Utils.cpp

Go to the documentation of this file.

//===- llvm/CodeGen/GlobalISel/Utils.cpp -------------------------*- C++ -*-==//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
/// \file This file implements the utility functions used by the GlobalISel
/// pipeline.
//===----------------------------------------------------------------------===//
 
#include "llvm/CodeGen/GlobalISel/Utils.h"
#include "llvm/ADT/APFloat.h"
#include "llvm/ADT/APInt.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/CodeGen/CodeGenCommonISel.h"
#include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
#include "llvm/CodeGen/GlobalISel/GISelValueTracking.h"
#include "llvm/CodeGen/GlobalISel/GenericMachineInstrs.h"
#include "llvm/CodeGen/GlobalISel/LostDebugLocObserver.h"
#include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
#include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
#include "llvm/CodeGen/LowLevelTypeUtils.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineOptimizationRemarkEmitter.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/MachineSizeOpts.h"
#include "llvm/CodeGen/RegisterBankInfo.h"
#include "llvm/CodeGen/StackProtector.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetLowering.h"
#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/CodeGen/TargetPassConfig.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/IR/Constants.h"
#include "llvm/Support/UndefPoison.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Transforms/Utils/SizeOpts.h"
#include <limits>
#include <numeric>
#include <optional>
#include <tuple>
 
#define DEBUG_TYPE "globalisel-utils"
 
using namespace llvm;
using namespace MIPatternMatch;
 
Register llvm::constrainRegToClass(MachineRegisterInfo &MRI,
                                   const TargetInstrInfo &TII,
                                   const RegisterBankInfo &RBI, Register Reg,
                                   const TargetRegisterClass &RegClass) {
  if (!RBI.constrainGenericRegister(Reg, RegClass, MRI))
    return MRI.createVirtualRegister(&RegClass);
 
  return Reg;
}
 
Register llvm::constrainOperandRegClass(
    const MachineFunction &MF, const TargetRegisterInfo &TRI,
    MachineRegisterInfo &MRI, const TargetInstrInfo &TII,
    const RegisterBankInfo &RBI, MachineInstr &InsertPt,
    const TargetRegisterClass &RegClass, MachineOperand &RegMO) {
  Register Reg = RegMO.getReg();
  // Assume physical registers are properly constrained.
  assert(Reg.isVirtual() && "PhysReg not implemented");
 
  // Save the old register class to check whether
  // the change notifications will be required.
  // TODO: A better approach would be to pass
  // the observers to constrainRegToClass().
  auto *OldRegClass = MRI.getRegClassOrNull(Reg);
  Register ConstrainedReg = constrainRegToClass(MRI, TII, RBI, Reg, RegClass);
  // If we created a new virtual register because the class is not compatible
  // then create a copy between the new and the old register.
  if (ConstrainedReg != Reg) {
    MachineBasicBlock::iterator InsertIt(&InsertPt);
    MachineBasicBlock &MBB = *InsertPt.getParent();
    // FIXME: The copy needs to have the classes constrained for its operands.
    // Use operand's regbank to get the class for old register (Reg).
    if (RegMO.isUse()) {
      BuildMI(MBB, InsertIt, InsertPt.getDebugLoc(),
              TII.get(TargetOpcode::COPY), ConstrainedReg)
          .addReg(Reg);
    } else {
      assert(RegMO.isDef() && "Must be a definition");
      BuildMI(MBB, std::next(InsertIt), InsertPt.getDebugLoc(),
              TII.get(TargetOpcode::COPY), Reg)
          .addReg(ConstrainedReg);
    }
    if (GISelChangeObserver *Observer = MF.getObserver()) {
      Observer->changingInstr(*RegMO.getParent());
    }
    RegMO.setReg(ConstrainedReg);
    if (GISelChangeObserver *Observer = MF.getObserver()) {
      Observer->changedInstr(*RegMO.getParent());
    }
  } else if (OldRegClass != MRI.getRegClassOrNull(Reg)) {
    if (GISelChangeObserver *Observer = MF.getObserver()) {
      if (!RegMO.isDef()) {
        MachineInstr *RegDef = MRI.getVRegDef(Reg);
        Observer->changedInstr(*RegDef);
      }
      Observer->changingAllUsesOfReg(MRI, Reg);
      Observer->finishedChangingAllUsesOfReg();
    }
  }
  return ConstrainedReg;
}
 
Register llvm::constrainOperandRegClass(
    const MachineFunction &MF, const TargetRegisterInfo &TRI,
    MachineRegisterInfo &MRI, const TargetInstrInfo &TII,
    const RegisterBankInfo &RBI, MachineInstr &InsertPt, const MCInstrDesc &II,
    MachineOperand &RegMO, unsigned OpIdx) {
  Register Reg = RegMO.getReg();
  // Assume physical registers are properly constrained.
  assert(Reg.isVirtual() && "PhysReg not implemented");
 
  const TargetRegisterClass *OpRC = TII.getRegClass(II, OpIdx);
  // Some of the target independent instructions, like COPY, may not impose any
  // register class constraints on some of their operands: If it's a use, we can
  // skip constraining as the instruction defining the register would constrain
  // it.
 
  if (OpRC) {
    // Obtain the RC from incoming regbank if it is a proper sub-class. Operands
    // can have multiple regbanks for a superclass that combine different
    // register types (E.g., AMDGPU's VGPR and AGPR). The regbank ambiguity
    // resolved by targets during regbankselect should not be overridden.
    if (const auto *SubRC = TRI.getCommonSubClass(
            OpRC, TRI.getConstrainedRegClassForOperand(RegMO, MRI)))
      OpRC = SubRC;
 
    OpRC = TRI.getAllocatableClass(OpRC);
  }
 
  if (!OpRC) {
    assert((!isTargetSpecificOpcode(II.getOpcode()) || RegMO.isUse()) &&
           "Register class constraint is required unless either the "
           "instruction is target independent or the operand is a use");
    // FIXME: Just bailing out like this here could be not enough, unless we
    // expect the users of this function to do the right thing for PHIs and
    // COPY:
    //   v1 = COPY v0
    //   v2 = COPY v1
    // v1 here may end up not being constrained at all. Please notice that to
    // reproduce the issue we likely need a destination pattern of a selection
    // rule producing such extra copies, not just an input GMIR with them as
    // every existing target using selectImpl handles copies before calling it
    // and they never reach this function.
    return Reg;
  }
  return constrainOperandRegClass(MF, TRI, MRI, TII, RBI, InsertPt, *OpRC,
                                  RegMO);
}
 
void llvm::constrainSelectedInstRegOperands(MachineInstr &I,
                                            const TargetInstrInfo &TII,
                                            const TargetRegisterInfo &TRI,
                                            const RegisterBankInfo &RBI) {
  assert(!isPreISelGenericOpcode(I.getOpcode()) &&
         "A selected instruction is expected");
  MachineBasicBlock &MBB = *I.getParent();
  MachineFunction &MF = *MBB.getParent();
  MachineRegisterInfo &MRI = MF.getRegInfo();
 
  for (unsigned OpI = 0, OpE = I.getNumExplicitOperands(); OpI != OpE; ++OpI) {
    MachineOperand &MO = I.getOperand(OpI);
 
    // There's nothing to be done on non-register operands.
    if (!MO.isReg())
      continue;
 
    LLVM_DEBUG(dbgs() << "Converting operand: " << MO << '\n');
 
    Register Reg = MO.getReg();
    // Physical registers don't need to be constrained.
    if (Reg.isPhysical())
      continue;
 
    // Register operands with a value of 0 (e.g. predicate operands) don't need
    // to be constrained.
    if (Reg == 0)
      continue;
 
    // If the operand is a vreg, we should constrain its regclass, and only
    // insert COPYs if that's impossible.
    // constrainOperandRegClass does that for us.
    constrainOperandRegClass(MF, TRI, MRI, TII, RBI, I, I.getDesc(), MO, OpI);
 
    // Tie uses to defs as indicated in MCInstrDesc if this hasn't already been
    // done.
    if (MO.isUse()) {
      int DefIdx = I.getDesc().getOperandConstraint(OpI, MCOI::TIED_TO);
      if (DefIdx != -1 && !I.isRegTiedToUseOperand(DefIdx))
        I.tieOperands(DefIdx, OpI);
    }
  }
}
 
bool llvm::canReplaceReg(Register DstReg, Register SrcReg,
                         MachineRegisterInfo &MRI) {
  // Give up if either DstReg or SrcReg  is a physical register.
  if (DstReg.isPhysical() || SrcReg.isPhysical())
    return false;
  // Give up if the types don't match.
  if (MRI.getType(DstReg) != MRI.getType(SrcReg))
    return false;
  // Replace if either DstReg has no constraints or the register
  // constraints match.
  const auto &DstRBC = MRI.getRegClassOrRegBank(DstReg);
  if (!DstRBC || DstRBC == MRI.getRegClassOrRegBank(SrcReg))
    return true;
 
  // Otherwise match if the Src is already a regclass that is covered by the Dst
  // RegBank.
  return isa<const RegisterBank *>(DstRBC) && MRI.getRegClassOrNull(SrcReg) &&
         cast<const RegisterBank *>(DstRBC)->covers(
             *MRI.getRegClassOrNull(SrcReg));
}
 
bool llvm::isTriviallyDead(const MachineInstr &MI,
                           const MachineRegisterInfo &MRI) {
  // Instructions without side-effects are dead iff they only define dead regs.
  // This function is hot and this loop returns early in the common case,
  // so only perform additional checks before this if absolutely necessary.
  for (const auto &MO : MI.all_defs()) {
    Register Reg = MO.getReg();
    if (Reg.isPhysical() || !MRI.use_nodbg_empty(Reg))
      return false;
  }
  return MI.wouldBeTriviallyDead();
}
 
static void reportGISelDiagnostic(DiagnosticSeverity Severity,
                                  MachineFunction &MF,
                                  MachineOptimizationRemarkEmitter &MORE,
                                  MachineOptimizationRemarkMissed &R) {
  bool IsGlobalISelAbortEnabled =
      MF.getTarget().Options.GlobalISelAbort == GlobalISelAbortMode::Enable;
  bool IsFatal = Severity == DS_Error && IsGlobalISelAbortEnabled;
  // Print the function name explicitly if we don't have a debug location (which
  // makes the diagnostic less useful) or if we're going to emit a raw error.
  if (!R.getLocation().isValid() || IsFatal)
    R << (" (in function: " + MF.getName() + ")").str();
 
  if (IsFatal)
    reportFatalUsageError(Twine(R.getMsg()));
  else
    MORE.emit(R);
}
 
void llvm::reportGISelWarning(MachineFunction &MF,
                              MachineOptimizationRemarkEmitter &MORE,
                              MachineOptimizationRemarkMissed &R) {
  reportGISelDiagnostic(DS_Warning, MF, MORE, R);
}
 
void llvm::reportGISelFailure(MachineFunction &MF,
                              MachineOptimizationRemarkEmitter &MORE,
                              MachineOptimizationRemarkMissed &R) {
  MF.getProperties().setFailedISel();
  reportGISelDiagnostic(DS_Error, MF, MORE, R);
}
 
void llvm::reportGISelFailure(MachineFunction &MF,
                              MachineOptimizationRemarkEmitter &MORE,
                              const char *PassName, StringRef Msg,
                              const MachineInstr &MI) {
  MachineOptimizationRemarkMissed R(PassName, "GISelFailure: ",
                                    MI.getDebugLoc(), MI.getParent());
  R << Msg;
  // Printing MI is expensive;  only do it if expensive remarks are enabled.
  if (MF.getTarget().Options.GlobalISelAbort == GlobalISelAbortMode::Enable ||
      MORE.allowExtraAnalysis(PassName))
    R << ": " << ore::MNV("Inst", MI);
  reportGISelFailure(MF, MORE, R);
}
 
unsigned llvm::getInverseGMinMaxOpcode(unsigned MinMaxOpc) {
  switch (MinMaxOpc) {
  case TargetOpcode::G_SMIN:
    return TargetOpcode::G_SMAX;
  case TargetOpcode::G_SMAX:
    return TargetOpcode::G_SMIN;
  case TargetOpcode::G_UMIN:
    return TargetOpcode::G_UMAX;
  case TargetOpcode::G_UMAX:
    return TargetOpcode::G_UMIN;
  default:
    llvm_unreachable("unrecognized opcode");
  }
}
 
std::optional<APInt> llvm::getIConstantVRegVal(Register VReg,
                                               const MachineRegisterInfo &MRI) {
  std::optional<ValueAndVReg> ValAndVReg = getIConstantVRegValWithLookThrough(
      VReg, MRI, /*LookThroughInstrs*/ false);
  assert((!ValAndVReg || ValAndVReg->VReg == VReg) &&
         "Value found while looking through instrs");
  if (!ValAndVReg)
    return std::nullopt;
  return ValAndVReg->Value;
}
 
const APInt &llvm::getIConstantFromReg(Register Reg,
                                       const MachineRegisterInfo &MRI) {
  MachineInstr *Const = MRI.getVRegDef(Reg);
  assert((Const && Const->getOpcode() == TargetOpcode::G_CONSTANT) &&
         "expected a G_CONSTANT on Reg");
  return Const->getOperand(1).getCImm()->getValue();
}
 
std::optional<int64_t>
llvm::getIConstantVRegSExtVal(Register VReg, const MachineRegisterInfo &MRI) {
  std::optional<APInt> Val = getIConstantVRegVal(VReg, MRI);
  if (Val && Val->getBitWidth() <= 64)
    return Val->getSExtValue();
  return std::nullopt;
}
 
namespace {
 
// This function is used in many places, and as such, it has some
// micro-optimizations to try and make it as fast as it can be.
//
// - We use template arguments to avoid an indirect call caused by passing a
// function_ref/std::function
// - GetAPCstValue does not return std::optional<APInt> as that's expensive.
// Instead it returns true/false and places the result in a pre-constructed
// APInt.
//
// Please change this function carefully and benchmark your changes.
template <bool (*IsConstantOpcode)(const MachineInstr *),
          bool (*GetAPCstValue)(const MachineInstr *MI, APInt &)>
std::optional<ValueAndVReg>
getConstantVRegValWithLookThrough(Register VReg, const MachineRegisterInfo &MRI,
                                  bool LookThroughInstrs = true,
                                  bool LookThroughAnyExt = false) {
  SmallVector<std::pair<unsigned, unsigned>, 4> SeenOpcodes;
  MachineInstr *MI;
 
  while ((MI = MRI.getVRegDef(VReg)) && !IsConstantOpcode(MI) &&
         LookThroughInstrs) {
    switch (MI->getOpcode()) {
    case TargetOpcode::G_ANYEXT:
      if (!LookThroughAnyExt)
        return std::nullopt;
      [[fallthrough]];
    case TargetOpcode::G_TRUNC:
    case TargetOpcode::G_SEXT:
    case TargetOpcode::G_ZEXT:
      SeenOpcodes.push_back(std::make_pair(
          MI->getOpcode(),
          MRI.getType(MI->getOperand(0).getReg()).getSizeInBits()));
      VReg = MI->getOperand(1).getReg();
      break;
    case TargetOpcode::COPY:
      VReg = MI->getOperand(1).getReg();
      if (VReg.isPhysical())
        return std::nullopt;
      break;
    case TargetOpcode::G_INTTOPTR:
      VReg = MI->getOperand(1).getReg();
      break;
    default:
      return std::nullopt;
    }
  }
  if (!MI || !IsConstantOpcode(MI))
    return std::nullopt;
 
  APInt Val;
  if (!GetAPCstValue(MI, Val))
    return std::nullopt;
  for (auto &Pair : reverse(SeenOpcodes)) {
    switch (Pair.first) {
    case TargetOpcode::G_TRUNC:
      Val = Val.trunc(Pair.second);
      break;
    case TargetOpcode::G_ANYEXT:
    case TargetOpcode::G_SEXT:
      Val = Val.sext(Pair.second);
      break;
    case TargetOpcode::G_ZEXT:
      Val = Val.zext(Pair.second);
      break;
    }
  }
 
  return ValueAndVReg{std::move(Val), VReg};
}
 
bool isIConstant(const MachineInstr *MI) {
  if (!MI)
    return false;
  return MI->getOpcode() == TargetOpcode::G_CONSTANT;
}
 
bool isFConstant(const MachineInstr *MI) {
  if (!MI)
    return false;
  return MI->getOpcode() == TargetOpcode::G_FCONSTANT;
}
 
bool isAnyConstant(const MachineInstr *MI) {
  if (!MI)
    return false;
  unsigned Opc = MI->getOpcode();
  return Opc == TargetOpcode::G_CONSTANT || Opc == TargetOpcode::G_FCONSTANT;
}
 
bool getCImmAsAPInt(const MachineInstr *MI, APInt &Result) {
  const MachineOperand &CstVal = MI->getOperand(1);
  if (!CstVal.isCImm())
    return false;
  Result = CstVal.getCImm()->getValue();
  return true;
}
 
bool getCImmOrFPImmAsAPInt(const MachineInstr *MI, APInt &Result) {
  const MachineOperand &CstVal = MI->getOperand(1);
  if (CstVal.isCImm())
    Result = CstVal.getCImm()->getValue();
  else if (CstVal.isFPImm())
    Result = CstVal.getFPImm()->getValueAPF().bitcastToAPInt();
  else
    return false;
  return true;
}
 
} // end anonymous namespace
 
std::optional<ValueAndVReg> llvm::getIConstantVRegValWithLookThrough(
    Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs) {
  return getConstantVRegValWithLookThrough<isIConstant, getCImmAsAPInt>(
      VReg, MRI, LookThroughInstrs);
}
 
std::optional<ValueAndVReg> llvm::getAnyConstantVRegValWithLookThrough(
    Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs,
    bool LookThroughAnyExt) {
  return getConstantVRegValWithLookThrough<isAnyConstant,
                                           getCImmOrFPImmAsAPInt>(
      VReg, MRI, LookThroughInstrs, LookThroughAnyExt);
}
 
std::optional<FPValueAndVReg> llvm::getFConstantVRegValWithLookThrough(
    Register VReg, const MachineRegisterInfo &MRI, bool LookThroughInstrs) {
  auto Reg =
      getConstantVRegValWithLookThrough<isFConstant, getCImmOrFPImmAsAPInt>(
          VReg, MRI, LookThroughInstrs);
  if (!Reg)
    return std::nullopt;
 
  APFloat FloatVal(getFltSemanticForLLT(LLT::scalar(Reg->Value.getBitWidth())),
                   Reg->Value);
  return FPValueAndVReg{FloatVal, Reg->VReg};
}
 
const ConstantFP *
llvm::getConstantFPVRegVal(Register VReg, const MachineRegisterInfo &MRI) {
  MachineInstr *MI = MRI.getVRegDef(VReg);
  if (TargetOpcode::G_FCONSTANT != MI->getOpcode())
    return nullptr;
  return MI->getOperand(1).getFPImm();
}
 
std::optional<DefinitionAndSourceRegister>
llvm::getDefSrcRegIgnoringCopies(Register Reg, const MachineRegisterInfo &MRI) {
  Register DefSrcReg = Reg;
  // This assumes that the code is in SSA form, so there should only be one
  // definition.
  auto DefIt = MRI.def_begin(Reg);
  if (DefIt == MRI.def_end())
    return {};
  MachineOperand &DefOpnd = *DefIt;
  MachineInstr *DefMI = DefOpnd.getParent();
  auto DstTy = MRI.getType(DefOpnd.getReg());
  if (!DstTy.isValid())
    return std::nullopt;
  unsigned Opc = DefMI->getOpcode();
  while (Opc == TargetOpcode::COPY || isPreISelGenericOptimizationHint(Opc)) {
    Register SrcReg = DefMI->getOperand(1).getReg();
    auto SrcTy = MRI.getType(SrcReg);
    if (!SrcTy.isValid())
      break;
    DefMI = MRI.getVRegDef(SrcReg);
    DefSrcReg = SrcReg;
    Opc = DefMI->getOpcode();
  }
  return DefinitionAndSourceRegister{DefMI, DefSrcReg};
}
 
MachineInstr *llvm::getDefIgnoringCopies(Register Reg,
                                         const MachineRegisterInfo &MRI) {
  std::optional<DefinitionAndSourceRegister> DefSrcReg =
      getDefSrcRegIgnoringCopies(Reg, MRI);
  return DefSrcReg ? DefSrcReg->MI : nullptr;
}
 
Register llvm::getSrcRegIgnoringCopies(Register Reg,
                                       const MachineRegisterInfo &MRI) {
  std::optional<DefinitionAndSourceRegister> DefSrcReg =
      getDefSrcRegIgnoringCopies(Reg, MRI);
  return DefSrcReg ? DefSrcReg->Reg : Register();
}
 
void llvm::extractParts(Register Reg, LLT Ty, int NumParts,
                        SmallVectorImpl<Register> &VRegs,
                        MachineIRBuilder &MIRBuilder,
                        MachineRegisterInfo &MRI) {
  for (int i = 0; i < NumParts; ++i)
    VRegs.push_back(MRI.createGenericVirtualRegister(Ty));
  MIRBuilder.buildUnmerge(VRegs, Reg);
}
 
bool llvm::extractParts(Register Reg, LLT RegTy, LLT MainTy, LLT &LeftoverTy,
                        SmallVectorImpl<Register> &VRegs,
                        SmallVectorImpl<Register> &LeftoverRegs,
                        MachineIRBuilder &MIRBuilder,
                        MachineRegisterInfo &MRI) {
  assert(!LeftoverTy.isValid() && "this is an out argument");
 
  unsigned RegSize = RegTy.getSizeInBits();
  unsigned MainSize = MainTy.getSizeInBits();
  unsigned NumParts = RegSize / MainSize;
  unsigned LeftoverSize = RegSize - NumParts * MainSize;
 
  // Use an unmerge when possible.
  if (LeftoverSize == 0) {
    for (unsigned I = 0; I < NumParts; ++I)
      VRegs.push_back(MRI.createGenericVirtualRegister(MainTy));
    MIRBuilder.buildUnmerge(VRegs, Reg);
    return true;
  }
 
  // Try to use unmerge for irregular vector split where possible
  // For example when splitting a <6 x i32> into <4 x i32> with <2 x i32>
  // leftover, it becomes:
  //  <2 x i32> %2, <2 x i32>%3, <2 x i32> %4 = G_UNMERGE_VALUE <6 x i32> %1
  //  <4 x i32> %5 = G_CONCAT_VECTOR <2 x i32> %2, <2 x i32> %3
  if (RegTy.isVector() && MainTy.isVector()) {
    unsigned RegNumElts = RegTy.getNumElements();
    unsigned MainNumElts = MainTy.getNumElements();
    unsigned LeftoverNumElts = RegNumElts % MainNumElts;
    // If can unmerge to LeftoverTy, do it
    if (MainNumElts % LeftoverNumElts == 0 &&
        RegNumElts % LeftoverNumElts == 0 &&
        RegTy.getScalarSizeInBits() == MainTy.getScalarSizeInBits() &&
        LeftoverNumElts > 1) {
      LeftoverTy = LLT::fixed_vector(LeftoverNumElts, RegTy.getElementType());
 
      // Unmerge the SrcReg to LeftoverTy vectors
      SmallVector<Register, 4> UnmergeValues;
      extractParts(Reg, LeftoverTy, RegNumElts / LeftoverNumElts, UnmergeValues,
                   MIRBuilder, MRI);
 
      // Find how many LeftoverTy makes one MainTy
      unsigned LeftoverPerMain = MainNumElts / LeftoverNumElts;
      unsigned NumOfLeftoverVal =
          ((RegNumElts % MainNumElts) / LeftoverNumElts);
 
      // Create as many MainTy as possible using unmerged value
      SmallVector<Register, 4> MergeValues;
      for (unsigned I = 0; I < UnmergeValues.size() - NumOfLeftoverVal; I++) {
        MergeValues.push_back(UnmergeValues[I]);
        if (MergeValues.size() == LeftoverPerMain) {
          VRegs.push_back(
              MIRBuilder.buildMergeLikeInstr(MainTy, MergeValues).getReg(0));
          MergeValues.clear();
        }
      }
      // Populate LeftoverRegs with the leftovers
      for (unsigned I = UnmergeValues.size() - NumOfLeftoverVal;
           I < UnmergeValues.size(); I++) {
        LeftoverRegs.push_back(UnmergeValues[I]);
      }
      return true;
    }
  }
  // Perform irregular split. Leftover is last element of RegPieces.
  if (MainTy.isVector()) {
    SmallVector<Register, 8> RegPieces;
    extractVectorParts(Reg, MainTy.getNumElements(), RegPieces, MIRBuilder,
                       MRI);
    for (unsigned i = 0; i < RegPieces.size() - 1; ++i)
      VRegs.push_back(RegPieces[i]);
    LeftoverRegs.push_back(RegPieces[RegPieces.size() - 1]);
    LeftoverTy = MRI.getType(LeftoverRegs[0]);
    return true;
  }
 
  LeftoverTy = LLT::integer(LeftoverSize);
  // For irregular sizes, extract the individual parts.
  for (unsigned I = 0; I != NumParts; ++I) {
    Register NewReg = MRI.createGenericVirtualRegister(MainTy);
    VRegs.push_back(NewReg);
    MIRBuilder.buildExtract(NewReg, Reg, MainSize * I);
  }
 
  for (unsigned Offset = MainSize * NumParts; Offset < RegSize;
       Offset += LeftoverSize) {
    Register NewReg = MRI.createGenericVirtualRegister(LeftoverTy);
    LeftoverRegs.push_back(NewReg);
    MIRBuilder.buildExtract(NewReg, Reg, Offset);
  }
 
  return true;
}
 
void llvm::extractVectorParts(Register Reg, unsigned NumElts,
                              SmallVectorImpl<Register> &VRegs,
                              MachineIRBuilder &MIRBuilder,
                              MachineRegisterInfo &MRI) {
  LLT RegTy = MRI.getType(Reg);
  assert(RegTy.isVector() && "Expected a vector type");
 
  LLT EltTy = RegTy.getElementType();
  LLT NarrowTy = (NumElts == 1) ? EltTy : LLT::fixed_vector(NumElts, EltTy);
  unsigned RegNumElts = RegTy.getNumElements();
  unsigned LeftoverNumElts = RegNumElts % NumElts;
  unsigned NumNarrowTyPieces = RegNumElts / NumElts;
 
  // Perfect split without leftover
  if (LeftoverNumElts == 0)
    return extractParts(Reg, NarrowTy, NumNarrowTyPieces, VRegs, MIRBuilder,
                        MRI);
 
  // Irregular split. Provide direct access to all elements for artifact
  // combiner using unmerge to elements. Then build vectors with NumElts
  // elements. Remaining element(s) will be (used to build vector) Leftover.
  SmallVector<Register, 8> Elts;
  extractParts(Reg, EltTy, RegNumElts, Elts, MIRBuilder, MRI);
 
  unsigned Offset = 0;
  // Requested sub-vectors of NarrowTy.
  for (unsigned i = 0; i < NumNarrowTyPieces; ++i, Offset += NumElts) {
    ArrayRef<Register> Pieces(&Elts[Offset], NumElts);
    VRegs.push_back(MIRBuilder.buildMergeLikeInstr(NarrowTy, Pieces).getReg(0));
  }
 
  // Leftover element(s).
  if (LeftoverNumElts == 1) {
    VRegs.push_back(Elts[Offset]);
  } else {
    LLT LeftoverTy = LLT::fixed_vector(LeftoverNumElts, EltTy);
    ArrayRef<Register> Pieces(&Elts[Offset], LeftoverNumElts);
    VRegs.push_back(
        MIRBuilder.buildMergeLikeInstr(LeftoverTy, Pieces).getReg(0));
  }
}
 
MachineInstr *llvm::getOpcodeDef(unsigned Opcode, Register Reg,
                                 const MachineRegisterInfo &MRI) {
  MachineInstr *DefMI = getDefIgnoringCopies(Reg, MRI);
  return DefMI && DefMI->getOpcode() == Opcode ? DefMI : nullptr;
}
 
APFloat llvm::getAPFloatFromSize(double Val, unsigned Size) {
  if (Size == 32)
    return APFloat(float(Val));
  if (Size == 64)
    return APFloat(Val);
  if (Size != 16)
    llvm_unreachable("Unsupported FPConstant size");
  bool Ignored;
  APFloat APF(Val);
  APF.convert(APFloat::IEEEhalf(), APFloat::rmNearestTiesToEven, &Ignored);
  return APF;
}
 
std::optional<APInt> llvm::ConstantFoldBinOp(unsigned Opcode,
                                             const Register Op1,
                                             const Register Op2,
                                             const MachineRegisterInfo &MRI) {
  auto MaybeOp2Cst = getAnyConstantVRegValWithLookThrough(Op2, MRI, false);
  if (!MaybeOp2Cst)
    return std::nullopt;
 
  auto MaybeOp1Cst = getAnyConstantVRegValWithLookThrough(Op1, MRI, false);
  if (!MaybeOp1Cst)
    return std::nullopt;
 
  const APInt &C1 = MaybeOp1Cst->Value;
  const APInt &C2 = MaybeOp2Cst->Value;
  switch (Opcode) {
  default:
    break;
  case TargetOpcode::G_ADD:
    return C1 + C2;
  case TargetOpcode::G_PTR_ADD:
    // Types can be of different width here.
    // Result needs to be the same width as C1, so trunc or sext C2.
    return C1 + C2.sextOrTrunc(C1.getBitWidth());
  case TargetOpcode::G_AND:
    return C1 & C2;
  case TargetOpcode::G_ASHR:
    return C1.ashr(C2);
  case TargetOpcode::G_LSHR:
    return C1.lshr(C2);
  case TargetOpcode::G_MUL:
    return C1 * C2;
  case TargetOpcode::G_OR:
    return C1 | C2;
  case TargetOpcode::G_SHL:
    return C1 << C2;
  case TargetOpcode::G_SUB:
    return C1 - C2;
  case TargetOpcode::G_XOR:
    return C1 ^ C2;
  case TargetOpcode::G_UDIV:
    if (!C2.getBoolValue())
      break;
    return C1.udiv(C2);
  case TargetOpcode::G_SDIV:
    if (!C2.getBoolValue())
      break;
    return C1.sdiv(C2);
  case TargetOpcode::G_UREM:
    if (!C2.getBoolValue())
      break;
    return C1.urem(C2);
  case TargetOpcode::G_SREM:
    if (!C2.getBoolValue())
      break;
    return C1.srem(C2);
  case TargetOpcode::G_SMIN:
    return APIntOps::smin(C1, C2);
  case TargetOpcode::G_SMAX:
    return APIntOps::smax(C1, C2);
  case TargetOpcode::G_UMIN:
    return APIntOps::umin(C1, C2);
  case TargetOpcode::G_UMAX:
    return APIntOps::umax(C1, C2);
  }
 
  return std::nullopt;
}
 
std::optional<APFloat>
llvm::ConstantFoldFPBinOp(unsigned Opcode, const Register Op1,
                          const Register Op2, const MachineRegisterInfo &MRI) {
  const ConstantFP *Op2Cst = getConstantFPVRegVal(Op2, MRI);
  if (!Op2Cst)
    return std::nullopt;
 
  const ConstantFP *Op1Cst = getConstantFPVRegVal(Op1, MRI);
  if (!Op1Cst)
    return std::nullopt;
 
  APFloat C1 = Op1Cst->getValueAPF();
  const APFloat &C2 = Op2Cst->getValueAPF();
  switch (Opcode) {
  case TargetOpcode::G_FADD:
    C1.add(C2, APFloat::rmNearestTiesToEven);
    return C1;
  case TargetOpcode::G_FSUB:
    C1.subtract(C2, APFloat::rmNearestTiesToEven);
    return C1;
  case TargetOpcode::G_FMUL:
    C1.multiply(C2, APFloat::rmNearestTiesToEven);
    return C1;
  case TargetOpcode::G_FDIV:
    C1.divide(C2, APFloat::rmNearestTiesToEven);
    return C1;
  case TargetOpcode::G_FREM:
    C1.mod(C2);
    return C1;
  case TargetOpcode::G_FCOPYSIGN:
    C1.copySign(C2);
    return C1;
  case TargetOpcode::G_FMINNUM:
    return minnum(C1, C2);
  case TargetOpcode::G_FMAXNUM:
    return maxnum(C1, C2);
  case TargetOpcode::G_FMINIMUM:
    return minimum(C1, C2);
  case TargetOpcode::G_FMAXIMUM:
    return maximum(C1, C2);
  case TargetOpcode::G_FMINIMUMNUM:
    return minimumnum(C1, C2);
  case TargetOpcode::G_FMAXIMUMNUM:
    return maximumnum(C1, C2);
  case TargetOpcode::G_FMINNUM_IEEE:
  case TargetOpcode::G_FMAXNUM_IEEE:
    // FIXME: These operations were unfortunately named. fminnum/fmaxnum do not
    // follow the IEEE behavior for signaling nans and follow libm's fmin/fmax,
    // and currently there isn't a nice wrapper in APFloat for the version with
    // correct snan handling.
    break;
  default:
    break;
  }
 
  return std::nullopt;
}
 
static GBuildVector *getBuildVectorLikeDef(Register Reg,
                                           const MachineRegisterInfo &MRI) {
  if (auto *BV = getOpcodeDef<GBuildVector>(Reg, MRI))
    return BV;
 
  auto *Bitcast = getOpcodeDef(TargetOpcode::G_BITCAST, Reg, MRI);
  if (!Bitcast)
    return nullptr;
 
  auto [Dst, DstTy, Src, SrcTy] = Bitcast->getFirst2RegLLTs();
  if (!SrcTy.isVector() || !DstTy.isVector())
    return nullptr;
  if (SrcTy.getElementCount() != DstTy.getElementCount())
    return nullptr;
  if (SrcTy.getScalarSizeInBits() != DstTy.getScalarSizeInBits())
    return nullptr;
 
  return getOpcodeDef<GBuildVector>(Src, MRI);
}
 
SmallVector<APInt>
llvm::ConstantFoldVectorBinop(unsigned Opcode, const Register Op1,
                              const Register Op2,
                              const MachineRegisterInfo &MRI) {
  auto *SrcVec2 = getBuildVectorLikeDef(Op2, MRI);
  if (!SrcVec2)
    return SmallVector<APInt>();
 
  auto *SrcVec1 = getBuildVectorLikeDef(Op1, MRI);
  if (!SrcVec1)
    return SmallVector<APInt>();
 
  SmallVector<APInt> FoldedElements;
  for (unsigned Idx = 0, E = SrcVec1->getNumSources(); Idx < E; ++Idx) {
    auto MaybeCst = ConstantFoldBinOp(Opcode, SrcVec1->getSourceReg(Idx),
                                      SrcVec2->getSourceReg(Idx), MRI);
    if (!MaybeCst)
      return SmallVector<APInt>();
    FoldedElements.push_back(*MaybeCst);
  }
  return FoldedElements;
}
 
Align llvm::inferAlignFromPtrInfo(MachineFunction &MF,
                                  const MachinePointerInfo &MPO) {
  auto PSV = dyn_cast_if_present<const PseudoSourceValue *>(MPO.V);
  if (auto FSPV = dyn_cast_or_null<FixedStackPseudoSourceValue>(PSV)) {
    MachineFrameInfo &MFI = MF.getFrameInfo();
    return commonAlignment(MFI.getObjectAlign(FSPV->getFrameIndex()),
                           MPO.Offset);
  }
 
  if (const Value *V = dyn_cast_if_present<const Value *>(MPO.V)) {
    const Module *M = MF.getFunction().getParent();
    return V->getPointerAlignment(M->getDataLayout());
  }
 
  return Align(1);
}
 
Register llvm::getFunctionLiveInPhysReg(MachineFunction &MF,
                                        const TargetInstrInfo &TII,
                                        MCRegister PhysReg,
                                        const TargetRegisterClass &RC,
                                        const DebugLoc &DL, LLT RegTy) {
  MachineBasicBlock &EntryMBB = MF.front();
  MachineRegisterInfo &MRI = MF.getRegInfo();
  Register LiveIn = MRI.getLiveInVirtReg(PhysReg);
  if (LiveIn) {
    MachineInstr *Def = MRI.getVRegDef(LiveIn);
    if (Def) {
      // FIXME: Should the verifier check this is in the entry block?
      assert(Def->getParent() == &EntryMBB && "live-in copy not in entry block");
      return LiveIn;
    }
 
    // It's possible the incoming argument register and copy was added during
    // lowering, but later deleted due to being/becoming dead. If this happens,
    // re-insert the copy.
  } else {
    // The live in register was not present, so add it.
    LiveIn = MF.addLiveIn(PhysReg, &RC);
    if (RegTy.isValid())
      MRI.setType(LiveIn, RegTy);
  }
 
  BuildMI(EntryMBB, EntryMBB.begin(), DL, TII.get(TargetOpcode::COPY), LiveIn)
    .addReg(PhysReg);
  if (!EntryMBB.isLiveIn(PhysReg))
    EntryMBB.addLiveIn(PhysReg);
  return LiveIn;
}
 
std::optional<APInt> llvm::ConstantFoldExtOp(unsigned Opcode,
                                             const Register Op1, uint64_t Imm,
                                             const MachineRegisterInfo &MRI) {
  auto MaybeOp1Cst = getIConstantVRegVal(Op1, MRI);
  if (MaybeOp1Cst) {
    switch (Opcode) {
    default:
      break;
    case TargetOpcode::G_SEXT_INREG: {
      LLT Ty = MRI.getType(Op1);
      return MaybeOp1Cst->trunc(Imm).sext(Ty.getScalarSizeInBits());
    }
    }
  }
  return std::nullopt;
}
 
std::optional<APInt> llvm::ConstantFoldCastOp(unsigned Opcode, LLT DstTy,
                                              const Register Op0,
                                              const MachineRegisterInfo &MRI) {
  std::optional<APInt> Val = getIConstantVRegVal(Op0, MRI);
  if (!Val)
    return Val;
 
  const unsigned DstSize = DstTy.getScalarSizeInBits();
 
  switch (Opcode) {
  case TargetOpcode::G_SEXT:
    return Val->sext(DstSize);
  case TargetOpcode::G_ZEXT:
  case TargetOpcode::G_ANYEXT:
    // TODO: DAG considers target preference when constant folding any_extend.
    return Val->zext(DstSize);
  default:
    break;
  }
 
  llvm_unreachable("unexpected cast opcode to constant fold");
}
 
std::optional<APFloat>
llvm::ConstantFoldIntToFloat(unsigned Opcode, LLT DstTy, Register Src,
                             const MachineRegisterInfo &MRI) {
  assert(Opcode == TargetOpcode::G_SITOFP || Opcode == TargetOpcode::G_UITOFP);
  if (auto MaybeSrcVal = getIConstantVRegVal(Src, MRI)) {
    APFloat DstVal(getFltSemanticForLLT(DstTy));
    DstVal.convertFromAPInt(*MaybeSrcVal, Opcode == TargetOpcode::G_SITOFP,
                            APFloat::rmNearestTiesToEven);
    return DstVal;
  }
  return std::nullopt;
}
 
SmallVector<APInt>
llvm::ConstantFoldUnaryIntOp(unsigned Opcode, LLT DstTy, Register Src,
                             const MachineRegisterInfo &MRI) {
  unsigned EltBits = DstTy.getScalarSizeInBits();
  auto Fold = [Opcode, EltBits](const APInt &V) -> APInt {
    switch (Opcode) {
    case TargetOpcode::G_CTLZ:
    case TargetOpcode::G_CTLZ_ZERO_POISON:
      return APInt(EltBits, V.countl_zero());
    case TargetOpcode::G_CTTZ:
    case TargetOpcode::G_CTTZ_ZERO_POISON:
      return APInt(EltBits, V.countr_zero());
    case TargetOpcode::G_CTPOP:
      return APInt(EltBits, V.popcount());
    case TargetOpcode::G_ABS:
      return V.abs();
    case TargetOpcode::G_BSWAP:
      return V.byteSwap();
    case TargetOpcode::G_BITREVERSE:
      return V.reverseBits();
    }
    llvm_unreachable("unexpected opcode in ConstantFoldUnaryIntOp");
  };
 
  auto tryFoldScalar = [&](Register R) -> std::optional<APInt> {
    if (auto MaybeCst = getIConstantVRegVal(R, MRI))
      return Fold(*MaybeCst);
    return std::nullopt;
  };
  if (MRI.getType(Src).isVector()) {
    auto *BV = getOpcodeDef<GBuildVector>(Src, MRI);
    if (!BV)
      return {};
    SmallVector<APInt> Folded;
    for (unsigned SrcIdx = 0; SrcIdx < BV->getNumSources(); ++SrcIdx) {
      if (auto MaybeFold = tryFoldScalar(BV->getSourceReg(SrcIdx))) {
        Folded.emplace_back(std::move(*MaybeFold));
        continue;
      }
      return {};
    }
    return Folded;
  }
  if (auto MaybeCst = tryFoldScalar(Src))
    return {std::move(*MaybeCst)};
  return {};
}
 
std::optional<SmallVector<APInt>>
llvm::ConstantFoldICmp(unsigned Pred, const Register Op1, const Register Op2,
                       unsigned DstScalarSizeInBits, unsigned ExtOp,
                       const MachineRegisterInfo &MRI) {
  assert(ExtOp == TargetOpcode::G_SEXT || ExtOp == TargetOpcode::G_ZEXT ||
         ExtOp == TargetOpcode::G_ANYEXT);
 
  const LLT Ty = MRI.getType(Op1);
 
  auto GetICmpResultCst = [&](bool IsTrue) {
    if (IsTrue)
      return ExtOp == TargetOpcode::G_SEXT
                 ? APInt::getAllOnes(DstScalarSizeInBits)
                 : APInt::getOneBitSet(DstScalarSizeInBits, 0);
    return APInt::getZero(DstScalarSizeInBits);
  };
 
  auto TryFoldScalar = [&](Register LHS, Register RHS) -> std::optional<APInt> {
    auto RHSCst = getIConstantVRegVal(RHS, MRI);
    if (!RHSCst)
      return std::nullopt;
    auto LHSCst = getIConstantVRegVal(LHS, MRI);
    if (!LHSCst)
      return std::nullopt;
 
    switch (Pred) {
    case CmpInst::Predicate::ICMP_EQ:
      return GetICmpResultCst(LHSCst->eq(*RHSCst));
    case CmpInst::Predicate::ICMP_NE:
      return GetICmpResultCst(LHSCst->ne(*RHSCst));
    case CmpInst::Predicate::ICMP_UGT:
      return GetICmpResultCst(LHSCst->ugt(*RHSCst));
    case CmpInst::Predicate::ICMP_UGE:
      return GetICmpResultCst(LHSCst->uge(*RHSCst));
    case CmpInst::Predicate::ICMP_ULT:
      return GetICmpResultCst(LHSCst->ult(*RHSCst));
    case CmpInst::Predicate::ICMP_ULE:
      return GetICmpResultCst(LHSCst->ule(*RHSCst));
    case CmpInst::Predicate::ICMP_SGT:
      return GetICmpResultCst(LHSCst->sgt(*RHSCst));
    case CmpInst::Predicate::ICMP_SGE:
      return GetICmpResultCst(LHSCst->sge(*RHSCst));
    case CmpInst::Predicate::ICMP_SLT:
      return GetICmpResultCst(LHSCst->slt(*RHSCst));
    case CmpInst::Predicate::ICMP_SLE:
      return GetICmpResultCst(LHSCst->sle(*RHSCst));
    default:
      return std::nullopt;
    }
  };
 
  SmallVector<APInt> FoldedICmps;
 
  if (Ty.isVector()) {
    // Try to constant fold each element.
    auto *BV1 = getOpcodeDef<GBuildVector>(Op1, MRI);
    auto *BV2 = getOpcodeDef<GBuildVector>(Op2, MRI);
    if (!BV1 || !BV2)
      return std::nullopt;
    assert(BV1->getNumSources() == BV2->getNumSources() && "Invalid vectors");
    for (unsigned I = 0; I < BV1->getNumSources(); ++I) {
      if (auto MaybeFold =
              TryFoldScalar(BV1->getSourceReg(I), BV2->getSourceReg(I))) {
        FoldedICmps.emplace_back(*MaybeFold);
        continue;
      }
      return std::nullopt;
    }
    return FoldedICmps;
  }
 
  if (auto MaybeCst = TryFoldScalar(Op1, Op2)) {
    FoldedICmps.emplace_back(*MaybeCst);
    return FoldedICmps;
  }
 
  return std::nullopt;
}
 
bool llvm::isKnownToBeAPowerOfTwo(Register Reg, const MachineRegisterInfo &MRI,
                                  GISelValueTracking *VT, bool OrNegative) {
  std::optional<DefinitionAndSourceRegister> DefSrcReg =
      getDefSrcRegIgnoringCopies(Reg, MRI);
  if (!DefSrcReg)
    return false;
 
  const MachineInstr &MI = *DefSrcReg->MI;
  const LLT Ty = MRI.getType(Reg);
 
  auto IsPow2 = [OrNegative](const APInt &V) {
    return V.isPowerOf2() || (OrNegative && V.isNegatedPowerOf2());
  };
 
  switch (MI.getOpcode()) {
  case TargetOpcode::G_CONSTANT: {
    unsigned BitWidth = Ty.getScalarSizeInBits();
    const ConstantInt *CI = MI.getOperand(1).getCImm();
    return IsPow2(CI->getValue().zextOrTrunc(BitWidth));
  }
  case TargetOpcode::G_SHL: {
    // A left-shift of a constant one will have exactly one bit set because
    // shifting the bit off the end is undefined.
 
    // TODO: Constant splat
    if (auto ConstLHS = getIConstantVRegVal(MI.getOperand(1).getReg(), MRI)) {
      if (*ConstLHS == 1)
        return true;
    }
 
    break;
  }
  case TargetOpcode::G_LSHR: {
    if (auto ConstLHS = getIConstantVRegVal(MI.getOperand(1).getReg(), MRI)) {
      if (ConstLHS->isSignMask())
        return true;
    }
 
    break;
  }
  case TargetOpcode::G_BUILD_VECTOR: {
    // TODO: Probably should have a recursion depth guard since you could have
    // bitcasted vector elements.
    for (const MachineOperand &MO : llvm::drop_begin(MI.operands()))
      if (!isKnownToBeAPowerOfTwo(MO.getReg(), MRI, VT, OrNegative))
        return false;
 
    return true;
  }
  case TargetOpcode::G_BUILD_VECTOR_TRUNC: {
    // Only handle constants since we would need to know if number of leading
    // zeros is greater than the truncation amount.
    const unsigned BitWidth = Ty.getScalarSizeInBits();
    for (const MachineOperand &MO : llvm::drop_begin(MI.operands())) {
      auto Const = getIConstantVRegVal(MO.getReg(), MRI);
      if (!Const || !IsPow2(Const->zextOrTrunc(BitWidth)))
        return false;
    }
 
    return true;
  }
  default:
    break;
  }
 
  if (!VT)
    return false;
 
  // More could be done here, though the above checks are enough
  // to handle some common cases.
 
  // Fall back to computeKnownBits to catch other known cases.
  KnownBits Known = VT->getKnownBits(Reg);
  return (Known.countMaxPopulation() == 1) && (Known.countMinPopulation() == 1);
}
 
void llvm::getSelectionDAGFallbackAnalysisUsage(AnalysisUsage &AU) {
  AU.addPreserved<StackProtector>();
}
 
LLT llvm::getLCMType(LLT OrigTy, LLT TargetTy) {
  if (OrigTy.getSizeInBits() == TargetTy.getSizeInBits())
    return OrigTy;
 
  if (OrigTy.isVector() && TargetTy.isVector()) {
    LLT OrigElt = OrigTy.getElementType();
    LLT TargetElt = TargetTy.getElementType();
 
    // TODO: The docstring for this function says the intention is to use this
    // function to build MERGE/UNMERGE instructions. It won't be the case that
    // we generate a MERGE/UNMERGE between fixed and scalable vector types. We
    // could implement getLCMType between the two in the future if there was a
    // need, but it is not worth it now as this function should not be used in
    // that way.
    assert(((OrigTy.isScalableVector() && !TargetTy.isFixedVector()) ||
            (OrigTy.isFixedVector() && !TargetTy.isScalableVector())) &&
           "getLCMType not implemented between fixed and scalable vectors.");
 
    if (OrigElt.getSizeInBits() == TargetElt.getSizeInBits()) {
      int GCDMinElts = std::gcd(OrigTy.getElementCount().getKnownMinValue(),
                                TargetTy.getElementCount().getKnownMinValue());
      // Prefer the original element type.
      ElementCount Mul = OrigTy.getElementCount().multiplyCoefficientBy(
          TargetTy.getElementCount().getKnownMinValue());
      return LLT::vector(Mul.divideCoefficientBy(GCDMinElts),
                         OrigTy.getElementType());
    }
    unsigned LCM = std::lcm(OrigTy.getSizeInBits().getKnownMinValue(),
                            TargetTy.getSizeInBits().getKnownMinValue());
    return LLT::vector(
        ElementCount::get(LCM / OrigElt.getSizeInBits(), OrigTy.isScalable()),
        OrigElt);
  }
 
  // One type is scalar, one type is vector
  if (OrigTy.isVector() || TargetTy.isVector()) {
    LLT VecTy = OrigTy.isVector() ? OrigTy : TargetTy;
    LLT ScalarTy = OrigTy.isVector() ? TargetTy : OrigTy;
    LLT EltTy = VecTy.getElementType();
    LLT OrigEltTy = OrigTy.isVector() ? OrigTy.getElementType() : OrigTy;
 
    // Prefer scalar type from OrigTy.
    if (EltTy.getSizeInBits() == ScalarTy.getSizeInBits())
      return LLT::vector(VecTy.getElementCount(), OrigEltTy);
 
    // Different size scalars. Create vector with the same total size.
    // LCM will take fixed/scalable from VecTy.
    unsigned LCM = std::lcm(EltTy.getSizeInBits().getFixedValue() *
                                VecTy.getElementCount().getKnownMinValue(),
                            ScalarTy.getSizeInBits().getFixedValue());
    // Prefer type from OrigTy
    return LLT::vector(ElementCount::get(LCM / OrigEltTy.getSizeInBits(),
                                         VecTy.getElementCount().isScalable()),
                       OrigEltTy);
  }
 
  // At this point, both types are scalars of different size
  unsigned LCM = std::lcm(OrigTy.getSizeInBits().getFixedValue(),
                          TargetTy.getSizeInBits().getFixedValue());
  // Preserve pointer types.
  if (LCM == OrigTy.getSizeInBits())
    return OrigTy;
  if (LCM == TargetTy.getSizeInBits())
    return TargetTy;
  return LLT::scalar(LCM);
}
 
LLT llvm::getCoverTy(LLT OrigTy, LLT TargetTy) {
 
  if ((OrigTy.isScalableVector() && TargetTy.isFixedVector()) ||
      (OrigTy.isFixedVector() && TargetTy.isScalableVector()))
    llvm_unreachable(
        "getCoverTy not implemented between fixed and scalable vectors.");
 
  if (!OrigTy.isVector() || !TargetTy.isVector() || OrigTy == TargetTy ||
      (OrigTy.getScalarSizeInBits() != TargetTy.getScalarSizeInBits()))
    return getLCMType(OrigTy, TargetTy);
 
  unsigned OrigTyNumElts = OrigTy.getElementCount().getKnownMinValue();
  unsigned TargetTyNumElts = TargetTy.getElementCount().getKnownMinValue();
  if (OrigTyNumElts % TargetTyNumElts == 0)
    return OrigTy;
 
  unsigned NumElts = alignTo(OrigTyNumElts, TargetTyNumElts);
  return LLT::scalarOrVector(ElementCount::getFixed(NumElts),
                             OrigTy.getElementType());
}
 
LLT llvm::getGCDType(LLT OrigTy, LLT TargetTy) {
  if (OrigTy.getSizeInBits() == TargetTy.getSizeInBits())
    return OrigTy;
 
  if (OrigTy.isVector() && TargetTy.isVector()) {
    LLT OrigElt = OrigTy.getElementType();
 
    // TODO: The docstring for this function says the intention is to use this
    // function to build MERGE/UNMERGE instructions. It won't be the case that
    // we generate a MERGE/UNMERGE between fixed and scalable vector types. We
    // could implement getGCDType between the two in the future if there was a
    // need, but it is not worth it now as this function should not be used in
    // that way.
    assert(((OrigTy.isScalableVector() && !TargetTy.isFixedVector()) ||
            (OrigTy.isFixedVector() && !TargetTy.isScalableVector())) &&
           "getGCDType not implemented between fixed and scalable vectors.");
 
    unsigned GCD = std::gcd(OrigTy.getSizeInBits().getKnownMinValue(),
                            TargetTy.getSizeInBits().getKnownMinValue());
    if (GCD == OrigElt.getSizeInBits())
      return LLT::scalarOrVector(ElementCount::get(1, OrigTy.isScalable()),
                                 OrigElt);
 
    // Cannot produce original element type, but both have vscale in common.
    if (GCD < OrigElt.getSizeInBits())
      return LLT::scalarOrVector(ElementCount::get(1, OrigTy.isScalable()),
                                 GCD);
 
    return LLT::vector(
        ElementCount::get(GCD / OrigElt.getSizeInBits().getFixedValue(),
                          OrigTy.isScalable()),
        OrigElt);
  }
 
  // If one type is vector and the element size matches the scalar size, then
  // the gcd is the scalar type.
  if (OrigTy.isVector() &&
      OrigTy.getElementType().getSizeInBits() == TargetTy.getSizeInBits())
    return OrigTy.getElementType();
  if (TargetTy.isVector() &&
      TargetTy.getElementType().getSizeInBits() == OrigTy.getSizeInBits())
    return OrigTy;
 
  // At this point, both types are either scalars of different type or one is a
  // vector and one is a scalar. If both types are scalars, the GCD type is the
  // GCD between the two scalar sizes. If one is vector and one is scalar, then
  // the GCD type is the GCD between the scalar and the vector element size.
  LLT OrigScalar = OrigTy.getScalarType();
  LLT TargetScalar = TargetTy.getScalarType();
  unsigned GCD = std::gcd(OrigScalar.getSizeInBits().getFixedValue(),
                          TargetScalar.getSizeInBits().getFixedValue());
  return LLT::integer(GCD);
}
 
std::optional<int> llvm::getSplatIndex(MachineInstr &MI) {
  assert(MI.getOpcode() == TargetOpcode::G_SHUFFLE_VECTOR &&
         "Only G_SHUFFLE_VECTOR can have a splat index!");
  ArrayRef<int> Mask = MI.getOperand(3).getShuffleMask();
  auto FirstDefinedIdx = find_if(Mask, [](int Elt) { return Elt >= 0; });
 
  // If all elements are undefined, this shuffle can be considered a splat.
  // Return 0 for better potential for callers to simplify.
  if (FirstDefinedIdx == Mask.end())
    return 0;
 
  // Make sure all remaining elements are either undef or the same
  // as the first non-undef value.
  int SplatValue = *FirstDefinedIdx;
  if (any_of(make_range(std::next(FirstDefinedIdx), Mask.end()),
             [&SplatValue](int Elt) { return Elt >= 0 && Elt != SplatValue; }))
    return std::nullopt;
 
  return SplatValue;
}
 
static bool isBuildVectorOp(unsigned Opcode) {
  return Opcode == TargetOpcode::G_BUILD_VECTOR ||
         Opcode == TargetOpcode::G_BUILD_VECTOR_TRUNC;
}
 
namespace {
 
std::optional<ValueAndVReg> getAnyConstantSplat(Register VReg,
                                                const MachineRegisterInfo &MRI,
                                                bool AllowUndef) {
  MachineInstr *MI = getDefIgnoringCopies(VReg, MRI);
  if (!MI)
    return std::nullopt;
 
  bool isConcatVectorsOp = MI->getOpcode() == TargetOpcode::G_CONCAT_VECTORS;
  if (!isBuildVectorOp(MI->getOpcode()) && !isConcatVectorsOp)
    return std::nullopt;
 
  std::optional<ValueAndVReg> SplatValAndReg;
  for (MachineOperand &Op : MI->uses()) {
    Register Element = Op.getReg();
    // If we have a G_CONCAT_VECTOR, we recursively look into the
    // vectors that we're concatenating to see if they're splats.
    auto ElementValAndReg =
        isConcatVectorsOp
            ? getAnyConstantSplat(Element, MRI, AllowUndef)
            : getAnyConstantVRegValWithLookThrough(Element, MRI, true, true);
 
    // If AllowUndef, treat undef as value that will result in a constant splat.
    if (!ElementValAndReg) {
      if (AllowUndef && isa<GImplicitDef>(MRI.getVRegDef(Element)))
        continue;
      return std::nullopt;
    }
 
    // Record splat value
    if (!SplatValAndReg)
      SplatValAndReg = ElementValAndReg;
 
    // Different constant than the one already recorded, not a constant splat.
    if (SplatValAndReg->Value != ElementValAndReg->Value)
      return std::nullopt;
  }
 
  return SplatValAndReg;
}
 
} // end anonymous namespace
 
bool llvm::isBuildVectorConstantSplat(const Register Reg,
                                      const MachineRegisterInfo &MRI,
                                      int64_t SplatValue, bool AllowUndef) {
  if (auto SplatValAndReg = getAnyConstantSplat(Reg, MRI, AllowUndef))
    return SplatValAndReg->Value.getSExtValue() == SplatValue;
 
  return false;
}
 
bool llvm::isBuildVectorConstantSplat(const Register Reg,
                                      const MachineRegisterInfo &MRI,
                                      const APInt &SplatValue,
                                      bool AllowUndef) {
  if (auto SplatValAndReg = getAnyConstantSplat(Reg, MRI, AllowUndef)) {
    if (SplatValAndReg->Value.getBitWidth() < SplatValue.getBitWidth())
      return APInt::isSameValue(
          SplatValAndReg->Value.sext(SplatValue.getBitWidth()), SplatValue);
    return APInt::isSameValue(
        SplatValAndReg->Value,
        SplatValue.sext(SplatValAndReg->Value.getBitWidth()));
  }
 
  return false;
}
 
bool llvm::isBuildVectorConstantSplat(const MachineInstr &MI,
                                      const MachineRegisterInfo &MRI,
                                      int64_t SplatValue, bool AllowUndef) {
  return isBuildVectorConstantSplat(MI.getOperand(0).getReg(), MRI, SplatValue,
                                    AllowUndef);
}
 
bool llvm::isBuildVectorConstantSplat(const MachineInstr &MI,
                                      const MachineRegisterInfo &MRI,
                                      const APInt &SplatValue,
                                      bool AllowUndef) {
  return isBuildVectorConstantSplat(MI.getOperand(0).getReg(), MRI, SplatValue,
                                    AllowUndef);
}
 
std::optional<APInt>
llvm::getIConstantSplatVal(const Register Reg, const MachineRegisterInfo &MRI) {
  if (auto SplatValAndReg =
          getAnyConstantSplat(Reg, MRI, /* AllowUndef */ false)) {
    if (std::optional<ValueAndVReg> ValAndVReg =
        getIConstantVRegValWithLookThrough(SplatValAndReg->VReg, MRI))
      return ValAndVReg->Value;
  }
 
  return std::nullopt;
}
 
std::optional<APInt>
llvm::getIConstantSplatVal(const MachineInstr &MI,
                           const MachineRegisterInfo &MRI) {
  return getIConstantSplatVal(MI.getOperand(0).getReg(), MRI);
}
 
std::optional<int64_t>
llvm::getIConstantSplatSExtVal(const Register Reg,
                               const MachineRegisterInfo &MRI) {
  if (auto SplatValAndReg =
          getAnyConstantSplat(Reg, MRI, /* AllowUndef */ false))
    return getIConstantVRegSExtVal(SplatValAndReg->VReg, MRI);
  return std::nullopt;
}
 
std::optional<int64_t>
llvm::getIConstantSplatSExtVal(const MachineInstr &MI,
                               const MachineRegisterInfo &MRI) {
  return getIConstantSplatSExtVal(MI.getOperand(0).getReg(), MRI);
}
 
std::optional<FPValueAndVReg>
llvm::getFConstantSplat(Register VReg, const MachineRegisterInfo &MRI,
                        bool AllowUndef) {
  if (auto SplatValAndReg = getAnyConstantSplat(VReg, MRI, AllowUndef))
    return getFConstantVRegValWithLookThrough(SplatValAndReg->VReg, MRI);
  return std::nullopt;
}
 
bool llvm::isBuildVectorAllZeros(const MachineInstr &MI,
                                 const MachineRegisterInfo &MRI,
                                 bool AllowUndef) {
  return isBuildVectorConstantSplat(MI, MRI, 0, AllowUndef);
}
 
bool llvm::isBuildVectorAllOnes(const MachineInstr &MI,
                                const MachineRegisterInfo &MRI,
                                bool AllowUndef) {
  return isBuildVectorConstantSplat(MI, MRI, -1, AllowUndef);
}
 
std::optional<RegOrConstant>
llvm::getVectorSplat(const MachineInstr &MI, const MachineRegisterInfo &MRI) {
  unsigned Opc = MI.getOpcode();
  if (!isBuildVectorOp(Opc))
    return std::nullopt;
  if (auto Splat = getIConstantSplatSExtVal(MI, MRI))
    return RegOrConstant(*Splat);
  auto Reg = MI.getOperand(1).getReg();
  if (any_of(drop_begin(MI.operands(), 2),
             [&Reg](const MachineOperand &Op) { return Op.getReg() != Reg; }))
    return std::nullopt;
  return RegOrConstant(Reg);
}
 
static bool isConstantScalar(const MachineInstr &MI,
                             const MachineRegisterInfo &MRI,
                             bool AllowFP = true,
                             bool AllowOpaqueConstants = true) {
  switch (MI.getOpcode()) {
  case TargetOpcode::G_CONSTANT:
  case TargetOpcode::G_IMPLICIT_DEF:
    return true;
  case TargetOpcode::G_FCONSTANT:
    return AllowFP;
  case TargetOpcode::G_GLOBAL_VALUE:
  case TargetOpcode::G_FRAME_INDEX:
  case TargetOpcode::G_BLOCK_ADDR:
  case TargetOpcode::G_JUMP_TABLE:
    return AllowOpaqueConstants;
  default:
    return false;
  }
}
 
bool llvm::isConstantOrConstantVector(MachineInstr &MI,
                                      const MachineRegisterInfo &MRI) {
  Register Def = MI.getOperand(0).getReg();
  if (auto C = getIConstantVRegValWithLookThrough(Def, MRI))
    return true;
  GBuildVector *BV = dyn_cast<GBuildVector>(&MI);
  if (!BV)
    return false;
  for (unsigned SrcIdx = 0; SrcIdx < BV->getNumSources(); ++SrcIdx) {
    if (getIConstantVRegValWithLookThrough(BV->getSourceReg(SrcIdx), MRI) ||
        getOpcodeDef<GImplicitDef>(BV->getSourceReg(SrcIdx), MRI))
      continue;
    return false;
  }
  return true;
}
 
bool llvm::isConstantOrConstantVector(const MachineInstr &MI,
                                      const MachineRegisterInfo &MRI,
                                      bool AllowFP, bool AllowOpaqueConstants) {
  if (isConstantScalar(MI, MRI, AllowFP, AllowOpaqueConstants))
    return true;
 
  if (!isBuildVectorOp(MI.getOpcode()))
    return false;
 
  const unsigned NumOps = MI.getNumOperands();
  for (unsigned I = 1; I != NumOps; ++I) {
    const MachineInstr *ElementDef = MRI.getVRegDef(MI.getOperand(I).getReg());
    if (!isConstantScalar(*ElementDef, MRI, AllowFP, AllowOpaqueConstants))
      return false;
  }
 
  return true;
}
 
std::optional<APInt>
llvm::isConstantOrConstantSplatVector(MachineInstr &MI,
                                      const MachineRegisterInfo &MRI) {
  Register Def = MI.getOperand(0).getReg();
  if (auto C = getIConstantVRegValWithLookThrough(Def, MRI))
    return C->Value;
  auto MaybeCst = getIConstantSplatSExtVal(MI, MRI);
  if (!MaybeCst)
    return std::nullopt;
  const unsigned ScalarSize = MRI.getType(Def).getScalarSizeInBits();
  return APInt(ScalarSize, *MaybeCst, true);
}
 
std::optional<APFloat>
llvm::isConstantOrConstantSplatVectorFP(MachineInstr &MI,
                                        const MachineRegisterInfo &MRI) {
  Register Def = MI.getOperand(0).getReg();
  if (auto FpConst = getFConstantVRegValWithLookThrough(Def, MRI))
    return FpConst->Value;
  auto MaybeCstFP = getFConstantSplat(Def, MRI, /*allowUndef=*/false);
  if (!MaybeCstFP)
    return std::nullopt;
  return MaybeCstFP->Value;
}
 
bool llvm::isNullOrNullSplat(const MachineInstr &MI,
                             const MachineRegisterInfo &MRI, bool AllowUndefs) {
  switch (MI.getOpcode()) {
  case TargetOpcode::G_IMPLICIT_DEF:
    return AllowUndefs;
  case TargetOpcode::G_CONSTANT:
    return MI.getOperand(1).getCImm()->isNullValue();
  case TargetOpcode::G_FCONSTANT: {
    const ConstantFP *FPImm = MI.getOperand(1).getFPImm();
    return FPImm->isZero() && !FPImm->isNegative();
  }
  default:
    if (!AllowUndefs) // TODO: isBuildVectorAllZeros assumes undef is OK already
      return false;
    return isBuildVectorAllZeros(MI, MRI);
  }
}
 
bool llvm::isAllOnesOrAllOnesSplat(const MachineInstr &MI,
                                   const MachineRegisterInfo &MRI,
                                   bool AllowUndefs) {
  switch (MI.getOpcode()) {
  case TargetOpcode::G_IMPLICIT_DEF:
    return AllowUndefs;
  case TargetOpcode::G_CONSTANT:
    return MI.getOperand(1).getCImm()->isAllOnesValue();
  default:
    if (!AllowUndefs) // TODO: isBuildVectorAllOnes assumes undef is OK already
      return false;
    return isBuildVectorAllOnes(MI, MRI);
  }
}
 
bool llvm::matchUnaryPredicate(
    const MachineRegisterInfo &MRI, Register Reg,
    std::function<bool(const Constant *ConstVal)> Match, bool AllowUndefs) {
 
  const MachineInstr *Def = getDefIgnoringCopies(Reg, MRI);
  if (AllowUndefs && Def->getOpcode() == TargetOpcode::G_IMPLICIT_DEF)
    return Match(nullptr);
 
  // TODO: Also handle fconstant
  if (Def->getOpcode() == TargetOpcode::G_CONSTANT)
    return Match(Def->getOperand(1).getCImm());
 
  if (Def->getOpcode() != TargetOpcode::G_BUILD_VECTOR)
    return false;
 
  for (unsigned I = 1, E = Def->getNumOperands(); I != E; ++I) {
    Register SrcElt = Def->getOperand(I).getReg();
    const MachineInstr *SrcDef = getDefIgnoringCopies(SrcElt, MRI);
    if (AllowUndefs && SrcDef->getOpcode() == TargetOpcode::G_IMPLICIT_DEF) {
      if (!Match(nullptr))
        return false;
      continue;
    }
 
    if (SrcDef->getOpcode() != TargetOpcode::G_CONSTANT ||
        !Match(SrcDef->getOperand(1).getCImm()))
      return false;
  }
 
  return true;
}
 
bool llvm::isConstTrueVal(const TargetLowering &TLI, int64_t Val, bool IsVector,
                          bool IsFP) {
  switch (TLI.getBooleanContents(IsVector, IsFP)) {
  case TargetLowering::UndefinedBooleanContent:
    return Val & 0x1;
  case TargetLowering::ZeroOrOneBooleanContent:
    return Val == 1;
  case TargetLowering::ZeroOrNegativeOneBooleanContent:
    return Val == -1;
  }
  llvm_unreachable("Invalid boolean contents");
}
 
bool llvm::isConstFalseVal(const TargetLowering &TLI, int64_t Val,
                           bool IsVector, bool IsFP) {
  switch (TLI.getBooleanContents(IsVector, IsFP)) {
  case TargetLowering::UndefinedBooleanContent:
    return ~Val & 0x1;
  case TargetLowering::ZeroOrOneBooleanContent:
  case TargetLowering::ZeroOrNegativeOneBooleanContent:
    return Val == 0;
  }
  llvm_unreachable("Invalid boolean contents");
}
 
int64_t llvm::getICmpTrueVal(const TargetLowering &TLI, bool IsVector,
                             bool IsFP) {
  switch (TLI.getBooleanContents(IsVector, IsFP)) {
  case TargetLowering::UndefinedBooleanContent:
  case TargetLowering::ZeroOrOneBooleanContent:
    return 1;
  case TargetLowering::ZeroOrNegativeOneBooleanContent:
    return -1;
  }
  llvm_unreachable("Invalid boolean contents");
}
 
void llvm::saveUsesAndErase(MachineInstr &MI, MachineRegisterInfo &MRI,
                            LostDebugLocObserver *LocObserver,
                            SmallInstListTy &DeadInstChain) {
  for (MachineOperand &Op : MI.uses()) {
    if (Op.isReg() && Op.getReg().isVirtual())
      DeadInstChain.insert(MRI.getVRegDef(Op.getReg()));
  }
  LLVM_DEBUG(dbgs() << MI << "Is dead; erasing.\n");
  DeadInstChain.remove(&MI);
  MI.eraseFromParent();
  if (LocObserver)
    LocObserver->checkpoint(false);
}
 
void llvm::eraseInstrs(ArrayRef<MachineInstr *> DeadInstrs,
                       MachineRegisterInfo &MRI,
                       LostDebugLocObserver *LocObserver) {
  SmallInstListTy DeadInstChain;
  for (MachineInstr *MI : DeadInstrs)
    saveUsesAndErase(*MI, MRI, LocObserver, DeadInstChain);
 
  while (!DeadInstChain.empty()) {
    MachineInstr *Inst = DeadInstChain.pop_back_val();
    if (!isTriviallyDead(*Inst, MRI))
      continue;
    saveUsesAndErase(*Inst, MRI, LocObserver, DeadInstChain);
  }
}
 
void llvm::eraseInstr(MachineInstr &MI, MachineRegisterInfo &MRI,
                      LostDebugLocObserver *LocObserver) {
  return eraseInstrs({&MI}, MRI, LocObserver);
}
 
void llvm::salvageDebugInfo(const MachineRegisterInfo &MRI, MachineInstr &MI) {
  for (auto &Def : MI.defs()) {
    assert(Def.isReg() && "Must be a reg");
 
    SmallVector<MachineOperand *, 16> DbgUsers;
    for (auto &MOUse : MRI.use_operands(Def.getReg())) {
      MachineInstr *DbgValue = MOUse.getParent();
      // Ignore partially formed DBG_VALUEs.
      if (DbgValue->isNonListDebugValue() && DbgValue->getNumOperands() == 4) {
        DbgUsers.push_back(&MOUse);
      }
    }
 
    if (!DbgUsers.empty()) {
      salvageDebugInfoForDbgValue(MRI, MI, DbgUsers);
    }
  }
}
 
bool llvm::isPreISelGenericFloatingPointOpcode(unsigned Opc) {
  switch (Opc) {
  case TargetOpcode::G_FABS:
  case TargetOpcode::G_FADD:
  case TargetOpcode::G_FCANONICALIZE:
  case TargetOpcode::G_FCEIL:
  case TargetOpcode::G_FCONSTANT:
  case TargetOpcode::G_FCOPYSIGN:
  case TargetOpcode::G_FCOS:
  case TargetOpcode::G_FDIV:
  case TargetOpcode::G_FEXP2:
  case TargetOpcode::G_FEXP:
  case TargetOpcode::G_FFLOOR:
  case TargetOpcode::G_FLOG10:
  case TargetOpcode::G_FLOG2:
  case TargetOpcode::G_FLOG:
  case TargetOpcode::G_FMA:
  case TargetOpcode::G_FMAD:
  case TargetOpcode::G_FMAXIMUM:
  case TargetOpcode::G_FMAXIMUMNUM:
  case TargetOpcode::G_FMAXNUM:
  case TargetOpcode::G_FMAXNUM_IEEE:
  case TargetOpcode::G_FMINIMUM:
  case TargetOpcode::G_FMINIMUMNUM:
  case TargetOpcode::G_FMINNUM:
  case TargetOpcode::G_FMINNUM_IEEE:
  case TargetOpcode::G_FMUL:
  case TargetOpcode::G_FNEARBYINT:
  case TargetOpcode::G_FNEG:
  case TargetOpcode::G_FPEXT:
  case TargetOpcode::G_FPEXTLOAD:
  case TargetOpcode::G_FPOW:
  case TargetOpcode::G_FPTRUNC:
  case TargetOpcode::G_FPTRUNCSTORE:
  case TargetOpcode::G_FREM:
  case TargetOpcode::G_FRINT:
  case TargetOpcode::G_FSIN:
  case TargetOpcode::G_FTAN:
  case TargetOpcode::G_FACOS:
  case TargetOpcode::G_FASIN:
  case TargetOpcode::G_FATAN:
  case TargetOpcode::G_FATAN2:
  case TargetOpcode::G_FCOSH:
  case TargetOpcode::G_FSINH:
  case TargetOpcode::G_FTANH:
  case TargetOpcode::G_FSQRT:
  case TargetOpcode::G_FSUB:
  case TargetOpcode::G_INTRINSIC_ROUND:
  case TargetOpcode::G_INTRINSIC_ROUNDEVEN:
  case TargetOpcode::G_INTRINSIC_TRUNC:
    return true;
  default:
    return false;
  }
}
 
/// Shifts return poison if shiftwidth is larger than the bitwidth.
static bool shiftAmountKnownInRange(Register ShiftAmount,
                                    const MachineRegisterInfo &MRI) {
  LLT Ty = MRI.getType(ShiftAmount);
 
  if (Ty.isScalableVector())
    return false; // Can't tell, just return false to be safe
 
  if (Ty.isScalar()) {
    std::optional<ValueAndVReg> Val =
        getIConstantVRegValWithLookThrough(ShiftAmount, MRI);
    if (!Val)
      return false;
    return Val->Value.ult(Ty.getScalarSizeInBits());
  }
 
  GBuildVector *BV = getOpcodeDef<GBuildVector>(ShiftAmount, MRI);
  if (!BV)
    return false;
 
  unsigned Sources = BV->getNumSources();
  for (unsigned I = 0; I < Sources; ++I) {
    std::optional<ValueAndVReg> Val =
        getIConstantVRegValWithLookThrough(BV->getSourceReg(I), MRI);
    if (!Val)
      return false;
    if (!Val->Value.ult(Ty.getScalarSizeInBits()))
      return false;
  }
 
  return true;
}
 
static bool canCreateUndefOrPoison(Register Reg, const MachineRegisterInfo &MRI,
                                   bool ConsiderFlagsAndMetadata,
                                   UndefPoisonKind Kind) {
  MachineInstr *RegDef = MRI.getVRegDef(Reg);
 
  if (ConsiderFlagsAndMetadata && includesPoison(Kind))
    if (auto *GMI = dyn_cast<GenericMachineInstr>(RegDef))
      if (GMI->hasPoisonGeneratingFlags())
        return true;
 
  // Check whether opcode is a poison/undef-generating operation.
  switch (RegDef->getOpcode()) {
  case TargetOpcode::G_BUILD_VECTOR:
  case TargetOpcode::G_CONSTANT_FOLD_BARRIER:
    return false;
  case TargetOpcode::G_SHL:
  case TargetOpcode::G_ASHR:
  case TargetOpcode::G_LSHR:
    return includesPoison(Kind) &&
           !shiftAmountKnownInRange(RegDef->getOperand(2).getReg(), MRI);
  case TargetOpcode::G_FPTOSI:
  case TargetOpcode::G_FPTOUI:
    // fptosi/ui yields poison if the resulting value does not fit in the
    // destination type.
    return true;
  case TargetOpcode::G_CTLZ:
  case TargetOpcode::G_CTTZ:
  case TargetOpcode::G_CTLS:
  case TargetOpcode::G_ABS:
  case TargetOpcode::G_CTPOP:
  case TargetOpcode::G_BSWAP:
  case TargetOpcode::G_BITREVERSE:
  case TargetOpcode::G_FSHL:
  case TargetOpcode::G_FSHR:
  case TargetOpcode::G_SMAX:
  case TargetOpcode::G_SMIN:
  case TargetOpcode::G_SCMP:
  case TargetOpcode::G_UMAX:
  case TargetOpcode::G_UMIN:
  case TargetOpcode::G_UCMP:
  case TargetOpcode::G_PTRMASK:
  case TargetOpcode::G_SADDO:
  case TargetOpcode::G_SSUBO:
  case TargetOpcode::G_UADDO:
  case TargetOpcode::G_USUBO:
  case TargetOpcode::G_SMULO:
  case TargetOpcode::G_UMULO:
  case TargetOpcode::G_SADDSAT:
  case TargetOpcode::G_UADDSAT:
  case TargetOpcode::G_SSUBSAT:
  case TargetOpcode::G_USUBSAT:
  case TargetOpcode::G_SBFX:
  case TargetOpcode::G_UBFX:
    return false;
  case TargetOpcode::G_SSHLSAT:
  case TargetOpcode::G_USHLSAT:
    return includesPoison(Kind) &&
           !shiftAmountKnownInRange(RegDef->getOperand(2).getReg(), MRI);
  case TargetOpcode::G_INSERT_VECTOR_ELT: {
    GInsertVectorElement *Insert = cast<GInsertVectorElement>(RegDef);
    if (includesPoison(Kind)) {
      std::optional<ValueAndVReg> Index =
          getIConstantVRegValWithLookThrough(Insert->getIndexReg(), MRI);
      if (!Index)
        return true;
      LLT VecTy = MRI.getType(Insert->getVectorReg());
      return Index->Value.uge(VecTy.getElementCount().getKnownMinValue());
    }
    return false;
  }
  case TargetOpcode::G_EXTRACT_VECTOR_ELT: {
    GExtractVectorElement *Extract = cast<GExtractVectorElement>(RegDef);
    if (includesPoison(Kind)) {
      std::optional<ValueAndVReg> Index =
          getIConstantVRegValWithLookThrough(Extract->getIndexReg(), MRI);
      if (!Index)
        return true;
      LLT VecTy = MRI.getType(Extract->getVectorReg());
      return Index->Value.uge(VecTy.getElementCount().getKnownMinValue());
    }
    return false;
  }
  case TargetOpcode::G_SHUFFLE_VECTOR: {
    GShuffleVector *Shuffle = cast<GShuffleVector>(RegDef);
    ArrayRef<int> Mask = Shuffle->getMask();
    return includesPoison(Kind) && is_contained(Mask, -1);
  }
  case TargetOpcode::G_FNEG:
  case TargetOpcode::G_PHI:
  case TargetOpcode::G_SELECT:
  case TargetOpcode::G_UREM:
  case TargetOpcode::G_SREM:
  case TargetOpcode::G_FREEZE:
  case TargetOpcode::G_ICMP:
  case TargetOpcode::G_FCMP:
  case TargetOpcode::G_FADD:
  case TargetOpcode::G_FSUB:
  case TargetOpcode::G_FMUL:
  case TargetOpcode::G_FDIV:
  case TargetOpcode::G_FREM:
  case TargetOpcode::G_PTR_ADD:
    return false;
  default:
    return !isa<GCastOp>(RegDef) && !isa<GBinOp>(RegDef);
  }
}
 
static bool isGuaranteedNotToBeUndefOrPoison(Register Reg,
                                             const MachineRegisterInfo &MRI,
                                             unsigned Depth,
                                             UndefPoisonKind Kind) {
  if (Depth >= MaxAnalysisRecursionDepth)
    return false;
 
  MachineInstr *RegDef = MRI.getVRegDef(Reg);
 
  switch (RegDef->getOpcode()) {
  case TargetOpcode::G_FREEZE:
    return true;
  case TargetOpcode::G_IMPLICIT_DEF:
    return !includesUndef(Kind);
  case TargetOpcode::G_CONSTANT:
  case TargetOpcode::G_FCONSTANT:
    return true;
  case TargetOpcode::G_BUILD_VECTOR: {
    GBuildVector *BV = cast<GBuildVector>(RegDef);
    unsigned NumSources = BV->getNumSources();
    for (unsigned I = 0; I < NumSources; ++I)
      if (!::isGuaranteedNotToBeUndefOrPoison(BV->getSourceReg(I), MRI,
                                              Depth + 1, Kind))
        return false;
    return true;
  }
  case TargetOpcode::G_PHI: {
    GPhi *Phi = cast<GPhi>(RegDef);
    unsigned NumIncoming = Phi->getNumIncomingValues();
    for (unsigned I = 0; I < NumIncoming; ++I)
      if (!::isGuaranteedNotToBeUndefOrPoison(Phi->getIncomingValue(I), MRI,
                                              Depth + 1, Kind))
        return false;
    return true;
  }
  default: {
    auto MOCheck = [&](const MachineOperand &MO) {
      if (!MO.isReg())
        return true;
      return ::isGuaranteedNotToBeUndefOrPoison(MO.getReg(), MRI, Depth + 1,
                                                Kind);
    };
    return !::canCreateUndefOrPoison(Reg, MRI,
                                     /*ConsiderFlagsAndMetadata=*/true, Kind) &&
           all_of(RegDef->uses(), MOCheck);
  }
  }
}
 
bool llvm::canCreateUndefOrPoison(Register Reg, const MachineRegisterInfo &MRI,
                                  bool ConsiderFlagsAndMetadata) {
  return ::canCreateUndefOrPoison(Reg, MRI, ConsiderFlagsAndMetadata,
                                  UndefPoisonKind::UndefOrPoison);
}
 
bool canCreatePoison(Register Reg, const MachineRegisterInfo &MRI,
                     bool ConsiderFlagsAndMetadata = true) {
  return ::canCreateUndefOrPoison(Reg, MRI, ConsiderFlagsAndMetadata,
                                  UndefPoisonKind::PoisonOnly);
}
 
bool llvm::isGuaranteedNotToBeUndefOrPoison(Register Reg,
                                            const MachineRegisterInfo &MRI,
                                            unsigned Depth) {
  return ::isGuaranteedNotToBeUndefOrPoison(Reg, MRI, Depth,
                                            UndefPoisonKind::UndefOrPoison);
}
 
bool llvm::isGuaranteedNotToBePoison(Register Reg,
                                     const MachineRegisterInfo &MRI,
                                     unsigned Depth) {
  return ::isGuaranteedNotToBeUndefOrPoison(Reg, MRI, Depth,
                                            UndefPoisonKind::PoisonOnly);
}
 
bool llvm::isGuaranteedNotToBeUndef(Register Reg,
                                    const MachineRegisterInfo &MRI,
                                    unsigned Depth) {
  return ::isGuaranteedNotToBeUndefOrPoison(Reg, MRI, Depth,
                                            UndefPoisonKind::UndefOnly);
}
 
Type *llvm::getTypeForLLT(LLT Ty, LLVMContext &C) {
  if (Ty.isVector())
    return VectorType::get(IntegerType::get(C, Ty.getScalarSizeInBits()),
                           Ty.getElementCount());
  return IntegerType::get(C, Ty.getSizeInBits());
}
 
bool llvm::isAssertMI(const MachineInstr &MI) {
  switch (MI.getOpcode()) {
  default:
    return false;
  case TargetOpcode::G_ASSERT_ALIGN:
  case TargetOpcode::G_ASSERT_SEXT:
  case TargetOpcode::G_ASSERT_ZEXT:
    return true;
  }
}
 
APInt llvm::GIConstant::getScalarValue() const {
  assert(Kind == GIConstantKind::Scalar && "Expected scalar constant");
 
  return Value;
}
 
std::optional<GIConstant>
llvm::GIConstant::getConstant(Register Const, const MachineRegisterInfo &MRI) {
  MachineInstr *Constant = getDefIgnoringCopies(Const, MRI);
 
  if (GSplatVector *Splat = dyn_cast<GSplatVector>(Constant)) {
    std::optional<ValueAndVReg> MayBeConstant =
        getIConstantVRegValWithLookThrough(Splat->getScalarReg(), MRI);
    if (!MayBeConstant)
      return std::nullopt;
    return GIConstant(MayBeConstant->Value, GIConstantKind::ScalableVector);
  }
 
  if (GBuildVector *Build = dyn_cast<GBuildVector>(Constant)) {
    SmallVector<APInt> Values;
    unsigned NumSources = Build->getNumSources();
    for (unsigned I = 0; I < NumSources; ++I) {
      Register SrcReg = Build->getSourceReg(I);
      std::optional<ValueAndVReg> MayBeConstant =
          getIConstantVRegValWithLookThrough(SrcReg, MRI);
      if (!MayBeConstant)
        return std::nullopt;
      Values.push_back(MayBeConstant->Value);
    }
    return GIConstant(Values);
  }
 
  std::optional<ValueAndVReg> MayBeConstant =
      getIConstantVRegValWithLookThrough(Const, MRI);
  if (!MayBeConstant)
    return std::nullopt;
 
  return GIConstant(MayBeConstant->Value, GIConstantKind::Scalar);
}
 
APFloat llvm::GFConstant::getScalarValue() const {
  assert(Kind == GFConstantKind::Scalar && "Expected scalar constant");
 
  return Values[0];
}
 
std::optional<GFConstant>
llvm::GFConstant::getConstant(Register Const, const MachineRegisterInfo &MRI) {
  MachineInstr *Constant = getDefIgnoringCopies(Const, MRI);
 
  if (GSplatVector *Splat = dyn_cast<GSplatVector>(Constant)) {
    std::optional<FPValueAndVReg> MayBeConstant =
        getFConstantVRegValWithLookThrough(Splat->getScalarReg(), MRI);
    if (!MayBeConstant)
      return std::nullopt;
    return GFConstant(MayBeConstant->Value, GFConstantKind::ScalableVector);
  }
 
  if (GBuildVector *Build = dyn_cast<GBuildVector>(Constant)) {
    SmallVector<APFloat> Values;
    unsigned NumSources = Build->getNumSources();
    for (unsigned I = 0; I < NumSources; ++I) {
      Register SrcReg = Build->getSourceReg(I);
      std::optional<FPValueAndVReg> MayBeConstant =
          getFConstantVRegValWithLookThrough(SrcReg, MRI);
      if (!MayBeConstant)
        return std::nullopt;
      Values.push_back(MayBeConstant->Value);
    }
    return GFConstant(Values);
  }
 
  std::optional<FPValueAndVReg> MayBeConstant =
      getFConstantVRegValWithLookThrough(Const, MRI);
  if (!MayBeConstant)
    return std::nullopt;
 
  return GFConstant(MayBeConstant->Value, GFConstantKind::Scalar);
}
 
// Returns a list of types to use for memory op lowering in MemOps. A partial
// port of findOptimalMemOpLowering in TargetLowering.
static bool findGISelOptimalMemOpLowering(std::vector<LLT> &MemOps,
                                          unsigned Limit, const MemOp &Op,
                                          unsigned DstAS, unsigned SrcAS,
                                          const AttributeList &FuncAttributes,
                                          const TargetLowering &TLI) {
  if (Op.isMemcpyWithFixedDstAlign() && Op.getSrcAlign() < Op.getDstAlign())
    return false;
 
  LLT Ty = TLI.getOptimalMemOpLLT(Op, FuncAttributes);
 
  if (Ty == LLT()) {
    // Use the largest scalar type whose alignment constraints are satisfied.
    // We only need to check DstAlign here as SrcAlign is always greater or
    // equal to DstAlign (or zero).
    Ty = LLT::integer(64);
    if (Op.isFixedDstAlign())
      while (Op.getDstAlign() < Ty.getSizeInBytes() &&
             !TLI.allowsMisalignedMemoryAccesses(Ty, DstAS, Op.getDstAlign()))
        Ty = LLT::integer(Ty.getSizeInBytes());
    assert(Ty.getSizeInBits() > 0 && "Could not find valid type");
    // FIXME: check for the largest legal type we can load/store to.
  }
 
  unsigned NumMemOps = 0;
  uint64_t Size = Op.size();
  while (Size) {
    unsigned TySize = Ty.getSizeInBytes();
    while (TySize > Size) {
      // For now, only use non-vector load / store's for the left-over pieces.
      LLT NewTy = Ty;
      // FIXME: check for mem op safety and legality of the types. Not all of
      // SDAGisms map cleanly to GISel concepts.
      if (NewTy.isVector())
        NewTy =
            NewTy.getSizeInBits() > 64 ? LLT::integer(64) : LLT::integer(32);
      NewTy = LLT::integer(llvm::bit_floor(NewTy.getSizeInBits() - 1));
      unsigned NewTySize = NewTy.getSizeInBytes();
      assert(NewTySize > 0 && "Could not find appropriate type");
 
      // If the new LLT cannot cover all of the remaining bits, then consider
      // issuing a (or a pair of) unaligned and overlapping load / store.
      unsigned Fast;
      // Need to get a VT equivalent for allowMisalignedMemoryAccesses().
      MVT VT = getMVTForLLT(Ty);
      if (NumMemOps && Op.allowOverlap() && NewTySize < Size &&
          TLI.allowsMisalignedMemoryAccesses(
              VT, DstAS, Op.isFixedDstAlign() ? Op.getDstAlign() : Align(1),
              MachineMemOperand::MONone, &Fast) &&
          Fast)
        TySize = Size;
      else {
        Ty = NewTy;
        TySize = NewTySize;
      }
    }
 
    if (++NumMemOps > Limit)
      return false;
 
    MemOps.push_back(Ty);
    Size -= TySize;
  }
 
  return true;
}
 
bool llvm::canLowerMemCpyFamily(const MachineInstr &MI,
                                const MachineRegisterInfo &MRI, unsigned MaxLen,
                                Register &Dst, Register &Src,
                                uint64_t &KnownLen, Align &Alignment,
                                bool &DstAlignCanChange,
                                std::vector<LLT> &MemOps) {
  const unsigned Opc = MI.getOpcode();
  assert((Opc == TargetOpcode::G_MEMCPY ||
          Opc == TargetOpcode::G_MEMCPY_INLINE ||
          Opc == TargetOpcode::G_MEMMOVE || Opc == TargetOpcode::G_MEMSET) &&
         "Expected memcpy like instruction");
 
  auto MMOIt = MI.memoperands_begin();
  const MachineMemOperand *MemOp = *MMOIt;
 
  Align DstAlign = MemOp->getBaseAlign();
  Align SrcAlign;
  Alignment = DstAlign;
  Register Len;
  std::tie(Dst, Src, Len) = MI.getFirst3Regs();
 
  if (Opc != TargetOpcode::G_MEMSET) {
    assert(MMOIt != MI.memoperands_end() && "Expected a second MMO on MI");
    MemOp = *(++MMOIt);
    SrcAlign = MemOp->getBaseAlign();
    Alignment = std::min(DstAlign, SrcAlign);
  }
 
  // See if this is a constant length copy.
  auto LenVRegAndVal = getIConstantVRegValWithLookThrough(Len, MRI);
  if (!LenVRegAndVal) {
    // FIXME: support dynamically sized G_MEMCPY_INLINE
    assert(Opc != TargetOpcode::G_MEMCPY_INLINE &&
           "inline memcpy with dynamic size is not yet supported");
    return false;
  }
 
  KnownLen = LenVRegAndVal->Value.getZExtValue();
  DstAlignCanChange = false;
 
  if (KnownLen == 0)
    return true;
 
  if (Opc != TargetOpcode::G_MEMCPY_INLINE && MaxLen && KnownLen > MaxLen)
    return false;
 
  bool IsVolatile = MemOp->isVolatile();
  const MachineFunction &MF = *MI.getParent()->getParent();
  const auto &TLI = *MF.getSubtarget().getTargetLowering();
  // On Darwin, -Os means optimize for size without hurting performance, so
  // only really optimize for size when -Oz (MinSize) is used.
  bool OptSize = MF.getTarget().getTargetTriple().isOSDarwin()
                     ? MF.getFunction().hasMinSize()
                     : MF.getFunction().hasOptSize();
 
  const MachineFrameInfo &MFI = MF.getFrameInfo();
  MachineInstr *FIDef = getOpcodeDef(TargetOpcode::G_FRAME_INDEX, Dst, MRI);
  if (FIDef && !MFI.isFixedObjectIndex(FIDef->getOperand(1).getIndex()))
    DstAlignCanChange = true;
 
  const auto &DstMMO = **MI.memoperands_begin();
  MachinePointerInfo DstPtrInfo = DstMMO.getPointerInfo();
 
  switch (Opc) {
  case TargetOpcode::G_MEMCPY_INLINE:
  case TargetOpcode::G_MEMCPY: {
    const auto &SrcMMO = **std::next(MI.memoperands_begin());
    MachinePointerInfo SrcPtrInfo = SrcMMO.getPointerInfo();
    uint64_t Limit = Opc == TargetOpcode::G_MEMCPY_INLINE
                         ? std::numeric_limits<uint64_t>::max()
                         : TLI.getMaxStoresPerMemcpy(OptSize);
    return findGISelOptimalMemOpLowering(
        MemOps, Limit,
        MemOp::Copy(KnownLen, DstAlignCanChange, std::min(DstAlign, SrcAlign),
                    SrcAlign, IsVolatile),
        DstPtrInfo.getAddrSpace(), SrcPtrInfo.getAddrSpace(),
        MF.getFunction().getAttributes(), TLI);
  }
  case TargetOpcode::G_MEMMOVE: {
    const auto &SrcMMO = **std::next(MI.memoperands_begin());
    MachinePointerInfo SrcPtrInfo = SrcMMO.getPointerInfo();
    unsigned Limit = TLI.getMaxStoresPerMemmove(OptSize);
    // FIXME: SelectionDAG always passes false for 'AllowOverlap', apparently
    // due to a bug in it's findOptimalMemOpLowering implementation. For now do
    // the same thing here.
    return findGISelOptimalMemOpLowering(
        MemOps, Limit,
        MemOp::Copy(KnownLen, DstAlignCanChange, std::min(DstAlign, SrcAlign),
                    SrcAlign, /*IsVolatile=*/true),
        DstPtrInfo.getAddrSpace(), SrcPtrInfo.getAddrSpace(),
        MF.getFunction().getAttributes(), TLI);
  }
  case TargetOpcode::G_MEMSET: {
    unsigned Limit = TLI.getMaxStoresPerMemset(OptSize);
    auto ValVRegAndVal = getIConstantVRegValWithLookThrough(Src, MRI);
    bool IsZeroVal = ValVRegAndVal && ValVRegAndVal->Value == 0;
    return findGISelOptimalMemOpLowering(
        MemOps, Limit,
        MemOp::Set(KnownLen, DstAlignCanChange, DstAlign,
                   /*IsZeroMemset=*/IsZeroVal,
                   /*IsVolatile=*/IsVolatile),
        DstPtrInfo.getAddrSpace(), ~0u, MF.getFunction().getAttributes(), TLI);
  }
  default:
    llvm_unreachable("Unexpected memcpy-family opcode");
  }
}