GVN.cpp

//===- GVN.cpp - Eliminate redundant values and loads ---------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This pass performs global value numbering to eliminate fully redundant
// instructions.  It also performs simple dead load elimination.
//
// Note that this pass does the value numbering itself; it does not use the
// ValueNumbering analysis passes.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "gvn"
#include "llvm/Transforms/Scalar.h"
#include "llvm/GlobalVariable.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/LLVMContext.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/Loads.h"
#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
#include "llvm/Analysis/PHITransAddr.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Assembly/Writer.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/SSAUpdater.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/DepthFirstIterator.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Support/Allocator.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/IRBuilder.h"
#include "llvm/Support/PatternMatch.h"
using namespace llvm;
using namespace PatternMatch;

STATISTIC(NumGVNInstr,  "Number of instructions deleted");
STATISTIC(NumGVNLoad,   "Number of loads deleted");
STATISTIC(NumGVNPRE,    "Number of instructions PRE'd");
STATISTIC(NumGVNBlocks, "Number of blocks merged");
STATISTIC(NumGVNSimpl,  "Number of instructions simplified");
STATISTIC(NumGVNEqProp, "Number of equalities propagated");
STATISTIC(NumPRELoad,   "Number of loads PRE'd");

static cl::opt<bool> EnablePRE("enable-pre",
                               cl::init(true), cl::Hidden);
static cl::opt<bool> EnableLoadPRE("enable-load-pre", cl::init(true));

//===----------------------------------------------------------------------===//
//                         ValueTable Class
//===----------------------------------------------------------------------===//

/// This class holds the mapping between values and value numbers.  It is used
/// as an efficient mechanism to determine the expression-wise equivalence of
/// two values.
namespace {
  struct Expression {
    uint32_t opcode;
    Type *type;
    SmallVector<uint32_t, 4> varargs;

    Expression(uint32_t o = ~2U) : opcode(o) { }

    bool operator==(const Expression &other) const {
      if (opcode != other.opcode)
        return false;
      if (opcode == ~0U || opcode == ~1U)
        return true;
      if (type != other.type)
        return false;
      if (varargs != other.varargs)
        return false;
      return true;
    }
  };

  class ValueTable {
    DenseMap<Value*, uint32_t> valueNumbering;
    DenseMap<Expression, uint32_t> expressionNumbering;
    AliasAnalysis *AA;
    MemoryDependenceAnalysis *MD;
    DominatorTree *DT;

    uint32_t nextValueNumber;

    Expression create_expression(Instruction* I);
    Expression create_extractvalue_expression(ExtractValueInst* EI);
    uint32_t lookup_or_add_call(CallInst* C);
  public:
    ValueTable() : nextValueNumber(1) { }
    uint32_t lookup_or_add(Value *V);
    uint32_t lookup(Value *V) const;
    void add(Value *V, uint32_t num);
    void clear();
    void erase(Value *v);
    void setAliasAnalysis(AliasAnalysis* A) { AA = A; }
    AliasAnalysis *getAliasAnalysis() const { return AA; }
    void setMemDep(MemoryDependenceAnalysis* M) { MD = M; }
    void setDomTree(DominatorTree* D) { DT = D; }
    uint32_t getNextUnusedValueNumber() { return nextValueNumber; }
    void verifyRemoved(const Value *) const;
  };
}

namespace llvm {
template <> struct DenseMapInfo<Expression> {
  static inline Expression getEmptyKey() {
    return ~0U;
  }

  static inline Expression getTombstoneKey() {
    return ~1U;
  }

  static unsigned getHashValue(const Expression e) {
    unsigned hash = e.opcode;

    hash = ((unsigned)((uintptr_t)e.type >> 4) ^
            (unsigned)((uintptr_t)e.type >> 9));

    for (SmallVector<uint32_t, 4>::const_iterator I = e.varargs.begin(),
         E = e.varargs.end(); I != E; ++I)
      hash = *I + hash * 37;
    
    return hash;
  }
  static bool isEqual(const Expression &LHS, const Expression &RHS) {
    return LHS == RHS;
  }
};

}

//===----------------------------------------------------------------------===//
//                     ValueTable Internal Functions
//===----------------------------------------------------------------------===//

Expression ValueTable::create_expression(Instruction *I) {
  Expression e;
  e.type = I->getType();
  e.opcode = I->getOpcode();
  for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
       OI != OE; ++OI)
    e.varargs.push_back(lookup_or_add(*OI));
  
  if (CmpInst *C = dyn_cast<CmpInst>(I)) {
    e.opcode = (C->getOpcode() << 8) | C->getPredicate();
  } else if (InsertValueInst *E = dyn_cast<InsertValueInst>(I)) {
    for (InsertValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end();
         II != IE; ++II)
      e.varargs.push_back(*II);
  }
  
  return e;
}

Expression ValueTable::create_extractvalue_expression(ExtractValueInst *EI) {
  assert(EI != 0 && "Not an ExtractValueInst?");
  Expression e;
  e.type = EI->getType();
  e.opcode = 0;

  IntrinsicInst *I = dyn_cast<IntrinsicInst>(EI->getAggregateOperand());
  if (I != 0 && EI->getNumIndices() == 1 && *EI->idx_begin() == 0 ) {
    // EI might be an extract from one of our recognised intrinsics. If it
    // is we'll synthesize a semantically equivalent expression instead on
    // an extract value expression.
    switch (I->getIntrinsicID()) {
      case Intrinsic::sadd_with_overflow:
      case Intrinsic::uadd_with_overflow:
        e.opcode = Instruction::Add;
        break;
      case Intrinsic::ssub_with_overflow:
      case Intrinsic::usub_with_overflow:
        e.opcode = Instruction::Sub;
        break;
      case Intrinsic::smul_with_overflow:
      case Intrinsic::umul_with_overflow:
        e.opcode = Instruction::Mul;
        break;
      default:
        break;
    }

    if (e.opcode != 0) {
      // Intrinsic recognized. Grab its args to finish building the expression.
      assert(I->getNumArgOperands() == 2 &&
             "Expect two args for recognised intrinsics.");
      e.varargs.push_back(lookup_or_add(I->getArgOperand(0)));
      e.varargs.push_back(lookup_or_add(I->getArgOperand(1)));
      return e;
    }
  }

  // Not a recognised intrinsic. Fall back to producing an extract value
  // expression.
  e.opcode = EI->getOpcode();
  for (Instruction::op_iterator OI = EI->op_begin(), OE = EI->op_end();
       OI != OE; ++OI)
    e.varargs.push_back(lookup_or_add(*OI));

  for (ExtractValueInst::idx_iterator II = EI->idx_begin(), IE = EI->idx_end();
         II != IE; ++II)
    e.varargs.push_back(*II);

  return e;
}

//===----------------------------------------------------------------------===//
//                     ValueTable External Functions
//===----------------------------------------------------------------------===//

/// add - Insert a value into the table with a specified value number.
void ValueTable::add(Value *V, uint32_t num) {
  valueNumbering.insert(std::make_pair(V, num));
}

uint32_t ValueTable::lookup_or_add_call(CallInst* C) {
  if (AA->doesNotAccessMemory(C)) {
    Expression exp = create_expression(C);
    uint32_t& e = expressionNumbering[exp];
    if (!e) e = nextValueNumber++;
    valueNumbering[C] = e;
    return e;
  } else if (AA->onlyReadsMemory(C)) {
    Expression exp = create_expression(C);
    uint32_t& e = expressionNumbering[exp];
    if (!e) {
      e = nextValueNumber++;
      valueNumbering[C] = e;
      return e;
    }
    if (!MD) {
      e = nextValueNumber++;
      valueNumbering[C] = e;
      return e;
    }

    MemDepResult local_dep = MD->getDependency(C);

    if (!local_dep.isDef() && !local_dep.isNonLocal()) {
      valueNumbering[C] =  nextValueNumber;
      return nextValueNumber++;
    }

    if (local_dep.isDef()) {
      CallInst* local_cdep = cast<CallInst>(local_dep.getInst());

      if (local_cdep->getNumArgOperands() != C->getNumArgOperands()) {
        valueNumbering[C] = nextValueNumber;
        return nextValueNumber++;
      }

      for (unsigned i = 0, e = C->getNumArgOperands(); i < e; ++i) {
        uint32_t c_vn = lookup_or_add(C->getArgOperand(i));
        uint32_t cd_vn = lookup_or_add(local_cdep->getArgOperand(i));
        if (c_vn != cd_vn) {
          valueNumbering[C] = nextValueNumber;
          return nextValueNumber++;
        }
      }

      uint32_t v = lookup_or_add(local_cdep);
      valueNumbering[C] = v;
      return v;
    }

    // Non-local case.
    const MemoryDependenceAnalysis::NonLocalDepInfo &deps =
      MD->getNonLocalCallDependency(CallSite(C));
    // FIXME: Move the checking logic to MemDep!
    CallInst* cdep = 0;

    // Check to see if we have a single dominating call instruction that is
    // identical to C.
    for (unsigned i = 0, e = deps.size(); i != e; ++i) {
      const NonLocalDepEntry *I = &deps[i];
      if (I->getResult().isNonLocal())
        continue;

      // We don't handle non-definitions.  If we already have a call, reject
      // instruction dependencies.
      if (!I->getResult().isDef() || cdep != 0) {
        cdep = 0;
        break;
      }

      CallInst *NonLocalDepCall = dyn_cast<CallInst>(I->getResult().getInst());
      // FIXME: All duplicated with non-local case.
      if (NonLocalDepCall && DT->properlyDominates(I->getBB(), C->getParent())){
        cdep = NonLocalDepCall;
        continue;
      }

      cdep = 0;
      break;
    }

    if (!cdep) {
      valueNumbering[C] = nextValueNumber;
      return nextValueNumber++;
    }

    if (cdep->getNumArgOperands() != C->getNumArgOperands()) {
      valueNumbering[C] = nextValueNumber;
      return nextValueNumber++;
    }
    for (unsigned i = 0, e = C->getNumArgOperands(); i < e; ++i) {
      uint32_t c_vn = lookup_or_add(C->getArgOperand(i));
      uint32_t cd_vn = lookup_or_add(cdep->getArgOperand(i));
      if (c_vn != cd_vn) {
        valueNumbering[C] = nextValueNumber;
        return nextValueNumber++;
      }
    }

    uint32_t v = lookup_or_add(cdep);
    valueNumbering[C] = v;
    return v;

  } else {
    valueNumbering[C] = nextValueNumber;
    return nextValueNumber++;
  }
}

/// lookup_or_add - Returns the value number for the specified value, assigning
/// it a new number if it did not have one before.
uint32_t ValueTable::lookup_or_add(Value *V) {
  DenseMap<Value*, uint32_t>::iterator VI = valueNumbering.find(V);
  if (VI != valueNumbering.end())
    return VI->second;

  if (!isa<Instruction>(V)) {
    valueNumbering[V] = nextValueNumber;
    return nextValueNumber++;
  }
  
  Instruction* I = cast<Instruction>(V);
  Expression exp;
  switch (I->getOpcode()) {
    case Instruction::Call:
      return lookup_or_add_call(cast<CallInst>(I));
    case Instruction::Add:
    case Instruction::FAdd:
    case Instruction::Sub:
    case Instruction::FSub:
    case Instruction::Mul:
    case Instruction::FMul:
    case Instruction::UDiv:
    case Instruction::SDiv:
    case Instruction::FDiv:
    case Instruction::URem:
    case Instruction::SRem:
    case Instruction::FRem:
    case Instruction::Shl:
    case Instruction::LShr:
    case Instruction::AShr:
    case Instruction::And:
    case Instruction::Or :
    case Instruction::Xor:
    case Instruction::ICmp:
    case Instruction::FCmp:
    case Instruction::Trunc:
    case Instruction::ZExt:
    case Instruction::SExt:
    case Instruction::FPToUI:
    case Instruction::FPToSI:
    case Instruction::UIToFP:
    case Instruction::SIToFP:
    case Instruction::FPTrunc:
    case Instruction::FPExt:
    case Instruction::PtrToInt:
    case Instruction::IntToPtr:
    case Instruction::BitCast:
    case Instruction::Select:
    case Instruction::ExtractElement:
    case Instruction::InsertElement:
    case Instruction::ShuffleVector:
    case Instruction::InsertValue:
    case Instruction::GetElementPtr:
      exp = create_expression(I);
      break;
    case Instruction::ExtractValue:
      exp = create_extractvalue_expression(cast<ExtractValueInst>(I));
      break;
    default:
      valueNumbering[V] = nextValueNumber;
      return nextValueNumber++;
  }

  uint32_t& e = expressionNumbering[exp];
  if (!e) e = nextValueNumber++;
  valueNumbering[V] = e;
  return e;
}

/// lookup - Returns the value number of the specified value. Fails if
/// the value has not yet been numbered.
uint32_t ValueTable::lookup(Value *V) const {
  DenseMap<Value*, uint32_t>::const_iterator VI = valueNumbering.find(V);
  assert(VI != valueNumbering.end() && "Value not numbered?");
  return VI->second;
}

/// clear - Remove all entries from the ValueTable.
void ValueTable::clear() {
  valueNumbering.clear();
  expressionNumbering.clear();
  nextValueNumber = 1;
}

/// erase - Remove a value from the value numbering.
void ValueTable::erase(Value *V) {
  valueNumbering.erase(V);
}

/// verifyRemoved - Verify that the value is removed from all internal data
/// structures.
void ValueTable::verifyRemoved(const Value *V) const {
  for (DenseMap<Value*, uint32_t>::const_iterator
         I = valueNumbering.begin(), E = valueNumbering.end(); I != E; ++I) {
    assert(I->first != V && "Inst still occurs in value numbering map!");
  }
}

//===----------------------------------------------------------------------===//
//                                GVN Pass
//===----------------------------------------------------------------------===//

namespace {

  class GVN : public FunctionPass {
    bool NoLoads;
    MemoryDependenceAnalysis *MD;
    DominatorTree *DT;
    const TargetData *TD;
    const TargetLibraryInfo *TLI;

    ValueTable VN;
    
    /// LeaderTable - A mapping from value numbers to lists of Value*'s that
    /// have that value number.  Use findLeader to query it.
    struct LeaderTableEntry {
      Value *Val;
      BasicBlock *BB;
      LeaderTableEntry *Next;
    };
    DenseMap<uint32_t, LeaderTableEntry> LeaderTable;
    BumpPtrAllocator TableAllocator;
    
    SmallVector<Instruction*, 8> InstrsToErase;
  public:
    static char ID; // Pass identification, replacement for typeid
    explicit GVN(bool noloads = false)
        : FunctionPass(ID), NoLoads(noloads), MD(0) {
      initializeGVNPass(*PassRegistry::getPassRegistry());
    }

    bool runOnFunction(Function &F);
    
    /// markInstructionForDeletion - This removes the specified instruction from
    /// our various maps and marks it for deletion.
    void markInstructionForDeletion(Instruction *I) {
      VN.erase(I);
      InstrsToErase.push_back(I);
    }
    
    const TargetData *getTargetData() const { return TD; }
    DominatorTree &getDominatorTree() const { return *DT; }
    AliasAnalysis *getAliasAnalysis() const { return VN.getAliasAnalysis(); }
    MemoryDependenceAnalysis &getMemDep() const { return *MD; }
  private:
    /// addToLeaderTable - Push a new Value to the LeaderTable onto the list for
    /// its value number.
    void addToLeaderTable(uint32_t N, Value *V, BasicBlock *BB) {
      LeaderTableEntry &Curr = LeaderTable[N];
      if (!Curr.Val) {
        Curr.Val = V;
        Curr.BB = BB;
        return;
      }
      
      LeaderTableEntry *Node = TableAllocator.Allocate<LeaderTableEntry>();
      Node->Val = V;
      Node->BB = BB;
      Node->Next = Curr.Next;
      Curr.Next = Node;
    }
    
    /// removeFromLeaderTable - Scan the list of values corresponding to a given
    /// value number, and remove the given value if encountered.
    void removeFromLeaderTable(uint32_t N, Value *V, BasicBlock *BB) {
      LeaderTableEntry* Prev = 0;
      LeaderTableEntry* Curr = &LeaderTable[N];

      while (Curr->Val != V || Curr->BB != BB) {
        Prev = Curr;
        Curr = Curr->Next;
      }
      
      if (Prev) {
        Prev->Next = Curr->Next;
      } else {
        if (!Curr->Next) {
          Curr->Val = 0;
          Curr->BB = 0;
        } else {
          LeaderTableEntry* Next = Curr->Next;
          Curr->Val = Next->Val;
          Curr->BB = Next->BB;
          Curr->Next = Next->Next;
        }
      }
    }

    // List of critical edges to be split between iterations.
    SmallVector<std::pair<TerminatorInst*, unsigned>, 4> toSplit;

    // This transformation requires dominator postdominator info
    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
      AU.addRequired<DominatorTree>();
      AU.addRequired<TargetLibraryInfo>();
      if (!NoLoads)
        AU.addRequired<MemoryDependenceAnalysis>();
      AU.addRequired<AliasAnalysis>();

      AU.addPreserved<DominatorTree>();
      AU.addPreserved<AliasAnalysis>();
    }
    

    // Helper fuctions
    // FIXME: eliminate or document these better
    bool processLoad(LoadInst *L);
    bool processInstruction(Instruction *I);
    bool processNonLocalLoad(LoadInst *L);
    bool processBlock(BasicBlock *BB);
    void dump(DenseMap<uint32_t, Value*> &d);
    bool iterateOnFunction(Function &F);
    bool performPRE(Function &F);
    Value *findLeader(BasicBlock *BB, uint32_t num);
    void cleanupGlobalSets();
    void verifyRemoved(const Instruction *I) const;
    bool splitCriticalEdges();
    unsigned replaceAllDominatedUsesWith(Value *From, Value *To,
                                         BasicBlock *Root);
    bool propagateEquality(Value *LHS, Value *RHS, BasicBlock *Root);
  };

  char GVN::ID = 0;
}

// createGVNPass - The public interface to this file...
FunctionPass *llvm::createGVNPass(bool NoLoads) {
  return new GVN(NoLoads);
}

INITIALIZE_PASS_BEGIN(GVN, "gvn", "Global Value Numbering", false, false)
INITIALIZE_PASS_DEPENDENCY(MemoryDependenceAnalysis)
INITIALIZE_PASS_DEPENDENCY(DominatorTree)
INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
INITIALIZE_PASS_END(GVN, "gvn", "Global Value Numbering", false, false)

void GVN::dump(DenseMap<uint32_t, Value*>& d) {
  errs() << "{\n";
  for (DenseMap<uint32_t, Value*>::iterator I = d.begin(),
       E = d.end(); I != E; ++I) {
      errs() << I->first << "\n";
      I->second->dump();
  }
  errs() << "}\n";
}

/// IsValueFullyAvailableInBlock - Return true if we can prove that the value
/// we're analyzing is fully available in the specified block.  As we go, keep
/// track of which blocks we know are fully alive in FullyAvailableBlocks.  This
/// map is actually a tri-state map with the following values:
///   0) we know the block *is not* fully available.
///   1) we know the block *is* fully available.
///   2) we do not know whether the block is fully available or not, but we are
///      currently speculating that it will be.
///   3) we are speculating for this block and have used that to speculate for
///      other blocks.
static bool IsValueFullyAvailableInBlock(BasicBlock *BB,
                            DenseMap<BasicBlock*, char> &FullyAvailableBlocks) {
  // Optimistically assume that the block is fully available and check to see
  // if we already know about this block in one lookup.
  std::pair<DenseMap<BasicBlock*, char>::iterator, char> IV =
    FullyAvailableBlocks.insert(std::make_pair(BB, 2));

  // If the entry already existed for this block, return the precomputed value.
  if (!IV.second) {
    // If this is a speculative "available" value, mark it as being used for
    // speculation of other blocks.
    if (IV.first->second == 2)
      IV.first->second = 3;
    return IV.first->second != 0;
  }

  // Otherwise, see if it is fully available in all predecessors.
  pred_iterator PI = pred_begin(BB), PE = pred_end(BB);

  // If this block has no predecessors, it isn't live-in here.
  if (PI == PE)
    goto SpeculationFailure;

  for (; PI != PE; ++PI)
    // If the value isn't fully available in one of our predecessors, then it
    // isn't fully available in this block either.  Undo our previous
    // optimistic assumption and bail out.
    if (!IsValueFullyAvailableInBlock(*PI, FullyAvailableBlocks))
      goto SpeculationFailure;

  return true;

// SpeculationFailure - If we get here, we found out that this is not, after
// all, a fully-available block.  We have a problem if we speculated on this and
// used the speculation to mark other blocks as available.
SpeculationFailure:
  char &BBVal = FullyAvailableBlocks[BB];

  // If we didn't speculate on this, just return with it set to false.
  if (BBVal == 2) {
    BBVal = 0;
    return false;
  }

  // If we did speculate on this value, we could have blocks set to 1 that are
  // incorrect.  Walk the (transitive) successors of this block and mark them as
  // 0 if set to one.
  SmallVector<BasicBlock*, 32> BBWorklist;
  BBWorklist.push_back(BB);

  do {
    BasicBlock *Entry = BBWorklist.pop_back_val();
    // Note that this sets blocks to 0 (unavailable) if they happen to not
    // already be in FullyAvailableBlocks.  This is safe.
    char &EntryVal = FullyAvailableBlocks[Entry];
    if (EntryVal == 0) continue;  // Already unavailable.

    // Mark as unavailable.
    EntryVal = 0;

    for (succ_iterator I = succ_begin(Entry), E = succ_end(Entry); I != E; ++I)
      BBWorklist.push_back(*I);
  } while (!BBWorklist.empty());

  return false;
}


/// CanCoerceMustAliasedValueToLoad - Return true if
/// CoerceAvailableValueToLoadType will succeed.
static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal,
                                            Type *LoadTy,
                                            const TargetData &TD) {
  // If the loaded or stored value is an first class array or struct, don't try
  // to transform them.  We need to be able to bitcast to integer.
  if (LoadTy->isStructTy() || LoadTy->isArrayTy() ||
      StoredVal->getType()->isStructTy() ||
      StoredVal->getType()->isArrayTy())
    return false;
  
  // The store has to be at least as big as the load.
  if (TD.getTypeSizeInBits(StoredVal->getType()) <
        TD.getTypeSizeInBits(LoadTy))
    return false;
  
  return true;
}
  

/// CoerceAvailableValueToLoadType - If we saw a store of a value to memory, and
/// then a load from a must-aliased pointer of a different type, try to coerce
/// the stored value.  LoadedTy is the type of the load we want to replace and
/// InsertPt is the place to insert new instructions.
///
/// If we can't do it, return null.
static Value *CoerceAvailableValueToLoadType(Value *StoredVal, 
                                             Type *LoadedTy,
                                             Instruction *InsertPt,
                                             const TargetData &TD) {
  if (!CanCoerceMustAliasedValueToLoad(StoredVal, LoadedTy, TD))
    return 0;
  
  // If this is already the right type, just return it.
  Type *StoredValTy = StoredVal->getType();
  
  uint64_t StoreSize = TD.getTypeSizeInBits(StoredValTy);
  uint64_t LoadSize = TD.getTypeSizeInBits(LoadedTy);
  
  // If the store and reload are the same size, we can always reuse it.
  if (StoreSize == LoadSize) {
    // Pointer to Pointer -> use bitcast.
    if (StoredValTy->isPointerTy() && LoadedTy->isPointerTy())
      return new BitCastInst(StoredVal, LoadedTy, "", InsertPt);
    
    // Convert source pointers to integers, which can be bitcast.
    if (StoredValTy->isPointerTy()) {
      StoredValTy = TD.getIntPtrType(StoredValTy->getContext());
      StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
    }
    
    Type *TypeToCastTo = LoadedTy;
    if (TypeToCastTo->isPointerTy())
      TypeToCastTo = TD.getIntPtrType(StoredValTy->getContext());
    
    if (StoredValTy != TypeToCastTo)
      StoredVal = new BitCastInst(StoredVal, TypeToCastTo, "", InsertPt);
    
    // Cast to pointer if the load needs a pointer type.
    if (LoadedTy->isPointerTy())
      StoredVal = new IntToPtrInst(StoredVal, LoadedTy, "", InsertPt);
    
    return StoredVal;
  }
  
  // If the loaded value is smaller than the available value, then we can
  // extract out a piece from it.  If the available value is too small, then we
  // can't do anything.
  assert(StoreSize >= LoadSize && "CanCoerceMustAliasedValueToLoad fail");
  
  // Convert source pointers to integers, which can be manipulated.
  if (StoredValTy->isPointerTy()) {
    StoredValTy = TD.getIntPtrType(StoredValTy->getContext());
    StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
  }
  
  // Convert vectors and fp to integer, which can be manipulated.
  if (!StoredValTy->isIntegerTy()) {
    StoredValTy = IntegerType::get(StoredValTy->getContext(), StoreSize);
    StoredVal = new BitCastInst(StoredVal, StoredValTy, "", InsertPt);
  }
  
  // If this is a big-endian system, we need to shift the value down to the low
  // bits so that a truncate will work.
  if (TD.isBigEndian()) {
    Constant *Val = ConstantInt::get(StoredVal->getType(), StoreSize-LoadSize);
    StoredVal = BinaryOperator::CreateLShr(StoredVal, Val, "tmp", InsertPt);
  }
  
  // Truncate the integer to the right size now.
  Type *NewIntTy = IntegerType::get(StoredValTy->getContext(), LoadSize);
  StoredVal = new TruncInst(StoredVal, NewIntTy, "trunc", InsertPt);
  
  if (LoadedTy == NewIntTy)
    return StoredVal;
  
  // If the result is a pointer, inttoptr.
  if (LoadedTy->isPointerTy())
    return new IntToPtrInst(StoredVal, LoadedTy, "inttoptr", InsertPt);
  
  // Otherwise, bitcast.
  return new BitCastInst(StoredVal, LoadedTy, "bitcast", InsertPt);
}

/// AnalyzeLoadFromClobberingWrite - This function is called when we have a
/// memdep query of a load that ends up being a clobbering memory write (store,
/// memset, memcpy, memmove).  This means that the write *may* provide bits used
/// by the load but we can't be sure because the pointers don't mustalias.
///
/// Check this case to see if there is anything more we can do before we give
/// up.  This returns -1 if we have to give up, or a byte number in the stored
/// value of the piece that feeds the load.
static int AnalyzeLoadFromClobberingWrite(Type *LoadTy, Value *LoadPtr,
                                          Value *WritePtr,
                                          uint64_t WriteSizeInBits,
                                          const TargetData &TD) {
  // If the loaded or stored value is a first class array or struct, don't try
  // to transform them.  We need to be able to bitcast to integer.
  if (LoadTy->isStructTy() || LoadTy->isArrayTy())
    return -1;
  
  int64_t StoreOffset = 0, LoadOffset = 0;
  Value *StoreBase = GetPointerBaseWithConstantOffset(WritePtr, StoreOffset,TD);
  Value *LoadBase = GetPointerBaseWithConstantOffset(LoadPtr, LoadOffset, TD);
  if (StoreBase != LoadBase)
    return -1;
  
  // If the load and store are to the exact same address, they should have been
  // a must alias.  AA must have gotten confused.
  // FIXME: Study to see if/when this happens.  One case is forwarding a memset
  // to a load from the base of the memset.
#if 0
  if (LoadOffset == StoreOffset) {
    dbgs() << "STORE/LOAD DEP WITH COMMON POINTER MISSED:\n"
    << "Base       = " << *StoreBase << "\n"
    << "Store Ptr  = " << *WritePtr << "\n"
    << "Store Offs = " << StoreOffset << "\n"
    << "Load Ptr   = " << *LoadPtr << "\n";
    abort();
  }
#endif
  
  // If the load and store don't overlap at all, the store doesn't provide
  // anything to the load.  In this case, they really don't alias at all, AA
  // must have gotten confused.
  uint64_t LoadSize = TD.getTypeSizeInBits(LoadTy);
  
  if ((WriteSizeInBits & 7) | (LoadSize & 7))
    return -1;
  uint64_t StoreSize = WriteSizeInBits >> 3;  // Convert to bytes.
  LoadSize >>= 3;
  
  
  bool isAAFailure = false;
  if (StoreOffset < LoadOffset)
    isAAFailure = StoreOffset+int64_t(StoreSize) <= LoadOffset;
  else
    isAAFailure = LoadOffset+int64_t(LoadSize) <= StoreOffset;

  if (isAAFailure) {
#if 0
    dbgs() << "STORE LOAD DEP WITH COMMON BASE:\n"
    << "Base       = " << *StoreBase << "\n"
    << "Store Ptr  = " << *WritePtr << "\n"
    << "Store Offs = " << StoreOffset << "\n"
    << "Load Ptr   = " << *LoadPtr << "\n";
    abort();
#endif
    return -1;
  }
  
  // If the Load isn't completely contained within the stored bits, we don't
  // have all the bits to feed it.  We could do something crazy in the future
  // (issue a smaller load then merge the bits in) but this seems unlikely to be
  // valuable.
  if (StoreOffset > LoadOffset ||
      StoreOffset+StoreSize < LoadOffset+LoadSize)
    return -1;
  
  // Okay, we can do this transformation.  Return the number of bytes into the
  // store that the load is.
  return LoadOffset-StoreOffset;
}  

/// AnalyzeLoadFromClobberingStore - This function is called when we have a
/// memdep query of a load that ends up being a clobbering store.
static int AnalyzeLoadFromClobberingStore(Type *LoadTy, Value *LoadPtr,
                                          StoreInst *DepSI,
                                          const TargetData &TD) {
  // Cannot handle reading from store of first-class aggregate yet.
  if (DepSI->getValueOperand()->getType()->isStructTy() ||
      DepSI->getValueOperand()->getType()->isArrayTy())
    return -1;

  Value *StorePtr = DepSI->getPointerOperand();
  uint64_t StoreSize =TD.getTypeSizeInBits(DepSI->getValueOperand()->getType());
  return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr,
                                        StorePtr, StoreSize, TD);
}

/// AnalyzeLoadFromClobberingLoad - This function is called when we have a
/// memdep query of a load that ends up being clobbered by another load.  See if
/// the other load can feed into the second load.
static int AnalyzeLoadFromClobberingLoad(Type *LoadTy, Value *LoadPtr,
                                         LoadInst *DepLI, const TargetData &TD){
  // Cannot handle reading from store of first-class aggregate yet.
  if (DepLI->getType()->isStructTy() || DepLI->getType()->isArrayTy())
    return -1;
  
  Value *DepPtr = DepLI->getPointerOperand();
  uint64_t DepSize = TD.getTypeSizeInBits(DepLI->getType());
  int R = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, DepSize, TD);
  if (R != -1) return R;
  
  // If we have a load/load clobber an DepLI can be widened to cover this load,
  // then we should widen it!
  int64_t LoadOffs = 0;
  const Value *LoadBase =
    GetPointerBaseWithConstantOffset(LoadPtr, LoadOffs, TD);
  unsigned LoadSize = TD.getTypeStoreSize(LoadTy);
  
  unsigned Size = MemoryDependenceAnalysis::
    getLoadLoadClobberFullWidthSize(LoadBase, LoadOffs, LoadSize, DepLI, TD);
  if (Size == 0) return -1;
  
  return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, Size*8, TD);
}



static int AnalyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
                                            MemIntrinsic *MI,
                                            const TargetData &TD) {
  // If the mem operation is a non-constant size, we can't handle it.
  ConstantInt *SizeCst = dyn_cast<ConstantInt>(MI->getLength());
  if (SizeCst == 0) return -1;
  uint64_t MemSizeInBits = SizeCst->getZExtValue()*8;

  // If this is memset, we just need to see if the offset is valid in the size
  // of the memset..
  if (MI->getIntrinsicID() == Intrinsic::memset)
    return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, MI->getDest(),
                                          MemSizeInBits, TD);
  
  // If we have a memcpy/memmove, the only case we can handle is if this is a
  // copy from constant memory.  In that case, we can read directly from the
  // constant memory.
  MemTransferInst *MTI = cast<MemTransferInst>(MI);
  
  Constant *Src = dyn_cast<Constant>(MTI->getSource());
  if (Src == 0) return -1;
  
  GlobalVariable *GV = dyn_cast<GlobalVariable>(GetUnderlyingObject(Src, &TD));
  if (GV == 0 || !GV->isConstant()) return -1;
  
  // See if the access is within the bounds of the transfer.
  int Offset = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr,
                                              MI->getDest(), MemSizeInBits, TD);
  if (Offset == -1)
    return Offset;
  
  // Otherwise, see if we can constant fold a load from the constant with the
  // offset applied as appropriate.
  Src = ConstantExpr::getBitCast(Src,
                                 llvm::Type::getInt8PtrTy(Src->getContext()));
  Constant *OffsetCst = 
    ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset);
  Src = ConstantExpr::getGetElementPtr(Src, OffsetCst);
  Src = ConstantExpr::getBitCast(Src, PointerType::getUnqual(LoadTy));
  if (ConstantFoldLoadFromConstPtr(Src, &TD))
    return Offset;
  return -1;
}
                                            

/// GetStoreValueForLoad - This function is called when we have a
/// memdep query of a load that ends up being a clobbering store.  This means
/// that the store provides bits used by the load but we the pointers don't
/// mustalias.  Check this case to see if there is anything more we can do
/// before we give up.
static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset,
                                   Type *LoadTy,
                                   Instruction *InsertPt, const TargetData &TD){
  LLVMContext &Ctx = SrcVal->getType()->getContext();
  
  uint64_t StoreSize = (TD.getTypeSizeInBits(SrcVal->getType()) + 7) / 8;
  uint64_t LoadSize = (TD.getTypeSizeInBits(LoadTy) + 7) / 8;
  
  IRBuilder<> Builder(InsertPt->getParent(), InsertPt);
  
  // Compute which bits of the stored value are being used by the load.  Convert
  // to an integer type to start with.
  if (SrcVal->getType()->isPointerTy())
    SrcVal = Builder.CreatePtrToInt(SrcVal, TD.getIntPtrType(Ctx));
  if (!SrcVal->getType()->isIntegerTy())
    SrcVal = Builder.CreateBitCast(SrcVal, IntegerType::get(Ctx, StoreSize*8));
  
  // Shift the bits to the least significant depending on endianness.
  unsigned ShiftAmt;
  if (TD.isLittleEndian())
    ShiftAmt = Offset*8;
  else
    ShiftAmt = (StoreSize-LoadSize-Offset)*8;
  
  if (ShiftAmt)
    SrcVal = Builder.CreateLShr(SrcVal, ShiftAmt);
  
  if (LoadSize != StoreSize)
    SrcVal = Builder.CreateTrunc(SrcVal, IntegerType::get(Ctx, LoadSize*8));
  
  return CoerceAvailableValueToLoadType(SrcVal, LoadTy, InsertPt, TD);
}

/// GetLoadValueForLoad - This function is called when we have a
/// memdep query of a load that ends up being a clobbering load.  This means
/// that the load *may* provide bits used by the load but we can't be sure
/// because the pointers don't mustalias.  Check this case to see if there is
/// anything more we can do before we give up.
static Value *GetLoadValueForLoad(LoadInst *SrcVal, unsigned Offset,
                                  Type *LoadTy, Instruction *InsertPt,
                                  GVN &gvn) {
  const TargetData &TD = *gvn.getTargetData();
  // If Offset+LoadTy exceeds the size of SrcVal, then we must be wanting to
  // widen SrcVal out to a larger load.
  unsigned SrcValSize = TD.getTypeStoreSize(SrcVal->getType());
  unsigned LoadSize = TD.getTypeStoreSize(LoadTy);
  if (Offset+LoadSize > SrcValSize) {
    assert(SrcVal->isSimple() && "Cannot widen volatile/atomic load!");
    assert(SrcVal->getType()->isIntegerTy() && "Can't widen non-integer load");
    // If we have a load/load clobber an DepLI can be widened to cover this
    // load, then we should widen it to the next power of 2 size big enough!
    unsigned NewLoadSize = Offset+LoadSize;
    if (!isPowerOf2_32(NewLoadSize))
      NewLoadSize = NextPowerOf2(NewLoadSize);