IndVarSimplify.cpp

    // No further widening is needed. The deceased [sz]ext had done it for us.
    return 0;
  }

  // Does this user itself evaluate to a recurrence after widening?
  const SCEVAddRecExpr *WideAddRec = GetWideRecurrence(NarrowUse);
  if (!WideAddRec) {
    // This user does not evaluate to a recurence after widening, so don't
    // follow it. Instead insert a Trunc to kill off the original use,
    // eventually isolating the original narrow IV so it can be removed.
    IRBuilder<> Builder(NarrowDefUse);
    Value *Trunc = Builder.CreateTrunc(WideDef, NarrowDef->getType());
    NarrowUse->replaceUsesOfWith(NarrowDef, Trunc);
    return 0;
  }
  // We assume that block terminators are not SCEVable. We wouldn't want to
  // insert a Trunc after a terminator if there happens to be a critical edge.
  assert(NarrowUse != NarrowUse->getParent()->getTerminator() &&
         "SCEV is not expected to evaluate a block terminator");

  // Reuse the IV increment that SCEVExpander created as long as it dominates
  // NarrowUse.
  Instruction *WideUse = 0;
  if (WideAddRec == WideIncExpr && HoistStep(WideInc, NarrowUse, DT)) {
    WideUse = WideInc;
  }
  else {
    WideUse = CloneIVUser(NarrowUse, NarrowDef, WideDef);
    if (!WideUse)
      return 0;
  }
  // Evaluation of WideAddRec ensured that the narrow expression could be
  // extended outside the loop without overflow. This suggests that the wide use
  // evaluates to the same expression as the extended narrow use, but doesn't
  // absolutely guarantee it. Hence the following failsafe check. In rare cases
  // where it fails, we simply throw away the newly created wide use.
  if (WideAddRec != SE->getSCEV(WideUse)) {
    DEBUG(dbgs() << "Wide use expression mismatch: " << *WideUse
          << ": " << *SE->getSCEV(WideUse) << " != " << *WideAddRec << "\n");
    DeadInsts.push_back(WideUse);
    return 0;
  }

  // Returning WideUse pushes it on the worklist.
  return WideUse;
}

/// pushNarrowIVUsers - Add eligible users of NarrowDef to NarrowIVUsers.
///
void WidenIV::pushNarrowIVUsers(Instruction *NarrowDef, Instruction *WideDef) {
  for (Value::use_iterator UI = NarrowDef->use_begin(),
         UE = NarrowDef->use_end(); UI != UE; ++UI) {
    Use &U = UI.getUse();

    // Handle data flow merges and bizarre phi cycles.
    if (!Widened.insert(cast<Instruction>(U.getUser())))
      continue;

    NarrowIVUsers.push_back(std::make_pair(&UI.getUse(), WideDef));
  }
}

/// CreateWideIV - Process a single induction variable. First use the
/// SCEVExpander to create a wide induction variable that evaluates to the same
/// recurrence as the original narrow IV. Then use a worklist to forward
/// traverse the narrow IV's def-use chain. After WidenIVUse has processed all
/// interesting IV users, the narrow IV will be isolated for removal by
/// DeleteDeadPHIs.
///
/// It would be simpler to delete uses as they are processed, but we must avoid
/// invalidating SCEV expressions.
///
PHINode *WidenIV::CreateWideIV(SCEVExpander &Rewriter) {
  // Is this phi an induction variable?
  const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(SE->getSCEV(OrigPhi));
  if (!AddRec)
    return NULL;

  // Widen the induction variable expression.
  const SCEV *WideIVExpr = IsSigned ?
    SE->getSignExtendExpr(AddRec, WideType) :
    SE->getZeroExtendExpr(AddRec, WideType);

  assert(SE->getEffectiveSCEVType(WideIVExpr->getType()) == WideType &&
         "Expect the new IV expression to preserve its type");

  // Can the IV be extended outside the loop without overflow?
  AddRec = dyn_cast<SCEVAddRecExpr>(WideIVExpr);
  if (!AddRec || AddRec->getLoop() != L)
    return NULL;

  // An AddRec must have loop-invariant operands. Since this AddRec is
  // materialized by a loop header phi, the expression cannot have any post-loop
  // operands, so they must dominate the loop header.
  assert(SE->properlyDominates(AddRec->getStart(), L->getHeader()) &&
         SE->properlyDominates(AddRec->getStepRecurrence(*SE), L->getHeader())
         && "Loop header phi recurrence inputs do not dominate the loop");

  // The rewriter provides a value for the desired IV expression. This may
  // either find an existing phi or materialize a new one. Either way, we
  // expect a well-formed cyclic phi-with-increments. i.e. any operand not part
  // of the phi-SCC dominates the loop entry.
  Instruction *InsertPt = L->getHeader()->begin();
  WidePhi = cast<PHINode>(Rewriter.expandCodeFor(AddRec, WideType, InsertPt));

  // Remembering the WideIV increment generated by SCEVExpander allows
  // WidenIVUse to reuse it when widening the narrow IV's increment. We don't
  // employ a general reuse mechanism because the call above is the only call to
  // SCEVExpander. Henceforth, we produce 1-to-1 narrow to wide uses.
  if (BasicBlock *LatchBlock = L->getLoopLatch()) {
    WideInc =
      cast<Instruction>(WidePhi->getIncomingValueForBlock(LatchBlock));
    WideIncExpr = SE->getSCEV(WideInc);
  }

  DEBUG(dbgs() << "Wide IV: " << *WidePhi << "\n");
  ++NumWidened;

  // Traverse the def-use chain using a worklist starting at the original IV.
  assert(Widened.empty() && NarrowIVUsers.empty() && "expect initial state" );

  Widened.insert(OrigPhi);
  pushNarrowIVUsers(OrigPhi, WidePhi);

  while (!NarrowIVUsers.empty()) {
    Use *UsePtr;
    Instruction *WideDef;
    tie(UsePtr, WideDef) = NarrowIVUsers.pop_back_val();
    Use &NarrowDefUse = *UsePtr;

    // Process a def-use edge. This may replace the use, so don't hold a
    // use_iterator across it.
    Instruction *NarrowDef = cast<Instruction>(NarrowDefUse.get());
    Instruction *WideUse = WidenIVUse(NarrowDefUse, NarrowDef, WideDef);

    // Follow all def-use edges from the previous narrow use.
    if (WideUse)
      pushNarrowIVUsers(cast<Instruction>(NarrowDefUse.getUser()), WideUse);

    // WidenIVUse may have removed the def-use edge.
    if (NarrowDef->use_empty())
      DeadInsts.push_back(NarrowDef);
  }
  return WidePhi;
}

//===----------------------------------------------------------------------===//
//  Simplification of IV users based on SCEV evaluation.
//===----------------------------------------------------------------------===//

void IndVarSimplify::EliminateIVComparison(ICmpInst *ICmp, Value *IVOperand) {
  unsigned IVOperIdx = 0;
  ICmpInst::Predicate Pred = ICmp->getPredicate();
  if (IVOperand != ICmp->getOperand(0)) {
    // Swapped
    assert(IVOperand == ICmp->getOperand(1) && "Can't find IVOperand");
    IVOperIdx = 1;
    Pred = ICmpInst::getSwappedPredicate(Pred);
  }

  // Get the SCEVs for the ICmp operands.
  const SCEV *S = SE->getSCEV(ICmp->getOperand(IVOperIdx));
  const SCEV *X = SE->getSCEV(ICmp->getOperand(1 - IVOperIdx));

  // Simplify unnecessary loops away.
  const Loop *ICmpLoop = LI->getLoopFor(ICmp->getParent());
  S = SE->getSCEVAtScope(S, ICmpLoop);
  X = SE->getSCEVAtScope(X, ICmpLoop);

  // If the condition is always true or always false, replace it with
  // a constant value.
  if (SE->isKnownPredicate(Pred, S, X))
    ICmp->replaceAllUsesWith(ConstantInt::getTrue(ICmp->getContext()));
  else if (SE->isKnownPredicate(ICmpInst::getInversePredicate(Pred), S, X))
    ICmp->replaceAllUsesWith(ConstantInt::getFalse(ICmp->getContext()));
  else
    return;

  DEBUG(dbgs() << "INDVARS: Eliminated comparison: " << *ICmp << '\n');
  ++NumElimCmp;
  Changed = true;
  DeadInsts.push_back(ICmp);
}

void IndVarSimplify::EliminateIVRemainder(BinaryOperator *Rem,
                                          Value *IVOperand,
                                          bool IsSigned) {
  // We're only interested in the case where we know something about
  // the numerator.
  if (IVOperand != Rem->getOperand(0))
    return;

  // Get the SCEVs for the ICmp operands.
  const SCEV *S = SE->getSCEV(Rem->getOperand(0));
  const SCEV *X = SE->getSCEV(Rem->getOperand(1));

  // Simplify unnecessary loops away.
  const Loop *ICmpLoop = LI->getLoopFor(Rem->getParent());
  S = SE->getSCEVAtScope(S, ICmpLoop);
  X = SE->getSCEVAtScope(X, ICmpLoop);

  // i % n  -->  i  if i is in [0,n).
  if ((!IsSigned || SE->isKnownNonNegative(S)) &&
      SE->isKnownPredicate(IsSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT,
                           S, X))
    Rem->replaceAllUsesWith(Rem->getOperand(0));
  else {
    // (i+1) % n  -->  (i+1)==n?0:(i+1)  if i is in [0,n).
    const SCEV *LessOne =
      SE->getMinusSCEV(S, SE->getConstant(S->getType(), 1));
    if (IsSigned && !SE->isKnownNonNegative(LessOne))
      return;

    if (!SE->isKnownPredicate(IsSigned ?
                              ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT,
                              LessOne, X))
      return;

    ICmpInst *ICmp = new ICmpInst(Rem, ICmpInst::ICMP_EQ,
                                  Rem->getOperand(0), Rem->getOperand(1),
                                  "tmp");
    SelectInst *Sel =
      SelectInst::Create(ICmp,
                         ConstantInt::get(Rem->getType(), 0),
                         Rem->getOperand(0), "tmp", Rem);
    Rem->replaceAllUsesWith(Sel);
  }

  // Inform IVUsers about the new users.
  if (IU) {
    if (Instruction *I = dyn_cast<Instruction>(Rem->getOperand(0)))
      IU->AddUsersIfInteresting(I);
  }
  DEBUG(dbgs() << "INDVARS: Simplified rem: " << *Rem << '\n');
  ++NumElimRem;
  Changed = true;
  DeadInsts.push_back(Rem);
}

/// EliminateIVUser - Eliminate an operation that consumes a simple IV and has
/// no observable side-effect given the range of IV values.
bool IndVarSimplify::EliminateIVUser(Instruction *UseInst,
                                     Instruction *IVOperand) {
  if (ICmpInst *ICmp = dyn_cast<ICmpInst>(UseInst)) {
    EliminateIVComparison(ICmp, IVOperand);
    return true;
  }
  if (BinaryOperator *Rem = dyn_cast<BinaryOperator>(UseInst)) {
    bool IsSigned = Rem->getOpcode() == Instruction::SRem;
    if (IsSigned || Rem->getOpcode() == Instruction::URem) {
      EliminateIVRemainder(Rem, IVOperand, IsSigned);
      return true;
    }
  }

  // Eliminate any operation that SCEV can prove is an identity function.
  if (!SE->isSCEVable(UseInst->getType()) ||
      (UseInst->getType() != IVOperand->getType()) ||
      (SE->getSCEV(UseInst) != SE->getSCEV(IVOperand)))
    return false;

  DEBUG(dbgs() << "INDVARS: Eliminated identity: " << *UseInst << '\n');

  UseInst->replaceAllUsesWith(IVOperand);
  ++NumElimIdentity;
  Changed = true;
  DeadInsts.push_back(UseInst);
  return true;
}

/// pushIVUsers - Add all uses of Def to the current IV's worklist.
///
static void pushIVUsers(
  Instruction *Def,
  SmallPtrSet<Instruction*,16> &Simplified,
  SmallVectorImpl< std::pair<Instruction*,Instruction*> > &SimpleIVUsers) {

  for (Value::use_iterator UI = Def->use_begin(), E = Def->use_end();
       UI != E; ++UI) {
    Instruction *User = cast<Instruction>(*UI);

    // Avoid infinite or exponential worklist processing.
    // Also ensure unique worklist users.
    // If Def is a LoopPhi, it may not be in the Simplified set, so check for
    // self edges first.
    if (User != Def && Simplified.insert(User))
      SimpleIVUsers.push_back(std::make_pair(User, Def));
  }
}

/// isSimpleIVUser - Return true if this instruction generates a simple SCEV
/// expression in terms of that IV.
///
/// This is similar to IVUsers' isInsteresting() but processes each instruction
/// non-recursively when the operand is already known to be a simpleIVUser.
///
static bool isSimpleIVUser(Instruction *I, const Loop *L, ScalarEvolution *SE) {
  if (!SE->isSCEVable(I->getType()))
    return false;

  // Get the symbolic expression for this instruction.
  const SCEV *S = SE->getSCEV(I);

  // We assume that terminators are not SCEVable.
  assert((!S || I != I->getParent()->getTerminator()) &&
         "can't fold terminators");

  // Only consider affine recurrences.
  const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S);
  if (AR && AR->getLoop() == L)
    return true;

  return false;
}

/// SimplifyIVUsersNoRewrite - Iteratively perform simplification on a worklist
/// of IV users. Each successive simplification may push more users which may
/// themselves be candidates for simplification.
///
/// The "NoRewrite" algorithm does not require IVUsers analysis. Instead, it
/// simplifies instructions in-place during analysis. Rather than rewriting
/// induction variables bottom-up from their users, it transforms a chain of
/// IVUsers top-down, updating the IR only when it encouters a clear
/// optimization opportunitiy. A SCEVExpander "Rewriter" instance is still
/// needed, but only used to generate a new IV (phi) of wider type for sign/zero
/// extend elimination.
///
/// Once DisableIVRewrite is default, LSR will be the only client of IVUsers.
///
void IndVarSimplify::SimplifyIVUsersNoRewrite(Loop *L, SCEVExpander &Rewriter) {
  std::map<PHINode *, WideIVInfo> WideIVMap;

  SmallVector<PHINode*, 8> LoopPhis;
  for (BasicBlock::iterator I = L->getHeader()->begin(); isa<PHINode>(I); ++I) {
    LoopPhis.push_back(cast<PHINode>(I));
  }
  // Each round of simplification iterates through the SimplifyIVUsers worklist
  // for all current phis, then determines whether any IVs can be
  // widened. Widening adds new phis to LoopPhis, inducing another round of
  // simplification on the wide IVs.
  while (!LoopPhis.empty()) {
    // Evaluate as many IV expressions as possible before widening any IVs. This
    // forces SCEV to set no-wrap flags before evaluating sign/zero
    // extension. The first time SCEV attempts to normalize sign/zero extension,
    // the result becomes final. So for the most predictable results, we delay
    // evaluation of sign/zero extend evaluation until needed, and avoid running
    // other SCEV based analysis prior to SimplifyIVUsersNoRewrite.
    do {
      PHINode *CurrIV = LoopPhis.pop_back_val();

      // Information about sign/zero extensions of CurrIV.
      WideIVInfo WI;

      // Instructions processed by SimplifyIVUsers for CurrIV.
      SmallPtrSet<Instruction*,16> Simplified;

      // Use-def pairs if IV users waiting to be processed for CurrIV.
      SmallVector<std::pair<Instruction*, Instruction*>, 8> SimpleIVUsers;

      // Push users of the current LoopPhi. In rare cases, pushIVUsers may be
      // called multiple times for the same LoopPhi. This is the proper thing to
      // do for loop header phis that use each other.
      pushIVUsers(CurrIV, Simplified, SimpleIVUsers);

      while (!SimpleIVUsers.empty()) {
        Instruction *UseInst, *Operand;
        tie(UseInst, Operand) = SimpleIVUsers.pop_back_val();
        // Bypass back edges to avoid extra work.
        if (UseInst == CurrIV) continue;

        if (EliminateIVUser(UseInst, Operand)) {
          pushIVUsers(Operand, Simplified, SimpleIVUsers);
          continue;
        }
        if (CastInst *Cast = dyn_cast<CastInst>(UseInst)) {
          bool IsSigned = Cast->getOpcode() == Instruction::SExt;
          if (IsSigned || Cast->getOpcode() == Instruction::ZExt) {
            CollectExtend(Cast, IsSigned, WI, SE, TD);
          }
          continue;
        }
        if (isSimpleIVUser(UseInst, L, SE)) {
          pushIVUsers(UseInst, Simplified, SimpleIVUsers);
        }
      }
      if (WI.WidestNativeType) {
        WideIVMap[CurrIV] = WI;
      }
    } while(!LoopPhis.empty());

    for (std::map<PHINode *, WideIVInfo>::const_iterator I = WideIVMap.begin(),
           E = WideIVMap.end(); I != E; ++I) {
      WidenIV Widener(I->first, I->second, LI, SE, DT, DeadInsts);
      if (PHINode *WidePhi = Widener.CreateWideIV(Rewriter)) {
        Changed = true;
        LoopPhis.push_back(WidePhi);
      }
    }
    WideIVMap.clear();
  }
}

/// SimplifyCongruentIVs - Check for congruent phis in this loop header and
/// populate ExprToIVMap for use later.
///
void IndVarSimplify::SimplifyCongruentIVs(Loop *L) {
  DenseMap<const SCEV *, PHINode *> ExprToIVMap;
  for (BasicBlock::iterator I = L->getHeader()->begin(); isa<PHINode>(I); ++I) {
    PHINode *Phi = cast<PHINode>(I);
    if (!SE->isSCEVable(Phi->getType()))
      continue;

    const SCEV *S = SE->getSCEV(Phi);
    DenseMap<const SCEV *, PHINode *>::const_iterator Pos;
    bool Inserted;
    tie(Pos, Inserted) = ExprToIVMap.insert(std::make_pair(S, Phi));
    if (Inserted)
      continue;
    PHINode *OrigPhi = Pos->second;
    // Replacing the congruent phi is sufficient because acyclic redundancy
    // elimination, CSE/GVN, should handle the rest. However, once SCEV proves
    // that a phi is congruent, it's almost certain to be the head of an IV
    // user cycle that is isomorphic with the original phi. So it's worth
    // eagerly cleaning up the common case of a single IV increment.
    if (BasicBlock *LatchBlock = L->getLoopLatch()) {
      Instruction *OrigInc =
        cast<Instruction>(OrigPhi->getIncomingValueForBlock(LatchBlock));
      Instruction *IsomorphicInc =
        cast<Instruction>(Phi->getIncomingValueForBlock(LatchBlock));
      if (OrigInc != IsomorphicInc &&
          SE->getSCEV(OrigInc) == SE->getSCEV(IsomorphicInc) &&
          HoistStep(OrigInc, IsomorphicInc, DT)) {
        DEBUG(dbgs() << "INDVARS: Eliminated congruent iv.inc: "
              << *IsomorphicInc << '\n');
        IsomorphicInc->replaceAllUsesWith(OrigInc);
        DeadInsts.push_back(IsomorphicInc);
      }
    }
    DEBUG(dbgs() << "INDVARS: Eliminated congruent iv: " << *Phi << '\n');
    ++NumElimIV;
    Phi->replaceAllUsesWith(OrigPhi);
    DeadInsts.push_back(Phi);
  }
}

//===----------------------------------------------------------------------===//
//  LinearFunctionTestReplace and its kin. Rewrite the loop exit condition.
//===----------------------------------------------------------------------===//

// Check for expressions that ScalarEvolution generates to compute
// BackedgeTakenInfo. If these expressions have not been reduced, then expanding
// them may incur additional cost (albeit in the loop preheader).
static bool isHighCostExpansion(const SCEV *S, BranchInst *BI,
                                ScalarEvolution *SE) {
  // If the backedge-taken count is a UDiv, it's very likely a UDiv that
  // ScalarEvolution's HowFarToZero or HowManyLessThans produced to compute a
  // precise expression, rather than a UDiv from the user's code. If we can't
  // find a UDiv in the code with some simple searching, assume the former and
  // forego rewriting the loop.
  if (isa<SCEVUDivExpr>(S)) {
    ICmpInst *OrigCond = dyn_cast<ICmpInst>(BI->getCondition());
    if (!OrigCond) return true;
    const SCEV *R = SE->getSCEV(OrigCond->getOperand(1));
    R = SE->getMinusSCEV(R, SE->getConstant(R->getType(), 1));
    if (R != S) {
      const SCEV *L = SE->getSCEV(OrigCond->getOperand(0));
      L = SE->getMinusSCEV(L, SE->getConstant(L->getType(), 1));
      if (L != S)
        return true;
    }
  }

  if (!DisableIVRewrite)
    return false;

  // Recurse past add expressions, which commonly occur in the
  // BackedgeTakenCount. They may already exist in program code, and if not,
  // they are not too expensive rematerialize.
  if (const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(S)) {
    for (SCEVAddExpr::op_iterator I = Add->op_begin(), E = Add->op_end();
         I != E; ++I) {
      if (isHighCostExpansion(*I, BI, SE))
        return true;
    }
    return false;
  }

  // HowManyLessThans uses a Max expression whenever the loop is not guarded by
  // the exit condition.
  if (isa<SCEVSMaxExpr>(S) || isa<SCEVUMaxExpr>(S))
    return true;

  // If we haven't recognized an expensive SCEV patter, assume its an expression
  // produced by program code.
  return false;
}

/// canExpandBackedgeTakenCount - Return true if this loop's backedge taken
/// count expression can be safely and cheaply expanded into an instruction
/// sequence that can be used by LinearFunctionTestReplace.
static bool canExpandBackedgeTakenCount(Loop *L, ScalarEvolution *SE) {
  const SCEV *BackedgeTakenCount = SE->getBackedgeTakenCount(L);
  if (isa<SCEVCouldNotCompute>(BackedgeTakenCount) ||
      BackedgeTakenCount->isZero())
    return false;

  if (!L->getExitingBlock())
    return false;

  // Can't rewrite non-branch yet.
  BranchInst *BI = dyn_cast<BranchInst>(L->getExitingBlock()->getTerminator());
  if (!BI)
    return false;

  if (isHighCostExpansion(BackedgeTakenCount, BI, SE))
    return false;

  return true;
}

/// getBackedgeIVType - Get the widest type used by the loop test after peeking
/// through Truncs.
///
/// TODO: Unnecessary if LFTR does not force a canonical IV.
static Type *getBackedgeIVType(Loop *L) {
  if (!L->getExitingBlock())
    return 0;

  // Can't rewrite non-branch yet.
  BranchInst *BI = dyn_cast<BranchInst>(L->getExitingBlock()->getTerminator());
  if (!BI)
    return 0;

  ICmpInst *Cond = dyn_cast<ICmpInst>(BI->getCondition());
  if (!Cond)
    return 0;

  Type *Ty = 0;
  for(User::op_iterator OI = Cond->op_begin(), OE = Cond->op_end();
      OI != OE; ++OI) {
    assert((!Ty || Ty == (*OI)->getType()) && "bad icmp operand types");
    TruncInst *Trunc = dyn_cast<TruncInst>(*OI);
    if (!Trunc)
      continue;

    return Trunc->getSrcTy();
  }
  return Ty;
}

/// LinearFunctionTestReplace - This method rewrites the exit condition of the
/// loop to be a canonical != comparison against the incremented loop induction
/// variable.  This pass is able to rewrite the exit tests of any loop where the
/// SCEV analysis can determine a loop-invariant trip count of the loop, which
/// is actually a much broader range than just linear tests.
ICmpInst *IndVarSimplify::
LinearFunctionTestReplace(Loop *L,
                          const SCEV *BackedgeTakenCount,
                          PHINode *IndVar,
                          SCEVExpander &Rewriter) {
  assert(canExpandBackedgeTakenCount(L, SE) && "precondition");
  BranchInst *BI = cast<BranchInst>(L->getExitingBlock()->getTerminator());

  // If the exiting block is not the same as the backedge block, we must compare
  // against the preincremented value, otherwise we prefer to compare against
  // the post-incremented value.
  Value *CmpIndVar;
  const SCEV *RHS = BackedgeTakenCount;
  if (L->getExitingBlock() == L->getLoopLatch()) {
    // Add one to the "backedge-taken" count to get the trip count.
    // If this addition may overflow, we have to be more pessimistic and
    // cast the induction variable before doing the add.
    const SCEV *Zero = SE->getConstant(BackedgeTakenCount->getType(), 0);
    const SCEV *N =
      SE->getAddExpr(BackedgeTakenCount,
                     SE->getConstant(BackedgeTakenCount->getType(), 1));
    if ((isa<SCEVConstant>(N) && !N->isZero()) ||
        SE->isLoopEntryGuardedByCond(L, ICmpInst::ICMP_NE, N, Zero)) {
      // No overflow. Cast the sum.
      RHS = SE->getTruncateOrZeroExtend(N, IndVar->getType());
    } else {
      // Potential overflow. Cast before doing the add.
      RHS = SE->getTruncateOrZeroExtend(BackedgeTakenCount,
                                        IndVar->getType());
      RHS = SE->getAddExpr(RHS,
                           SE->getConstant(IndVar->getType(), 1));
    }

    // The BackedgeTaken expression contains the number of times that the
    // backedge branches to the loop header.  This is one less than the
    // number of times the loop executes, so use the incremented indvar.
    CmpIndVar = IndVar->getIncomingValueForBlock(L->getExitingBlock());
  } else {
    // We have to use the preincremented value...
    RHS = SE->getTruncateOrZeroExtend(BackedgeTakenCount,
                                      IndVar->getType());
    CmpIndVar = IndVar;
  }

  // Expand the code for the iteration count.
  assert(SE->isLoopInvariant(RHS, L) &&
         "Computed iteration count is not loop invariant!");
  Value *ExitCnt = Rewriter.expandCodeFor(RHS, IndVar->getType(), BI);

  // Insert a new icmp_ne or icmp_eq instruction before the branch.
  ICmpInst::Predicate Opcode;
  if (L->contains(BI->getSuccessor(0)))
    Opcode = ICmpInst::ICMP_NE;
  else
    Opcode = ICmpInst::ICMP_EQ;

  DEBUG(dbgs() << "INDVARS: Rewriting loop exit condition to:\n"
               << "      LHS:" << *CmpIndVar << '\n'
               << "       op:\t"
               << (Opcode == ICmpInst::ICMP_NE ? "!=" : "==") << "\n"
               << "      RHS:\t" << *RHS << "\n");

  ICmpInst *Cond = new ICmpInst(BI, Opcode, CmpIndVar, ExitCnt, "exitcond");
  Cond->setDebugLoc(BI->getDebugLoc());
  Value *OrigCond = BI->getCondition();
  // It's tempting to use replaceAllUsesWith here to fully replace the old
  // comparison, but that's not immediately safe, since users of the old
  // comparison may not be dominated by the new comparison. Instead, just
  // update the branch to use the new comparison; in the common case this
  // will make old comparison dead.
  BI->setCondition(Cond);
  DeadInsts.push_back(OrigCond);

  ++NumLFTR;
  Changed = true;
  return Cond;
}

//===----------------------------------------------------------------------===//
//  SinkUnusedInvariants. A late subpass to cleanup loop preheaders.
//===----------------------------------------------------------------------===//

/// If there's a single exit block, sink any loop-invariant values that
/// were defined in the preheader but not used inside the loop into the
/// exit block to reduce register pressure in the loop.
void IndVarSimplify::SinkUnusedInvariants(Loop *L) {
  BasicBlock *ExitBlock = L->getExitBlock();
  if (!ExitBlock) return;

  BasicBlock *Preheader = L->getLoopPreheader();
  if (!Preheader) return;

  Instruction *InsertPt = ExitBlock->getFirstNonPHI();
  BasicBlock::iterator I = Preheader->getTerminator();
  while (I != Preheader->begin()) {
    --I;
    // New instructions were inserted at the end of the preheader.
    if (isa<PHINode>(I))
      break;

    // Don't move instructions which might have side effects, since the side
    // effects need to complete before instructions inside the loop.  Also don't
    // move instructions which might read memory, since the loop may modify
    // memory. Note that it's okay if the instruction might have undefined
    // behavior: LoopSimplify guarantees that the preheader dominates the exit
    // block.
    if (I->mayHaveSideEffects() || I->mayReadFromMemory())
      continue;

    // Skip debug info intrinsics.
    if (isa<DbgInfoIntrinsic>(I))
      continue;

    // Don't sink static AllocaInsts out of the entry block, which would
    // turn them into dynamic allocas!
    if (AllocaInst *AI = dyn_cast<AllocaInst>(I))
      if (AI->isStaticAlloca())
        continue;

    // Determine if there is a use in or before the loop (direct or
    // otherwise).
    bool UsedInLoop = false;
    for (Value::use_iterator UI = I->use_begin(), UE = I->use_end();
         UI != UE; ++UI) {
      User *U = *UI;
      BasicBlock *UseBB = cast<Instruction>(U)->getParent();
      if (PHINode *P = dyn_cast<PHINode>(U)) {
        unsigned i =
          PHINode::getIncomingValueNumForOperand(UI.getOperandNo());
        UseBB = P->getIncomingBlock(i);
      }
      if (UseBB == Preheader || L->contains(UseBB)) {
        UsedInLoop = true;
        break;
      }
    }

    // If there is, the def must remain in the preheader.
    if (UsedInLoop)
      continue;

    // Otherwise, sink it to the exit block.
    Instruction *ToMove = I;
    bool Done = false;

    if (I != Preheader->begin()) {
      // Skip debug info intrinsics.
      do {
        --I;
      } while (isa<DbgInfoIntrinsic>(I) && I != Preheader->begin());

      if (isa<DbgInfoIntrinsic>(I) && I == Preheader->begin())
        Done = true;
    } else {
      Done = true;
    }

    ToMove->moveBefore(InsertPt);
    if (Done) break;
    InsertPt = ToMove;
  }
}

//===----------------------------------------------------------------------===//
//  IndVarSimplify driver. Manage several subpasses of IV simplification.
//===----------------------------------------------------------------------===//

bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
  // If LoopSimplify form is not available, stay out of trouble. Some notes:
  //  - LSR currently only supports LoopSimplify-form loops. Indvars'
  //    canonicalization can be a pessimization without LSR to "clean up"
  //    afterwards.
  //  - We depend on having a preheader; in particular,
  //    Loop::getCanonicalInductionVariable only supports loops with preheaders,
  //    and we're in trouble if we can't find the induction variable even when
  //    we've manually inserted one.
  if (!L->isLoopSimplifyForm())
    return false;

  if (!DisableIVRewrite)
    IU = &getAnalysis<IVUsers>();
  LI = &getAnalysis<LoopInfo>();
  SE = &getAnalysis<ScalarEvolution>();
  DT = &getAnalysis<DominatorTree>();
  TD = getAnalysisIfAvailable<TargetData>();

  DeadInsts.clear();
  Changed = false;

  // If there are any floating-point recurrences, attempt to
  // transform them to use integer recurrences.
  RewriteNonIntegerIVs(L);

  const SCEV *BackedgeTakenCount = SE->getBackedgeTakenCount(L);

  // Create a rewriter object which we'll use to transform the code with.
  SCEVExpander Rewriter(*SE, "indvars");

  // Eliminate redundant IV users.
  //
  // Simplification works best when run before other consumers of SCEV. We
  // attempt to avoid evaluating SCEVs for sign/zero extend operations until
  // other expressions involving loop IVs have been evaluated. This helps SCEV
  // set no-wrap flags before normalizing sign/zero extension.
  if (DisableIVRewrite) {
    Rewriter.disableCanonicalMode();
    SimplifyIVUsersNoRewrite(L, Rewriter);
  }

  // Check to see if this loop has a computable loop-invariant execution count.
  // If so, this means that we can compute the final value of any expressions
  // that are recurrent in the loop, and substitute the exit values from the
  // loop into any instructions outside of the loop that use the final values of
  // the current expressions.
  //
  if (!isa<SCEVCouldNotCompute>(BackedgeTakenCount))
    RewriteLoopExitValues(L, Rewriter);

  // Eliminate redundant IV users.
  if (!DisableIVRewrite)
    SimplifyIVUsers(Rewriter);

  // Eliminate redundant IV cycles.
  if (DisableIVRewrite)
    SimplifyCongruentIVs(L);

  // Compute the type of the largest recurrence expression, and decide whether
  // a canonical induction variable should be inserted.
  Type *LargestType = 0;
  bool NeedCannIV = false;
  bool ExpandBECount = canExpandBackedgeTakenCount(L, SE);
  if (ExpandBECount) {
    // If we have a known trip count and a single exit block, we'll be
    // rewriting the loop exit test condition below, which requires a
    // canonical induction variable.
    NeedCannIV = true;
    Type *Ty = BackedgeTakenCount->getType();
    if (DisableIVRewrite) {
      // In this mode, SimplifyIVUsers may have already widened the IV used by
      // the backedge test and inserted a Trunc on the compare's operand. Get
      // the wider type to avoid creating a redundant narrow IV only used by the
      // loop test.
      LargestType = getBackedgeIVType(L);
    }
    if (!LargestType ||
        SE->getTypeSizeInBits(Ty) >
        SE->getTypeSizeInBits(LargestType))
      LargestType = SE->getEffectiveSCEVType(Ty);
  }
  if (!DisableIVRewrite) {
    for (IVUsers::const_iterator I = IU->begin(), E = IU->end(); I != E; ++I) {
      NeedCannIV = true;
      Type *Ty =
        SE->getEffectiveSCEVType(I->getOperandValToReplace()->getType());
      if (!LargestType ||
          SE->getTypeSizeInBits(Ty) >
          SE->getTypeSizeInBits(LargestType))
        LargestType = Ty;
    }
  }

  // Now that we know the largest of the induction variable expressions
  // in this loop, insert a canonical induction variable of the largest size.
  PHINode *IndVar = 0;
  if (NeedCannIV) {
    // Check to see if the loop already has any canonical-looking induction
    // variables. If any are present and wider than the planned canonical
    // induction variable, temporarily remove them, so that the Rewriter
    // doesn't attempt to reuse them.
    SmallVector<PHINode *, 2> OldCannIVs;
    while (PHINode *OldCannIV = L->getCanonicalInductionVariable()) {
      if (SE->getTypeSizeInBits(OldCannIV->getType()) >
          SE->getTypeSizeInBits(LargestType))
        OldCannIV->removeFromParent();
      else
        break;
      OldCannIVs.push_back(OldCannIV);
    }

    IndVar = Rewriter.getOrInsertCanonicalInductionVariable(L, LargestType);

    ++NumInserted;
    Changed = true;
    DEBUG(dbgs() << "INDVARS: New CanIV: " << *IndVar << '\n');

    // Now that the official induction variable is established, reinsert
    // any old canonical-looking variables after it so that the IR remains
    // consistent. They will be deleted as part of the dead-PHI deletion at
    // the end of the pass.
    while (!OldCannIVs.empty()) {
      PHINode *OldCannIV = OldCannIVs.pop_back_val();
      OldCannIV->insertBefore(L->getHeader()->getFirstNonPHI());
    }
  }

  // If we have a trip count expression, rewrite the loop's exit condition
  // using it.  We can currently only handle loops with a single exit.
  ICmpInst *NewICmp = 0;
  if (ExpandBECount) {
    assert(canExpandBackedgeTakenCount(L, SE) &&
           "canonical IV disrupted BackedgeTaken expansion");
    assert(NeedCannIV &&
           "LinearFunctionTestReplace requires a canonical induction variable");
    // Check preconditions for proper SCEVExpander operation. SCEV does not
    // express SCEVExpander's dependencies, such as LoopSimplify. Instead any
    // pass that uses the SCEVExpander must do it. This does not work well for
    // loop passes because SCEVExpander makes assumptions about all loops, while
    // LoopPassManager only forces the current loop to be simplified.
    //
    // FIXME: SCEV expansion has no way to bail out, so the caller must
    // explicitly check any assumptions made by SCEV. Brittle.
    const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(BackedgeTakenCount);
    if (!AR || AR->getLoop()->getLoopPreheader())
      NewICmp =
        LinearFunctionTestReplace(L, BackedgeTakenCount, IndVar, Rewriter);
  }
  // Rewrite IV-derived expressions.
  if (!DisableIVRewrite)
    RewriteIVExpressions(L, Rewriter);

  // Clear the rewriter cache, because values that are in the rewriter's cache
  // can be deleted in the loop below, causing the AssertingVH in the cache to
  // trigger.
  Rewriter.clear();

  // Now that we're done iterating through lists, clean up any instructions
  // which are now dead.
  while (!DeadInsts.empty())
    if (Instruction *Inst =
          dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val()))
      RecursivelyDeleteTriviallyDeadInstructions(Inst);

  // The Rewriter may not be used from this point on.

  // Loop-invariant instructions in the preheader that aren't used in the
  // loop may be sunk below the loop to reduce register pressure.
  SinkUnusedInvariants(L);

  // For completeness, inform IVUsers of the IV use in the newly-created
  // loop exit test instruction.
  if (NewICmp && IU)
    IU->AddUsersIfInteresting(cast<Instruction>(NewICmp->getOperand(0)));

  // Clean up dead instructions.
  Changed |= DeleteDeadPHIs(L->getHeader());
  // Check a post-condition.
  assert(L->isLCSSAForm(*DT) &&
         "Indvars did not leave the loop in lcssa form!");

  // Verify that LFTR, and any other change have not interfered with SCEV's
  // ability to compute trip count.
#ifndef NDEBUG
  if (DisableIVRewrite && !isa<SCEVCouldNotCompute>(BackedgeTakenCount)) {
    SE->forgetLoop(L);
    const SCEV *NewBECount = SE->getBackedgeTakenCount(L);
    if (SE->getTypeSizeInBits(BackedgeTakenCount->getType()) <
        SE->getTypeSizeInBits(NewBECount->getType()))
      NewBECount = SE->getTruncateOrNoop(NewBECount,
                                         BackedgeTakenCount->getType());
    else
      BackedgeTakenCount = SE->getTruncateOrNoop(BackedgeTakenCount,
                                                 NewBECount->getType());
    assert(BackedgeTakenCount == NewBECount && "indvars must preserve SCEV");
  }
#endif

  return Changed;
}