Home | History | Annotate | Download | only in Analysis
      1 //===- MemorySSA.h - Build Memory SSA ---------------------------*- C++ -*-===//
      2 //
      3 //                     The LLVM Compiler Infrastructure
      4 //
      5 // This file is distributed under the University of Illinois Open Source
      6 // License. See LICENSE.TXT for details.
      7 //
      8 //===----------------------------------------------------------------------===//
      9 //
     10 /// \file
     11 /// \brief This file exposes an interface to building/using memory SSA to
     12 /// walk memory instructions using a use/def graph.
     13 ///
     14 /// Memory SSA class builds an SSA form that links together memory access
     15 /// instructions such as loads, stores, atomics, and calls. Additionally, it
     16 /// does a trivial form of "heap versioning" Every time the memory state changes
     17 /// in the program, we generate a new heap version. It generates
     18 /// MemoryDef/Uses/Phis that are overlayed on top of the existing instructions.
     19 ///
     20 /// As a trivial example,
     21 /// define i32 @main() #0 {
     22 /// entry:
     23 ///   %call = call noalias i8* @_Znwm(i64 4) #2
     24 ///   %0 = bitcast i8* %call to i32*
     25 ///   %call1 = call noalias i8* @_Znwm(i64 4) #2
     26 ///   %1 = bitcast i8* %call1 to i32*
     27 ///   store i32 5, i32* %0, align 4
     28 ///   store i32 7, i32* %1, align 4
     29 ///   %2 = load i32* %0, align 4
     30 ///   %3 = load i32* %1, align 4
     31 ///   %add = add nsw i32 %2, %3
     32 ///   ret i32 %add
     33 /// }
     34 ///
     35 /// Will become
     36 /// define i32 @main() #0 {
     37 /// entry:
     38 ///   ; 1 = MemoryDef(0)
     39 ///   %call = call noalias i8* @_Znwm(i64 4) #3
     40 ///   %2 = bitcast i8* %call to i32*
     41 ///   ; 2 = MemoryDef(1)
     42 ///   %call1 = call noalias i8* @_Znwm(i64 4) #3
     43 ///   %4 = bitcast i8* %call1 to i32*
     44 ///   ; 3 = MemoryDef(2)
     45 ///   store i32 5, i32* %2, align 4
     46 ///   ; 4 = MemoryDef(3)
     47 ///   store i32 7, i32* %4, align 4
     48 ///   ; MemoryUse(3)
     49 ///   %7 = load i32* %2, align 4
     50 ///   ; MemoryUse(4)
     51 ///   %8 = load i32* %4, align 4
     52 ///   %add = add nsw i32 %7, %8
     53 ///   ret i32 %add
     54 /// }
     55 ///
     56 /// Given this form, all the stores that could ever effect the load at %8 can be
     57 /// gotten by using the MemoryUse associated with it, and walking from use to
     58 /// def until you hit the top of the function.
     59 ///
     60 /// Each def also has a list of users associated with it, so you can walk from
     61 /// both def to users, and users to defs. Note that we disambiguate MemoryUses,
     62 /// but not the RHS of MemoryDefs. You can see this above at %7, which would
     63 /// otherwise be a MemoryUse(4). Being disambiguated means that for a given
     64 /// store, all the MemoryUses on its use lists are may-aliases of that store
     65 /// (but the MemoryDefs on its use list may not be).
     66 ///
     67 /// MemoryDefs are not disambiguated because it would require multiple reaching
     68 /// definitions, which would require multiple phis, and multiple memoryaccesses
     69 /// per instruction.
     70 //
     71 //===----------------------------------------------------------------------===//
     72 
     73 #ifndef LLVM_ANALYSIS_MEMORYSSA_H
     74 #define LLVM_ANALYSIS_MEMORYSSA_H
     75 
     76 #include "llvm/ADT/DenseMap.h"
     77 #include "llvm/ADT/GraphTraits.h"
     78 #include "llvm/ADT/SmallPtrSet.h"
     79 #include "llvm/ADT/SmallVector.h"
     80 #include "llvm/ADT/ilist.h"
     81 #include "llvm/ADT/ilist_node.h"
     82 #include "llvm/ADT/iterator.h"
     83 #include "llvm/ADT/iterator_range.h"
     84 #include "llvm/ADT/simple_ilist.h"
     85 #include "llvm/Analysis/AliasAnalysis.h"
     86 #include "llvm/Analysis/MemoryLocation.h"
     87 #include "llvm/Analysis/PHITransAddr.h"
     88 #include "llvm/IR/BasicBlock.h"
     89 #include "llvm/IR/DerivedUser.h"
     90 #include "llvm/IR/Dominators.h"
     91 #include "llvm/IR/Module.h"
     92 #include "llvm/IR/Type.h"
     93 #include "llvm/IR/Use.h"
     94 #include "llvm/IR/User.h"
     95 #include "llvm/IR/Value.h"
     96 #include "llvm/Pass.h"
     97 #include "llvm/Support/Casting.h"
     98 #include <algorithm>
     99 #include <cassert>
    100 #include <cstddef>
    101 #include <iterator>
    102 #include <memory>
    103 #include <utility>
    104 
    105 namespace llvm {
    106 
    107 class Function;
    108 class Instruction;
    109 class MemoryAccess;
    110 class MemorySSAWalker;
    111 class LLVMContext;
    112 class raw_ostream;
    113 
    114 namespace MSSAHelpers {
    115 
    116 struct AllAccessTag {};
    117 struct DefsOnlyTag {};
    118 
    119 } // end namespace MSSAHelpers
    120 
    121 enum {
    122   // Used to signify what the default invalid ID is for MemoryAccess's
    123   // getID()
    124   INVALID_MEMORYACCESS_ID = 0
    125 };
    126 
    127 template <class T> class memoryaccess_def_iterator_base;
    128 using memoryaccess_def_iterator = memoryaccess_def_iterator_base<MemoryAccess>;
    129 using const_memoryaccess_def_iterator =
    130     memoryaccess_def_iterator_base<const MemoryAccess>;
    131 
    132 // \brief The base for all memory accesses. All memory accesses in a block are
    133 // linked together using an intrusive list.
    134 class MemoryAccess
    135     : public DerivedUser,
    136       public ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>,
    137       public ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>> {
    138 public:
    139   using AllAccessType =
    140       ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>;
    141   using DefsOnlyType =
    142       ilist_node<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>>;
    143 
    144   MemoryAccess(const MemoryAccess &) = delete;
    145   MemoryAccess &operator=(const MemoryAccess &) = delete;
    146 
    147   void *operator new(size_t) = delete;
    148 
    149   // Methods for support type inquiry through isa, cast, and
    150   // dyn_cast
    151   static bool classof(const Value *V) {
    152     unsigned ID = V->getValueID();
    153     return ID == MemoryUseVal || ID == MemoryPhiVal || ID == MemoryDefVal;
    154   }
    155 
    156   BasicBlock *getBlock() const { return Block; }
    157 
    158   void print(raw_ostream &OS) const;
    159   void dump() const;
    160 
    161   /// \brief The user iterators for a memory access
    162   using iterator = user_iterator;
    163   using const_iterator = const_user_iterator;
    164 
    165   /// \brief This iterator walks over all of the defs in a given
    166   /// MemoryAccess. For MemoryPhi nodes, this walks arguments. For
    167   /// MemoryUse/MemoryDef, this walks the defining access.
    168   memoryaccess_def_iterator defs_begin();
    169   const_memoryaccess_def_iterator defs_begin() const;
    170   memoryaccess_def_iterator defs_end();
    171   const_memoryaccess_def_iterator defs_end() const;
    172 
    173   /// \brief Get the iterators for the all access list and the defs only list
    174   /// We default to the all access list.
    175   AllAccessType::self_iterator getIterator() {
    176     return this->AllAccessType::getIterator();
    177   }
    178   AllAccessType::const_self_iterator getIterator() const {
    179     return this->AllAccessType::getIterator();
    180   }
    181   AllAccessType::reverse_self_iterator getReverseIterator() {
    182     return this->AllAccessType::getReverseIterator();
    183   }
    184   AllAccessType::const_reverse_self_iterator getReverseIterator() const {
    185     return this->AllAccessType::getReverseIterator();
    186   }
    187   DefsOnlyType::self_iterator getDefsIterator() {
    188     return this->DefsOnlyType::getIterator();
    189   }
    190   DefsOnlyType::const_self_iterator getDefsIterator() const {
    191     return this->DefsOnlyType::getIterator();
    192   }
    193   DefsOnlyType::reverse_self_iterator getReverseDefsIterator() {
    194     return this->DefsOnlyType::getReverseIterator();
    195   }
    196   DefsOnlyType::const_reverse_self_iterator getReverseDefsIterator() const {
    197     return this->DefsOnlyType::getReverseIterator();
    198   }
    199 
    200 protected:
    201   friend class MemoryDef;
    202   friend class MemoryPhi;
    203   friend class MemorySSA;
    204   friend class MemoryUse;
    205   friend class MemoryUseOrDef;
    206 
    207   /// \brief Used by MemorySSA to change the block of a MemoryAccess when it is
    208   /// moved.
    209   void setBlock(BasicBlock *BB) { Block = BB; }
    210 
    211   /// \brief Used for debugging and tracking things about MemoryAccesses.
    212   /// Guaranteed unique among MemoryAccesses, no guarantees otherwise.
    213   inline unsigned getID() const;
    214 
    215   MemoryAccess(LLVMContext &C, unsigned Vty, DeleteValueTy DeleteValue,
    216                BasicBlock *BB, unsigned NumOperands)
    217       : DerivedUser(Type::getVoidTy(C), Vty, nullptr, NumOperands, DeleteValue),
    218         Block(BB) {}
    219 
    220 private:
    221   BasicBlock *Block;
    222 };
    223 
    224 inline raw_ostream &operator<<(raw_ostream &OS, const MemoryAccess &MA) {
    225   MA.print(OS);
    226   return OS;
    227 }
    228 
    229 /// \brief Class that has the common methods + fields of memory uses/defs. It's
    230 /// a little awkward to have, but there are many cases where we want either a
    231 /// use or def, and there are many cases where uses are needed (defs aren't
    232 /// acceptable), and vice-versa.
    233 ///
    234 /// This class should never be instantiated directly; make a MemoryUse or
    235 /// MemoryDef instead.
    236 class MemoryUseOrDef : public MemoryAccess {
    237 public:
    238   void *operator new(size_t) = delete;
    239 
    240   DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess);
    241 
    242   /// \brief Get the instruction that this MemoryUse represents.
    243   Instruction *getMemoryInst() const { return MemoryInst; }
    244 
    245   /// \brief Get the access that produces the memory state used by this Use.
    246   MemoryAccess *getDefiningAccess() const { return getOperand(0); }
    247 
    248   static bool classof(const Value *MA) {
    249     return MA->getValueID() == MemoryUseVal || MA->getValueID() == MemoryDefVal;
    250   }
    251 
    252   // Sadly, these have to be public because they are needed in some of the
    253   // iterators.
    254   inline bool isOptimized() const;
    255   inline MemoryAccess *getOptimized() const;
    256   inline void setOptimized(MemoryAccess *);
    257 
    258   /// \brief Reset the ID of what this MemoryUse was optimized to, causing it to
    259   /// be rewalked by the walker if necessary.
    260   /// This really should only be called by tests.
    261   inline void resetOptimized();
    262 
    263 protected:
    264   friend class MemorySSA;
    265   friend class MemorySSAUpdater;
    266 
    267   MemoryUseOrDef(LLVMContext &C, MemoryAccess *DMA, unsigned Vty,
    268                  DeleteValueTy DeleteValue, Instruction *MI, BasicBlock *BB)
    269       : MemoryAccess(C, Vty, DeleteValue, BB, 1), MemoryInst(MI) {
    270     setDefiningAccess(DMA);
    271   }
    272 
    273   void setDefiningAccess(MemoryAccess *DMA, bool Optimized = false) {
    274     if (!Optimized) {
    275       setOperand(0, DMA);
    276       return;
    277     }
    278     setOptimized(DMA);
    279   }
    280 
    281 private:
    282   Instruction *MemoryInst;
    283 };
    284 
    285 template <>
    286 struct OperandTraits<MemoryUseOrDef>
    287     : public FixedNumOperandTraits<MemoryUseOrDef, 1> {};
    288 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryUseOrDef, MemoryAccess)
    289 
    290 /// \brief Represents read-only accesses to memory
    291 ///
    292 /// In particular, the set of Instructions that will be represented by
    293 /// MemoryUse's is exactly the set of Instructions for which
    294 /// AliasAnalysis::getModRefInfo returns "Ref".
    295 class MemoryUse final : public MemoryUseOrDef {
    296 public:
    297   DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess);
    298 
    299   MemoryUse(LLVMContext &C, MemoryAccess *DMA, Instruction *MI, BasicBlock *BB)
    300       : MemoryUseOrDef(C, DMA, MemoryUseVal, deleteMe, MI, BB) {}
    301 
    302   // allocate space for exactly one operand
    303   void *operator new(size_t s) { return User::operator new(s, 1); }
    304 
    305   static bool classof(const Value *MA) {
    306     return MA->getValueID() == MemoryUseVal;
    307   }
    308 
    309   void print(raw_ostream &OS) const;
    310 
    311   void setOptimized(MemoryAccess *DMA) {
    312     OptimizedID = DMA->getID();
    313     setOperand(0, DMA);
    314   }
    315 
    316   bool isOptimized() const {
    317     return getDefiningAccess() && OptimizedID == getDefiningAccess()->getID();
    318   }
    319 
    320   MemoryAccess *getOptimized() const {
    321     return getDefiningAccess();
    322   }
    323 
    324   void resetOptimized() {
    325     OptimizedID = INVALID_MEMORYACCESS_ID;
    326   }
    327 
    328 protected:
    329   friend class MemorySSA;
    330 
    331 private:
    332   static void deleteMe(DerivedUser *Self);
    333 
    334   unsigned int OptimizedID = 0;
    335 };
    336 
    337 template <>
    338 struct OperandTraits<MemoryUse> : public FixedNumOperandTraits<MemoryUse, 1> {};
    339 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryUse, MemoryAccess)
    340 
    341 /// \brief Represents a read-write access to memory, whether it is a must-alias,
    342 /// or a may-alias.
    343 ///
    344 /// In particular, the set of Instructions that will be represented by
    345 /// MemoryDef's is exactly the set of Instructions for which
    346 /// AliasAnalysis::getModRefInfo returns "Mod" or "ModRef".
    347 /// Note that, in order to provide def-def chains, all defs also have a use
    348 /// associated with them. This use points to the nearest reaching
    349 /// MemoryDef/MemoryPhi.
    350 class MemoryDef final : public MemoryUseOrDef {
    351 public:
    352   friend class MemorySSA;
    353 
    354   DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess);
    355 
    356   MemoryDef(LLVMContext &C, MemoryAccess *DMA, Instruction *MI, BasicBlock *BB,
    357             unsigned Ver)
    358       : MemoryUseOrDef(C, DMA, MemoryDefVal, deleteMe, MI, BB), ID(Ver) {}
    359 
    360   // allocate space for exactly one operand
    361   void *operator new(size_t s) { return User::operator new(s, 1); }
    362 
    363   static bool classof(const Value *MA) {
    364     return MA->getValueID() == MemoryDefVal;
    365   }
    366 
    367   void setOptimized(MemoryAccess *MA) {
    368     Optimized = MA;
    369     OptimizedID = getDefiningAccess()->getID();
    370   }
    371 
    372   MemoryAccess *getOptimized() const { return Optimized; }
    373 
    374   bool isOptimized() const {
    375     return getOptimized() && getDefiningAccess() &&
    376            OptimizedID == getDefiningAccess()->getID();
    377   }
    378 
    379   void resetOptimized() {
    380     OptimizedID = INVALID_MEMORYACCESS_ID;
    381   }
    382 
    383   void print(raw_ostream &OS) const;
    384 
    385   unsigned getID() const { return ID; }
    386 
    387 private:
    388   static void deleteMe(DerivedUser *Self);
    389 
    390   const unsigned ID;
    391   MemoryAccess *Optimized = nullptr;
    392   unsigned int OptimizedID = INVALID_MEMORYACCESS_ID;
    393 };
    394 
    395 template <>
    396 struct OperandTraits<MemoryDef> : public FixedNumOperandTraits<MemoryDef, 1> {};
    397 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryDef, MemoryAccess)
    398 
    399 /// \brief Represents phi nodes for memory accesses.
    400 ///
    401 /// These have the same semantic as regular phi nodes, with the exception that
    402 /// only one phi will ever exist in a given basic block.
    403 /// Guaranteeing one phi per block means guaranteeing there is only ever one
    404 /// valid reaching MemoryDef/MemoryPHI along each path to the phi node.
    405 /// This is ensured by not allowing disambiguation of the RHS of a MemoryDef or
    406 /// a MemoryPhi's operands.
    407 /// That is, given
    408 /// if (a) {
    409 ///   store %a
    410 ///   store %b
    411 /// }
    412 /// it *must* be transformed into
    413 /// if (a) {
    414 ///    1 = MemoryDef(liveOnEntry)
    415 ///    store %a
    416 ///    2 = MemoryDef(1)
    417 ///    store %b
    418 /// }
    419 /// and *not*
    420 /// if (a) {
    421 ///    1 = MemoryDef(liveOnEntry)
    422 ///    store %a
    423 ///    2 = MemoryDef(liveOnEntry)
    424 ///    store %b
    425 /// }
    426 /// even if the two stores do not conflict. Otherwise, both 1 and 2 reach the
    427 /// end of the branch, and if there are not two phi nodes, one will be
    428 /// disconnected completely from the SSA graph below that point.
    429 /// Because MemoryUse's do not generate new definitions, they do not have this
    430 /// issue.
    431 class MemoryPhi final : public MemoryAccess {
    432   // allocate space for exactly zero operands
    433   void *operator new(size_t s) { return User::operator new(s); }
    434 
    435 public:
    436   /// Provide fast operand accessors
    437   DECLARE_TRANSPARENT_OPERAND_ACCESSORS(MemoryAccess);
    438 
    439   MemoryPhi(LLVMContext &C, BasicBlock *BB, unsigned Ver, unsigned NumPreds = 0)
    440       : MemoryAccess(C, MemoryPhiVal, deleteMe, BB, 0), ID(Ver),
    441         ReservedSpace(NumPreds) {
    442     allocHungoffUses(ReservedSpace);
    443   }
    444 
    445   // Block iterator interface. This provides access to the list of incoming
    446   // basic blocks, which parallels the list of incoming values.
    447   using block_iterator = BasicBlock **;
    448   using const_block_iterator = BasicBlock *const *;
    449 
    450   block_iterator block_begin() {
    451     auto *Ref = reinterpret_cast<Use::UserRef *>(op_begin() + ReservedSpace);
    452     return reinterpret_cast<block_iterator>(Ref + 1);
    453   }
    454 
    455   const_block_iterator block_begin() const {
    456     const auto *Ref =
    457         reinterpret_cast<const Use::UserRef *>(op_begin() + ReservedSpace);
    458     return reinterpret_cast<const_block_iterator>(Ref + 1);
    459   }
    460 
    461   block_iterator block_end() { return block_begin() + getNumOperands(); }
    462 
    463   const_block_iterator block_end() const {
    464     return block_begin() + getNumOperands();
    465   }
    466 
    467   iterator_range<block_iterator> blocks() {
    468     return make_range(block_begin(), block_end());
    469   }
    470 
    471   iterator_range<const_block_iterator> blocks() const {
    472     return make_range(block_begin(), block_end());
    473   }
    474 
    475   op_range incoming_values() { return operands(); }
    476 
    477   const_op_range incoming_values() const { return operands(); }
    478 
    479   /// \brief Return the number of incoming edges
    480   unsigned getNumIncomingValues() const { return getNumOperands(); }
    481 
    482   /// \brief Return incoming value number x
    483   MemoryAccess *getIncomingValue(unsigned I) const { return getOperand(I); }
    484   void setIncomingValue(unsigned I, MemoryAccess *V) {
    485     assert(V && "PHI node got a null value!");
    486     setOperand(I, V);
    487   }
    488 
    489   static unsigned getOperandNumForIncomingValue(unsigned I) { return I; }
    490   static unsigned getIncomingValueNumForOperand(unsigned I) { return I; }
    491 
    492   /// \brief Return incoming basic block number @p i.
    493   BasicBlock *getIncomingBlock(unsigned I) const { return block_begin()[I]; }
    494 
    495   /// \brief Return incoming basic block corresponding
    496   /// to an operand of the PHI.
    497   BasicBlock *getIncomingBlock(const Use &U) const {
    498     assert(this == U.getUser() && "Iterator doesn't point to PHI's Uses?");
    499     return getIncomingBlock(unsigned(&U - op_begin()));
    500   }
    501 
    502   /// \brief Return incoming basic block corresponding
    503   /// to value use iterator.
    504   BasicBlock *getIncomingBlock(MemoryAccess::const_user_iterator I) const {
    505     return getIncomingBlock(I.getUse());
    506   }
    507 
    508   void setIncomingBlock(unsigned I, BasicBlock *BB) {
    509     assert(BB && "PHI node got a null basic block!");
    510     block_begin()[I] = BB;
    511   }
    512 
    513   /// \brief Add an incoming value to the end of the PHI list
    514   void addIncoming(MemoryAccess *V, BasicBlock *BB) {
    515     if (getNumOperands() == ReservedSpace)
    516       growOperands(); // Get more space!
    517     // Initialize some new operands.
    518     setNumHungOffUseOperands(getNumOperands() + 1);
    519     setIncomingValue(getNumOperands() - 1, V);
    520     setIncomingBlock(getNumOperands() - 1, BB);
    521   }
    522 
    523   /// \brief Return the first index of the specified basic
    524   /// block in the value list for this PHI.  Returns -1 if no instance.
    525   int getBasicBlockIndex(const BasicBlock *BB) const {
    526     for (unsigned I = 0, E = getNumOperands(); I != E; ++I)
    527       if (block_begin()[I] == BB)
    528         return I;
    529     return -1;
    530   }
    531 
    532   Value *getIncomingValueForBlock(const BasicBlock *BB) const {
    533     int Idx = getBasicBlockIndex(BB);
    534     assert(Idx >= 0 && "Invalid basic block argument!");
    535     return getIncomingValue(Idx);
    536   }
    537 
    538   static bool classof(const Value *V) {
    539     return V->getValueID() == MemoryPhiVal;
    540   }
    541 
    542   void print(raw_ostream &OS) const;
    543 
    544   unsigned getID() const { return ID; }
    545 
    546 protected:
    547   friend class MemorySSA;
    548 
    549   /// \brief this is more complicated than the generic
    550   /// User::allocHungoffUses, because we have to allocate Uses for the incoming
    551   /// values and pointers to the incoming blocks, all in one allocation.
    552   void allocHungoffUses(unsigned N) {
    553     User::allocHungoffUses(N, /* IsPhi */ true);
    554   }
    555 
    556 private:
    557   // For debugging only
    558   const unsigned ID;
    559   unsigned ReservedSpace;
    560 
    561   /// \brief This grows the operand list in response to a push_back style of
    562   /// operation.  This grows the number of ops by 1.5 times.
    563   void growOperands() {
    564     unsigned E = getNumOperands();
    565     // 2 op PHI nodes are VERY common, so reserve at least enough for that.
    566     ReservedSpace = std::max(E + E / 2, 2u);
    567     growHungoffUses(ReservedSpace, /* IsPhi */ true);
    568   }
    569 
    570   static void deleteMe(DerivedUser *Self);
    571 };
    572 
    573 inline unsigned MemoryAccess::getID() const {
    574   assert((isa<MemoryDef>(this) || isa<MemoryPhi>(this)) &&
    575          "only memory defs and phis have ids");
    576   if (const auto *MD = dyn_cast<MemoryDef>(this))
    577     return MD->getID();
    578   return cast<MemoryPhi>(this)->getID();
    579 }
    580 
    581 inline bool MemoryUseOrDef::isOptimized() const {
    582   if (const auto *MD = dyn_cast<MemoryDef>(this))
    583     return MD->isOptimized();
    584   return cast<MemoryUse>(this)->isOptimized();
    585 }
    586 
    587 inline MemoryAccess *MemoryUseOrDef::getOptimized() const {
    588   if (const auto *MD = dyn_cast<MemoryDef>(this))
    589     return MD->getOptimized();
    590   return cast<MemoryUse>(this)->getOptimized();
    591 }
    592 
    593 inline void MemoryUseOrDef::setOptimized(MemoryAccess *MA) {
    594   if (auto *MD = dyn_cast<MemoryDef>(this))
    595     MD->setOptimized(MA);
    596   else
    597     cast<MemoryUse>(this)->setOptimized(MA);
    598 }
    599 
    600 inline void MemoryUseOrDef::resetOptimized() {
    601   if (auto *MD = dyn_cast<MemoryDef>(this))
    602     MD->resetOptimized();
    603   else
    604     cast<MemoryUse>(this)->resetOptimized();
    605 }
    606 
    607 template <> struct OperandTraits<MemoryPhi> : public HungoffOperandTraits<2> {};
    608 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(MemoryPhi, MemoryAccess)
    609 
    610 /// \brief Encapsulates MemorySSA, including all data associated with memory
    611 /// accesses.
    612 class MemorySSA {
    613 public:
    614   MemorySSA(Function &, AliasAnalysis *, DominatorTree *);
    615   ~MemorySSA();
    616 
    617   MemorySSAWalker *getWalker();
    618 
    619   /// \brief Given a memory Mod/Ref'ing instruction, get the MemorySSA
    620   /// access associated with it. If passed a basic block gets the memory phi
    621   /// node that exists for that block, if there is one. Otherwise, this will get
    622   /// a MemoryUseOrDef.
    623   MemoryUseOrDef *getMemoryAccess(const Instruction *) const;
    624   MemoryPhi *getMemoryAccess(const BasicBlock *BB) const;
    625 
    626   void dump() const;
    627   void print(raw_ostream &) const;
    628 
    629   /// \brief Return true if \p MA represents the live on entry value
    630   ///
    631   /// Loads and stores from pointer arguments and other global values may be
    632   /// defined by memory operations that do not occur in the current function, so
    633   /// they may be live on entry to the function. MemorySSA represents such
    634   /// memory state by the live on entry definition, which is guaranteed to occur
    635   /// before any other memory access in the function.
    636   inline bool isLiveOnEntryDef(const MemoryAccess *MA) const {
    637     return MA == LiveOnEntryDef.get();
    638   }
    639 
    640   inline MemoryAccess *getLiveOnEntryDef() const {
    641     return LiveOnEntryDef.get();
    642   }
    643 
    644   // Sadly, iplists, by default, owns and deletes pointers added to the
    645   // list. It's not currently possible to have two iplists for the same type,
    646   // where one owns the pointers, and one does not. This is because the traits
    647   // are per-type, not per-tag.  If this ever changes, we should make the
    648   // DefList an iplist.
    649   using AccessList = iplist<MemoryAccess, ilist_tag<MSSAHelpers::AllAccessTag>>;
    650   using DefsList =
    651       simple_ilist<MemoryAccess, ilist_tag<MSSAHelpers::DefsOnlyTag>>;
    652 
    653   /// \brief Return the list of MemoryAccess's for a given basic block.
    654   ///
    655   /// This list is not modifiable by the user.
    656   const AccessList *getBlockAccesses(const BasicBlock *BB) const {
    657     return getWritableBlockAccesses(BB);
    658   }
    659 
    660   /// \brief Return the list of MemoryDef's and MemoryPhi's for a given basic
    661   /// block.
    662   ///
    663   /// This list is not modifiable by the user.
    664   const DefsList *getBlockDefs(const BasicBlock *BB) const {
    665     return getWritableBlockDefs(BB);
    666   }
    667 
    668   /// \brief Given two memory accesses in the same basic block, determine
    669   /// whether MemoryAccess \p A dominates MemoryAccess \p B.
    670   bool locallyDominates(const MemoryAccess *A, const MemoryAccess *B) const;
    671 
    672   /// \brief Given two memory accesses in potentially different blocks,
    673   /// determine whether MemoryAccess \p A dominates MemoryAccess \p B.
    674   bool dominates(const MemoryAccess *A, const MemoryAccess *B) const;
    675 
    676   /// \brief Given a MemoryAccess and a Use, determine whether MemoryAccess \p A
    677   /// dominates Use \p B.
    678   bool dominates(const MemoryAccess *A, const Use &B) const;
    679 
    680   /// \brief Verify that MemorySSA is self consistent (IE definitions dominate
    681   /// all uses, uses appear in the right places).  This is used by unit tests.
    682   void verifyMemorySSA() const;
    683 
    684   /// Used in various insertion functions to specify whether we are talking
    685   /// about the beginning or end of a block.
    686   enum InsertionPlace { Beginning, End };
    687 
    688 protected:
    689   // Used by Memory SSA annotater, dumpers, and wrapper pass
    690   friend class MemorySSAAnnotatedWriter;
    691   friend class MemorySSAPrinterLegacyPass;
    692   friend class MemorySSAUpdater;
    693 
    694   void verifyDefUses(Function &F) const;
    695   void verifyDomination(Function &F) const;
    696   void verifyOrdering(Function &F) const;
    697 
    698   // This is used by the use optimizer and updater.
    699   AccessList *getWritableBlockAccesses(const BasicBlock *BB) const {
    700     auto It = PerBlockAccesses.find(BB);
    701     return It == PerBlockAccesses.end() ? nullptr : It->second.get();
    702   }
    703 
    704   // This is used by the use optimizer and updater.
    705   DefsList *getWritableBlockDefs(const BasicBlock *BB) const {
    706     auto It = PerBlockDefs.find(BB);
    707     return It == PerBlockDefs.end() ? nullptr : It->second.get();
    708   }
    709 
    710   // These is used by the updater to perform various internal MemorySSA
    711   // machinsations.  They do not always leave the IR in a correct state, and
    712   // relies on the updater to fixup what it breaks, so it is not public.
    713 
    714   void moveTo(MemoryUseOrDef *What, BasicBlock *BB, AccessList::iterator Where);
    715   void moveTo(MemoryUseOrDef *What, BasicBlock *BB, InsertionPlace Point);
    716 
    717   // Rename the dominator tree branch rooted at BB.
    718   void renamePass(BasicBlock *BB, MemoryAccess *IncomingVal,
    719                   SmallPtrSetImpl<BasicBlock *> &Visited) {
    720     renamePass(DT->getNode(BB), IncomingVal, Visited, true, true);
    721   }
    722 
    723   void removeFromLookups(MemoryAccess *);
    724   void removeFromLists(MemoryAccess *, bool ShouldDelete = true);
    725   void insertIntoListsForBlock(MemoryAccess *, const BasicBlock *,
    726                                InsertionPlace);
    727   void insertIntoListsBefore(MemoryAccess *, const BasicBlock *,
    728                              AccessList::iterator);
    729   MemoryUseOrDef *createDefinedAccess(Instruction *, MemoryAccess *);
    730 
    731 private:
    732   class CachingWalker;
    733   class OptimizeUses;
    734 
    735   CachingWalker *getWalkerImpl();
    736   void buildMemorySSA();
    737   void optimizeUses();
    738 
    739   void verifyUseInDefs(MemoryAccess *, MemoryAccess *) const;
    740 
    741   using AccessMap = DenseMap<const BasicBlock *, std::unique_ptr<AccessList>>;
    742   using DefsMap = DenseMap<const BasicBlock *, std::unique_ptr<DefsList>>;
    743 
    744   void
    745   determineInsertionPoint(const SmallPtrSetImpl<BasicBlock *> &DefiningBlocks);
    746   void markUnreachableAsLiveOnEntry(BasicBlock *BB);
    747   bool dominatesUse(const MemoryAccess *, const MemoryAccess *) const;
    748   MemoryPhi *createMemoryPhi(BasicBlock *BB);
    749   MemoryUseOrDef *createNewAccess(Instruction *);
    750   MemoryAccess *findDominatingDef(BasicBlock *, enum InsertionPlace);
    751   void placePHINodes(const SmallPtrSetImpl<BasicBlock *> &,
    752                      const DenseMap<const BasicBlock *, unsigned int> &);
    753   MemoryAccess *renameBlock(BasicBlock *, MemoryAccess *, bool);
    754   void renameSuccessorPhis(BasicBlock *, MemoryAccess *, bool);
    755   void renamePass(DomTreeNode *, MemoryAccess *IncomingVal,
    756                   SmallPtrSetImpl<BasicBlock *> &Visited,
    757                   bool SkipVisited = false, bool RenameAllUses = false);
    758   AccessList *getOrCreateAccessList(const BasicBlock *);
    759   DefsList *getOrCreateDefsList(const BasicBlock *);
    760   void renumberBlock(const BasicBlock *) const;
    761   AliasAnalysis *AA;
    762   DominatorTree *DT;
    763   Function &F;
    764 
    765   // Memory SSA mappings
    766   DenseMap<const Value *, MemoryAccess *> ValueToMemoryAccess;
    767 
    768   // These two mappings contain the main block to access/def mappings for
    769   // MemorySSA. The list contained in PerBlockAccesses really owns all the
    770   // MemoryAccesses.
    771   // Both maps maintain the invariant that if a block is found in them, the
    772   // corresponding list is not empty, and if a block is not found in them, the
    773   // corresponding list is empty.
    774   AccessMap PerBlockAccesses;
    775   DefsMap PerBlockDefs;
    776   std::unique_ptr<MemoryAccess> LiveOnEntryDef;
    777 
    778   // Domination mappings
    779   // Note that the numbering is local to a block, even though the map is
    780   // global.
    781   mutable SmallPtrSet<const BasicBlock *, 16> BlockNumberingValid;
    782   mutable DenseMap<const MemoryAccess *, unsigned long> BlockNumbering;
    783 
    784   // Memory SSA building info
    785   std::unique_ptr<CachingWalker> Walker;
    786   unsigned NextID;
    787 };
    788 
    789 // Internal MemorySSA utils, for use by MemorySSA classes and walkers
    790 class MemorySSAUtil {
    791 protected:
    792   friend class GVNHoist;
    793   friend class MemorySSAWalker;
    794 
    795   // This function should not be used by new passes.
    796   static bool defClobbersUseOrDef(MemoryDef *MD, const MemoryUseOrDef *MU,
    797                                   AliasAnalysis &AA);
    798 };
    799 
    800 // This pass does eager building and then printing of MemorySSA. It is used by
    801 // the tests to be able to build, dump, and verify Memory SSA.
    802 class MemorySSAPrinterLegacyPass : public FunctionPass {
    803 public:
    804   MemorySSAPrinterLegacyPass();
    805 
    806   bool runOnFunction(Function &) override;
    807   void getAnalysisUsage(AnalysisUsage &AU) const override;
    808 
    809   static char ID;
    810 };
    811 
    812 /// An analysis that produces \c MemorySSA for a function.
    813 ///
    814 class MemorySSAAnalysis : public AnalysisInfoMixin<MemorySSAAnalysis> {
    815   friend AnalysisInfoMixin<MemorySSAAnalysis>;
    816 
    817   static AnalysisKey Key;
    818 
    819 public:
    820   // Wrap MemorySSA result to ensure address stability of internal MemorySSA
    821   // pointers after construction.  Use a wrapper class instead of plain
    822   // unique_ptr<MemorySSA> to avoid build breakage on MSVC.
    823   struct Result {
    824     Result(std::unique_ptr<MemorySSA> &&MSSA) : MSSA(std::move(MSSA)) {}
    825 
    826     MemorySSA &getMSSA() { return *MSSA.get(); }
    827 
    828     std::unique_ptr<MemorySSA> MSSA;
    829   };
    830 
    831   Result run(Function &F, FunctionAnalysisManager &AM);
    832 };
    833 
    834 /// \brief Printer pass for \c MemorySSA.
    835 class MemorySSAPrinterPass : public PassInfoMixin<MemorySSAPrinterPass> {
    836   raw_ostream &OS;
    837 
    838 public:
    839   explicit MemorySSAPrinterPass(raw_ostream &OS) : OS(OS) {}
    840 
    841   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
    842 };
    843 
    844 /// \brief Verifier pass for \c MemorySSA.
    845 struct MemorySSAVerifierPass : PassInfoMixin<MemorySSAVerifierPass> {
    846   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
    847 };
    848 
    849 /// \brief Legacy analysis pass which computes \c MemorySSA.
    850 class MemorySSAWrapperPass : public FunctionPass {
    851 public:
    852   MemorySSAWrapperPass();
    853 
    854   static char ID;
    855 
    856   bool runOnFunction(Function &) override;
    857   void releaseMemory() override;
    858   MemorySSA &getMSSA() { return *MSSA; }
    859   const MemorySSA &getMSSA() const { return *MSSA; }
    860 
    861   void getAnalysisUsage(AnalysisUsage &AU) const override;
    862 
    863   void verifyAnalysis() const override;
    864   void print(raw_ostream &OS, const Module *M = nullptr) const override;
    865 
    866 private:
    867   std::unique_ptr<MemorySSA> MSSA;
    868 };
    869 
    870 /// \brief This is the generic walker interface for walkers of MemorySSA.
    871 /// Walkers are used to be able to further disambiguate the def-use chains
    872 /// MemorySSA gives you, or otherwise produce better info than MemorySSA gives
    873 /// you.
    874 /// In particular, while the def-use chains provide basic information, and are
    875 /// guaranteed to give, for example, the nearest may-aliasing MemoryDef for a
    876 /// MemoryUse as AliasAnalysis considers it, a user mant want better or other
    877 /// information. In particular, they may want to use SCEV info to further
    878 /// disambiguate memory accesses, or they may want the nearest dominating
    879 /// may-aliasing MemoryDef for a call or a store. This API enables a
    880 /// standardized interface to getting and using that info.
    881 class MemorySSAWalker {
    882 public:
    883   MemorySSAWalker(MemorySSA *);
    884   virtual ~MemorySSAWalker() = default;
    885 
    886   using MemoryAccessSet = SmallVector<MemoryAccess *, 8>;
    887 
    888   /// \brief Given a memory Mod/Ref/ModRef'ing instruction, calling this
    889   /// will give you the nearest dominating MemoryAccess that Mod's the location
    890   /// the instruction accesses (by skipping any def which AA can prove does not
    891   /// alias the location(s) accessed by the instruction given).
    892   ///
    893   /// Note that this will return a single access, and it must dominate the
    894   /// Instruction, so if an operand of a MemoryPhi node Mod's the instruction,
    895   /// this will return the MemoryPhi, not the operand. This means that
    896   /// given:
    897   /// if (a) {
    898   ///   1 = MemoryDef(liveOnEntry)
    899   ///   store %a
    900   /// } else {
    901   ///   2 = MemoryDef(liveOnEntry)
    902   ///   store %b
    903   /// }
    904   /// 3 = MemoryPhi(2, 1)
    905   /// MemoryUse(3)
    906   /// load %a
    907   ///
    908   /// calling this API on load(%a) will return the MemoryPhi, not the MemoryDef
    909   /// in the if (a) branch.
    910   MemoryAccess *getClobberingMemoryAccess(const Instruction *I) {
    911     MemoryAccess *MA = MSSA->getMemoryAccess(I);
    912     assert(MA && "Handed an instruction that MemorySSA doesn't recognize?");
    913     return getClobberingMemoryAccess(MA);
    914   }
    915 
    916   /// Does the same thing as getClobberingMemoryAccess(const Instruction *I),
    917   /// but takes a MemoryAccess instead of an Instruction.
    918   virtual MemoryAccess *getClobberingMemoryAccess(MemoryAccess *) = 0;
    919 
    920   /// \brief Given a potentially clobbering memory access and a new location,
    921   /// calling this will give you the nearest dominating clobbering MemoryAccess
    922   /// (by skipping non-aliasing def links).
    923   ///
    924   /// This version of the function is mainly used to disambiguate phi translated
    925   /// pointers, where the value of a pointer may have changed from the initial
    926   /// memory access. Note that this expects to be handed either a MemoryUse,
    927   /// or an already potentially clobbering access. Unlike the above API, if
    928   /// given a MemoryDef that clobbers the pointer as the starting access, it
    929   /// will return that MemoryDef, whereas the above would return the clobber
    930   /// starting from the use side of  the memory def.
    931   virtual MemoryAccess *getClobberingMemoryAccess(MemoryAccess *,
    932                                                   const MemoryLocation &) = 0;
    933 
    934   /// \brief Given a memory access, invalidate anything this walker knows about
    935   /// that access.
    936   /// This API is used by walkers that store information to perform basic cache
    937   /// invalidation.  This will be called by MemorySSA at appropriate times for
    938   /// the walker it uses or returns.
    939   virtual void invalidateInfo(MemoryAccess *) {}
    940 
    941   virtual void verify(const MemorySSA *MSSA) { assert(MSSA == this->MSSA); }
    942 
    943 protected:
    944   friend class MemorySSA; // For updating MSSA pointer in MemorySSA move
    945                           // constructor.
    946   MemorySSA *MSSA;
    947 };
    948 
    949 /// \brief A MemorySSAWalker that does no alias queries, or anything else. It
    950 /// simply returns the links as they were constructed by the builder.
    951 class DoNothingMemorySSAWalker final : public MemorySSAWalker {
    952 public:
    953   // Keep the overrides below from hiding the Instruction overload of
    954   // getClobberingMemoryAccess.
    955   using MemorySSAWalker::getClobberingMemoryAccess;
    956 
    957   MemoryAccess *getClobberingMemoryAccess(MemoryAccess *) override;
    958   MemoryAccess *getClobberingMemoryAccess(MemoryAccess *,
    959                                           const MemoryLocation &) override;
    960 };
    961 
    962 using MemoryAccessPair = std::pair<MemoryAccess *, MemoryLocation>;
    963 using ConstMemoryAccessPair = std::pair<const MemoryAccess *, MemoryLocation>;
    964 
    965 /// \brief Iterator base class used to implement const and non-const iterators
    966 /// over the defining accesses of a MemoryAccess.
    967 template <class T>
    968 class memoryaccess_def_iterator_base
    969     : public iterator_facade_base<memoryaccess_def_iterator_base<T>,
    970                                   std::forward_iterator_tag, T, ptrdiff_t, T *,
    971                                   T *> {
    972   using BaseT = typename memoryaccess_def_iterator_base::iterator_facade_base;
    973 
    974 public:
    975   memoryaccess_def_iterator_base(T *Start) : Access(Start) {}
    976   memoryaccess_def_iterator_base() = default;
    977 
    978   bool operator==(const memoryaccess_def_iterator_base &Other) const {
    979     return Access == Other.Access && (!Access || ArgNo == Other.ArgNo);
    980   }
    981 
    982   // This is a bit ugly, but for MemoryPHI's, unlike PHINodes, you can't get the
    983   // block from the operand in constant time (In a PHINode, the uselist has
    984   // both, so it's just subtraction). We provide it as part of the
    985   // iterator to avoid callers having to linear walk to get the block.
    986   // If the operation becomes constant time on MemoryPHI's, this bit of
    987   // abstraction breaking should be removed.
    988   BasicBlock *getPhiArgBlock() const {
    989     MemoryPhi *MP = dyn_cast<MemoryPhi>(Access);
    990     assert(MP && "Tried to get phi arg block when not iterating over a PHI");
    991     return MP->getIncomingBlock(ArgNo);
    992   }
    993 
    994   typename BaseT::iterator::pointer operator*() const {
    995     assert(Access && "Tried to access past the end of our iterator");
    996     // Go to the first argument for phis, and the defining access for everything
    997     // else.
    998     if (MemoryPhi *MP = dyn_cast<MemoryPhi>(Access))
    999       return MP->getIncomingValue(ArgNo);
   1000     return cast<MemoryUseOrDef>(Access)->getDefiningAccess();
   1001   }
   1002 
   1003   using BaseT::operator++;
   1004   memoryaccess_def_iterator &operator++() {
   1005     assert(Access && "Hit end of iterator");
   1006     if (MemoryPhi *MP = dyn_cast<MemoryPhi>(Access)) {
   1007       if (++ArgNo >= MP->getNumIncomingValues()) {
   1008         ArgNo = 0;
   1009         Access = nullptr;
   1010       }
   1011     } else {
   1012       Access = nullptr;
   1013     }
   1014     return *this;
   1015   }
   1016 
   1017 private:
   1018   T *Access = nullptr;
   1019   unsigned ArgNo = 0;
   1020 };
   1021 
   1022 inline memoryaccess_def_iterator MemoryAccess::defs_begin() {
   1023   return memoryaccess_def_iterator(this);
   1024 }
   1025 
   1026 inline const_memoryaccess_def_iterator MemoryAccess::defs_begin() const {
   1027   return const_memoryaccess_def_iterator(this);
   1028 }
   1029 
   1030 inline memoryaccess_def_iterator MemoryAccess::defs_end() {
   1031   return memoryaccess_def_iterator();
   1032 }
   1033 
   1034 inline const_memoryaccess_def_iterator MemoryAccess::defs_end() const {
   1035   return const_memoryaccess_def_iterator();
   1036 }
   1037 
   1038 /// \brief GraphTraits for a MemoryAccess, which walks defs in the normal case,
   1039 /// and uses in the inverse case.
   1040 template <> struct GraphTraits<MemoryAccess *> {
   1041   using NodeRef = MemoryAccess *;
   1042   using ChildIteratorType = memoryaccess_def_iterator;
   1043 
   1044   static NodeRef getEntryNode(NodeRef N) { return N; }
   1045   static ChildIteratorType child_begin(NodeRef N) { return N->defs_begin(); }
   1046   static ChildIteratorType child_end(NodeRef N) { return N->defs_end(); }
   1047 };
   1048 
   1049 template <> struct GraphTraits<Inverse<MemoryAccess *>> {
   1050   using NodeRef = MemoryAccess *;
   1051   using ChildIteratorType = MemoryAccess::iterator;
   1052 
   1053   static NodeRef getEntryNode(NodeRef N) { return N; }
   1054   static ChildIteratorType child_begin(NodeRef N) { return N->user_begin(); }
   1055   static ChildIteratorType child_end(NodeRef N) { return N->user_end(); }
   1056 };
   1057 
   1058 /// \brief Provide an iterator that walks defs, giving both the memory access,
   1059 /// and the current pointer location, updating the pointer location as it
   1060 /// changes due to phi node translation.
   1061 ///
   1062 /// This iterator, while somewhat specialized, is what most clients actually
   1063 /// want when walking upwards through MemorySSA def chains. It takes a pair of
   1064 /// <MemoryAccess,MemoryLocation>, and walks defs, properly translating the
   1065 /// memory location through phi nodes for the user.
   1066 class upward_defs_iterator
   1067     : public iterator_facade_base<upward_defs_iterator,
   1068                                   std::forward_iterator_tag,
   1069                                   const MemoryAccessPair> {
   1070   using BaseT = upward_defs_iterator::iterator_facade_base;
   1071 
   1072 public:
   1073   upward_defs_iterator(const MemoryAccessPair &Info)
   1074       : DefIterator(Info.first), Location(Info.second),
   1075         OriginalAccess(Info.first) {
   1076     CurrentPair.first = nullptr;
   1077 
   1078     WalkingPhi = Info.first && isa<MemoryPhi>(Info.first);
   1079     fillInCurrentPair();
   1080   }
   1081 
   1082   upward_defs_iterator() { CurrentPair.first = nullptr; }
   1083 
   1084   bool operator==(const upward_defs_iterator &Other) const {
   1085     return DefIterator == Other.DefIterator;
   1086   }
   1087 
   1088   BaseT::iterator::reference operator*() const {
   1089     assert(DefIterator != OriginalAccess->defs_end() &&
   1090            "Tried to access past the end of our iterator");
   1091     return CurrentPair;
   1092   }
   1093 
   1094   using BaseT::operator++;
   1095   upward_defs_iterator &operator++() {
   1096     assert(DefIterator != OriginalAccess->defs_end() &&
   1097            "Tried to access past the end of the iterator");
   1098     ++DefIterator;
   1099     if (DefIterator != OriginalAccess->defs_end())
   1100       fillInCurrentPair();
   1101     return *this;
   1102   }
   1103 
   1104   BasicBlock *getPhiArgBlock() const { return DefIterator.getPhiArgBlock(); }
   1105 
   1106 private:
   1107   void fillInCurrentPair() {
   1108     CurrentPair.first = *DefIterator;
   1109     if (WalkingPhi && Location.Ptr) {
   1110       PHITransAddr Translator(
   1111           const_cast<Value *>(Location.Ptr),
   1112           OriginalAccess->getBlock()->getModule()->getDataLayout(), nullptr);
   1113       if (!Translator.PHITranslateValue(OriginalAccess->getBlock(),
   1114                                         DefIterator.getPhiArgBlock(), nullptr,
   1115                                         false))
   1116         if (Translator.getAddr() != Location.Ptr) {
   1117           CurrentPair.second = Location.getWithNewPtr(Translator.getAddr());
   1118           return;
   1119         }
   1120     }
   1121     CurrentPair.second = Location;
   1122   }
   1123 
   1124   MemoryAccessPair CurrentPair;
   1125   memoryaccess_def_iterator DefIterator;
   1126   MemoryLocation Location;
   1127   MemoryAccess *OriginalAccess = nullptr;
   1128   bool WalkingPhi = false;
   1129 };
   1130 
   1131 inline upward_defs_iterator upward_defs_begin(const MemoryAccessPair &Pair) {
   1132   return upward_defs_iterator(Pair);
   1133 }
   1134 
   1135 inline upward_defs_iterator upward_defs_end() { return upward_defs_iterator(); }
   1136 
   1137 inline iterator_range<upward_defs_iterator>
   1138 upward_defs(const MemoryAccessPair &Pair) {
   1139   return make_range(upward_defs_begin(Pair), upward_defs_end());
   1140 }
   1141 
   1142 /// Walks the defining accesses of MemoryDefs. Stops after we hit something that
   1143 /// has no defining use (e.g. a MemoryPhi or liveOnEntry). Note that, when
   1144 /// comparing against a null def_chain_iterator, this will compare equal only
   1145 /// after walking said Phi/liveOnEntry.
   1146 ///
   1147 /// The UseOptimizedChain flag specifies whether to walk the clobbering
   1148 /// access chain, or all the accesses.
   1149 ///
   1150 /// Normally, MemoryDef are all just def/use linked together, so a def_chain on
   1151 /// a MemoryDef will walk all MemoryDefs above it in the program until it hits
   1152 /// a phi node.  The optimized chain walks the clobbering access of a store.
   1153 /// So if you are just trying to find, given a store, what the next
   1154 /// thing that would clobber the same memory is, you want the optimized chain.
   1155 template <class T, bool UseOptimizedChain = false>
   1156 struct def_chain_iterator
   1157     : public iterator_facade_base<def_chain_iterator<T, UseOptimizedChain>,
   1158                                   std::forward_iterator_tag, MemoryAccess *> {
   1159   def_chain_iterator() : MA(nullptr) {}
   1160   def_chain_iterator(T MA) : MA(MA) {}
   1161 
   1162   T operator*() const { return MA; }
   1163 
   1164   def_chain_iterator &operator++() {
   1165     // N.B. liveOnEntry has a null defining access.
   1166     if (auto *MUD = dyn_cast<MemoryUseOrDef>(MA)) {
   1167       if (UseOptimizedChain && MUD->isOptimized())
   1168         MA = MUD->getOptimized();
   1169       else
   1170         MA = MUD->getDefiningAccess();
   1171     } else {
   1172       MA = nullptr;
   1173     }
   1174 
   1175     return *this;
   1176   }
   1177 
   1178   bool operator==(const def_chain_iterator &O) const { return MA == O.MA; }
   1179 
   1180 private:
   1181   T MA;
   1182 };
   1183 
   1184 template <class T>
   1185 inline iterator_range<def_chain_iterator<T>>
   1186 def_chain(T MA, MemoryAccess *UpTo = nullptr) {
   1187 #ifdef EXPENSIVE_CHECKS
   1188   assert((!UpTo || find(def_chain(MA), UpTo) != def_chain_iterator<T>()) &&
   1189          "UpTo isn't in the def chain!");
   1190 #endif
   1191   return make_range(def_chain_iterator<T>(MA), def_chain_iterator<T>(UpTo));
   1192 }
   1193 
   1194 template <class T>
   1195 inline iterator_range<def_chain_iterator<T, true>> optimized_def_chain(T MA) {
   1196   return make_range(def_chain_iterator<T, true>(MA),
   1197                     def_chain_iterator<T, true>(nullptr));
   1198 }
   1199 
   1200 } // end namespace llvm
   1201 
   1202 #endif // LLVM_ANALYSIS_MEMORYSSA_H
   1203