperPass() override = default; 
-   
-   static char ID; // Pass identification, replacement for typeid 
-   
-   ScopInfo *getSI() { return Result.get(); } 
-   const ScopInfo *getSI() const { return Result.get(); } 
-   
-   /// Calculate all the polyhedral scops for a given function. 
-   bool runOnFunction(Function &F) override; 
-   
-   void releaseMemory() override { Result.reset(); } 
-   
-   void print(raw_ostream &O, const Module *M = nullptr) const override; 
-   
-   void getAnalysisUsage(AnalysisUsage &AU) const override; 
- }; 
-   
- llvm::Pass *createScopInfoPrinterLegacyFunctionPass(llvm::raw_ostream &OS); 
- } // end namespace polly 
-   
- namespace llvm { 
- void initializeScopInfoRegionPassPass(PassRegistry &); 
- void initializeScopInfoPrinterLegacyRegionPassPass(PassRegistry &); 
- void initializeScopInfoWrapperPassPass(PassRegistry &); 
- void initializeScopInfoPrinterLegacyFunctionPassPass(PassRegistry &); 
- } // end namespace llvm 
-   
- #endif // POLLY_SCOPINFO_H 
-