Target/Mips/MipsDelaySlotFiller.cpp

//===- MipsDelaySlotFiller.cpp - Mips Delay Slot Filler -------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// Simple pass to fill delay slots with useful instructions.
//
//===----------------------------------------------------------------------===//

#include "MCTargetDesc/MipsMCNaCl.h"
#include "Mips.h"
#include "MipsInstrInfo.h"
#include "MipsRegisterInfo.h"
#include "MipsSubtarget.h"
#include "llvm/ADT/BitVector.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/PointerUnion.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/MC/MCInstrDesc.h"
#include "llvm/MC/MCRegisterInfo.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CodeGen.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Target/TargetMachine.h"
#include <algorithm>
#include <cassert>
#include <iterator>
#include <memory>
#include <utility>

using namespace llvm;

#define DEBUG_TYPE "mips-delay-slot-filler"

STATISTIC(FilledSlots, "Number of delay slots filled");
STATISTIC(UsefulSlots, "Number of delay slots filled with instructions that"
                       " are not NOP.");

static cl::opt<bool> DisableDelaySlotFiller(
  "disable-mips-delay-filler",
  cl::init(false),
  cl::desc("Fill all delay slots with NOPs."),
  cl::Hidden);

static cl::opt<bool> DisableForwardSearch(
  "disable-mips-df-forward-search",
  cl::init(true),
  cl::desc("Disallow MIPS delay filler to search forward."),
  cl::Hidden);

static cl::opt<bool> DisableSuccBBSearch(
  "disable-mips-df-succbb-search",
  cl::init(true),
  cl::desc("Disallow MIPS delay filler to search successor basic blocks."),
  cl::Hidden);

static cl::opt<bool> DisableBackwardSearch(
  "disable-mips-df-backward-search",
  cl::init(false),
  cl::desc("Disallow MIPS delay filler to search backward."),
  cl::Hidden);

enum CompactBranchPolicy {
  CB_Never,   ///< The policy 'never' may in some circumstances or for some
              ///< ISAs not be absolutely adhered to.
  CB_Optimal, ///< Optimal is the default and will produce compact branches
              ///< when delay slots cannot be filled.
  CB_Always   ///< 'always' may in some circumstances may not be
              ///< absolutely adhered to there may not be a corresponding
              ///< compact form of a branch.
};

static cl::opt<CompactBranchPolicy> MipsCompactBranchPolicy(
    "mips-compact-branches", cl::Optional, cl::init(CB_Optimal),
    cl::desc("MIPS Specific: Compact branch policy."),
    cl::values(clEnumValN(CB_Never, "never",
                          "Do not use compact branches if possible."),
               clEnumValN(CB_Optimal, "optimal",
                          "Use compact branches where appropriate (default)."),
               clEnumValN(CB_Always, "always",
                          "Always use compact branches if possible.")));

namespace {

  using Iter = MachineBasicBlock::iterator;
  using ReverseIter = MachineBasicBlock::reverse_iterator;
  using BB2BrMap = SmallDenseMap<MachineBasicBlock *, MachineInstr *, 2>;

  class RegDefsUses {
  public:
    RegDefsUses(const TargetRegisterInfo &TRI);

    void init(const MachineInstr &MI);

    /// This function sets all caller-saved registers in Defs.
    void setCallerSaved(const MachineInstr &MI);

    /// This function sets all unallocatable registers in Defs.
    void setUnallocatableRegs(const MachineFunction &MF);

    /// Set bits in Uses corresponding to MBB's live-out registers except for
    /// the registers that are live-in to SuccBB.
    void addLiveOut(const MachineBasicBlock &MBB,
                    const MachineBasicBlock &SuccBB);

    bool update(const MachineInstr &MI, unsigned Begin, unsigned End);

  private:
    bool checkRegDefsUses(BitVector &NewDefs, BitVector &NewUses, unsigned Reg,
                          bool IsDef) const;

    /// Returns true if Reg or its alias is in RegSet.
    bool isRegInSet(const BitVector &RegSet, unsigned Reg) const;

    const TargetRegisterInfo &TRI;
    BitVector Defs, Uses;
  };

  /// Base class for inspecting loads and stores.
  class InspectMemInstr {
  public:
    InspectMemInstr(bool ForbidMemInstr_) : ForbidMemInstr(ForbidMemInstr_) {}
    virtual ~InspectMemInstr() = default;

    /// Return true if MI cannot be moved to delay slot.
    bool hasHazard(const MachineInstr &MI);

  protected:
    /// Flags indicating whether loads or stores have been seen.
    bool OrigSeenLoad = false;
    bool OrigSeenStore = false;
    bool SeenLoad = false;
    bool SeenStore = false;

    /// Memory instructions are not allowed to move to delay slot if this flag
    /// is true.
    bool ForbidMemInstr;

  private:
    virtual bool hasHazard_(const MachineInstr &MI) = 0;
  };

  /// This subclass rejects any memory instructions.
  class NoMemInstr : public InspectMemInstr {
  public:
    NoMemInstr() : InspectMemInstr(true) {}

  private:
    bool hasHazard_(const MachineInstr &MI) override { return true; }
  };

  /// This subclass accepts loads from stacks and constant loads.
  class LoadFromStackOrConst : public InspectMemInstr {
  public:
    LoadFromStackOrConst() : InspectMemInstr(false) {}

  private:
    bool hasHazard_(const MachineInstr &MI) override;
  };

  /// This subclass uses memory dependence information to determine whether a
  /// memory instruction can be moved to a delay slot.
  class MemDefsUses : public InspectMemInstr {
  public:
    explicit MemDefsUses(const MachineFrameInfo *MFI);

  private:
    using ValueType = PointerUnion<const Value *, const PseudoSourceValue *>;

    bool hasHazard_(const MachineInstr &MI) override;

    /// Update Defs and Uses. Return true if there exist dependences that
    /// disqualify the delay slot candidate between V and values in Uses and
    /// Defs.
    bool updateDefsUses(ValueType V, bool MayStore);

    /// Get the list of underlying objects of MI's memory operand.
    bool getUnderlyingObjects(const MachineInstr &MI,
                              SmallVectorImpl<ValueType> &Objects) const;

    const MachineFrameInfo *MFI;
    SmallPtrSet<ValueType, 4> Uses, Defs;

    /// Flags indicating whether loads or stores with no underlying objects have
    /// been seen.
    bool SeenNoObjLoad = false;
    bool SeenNoObjStore = false;
  };

  class MipsDelaySlotFiller : public MachineFunctionPass {
  public:
    MipsDelaySlotFiller() : MachineFunctionPass(ID) {
      initializeMipsDelaySlotFillerPass(*PassRegistry::getPassRegistry());
    }

    StringRef getPassName() const override { return "Mips Delay Slot Filler"; }

    bool runOnMachineFunction(MachineFunction &F) override {
      TM = &F.getTarget();
      bool Changed = false;
      for (MachineBasicBlock &MBB : F)
        Changed |= runOnMachineBasicBlock(MBB);

      // This pass invalidates liveness information when it reorders
      // instructions to fill delay slot. Without this, -verify-machineinstrs
      // will fail.
      if (Changed)
        F.getRegInfo().invalidateLiveness();

      return Changed;
    }

    MachineFunctionProperties getRequiredProperties() const override {
      return MachineFunctionProperties().set(
          MachineFunctionProperties::Property::NoVRegs);
    }

    void getAnalysisUsage(AnalysisUsage &AU) const override {
      AU.addRequired<MachineBranchProbabilityInfo>();
      MachineFunctionPass::getAnalysisUsage(AU);
    }

    static char ID;

  private:
    bool runOnMachineBasicBlock(MachineBasicBlock &MBB);

    Iter replaceWithCompactBranch(MachineBasicBlock &MBB, Iter Branch,
                                  const DebugLoc &DL);

    /// This function checks if it is valid to move Candidate to the delay slot
    /// and returns true if it isn't. It also updates memory and register
    /// dependence information.
    bool delayHasHazard(const MachineInstr &Candidate, RegDefsUses &RegDU,
                        InspectMemInstr &IM) const;

    /// This function searches range [Begin, End) for an instruction that can be
    /// moved to the delay slot. Returns true on success.
    template<typename IterTy>
    bool searchRange(MachineBasicBlock &MBB, IterTy Begin, IterTy End,
                     RegDefsUses &RegDU, InspectMemInstr &IM, Iter Slot,
                     IterTy &Filler) const;

    /// This function searches in the backward direction for an instruction that
    /// can be moved to the delay slot. Returns true on success.
    bool searchBackward(MachineBasicBlock &MBB, MachineInstr &Slot) const;

    /// This function searches MBB in the forward direction for an instruction
    /// that can be moved to the delay slot. Returns true on success.
    bool searchForward(MachineBasicBlock &MBB, Iter Slot) const;

    /// This function searches one of MBB's successor blocks for an instruction
    /// that can be moved to the delay slot and inserts clones of the
    /// instruction into the successor's predecessor blocks.
    bool searchSuccBBs(MachineBasicBlock &MBB, Iter Slot) const;

    /// Pick a successor block of MBB. Return NULL if MBB doesn't have a
    /// successor block that is not a landing pad.
    MachineBasicBlock *selectSuccBB(MachineBasicBlock &B) const;

    /// This function analyzes MBB and returns an instruction with an unoccupied
    /// slot that branches to Dst.
    std::pair<MipsInstrInfo::BranchType, MachineInstr *>
    getBranch(MachineBasicBlock &MBB, const MachineBasicBlock &Dst) const;

    /// Examine Pred and see if it is possible to insert an instruction into
    /// one of its branches delay slot or its end.
    bool examinePred(MachineBasicBlock &Pred, const MachineBasicBlock &Succ,
                     RegDefsUses &RegDU, bool &HasMultipleSuccs,
                     BB2BrMap &BrMap) const;

    bool terminateSearch(const MachineInstr &Candidate) const;

    const TargetMachine *TM = nullptr;
  };

} // end anonymous namespace

char MipsDelaySlotFiller::ID = 0;

static bool hasUnoccupiedSlot(const MachineInstr *MI) {
  return MI->hasDelaySlot() && !MI->isBundledWithSucc();
}

INITIALIZE_PASS(MipsDelaySlotFiller, DEBUG_TYPE,
                "Fill delay slot for MIPS", false, false)

/// This function inserts clones of Filler into predecessor blocks.
static void insertDelayFiller(Iter Filler, const BB2BrMap &BrMap) {
  MachineFunction *MF = Filler->getParent()->getParent();

  for (const auto &I : BrMap) {
    if (I.second) {
      MIBundleBuilder(I.second).append(MF->CloneMachineInstr(&*Filler));
      ++UsefulSlots;
    } else {
      I.first->push_back(MF->CloneMachineInstr(&*Filler));
    }
  }
}

/// This function adds registers Filler defines to MBB's live-in register list.
static void addLiveInRegs(Iter Filler, MachineBasicBlock &MBB) {
  for (unsigned I = 0, E = Filler->getNumOperands(); I != E; ++I) {
    const MachineOperand &MO = Filler->getOperand(I);
    unsigned R;

    if (!MO.isReg() || !MO.isDef() || !(R = MO.getReg()))
      continue;

#ifndef NDEBUG
    const MachineFunction &MF = *MBB.getParent();
    assert(MF.getSubtarget().getRegisterInfo()->getAllocatableSet(MF).test(R) &&
           "Shouldn't move an instruction with unallocatable registers across "
           "basic block boundaries.");
#endif

    if (!MBB.isLiveIn(R))
      MBB.addLiveIn(R);
  }
}

RegDefsUses::RegDefsUses(const TargetRegisterInfo &TRI)
    : TRI(TRI), Defs(TRI.getNumRegs(), false), Uses(TRI.getNumRegs(), false) {}

void RegDefsUses::init(const MachineInstr &MI) {
  // Add all register operands which are explicit and non-variadic.
  update(MI, 0, MI.getDesc().getNumOperands());

  // If MI is a call, add RA to Defs to prevent users of RA from going into
  // delay slot.
  if (MI.isCall())
    Defs.set(Mips::RA);

  // Add all implicit register operands of branch instructions except
  // register AT.
  if (MI.isBranch()) {
    update(MI, MI.getDesc().getNumOperands(), MI.getNumOperands());
    Defs.reset(Mips::AT);
  }
}

void RegDefsUses::setCallerSaved(const MachineInstr &MI) {
  assert(MI.isCall());

  // Add RA/RA_64 to Defs to prevent users of RA/RA_64 from going into
  // the delay slot. The reason is that RA/RA_64 must not be changed
  // in the delay slot so that the callee can return to the caller.
  if (MI.definesRegister(Mips::RA) || MI.definesRegister(Mips::RA_64)) {
    Defs.set(Mips::RA);
    Defs.set(Mips::RA_64);
  }

  // If MI is a call, add all caller-saved registers to Defs.
  BitVector CallerSavedRegs(TRI.getNumRegs(), true);

  CallerSavedRegs.reset(Mips::ZERO);
  CallerSavedRegs.reset(Mips::ZERO_64);

  for (const MCPhysReg *R = TRI.getCalleeSavedRegs(MI.getParent()->getParent());
       *R; ++R)
    for (MCRegAliasIterator AI(*R, &TRI, true); AI.isValid(); ++AI)
      CallerSavedRegs.reset(*AI);

  Defs |= CallerSavedRegs;
}

void RegDefsUses::setUnallocatableRegs(const MachineFunction &MF) {
  BitVector AllocSet = TRI.getAllocatableSet(MF);

  for (unsigned R : AllocSet.set_bits())
    for (MCRegAliasIterator AI(R, &TRI, false); AI.isValid(); ++AI)
      AllocSet.set(*AI);

  AllocSet.set(Mips::ZERO);
  AllocSet.set(Mips::ZERO_64);

  Defs |= AllocSet.flip();
}

void RegDefsUses::addLiveOut(const MachineBasicBlock &MBB,
                             const MachineBasicBlock &SuccBB) {
  for (const MachineBasicBlock *S : MBB.successors())
    if (S != &SuccBB)
      for (const auto &LI : S->liveins())
        Uses.set(LI.PhysReg);
}

bool RegDefsUses::update(const MachineInstr &MI, unsigned Begin, unsigned End) {
  BitVector NewDefs(TRI.getNumRegs()), NewUses(TRI.getNumRegs());
  bool HasHazard = false;

  for (unsigned I = Begin; I != End; ++I) {
    const MachineOperand &MO = MI.getOperand(I);

    if (MO.isReg() && MO.getReg()) {
      if (checkRegDefsUses(NewDefs, NewUses, MO.getReg(), MO.isDef())) {
        LLVM_DEBUG(dbgs() << DEBUG_TYPE ": found register hazard for operand "
                          << I << ": ";
                   MO.dump());
        HasHazard = true;
      }
    }
  }

  Defs |= NewDefs;
  Uses |= NewUses;

  return HasHazard;
}

bool RegDefsUses::checkRegDefsUses(BitVector &NewDefs, BitVector &NewUses,
                                   unsigned Reg, bool IsDef) const {
  if (IsDef) {
    NewDefs.set(Reg);
    // check whether Reg has already been defined or used.
    return (isRegInSet(Defs, Reg) || isRegInSet(Uses, Reg));
  }

  NewUses.set(Reg);
  // check whether Reg has already been defined.
  return isRegInSet(Defs, Reg);
}

bool RegDefsUses::isRegInSet(const BitVector &RegSet, unsigned Reg) const {
  // Check Reg and all aliased Registers.
  for (MCRegAliasIterator AI(Reg, &TRI, true); AI.isValid(); ++AI)
    if (RegSet.test(*AI))
      return true;
  return false;
}

bool InspectMemInstr::hasHazard(const MachineInstr &MI) {
  if (!MI.mayStore() && !MI.mayLoad())
    return false;

  if (ForbidMemInstr)
    return true;

  OrigSeenLoad = SeenLoad;
  OrigSeenStore = SeenStore;
  SeenLoad |= MI.mayLoad();
  SeenStore |= MI.mayStore();

  // If MI is an ordered or volatile memory reference, disallow moving
  // subsequent loads and stores to delay slot.
  if (MI.hasOrderedMemoryRef() && (OrigSeenLoad || OrigSeenStore)) {
    ForbidMemInstr = true;
    return true;
  }

  return hasHazard_(MI);
}

bool LoadFromStackOrConst::hasHazard_(const MachineInstr &MI) {
  if (MI.mayStore())
    return true;

  if (!MI.hasOneMemOperand() || !(*MI.memoperands_begin())->getPseudoValue())
    return true;

  if (const PseudoSourceValue *PSV =
      (*MI.memoperands_begin())->getPseudoValue()) {
    if (isa<FixedStackPseudoSourceValue>(PSV))
      return false;
    return !PSV->isConstant(nullptr) && !PSV->isStack();
  }

  return true;
}

MemDefsUses::MemDefsUses(const MachineFrameInfo *MFI_)
    : InspectMemInstr(false), MFI(MFI_) {}

bool MemDefsUses::hasHazard_(const MachineInstr &MI) {
  bool HasHazard = false;

  // Check underlying object list.
  SmallVector<ValueType, 4> Objs;
  if (getUnderlyingObjects(MI, Objs)) {
    for (ValueType VT : Objs)
      HasHazard |= updateDefsUses(VT, MI.mayStore());
    return HasHazard;
  }

  // No underlying objects found.
  HasHazard = MI.mayStore() && (OrigSeenLoad || OrigSeenStore);
  HasHazard |= MI.mayLoad() || OrigSeenStore;

  SeenNoObjLoad |= MI.mayLoad();
  SeenNoObjStore |= MI.mayStore();

  return HasHazard;
}

bool MemDefsUses::updateDefsUses(ValueType V, bool MayStore) {
  if (MayStore)
    return !Defs.insert(V).second || Uses.count(V) || SeenNoObjStore ||
           SeenNoObjLoad;

  Uses.insert(V);
  return Defs.count(V) || SeenNoObjStore;
}

bool MemDefsUses::
getUnderlyingObjects(const MachineInstr &MI,
                     SmallVectorImpl<ValueType> &Objects) const {
  if (!MI.hasOneMemOperand())
    return false;

  auto & MMO = **MI.memoperands_begin();

  if (const PseudoSourceValue *PSV = MMO.getPseudoValue()) {
    if (!PSV->isAliased(MFI))
      return false;
    Objects.push_back(PSV);
    return true;
  }

  if (const Value *V = MMO.getValue()) {
    SmallVector<const Value *, 4> Objs;
    ::getUnderlyingObjects(V, Objs);

    for (const Value *UValue : Objs) {
      if (!isIdentifiedObject(V))
        return false;

      Objects.push_back(UValue);
    }
    return true;
  }

  return false;
}

// Replace Branch with the compact branch instruction.
Iter MipsDelaySlotFiller::replaceWithCompactBranch(MachineBasicBlock &MBB,
                                                   Iter Branch,
                                                   const DebugLoc &DL) {
  const MipsSubtarget &STI = MBB.getParent()->getSubtarget<MipsSubtarget>();
  const MipsInstrInfo *TII = STI.getInstrInfo();

  unsigned NewOpcode = TII->getEquivalentCompactForm(Branch);
  Branch = TII->genInstrWithNewOpc(NewOpcode, Branch);

  auto *ToErase = cast<MachineInstr>(&*std::next(Branch));
  // Update call site info for the Branch.
  if (ToErase->shouldUpdateCallSiteInfo())
    ToErase->getMF()->moveCallSiteInfo(ToErase, cast<MachineInstr>(&*Branch));
  ToErase->eraseFromParent();
  return Branch;
}

// For given opcode returns opcode of corresponding instruction with short
// delay slot.
// For the pseudo TAILCALL*_MM instructions return the short delay slot
// form. Unfortunately, TAILCALL<->b16 is denied as b16 has a limited range
// that is too short to make use of for tail calls.
static int getEquivalentCallShort(int Opcode) {
  switch (Opcode) {
  case Mips::BGEZAL:
    return Mips::BGEZALS_MM;
  case Mips::BLTZAL:
    return Mips::BLTZALS_MM;
  case Mips::JAL:
  case Mips::JAL_MM:
    return Mips::JALS_MM;
  case Mips::JALR:
    return Mips::JALRS_MM;
  case Mips::JALR16_MM:
    return Mips::JALRS16_MM;
  case Mips::TAILCALL_MM:
    llvm_unreachable("Attempting to shorten the TAILCALL_MM pseudo!");
  case Mips::TAILCALLREG:
    return Mips::JR16_MM;
  default:
    llvm_unreachable("Unexpected call instruction for microMIPS.");
  }
}

/// runOnMachineBasicBlock - Fill in delay slots for the given basic block.
/// We assume there is only one delay slot per delayed instruction.
bool MipsDelaySlotFiller::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
  bool Changed = false;
  const MipsSubtarget &STI = MBB.getParent()->getSubtarget<MipsSubtarget>();
  bool InMicroMipsMode = STI.inMicroMipsMode();
  const MipsInstrInfo *TII = STI.getInstrInfo();

  for (Iter I = MBB.begin(); I != MBB.end(); ++I) {
    if (!hasUnoccupiedSlot(&*I))
      continue;

    // Delay slot filling is disabled at -O0, or in microMIPS32R6.
    if (!DisableDelaySlotFiller && (TM->getOptLevel() != CodeGenOpt::None) &&
        !(InMicroMipsMode && STI.hasMips32r6())) {

      bool Filled = false;

      if (MipsCompactBranchPolicy.getValue() != CB_Always ||
           !TII->getEquivalentCompactForm(I)) {
        if (searchBackward(MBB, *I)) {
          LLVM_DEBUG(dbgs() << DEBUG_TYPE ": found instruction for delay slot"
                                          " in backwards search.\n");
          Filled = true;
        } else if (I->isTerminator()) {
          if (searchSuccBBs(MBB, I)) {
            Filled = true;
            LLVM_DEBUG(dbgs() << DEBUG_TYPE ": found instruction for delay slot"
                                            " in successor BB search.\n");
          }
        } else if (searchForward(MBB, I)) {
          LLVM_DEBUG(dbgs() << DEBUG_TYPE ": found instruction for delay slot"
                                          " in forwards search.\n");
          Filled = true;
        }
      }

      if (Filled) {
        // Get instruction with delay slot.
        MachineBasicBlock::instr_iterator DSI = I.getInstrIterator();

        if (InMicroMipsMode && TII->getInstSizeInBytes(*std::next(DSI)) == 2 &&
            DSI->isCall()) {
          // If instruction in delay slot is 16b change opcode to
          // corresponding instruction with short delay slot.

          // TODO: Implement an instruction mapping table of 16bit opcodes to
          // 32bit opcodes so that an instruction can be expanded. This would
          // save 16 bits as a TAILCALL_MM pseudo requires a fullsized nop.
          // TODO: Permit b16 when branching backwards to the same function
          // if it is in range.
          DSI->setDesc(TII->get(getEquivalentCallShort(DSI->getOpcode())));
        }
        ++FilledSlots;
        Changed = true;
        continue;
      }
    }

    // For microMIPS if instruction is BEQ or BNE with one ZERO register, then
    // instead of adding NOP replace this instruction with the corresponding
    // compact branch instruction, i.e. BEQZC or BNEZC. Additionally
    // PseudoReturn and PseudoIndirectBranch are expanded to JR_MM, so they can
    // be replaced with JRC16_MM.

    // For MIPSR6 attempt to produce the corresponding compact (no delay slot)
    // form of the CTI. For indirect jumps this will not require inserting a
    // NOP and for branches will hopefully avoid requiring a NOP.
    if ((InMicroMipsMode ||
         (STI.hasMips32r6() && MipsCompactBranchPolicy != CB_Never)) &&
        TII->getEquivalentCompactForm(I)) {
      I = replaceWithCompactBranch(MBB, I, I->getDebugLoc());
      Changed = true;
      continue;
    }

    // Bundle the NOP to the instruction with the delay slot.
    LLVM_DEBUG(dbgs() << DEBUG_TYPE << ": could not fill delay slot for ";
               I->dump());
    TII->insertNop(MBB, std::next(I), I->getDebugLoc());
    MIBundleBuilder(MBB, I, std::next(I, 2));
    ++FilledSlots;
    Changed = true;
  }

  return Changed;
}

template <typename IterTy>
bool MipsDelaySlotFiller::searchRange(MachineBasicBlock &MBB, IterTy Begin,
                                      IterTy End, RegDefsUses &RegDU,
                                      InspectMemInstr &IM, Iter Slot,
                                      IterTy &Filler) const {
  for (IterTy I = Begin; I != End;) {
    IterTy CurrI = I;
    ++I;
    LLVM_DEBUG(dbgs() << DEBUG_TYPE ": checking instruction: "; CurrI->dump());
    // skip debug value
    if (CurrI->isDebugInstr()) {
      LLVM_DEBUG(dbgs() << DEBUG_TYPE ": ignoring debug instruction: ";
                 CurrI->dump());
      continue;
    }

    if (CurrI->isBundle()) {
      LLVM_DEBUG(dbgs() << DEBUG_TYPE ": ignoring BUNDLE instruction: ";
                 CurrI->dump());
      // However, we still need to update the register def-use information.
      RegDU.update(*CurrI, 0, CurrI->getNumOperands());
      continue;
    }

    if (terminateSearch(*CurrI)) {
      LLVM_DEBUG(dbgs() << DEBUG_TYPE ": should terminate search: ";
                 CurrI->dump());
      break;
    }

    assert((!CurrI->isCall() && !CurrI->isReturn() && !CurrI->isBranch()) &&
           "Cannot put calls, returns or branches in delay slot.");

    if (CurrI->isKill()) {
      CurrI->eraseFromParent();
      continue;
    }

    if (delayHasHazard(*CurrI, RegDU, IM))
      continue;

    const MipsSubtarget &STI = MBB.getParent()->getSubtarget<MipsSubtarget>();
    if (STI.isTargetNaCl()) {
      // In NaCl, instructions that must be masked are forbidden in delay slots.
      // We only check for loads, stores and SP changes.  Calls, returns and
      // branches are not checked because non-NaCl targets never put them in
      // delay slots.
      unsigned AddrIdx;
      if ((isBasePlusOffsetMemoryAccess(CurrI->getOpcode(), &AddrIdx) &&
           baseRegNeedsLoadStoreMask(CurrI->getOperand(AddrIdx).getReg())) ||
          CurrI->modifiesRegister(Mips::SP, STI.getRegisterInfo()))
        continue;
    }

    bool InMicroMipsMode = STI.inMicroMipsMode();
    const MipsInstrInfo *TII = STI.getInstrInfo();
    unsigned Opcode = (*Slot).getOpcode();
    // This is complicated by the tail call optimization. For non-PIC code
    // there is only a 32bit sized unconditional branch which can be assumed
    // to be able to reach the target. b16 only has a range of +/- 1 KB.
    // It's entirely possible that the target function is reachable with b16
    // but we don't have enough information to make that decision.
     if (InMicroMipsMode && TII->getInstSizeInBytes(*CurrI) == 2 &&
        (Opcode == Mips::JR || Opcode == Mips::PseudoIndirectBranch ||
         Opcode == Mips::PseudoIndirectBranch_MM ||
         Opcode == Mips::PseudoReturn || Opcode == Mips::TAILCALL))
      continue;
     // Instructions LWP/SWP and MOVEP should not be in a delay slot as that
     // results in unpredictable behaviour
     if (InMicroMipsMode && (Opcode == Mips::LWP_MM || Opcode == Mips::SWP_MM ||
                             Opcode == Mips::MOVEP_MM))
       continue;

    Filler = CurrI;
    LLVM_DEBUG(dbgs() << DEBUG_TYPE ": found instruction for delay slot: ";
               CurrI->dump());

    return true;
  }

  return false;
}

bool MipsDelaySlotFiller::searchBackward(MachineBasicBlock &MBB,
                                         MachineInstr &Slot) const {
  if (DisableBackwardSearch)
    return false;

  auto *Fn = MBB.getParent();
  RegDefsUses RegDU(*Fn->getSubtarget().getRegisterInfo());
  MemDefsUses MemDU(&Fn->getFrameInfo());
  ReverseIter Filler;

  RegDU.init(Slot);

  MachineBasicBlock::iterator SlotI = Slot;
  if (!searchRange(MBB, ++SlotI.getReverse(), MBB.rend(), RegDU, MemDU, Slot,
                   Filler)) {
    LLVM_DEBUG(dbgs() << DEBUG_TYPE ": could not find instruction for delay "
                                    "slot using backwards search.\n");
    return false;
  }

  MBB.splice(std::next(SlotI), &MBB, Filler.getReverse());
  MIBundleBuilder(MBB, SlotI, std::next(SlotI, 2));
  ++UsefulSlots;
  return true;
}

bool MipsDelaySlotFiller::searchForward(MachineBasicBlock &MBB,
                                        Iter Slot) const {
  // Can handle only calls.
  if (DisableForwardSearch || !Slot->isCall())
    return false;

  RegDefsUses RegDU(*MBB.getParent()->getSubtarget().getRegisterInfo());
  NoMemInstr NM;
  Iter Filler;

  RegDU.setCallerSaved(*Slot);

  if (!searchRange(MBB, std::next(Slot), MBB.end(), RegDU, NM, Slot, Filler)) {
    LLVM_DEBUG(dbgs() << DEBUG_TYPE ": could not find instruction for delay "
                                    "slot using forwards search.\n");
    return false;
  }

  MBB.splice(std::next(Slot), &MBB, Filler);
  MIBundleBuilder(MBB, Slot, std::next(Slot, 2));
  ++UsefulSlots;
  return true;
}

bool MipsDelaySlotFiller::searchSuccBBs(MachineBasicBlock &MBB,
                                        Iter Slot) const {
  if (DisableSuccBBSearch)
    return false;

  MachineBasicBlock *SuccBB = selectSuccBB(MBB);

  if (!SuccBB)
    return false;

  RegDefsUses RegDU(*MBB.getParent()->getSubtarget().getRegisterInfo());
  bool HasMultipleSuccs = false;
  BB2BrMap BrMap;
  std::unique_ptr<InspectMemInstr> IM;
  Iter Filler;
  auto *Fn = MBB.getParent();

  // Iterate over SuccBB's predecessor list.
  for (MachineBasicBlock *Pred : SuccBB->predecessors())
    if (!examinePred(*Pred, *SuccBB, RegDU, HasMultipleSuccs, BrMap))
      return false;

  // Do not allow moving instructions which have unallocatable register operands
  // across basic block boundaries.
  RegDU.setUnallocatableRegs(*Fn);

  // Only allow moving loads from stack or constants if any of the SuccBB's
  // predecessors have multiple successors.
  if (HasMultipleSuccs) {
    IM.reset(new LoadFromStackOrConst());
  } else {
    const MachineFrameInfo &MFI = Fn->getFrameInfo();
    IM.reset(new MemDefsUses(&MFI));
  }

  if (!searchRange(MBB, SuccBB->begin(), SuccBB->end(), RegDU, *IM, Slot,
                   Filler))
    return false;

  insertDelayFiller(Filler, BrMap);
  addLiveInRegs(Filler, *SuccBB);
  Filler->eraseFromParent();

  return true;
}

MachineBasicBlock *
MipsDelaySlotFiller::selectSuccBB(MachineBasicBlock &B) const {
  if (B.succ_empty())
    return nullptr;

  // Select the successor with the larget edge weight.
  auto &Prob = getAnalysis<MachineBranchProbabilityInfo>();
  MachineBasicBlock *S = *std::max_element(
      B.succ_begin(), B.succ_end(),
      [&](const MachineBasicBlock *Dst0, const MachineBasicBlock *Dst1) {
        return Prob.getEdgeProbability(&B, Dst0) <
               Prob.getEdgeProbability(&B, Dst1);
      });
  return S->isEHPad() ? nullptr : S;
}

std::pair<MipsInstrInfo::BranchType, MachineInstr *>
MipsDelaySlotFiller::getBranch(MachineBasicBlock &MBB,
                               const MachineBasicBlock &Dst) const {
  const MipsInstrInfo *TII =
      MBB.getParent()->getSubtarget<MipsSubtarget>().getInstrInfo();
  MachineBasicBlock *TrueBB = nullptr, *FalseBB = nullptr;
  SmallVector<MachineInstr*, 2> BranchInstrs;
  SmallVector<MachineOperand, 2> Cond;

  MipsInstrInfo::BranchType R =
      TII->analyzeBranch(MBB, TrueBB, FalseBB, Cond, false, BranchInstrs);

  if ((R == MipsInstrInfo::BT_None) || (R == MipsInstrInfo::BT_NoBranch))
    return std::make_pair(R, nullptr);

  if (R != MipsInstrInfo::BT_CondUncond) {
    if (!hasUnoccupiedSlot(BranchInstrs[0]))
      return std::make_pair(MipsInstrInfo::BT_None, nullptr);

    assert(((R != MipsInstrInfo::BT_Uncond) || (TrueBB == &Dst)));

    return std::make_pair(R, BranchInstrs[0]);
  }

  assert((TrueBB == &Dst) || (FalseBB == &Dst));

  // Examine the conditional branch. See if its slot is occupied.
  if (hasUnoccupiedSlot(BranchInstrs[0]))
    return std::make_pair(MipsInstrInfo::BT_Cond, BranchInstrs[0]);

  // If that fails, try the unconditional branch.
  if (hasUnoccupiedSlot(BranchInstrs[1]) && (FalseBB == &Dst))
    return std::make_pair(MipsInstrInfo::BT_Uncond, BranchInstrs[1]);

  return std::make_pair(MipsInstrInfo::BT_None, nullptr);
}

bool MipsDelaySlotFiller::examinePred(MachineBasicBlock &Pred,
                                      const MachineBasicBlock &Succ,
                                      RegDefsUses &RegDU,
                                      bool &HasMultipleSuccs,
                                      BB2BrMap &BrMap) const {
  std::pair<MipsInstrInfo::BranchType, MachineInstr *> P =
      getBranch(Pred, Succ);

  // Return if either getBranch wasn't able to analyze the branches or there
  // were no branches with unoccupied slots.
  if (P.first == MipsInstrInfo::BT_None)
    return false;

  if ((P.first != MipsInstrInfo::BT_Uncond) &&
      (P.first != MipsInstrInfo::BT_NoBranch)) {
    HasMultipleSuccs = true;
    RegDU.addLiveOut(Pred, Succ);
  }

  BrMap[&Pred] = P.second;
  return true;
}

bool MipsDelaySlotFiller::delayHasHazard(const MachineInstr &Candidate,
                                         RegDefsUses &RegDU,
                                         InspectMemInstr &IM) const {
  assert(!Candidate.isKill() &&
         "KILL instructions should have been eliminated at this point.");

  bool HasHazard = Candidate.isImplicitDef();

  HasHazard |= IM.hasHazard(Candidate);
  HasHazard |= RegDU.update(Candidate, 0, Candidate.getNumOperands());

  return HasHazard;
}

bool MipsDelaySlotFiller::terminateSearch(const MachineInstr &Candidate) const {
  return (Candidate.isTerminator() || Candidate.isCall() ||
          Candidate.isPosition() || Candidate.isInlineAsm() ||
          Candidate.hasUnmodeledSideEffects());
}

/// createMipsDelaySlotFillerPass - Returns a pass that fills in delay
/// slots in Mips MachineFunctions
FunctionPass *llvm::createMipsDelaySlotFillerPass() {
  return new MipsDelaySlotFiller();
}