ValueTracking.cpp

Go to the documentation of this file.

//===- ValueTracking.cpp - Walk computations to compute properties --------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// This file contains routines that help analyze properties that chains of
// computations have.
//
//===----------------------------------------------------------------------===//
 
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/ADT/APFloat.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/ScopeExit.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/iterator_range.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/AssumeBundleQueries.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/DomConditionCache.h"
#include "llvm/Analysis/GuardUtils.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/Loads.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/OptimizationRemarkEmitter.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/VectorUtils.h"
#include "llvm/Analysis/WithCache.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/ConstantRange.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/DiagnosticInfo.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/EHPersonalities.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GetElementPtrTypeIterator.h"
#include "llvm/IR/GlobalAlias.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/IntrinsicsAArch64.h"
#include "llvm/IR/IntrinsicsAMDGPU.h"
#include "llvm/IR/IntrinsicsRISCV.h"
#include "llvm/IR/IntrinsicsX86.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/User.h"
#include "llvm/IR/Value.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/KnownBits.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/TargetParser/RISCVTargetParser.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <optional>
#include <utility>
 
using namespace llvm;
using namespace llvm::PatternMatch;
 
// Controls the number of uses of the value searched for possible
// dominating comparisons.
static cl::opt<unsigned> DomConditionsMaxUses("dom-conditions-max-uses",
                                              cl::Hidden, cl::init(20));
 
 
/// Returns the bitwidth of the given scalar or pointer type. For vector types,
/// returns the element type's bitwidth.
static unsigned getBitWidth(Type *Ty, const DataLayout &DL) {
  if (unsigned BitWidth = Ty->getScalarSizeInBits())
    return BitWidth;
 
  return DL.getPointerTypeSizeInBits(Ty);
}
 
// Given the provided Value and, potentially, a context instruction, return
// the preferred context instruction (if any).
static const Instruction *safeCxtI(const Value *V, const Instruction *CxtI) {
  // If we've been provided with a context instruction, then use that (provided
  // it has been inserted).
  if (CxtI && CxtI->getParent())
    return CxtI;
 
  // If the value is really an already-inserted instruction, then use that.
  CxtI = dyn_cast<Instruction>(V);
  if (CxtI && CxtI->getParent())
    return CxtI;
 
  return nullptr;
}
 
static const Instruction *safeCxtI(const Value *V1, const Value *V2, const Instruction *CxtI) {
  // If we've been provided with a context instruction, then use that (provided
  // it has been inserted).
  if (CxtI && CxtI->getParent())
    return CxtI;
 
  // If the value is really an already-inserted instruction, then use that.
  CxtI = dyn_cast<Instruction>(V1);
  if (CxtI && CxtI->getParent())
    return CxtI;
 
  CxtI = dyn_cast<Instruction>(V2);
  if (CxtI && CxtI->getParent())
    return CxtI;
 
  return nullptr;
}
 
static bool getShuffleDemandedElts(const ShuffleVectorInst *Shuf,
                                   const APInt &DemandedElts,
                                   APInt &DemandedLHS, APInt &DemandedRHS) {
  if (isa<ScalableVectorType>(Shuf->getType())) {
    assert(DemandedElts == APInt(1,1));
    DemandedLHS = DemandedRHS = DemandedElts;
    return true;
  }
 
  int NumElts =
      cast<FixedVectorType>(Shuf->getOperand(0)->getType())->getNumElements();
  return llvm::getShuffleDemandedElts(NumElts, Shuf->getShuffleMask(),
                                      DemandedElts, DemandedLHS, DemandedRHS);
}
 
static void computeKnownBits(const Value *V, const APInt &DemandedElts,
                             KnownBits &Known, unsigned Depth,
                             const SimplifyQuery &Q);
 
void llvm::computeKnownBits(const Value *V, KnownBits &Known, unsigned Depth,
                            const SimplifyQuery &Q) {
  // Since the number of lanes in a scalable vector is unknown at compile time,
  // we track one bit which is implicitly broadcast to all lanes.  This means
  // that all lanes in a scalable vector are considered demanded.
  auto *FVTy = dyn_cast<FixedVectorType>(V->getType());
  APInt DemandedElts =
      FVTy ? APInt::getAllOnes(FVTy->getNumElements()) : APInt(1, 1);
  ::computeKnownBits(V, DemandedElts, Known, Depth, Q);
}
 
void llvm::computeKnownBits(const Value *V, KnownBits &Known,
                            const DataLayout &DL, unsigned Depth,
                            AssumptionCache *AC, const Instruction *CxtI,
                            const DominatorTree *DT, bool UseInstrInfo) {
  computeKnownBits(
      V, Known, Depth,
      SimplifyQuery(DL, DT, AC, safeCxtI(V, CxtI), UseInstrInfo));
}
 
KnownBits llvm::computeKnownBits(const Value *V, const DataLayout &DL,
                                 unsigned Depth, AssumptionCache *AC,
                                 const Instruction *CxtI,
                                 const DominatorTree *DT, bool UseInstrInfo) {
  return computeKnownBits(
      V, Depth, SimplifyQuery(DL, DT, AC, safeCxtI(V, CxtI), UseInstrInfo));
}
 
KnownBits llvm::computeKnownBits(const Value *V, const APInt &DemandedElts,
                                 const DataLayout &DL, unsigned Depth,
                                 AssumptionCache *AC, const Instruction *CxtI,
                                 const DominatorTree *DT, bool UseInstrInfo) {
  return computeKnownBits(
      V, DemandedElts, Depth,
      SimplifyQuery(DL, DT, AC, safeCxtI(V, CxtI), UseInstrInfo));
}
 
static bool haveNoCommonBitsSetSpecialCases(const Value *LHS, const Value *RHS,
                                            const SimplifyQuery &SQ) {
  // Look for an inverted mask: (X & ~M) op (Y & M).
  {
    Value *M;
    if (match(LHS, m_c_And(m_Not(m_Value(M)), m_Value())) &&
        match(RHS, m_c_And(m_Specific(M), m_Value())) &&
        isGuaranteedNotToBeUndef(M, SQ.AC, SQ.CxtI, SQ.DT))
      return true;
  }
 
  // X op (Y & ~X)
  if (match(RHS, m_c_And(m_Not(m_Specific(LHS)), m_Value())) &&
      isGuaranteedNotToBeUndef(LHS, SQ.AC, SQ.CxtI, SQ.DT))
    return true;
 
  // X op ((X & Y) ^ Y) -- this is the canonical form of the previous pattern
  // for constant Y.
  Value *Y;
  if (match(RHS,
            m_c_Xor(m_c_And(m_Specific(LHS), m_Value(Y)), m_Deferred(Y))) &&
      isGuaranteedNotToBeUndef(LHS, SQ.AC, SQ.CxtI, SQ.DT) &&
      isGuaranteedNotToBeUndef(Y, SQ.AC, SQ.CxtI, SQ.DT))
    return true;
 
  // Peek through extends to find a 'not' of the other side:
  // (ext Y) op ext(~Y)
  if (match(LHS, m_ZExtOrSExt(m_Value(Y))) &&
      match(RHS, m_ZExtOrSExt(m_Not(m_Specific(Y)))) &&
      isGuaranteedNotToBeUndef(Y, SQ.AC, SQ.CxtI, SQ.DT))
    return true;
 
  // Look for: (A & B) op ~(A | B)
  {
    Value *A, *B;
    if (match(LHS, m_And(m_Value(A), m_Value(B))) &&
        match(RHS, m_Not(m_c_Or(m_Specific(A), m_Specific(B)))) &&
        isGuaranteedNotToBeUndef(A, SQ.AC, SQ.CxtI, SQ.DT) &&
        isGuaranteedNotToBeUndef(B, SQ.AC, SQ.CxtI, SQ.DT))
      return true;
  }
 
  return false;
}
 
bool llvm::haveNoCommonBitsSet(const WithCache<const Value *> &LHSCache,
                               const WithCache<const Value *> &RHSCache,
                               const SimplifyQuery &SQ) {
  const Value *LHS = LHSCache.getValue();
  const Value *RHS = RHSCache.getValue();
 
  assert(LHS->getType() == RHS->getType() &&
         "LHS and RHS should have the same type");
  assert(LHS->getType()->isIntOrIntVectorTy() &&
         "LHS and RHS should be integers");
 
  if (haveNoCommonBitsSetSpecialCases(LHS, RHS, SQ) ||
      haveNoCommonBitsSetSpecialCases(RHS, LHS, SQ))
    return true;
 
  return KnownBits::haveNoCommonBitsSet(LHSCache.getKnownBits(SQ),
                                        RHSCache.getKnownBits(SQ));
}
 
bool llvm::isOnlyUsedInZeroComparison(const Instruction *I) {
  return !I->user_empty() && all_of(I->users(), [](const User *U) {
    return match(U, m_ICmp(m_Value(), m_Zero()));
  });
}
 
bool llvm::isOnlyUsedInZeroEqualityComparison(const Instruction *I) {
  return !I->user_empty() && all_of(I->users(), [](const User *U) {
    ICmpInst::Predicate P;
    return match(U, m_ICmp(P, m_Value(), m_Zero())) && ICmpInst::isEquality(P);
  });
}
 
static bool isKnownToBeAPowerOfTwo(const Value *V, bool OrZero, unsigned Depth,
                                   const SimplifyQuery &Q);
 
bool llvm::isKnownToBeAPowerOfTwo(const Value *V, const DataLayout &DL,
                                  bool OrZero, unsigned Depth,
                                  AssumptionCache *AC, const Instruction *CxtI,
                                  const DominatorTree *DT, bool UseInstrInfo) {
  return ::isKnownToBeAPowerOfTwo(
      V, OrZero, Depth,
      SimplifyQuery(DL, DT, AC, safeCxtI(V, CxtI), UseInstrInfo));
}
 
static bool isKnownNonZero(const Value *V, const APInt &DemandedElts,
                           const SimplifyQuery &Q, unsigned Depth);
 
bool llvm::isKnownNonNegative(const Value *V, const SimplifyQuery &SQ,
                              unsigned Depth) {
  return computeKnownBits(V, Depth, SQ).isNonNegative();
}
 
bool llvm::isKnownPositive(const Value *V, const SimplifyQuery &SQ,
                           unsigned Depth) {
  if (auto *CI = dyn_cast<ConstantInt>(V))
    return CI->getValue().isStrictlyPositive();
 
  // If `isKnownNonNegative` ever becomes more sophisticated, make sure to keep
  // this updated.
  KnownBits Known = computeKnownBits(V, Depth, SQ);
  return Known.isNonNegative() &&
         (Known.isNonZero() || isKnownNonZero(V, SQ, Depth));
}
 
bool llvm::isKnownNegative(const Value *V, const SimplifyQuery &SQ,
                           unsigned Depth) {
  return computeKnownBits(V, Depth, SQ).isNegative();
}
 
static bool isKnownNonEqual(const Value *V1, const Value *V2,
                            const APInt &DemandedElts, unsigned Depth,
                            const SimplifyQuery &Q);
 
bool llvm::isKnownNonEqual(const Value *V1, const Value *V2,
                           const DataLayout &DL, AssumptionCache *AC,
                           const Instruction *CxtI, const DominatorTree *DT,
                           bool UseInstrInfo) {
  assert(V1->getType() == V2->getType() &&
         "Testing equality of non-equal types!");
  auto *FVTy = dyn_cast<FixedVectorType>(V1->getType());
  APInt DemandedElts =
      FVTy ? APInt::getAllOnes(FVTy->getNumElements()) : APInt(1, 1);
  return ::isKnownNonEqual(
      V1, V2, DemandedElts, 0,
      SimplifyQuery(DL, DT, AC, safeCxtI(V2, V1, CxtI), UseInstrInfo));
}
 
bool llvm::MaskedValueIsZero(const Value *V, const APInt &Mask,
                             const SimplifyQuery &SQ, unsigned Depth) {
  KnownBits Known(Mask.getBitWidth());
  computeKnownBits(V, Known, Depth, SQ);
  return Mask.isSubsetOf(Known.Zero);
}
 
static unsigned ComputeNumSignBits(const Value *V, const APInt &DemandedElts,
                                   unsigned Depth, const SimplifyQuery &Q);
 
static unsigned ComputeNumSignBits(const Value *V, unsigned Depth,
                                   const SimplifyQuery &Q) {
  auto *FVTy = dyn_cast<FixedVectorType>(V->getType());
  APInt DemandedElts =
      FVTy ? APInt::getAllOnes(FVTy->getNumElements()) : APInt(1, 1);
  return ComputeNumSignBits(V, DemandedElts, Depth, Q);
}
 
unsigned llvm::ComputeNumSignBits(const Value *V, const DataLayout &DL,
                                  unsigned Depth, AssumptionCache *AC,
                                  const Instruction *CxtI,
                                  const DominatorTree *DT, bool UseInstrInfo) {
  return ::ComputeNumSignBits(
      V, Depth, SimplifyQuery(DL, DT, AC, safeCxtI(V, CxtI), UseInstrInfo));
}
 
unsigned llvm::ComputeMaxSignificantBits(const Value *V, const DataLayout &DL,
                                         unsigned Depth, AssumptionCache *AC,
                                         const Instruction *CxtI,
                                         const DominatorTree *DT) {
  unsigned SignBits = ComputeNumSignBits(V, DL, Depth, AC, CxtI, DT);
  return V->getType()->getScalarSizeInBits() - SignBits + 1;
}
 
static void computeKnownBitsAddSub(bool Add, const Value *Op0, const Value *Op1,
                                   bool NSW, bool NUW,
                                   const APInt &DemandedElts,
                                   KnownBits &KnownOut, KnownBits &Known2,
                                   unsigned Depth, const SimplifyQuery &Q) {
  computeKnownBits(Op1, DemandedElts, KnownOut, Depth + 1, Q);
 
  // If one operand is unknown and we have no nowrap information,
  // the result will be unknown independently of the second operand.
  if (KnownOut.isUnknown() && !NSW && !NUW)
    return;
 
  computeKnownBits(Op0, DemandedElts, Known2, Depth + 1, Q);
  KnownOut = KnownBits::computeForAddSub(Add, NSW, NUW, Known2, KnownOut);
}
 
static void computeKnownBitsMul(const Value *Op0, const Value *Op1, bool NSW,
                                const APInt &DemandedElts, KnownBits &Known,
                                KnownBits &Known2, unsigned Depth,
                                const SimplifyQuery &Q) {
  computeKnownBits(Op1, DemandedElts, Known, Depth + 1, Q);
  computeKnownBits(Op0, DemandedElts, Known2, Depth + 1, Q);
 
  bool isKnownNegative = false;
  bool isKnownNonNegative = false;
  // If the multiplication is known not to overflow, compute the sign bit.
  if (NSW) {
    if (Op0 == Op1) {
      // The product of a number with itself is non-negative.
      isKnownNonNegative = true;
    } else {
      bool isKnownNonNegativeOp1 = Known.isNonNegative();
      bool isKnownNonNegativeOp0 = Known2.isNonNegative();
      bool isKnownNegativeOp1 = Known.isNegative();
      bool isKnownNegativeOp0 = Known2.isNegative();
      // The product of two numbers with the same sign is non-negative.
      isKnownNonNegative = (isKnownNegativeOp1 && isKnownNegativeOp0) ||
                           (isKnownNonNegativeOp1 && isKnownNonNegativeOp0);
      // The product of a negative number and a non-negative number is either
      // negative or zero.
      if (!isKnownNonNegative)
        isKnownNegative =
            (isKnownNegativeOp1 && isKnownNonNegativeOp0 &&
             Known2.isNonZero()) ||
            (isKnownNegativeOp0 && isKnownNonNegativeOp1 && Known.isNonZero());
    }
  }
 
  bool SelfMultiply = Op0 == Op1;
  if (SelfMultiply)
    SelfMultiply &=
        isGuaranteedNotToBeUndef(Op0, Q.AC, Q.CxtI, Q.DT, Depth + 1);
  Known = KnownBits::mul(Known, Known2, SelfMultiply);
 
  // Only make use of no-wrap flags if we failed to compute the sign bit
  // directly.  This matters if the multiplication always overflows, in
  // which case we prefer to follow the result of the direct computation,
  // though as the program is invoking undefined behaviour we can choose
  // whatever we like here.
  if (isKnownNonNegative && !Known.isNegative())
    Known.makeNonNegative();
  else if (isKnownNegative && !Known.isNonNegative())
    Known.makeNegative();
}
 
void llvm::computeKnownBitsFromRangeMetadata(const MDNode &Ranges,
                                             KnownBits &Known) {
  unsigned BitWidth = Known.getBitWidth();
  unsigned NumRanges = Ranges.getNumOperands() / 2;
  assert(NumRanges >= 1);
 
  Known.Zero.setAllBits();
  Known.One.setAllBits();
 
  for (unsigned i = 0; i < NumRanges; ++i) {
    ConstantInt *Lower =
        mdconst::extract<ConstantInt>(Ranges.getOperand(2 * i + 0));
    ConstantInt *Upper =
        mdconst::extract<ConstantInt>(Ranges.getOperand(2 * i + 1));
    ConstantRange Range(Lower->getValue(), Upper->getValue());
 
    // The first CommonPrefixBits of all values in Range are equal.
    unsigned CommonPrefixBits =
        (Range.getUnsignedMax() ^ Range.getUnsignedMin()).countl_zero();
    APInt Mask = APInt::getHighBitsSet(BitWidth, CommonPrefixBits);
    APInt UnsignedMax = Range.getUnsignedMax().zextOrTrunc(BitWidth);
    Known.One &= UnsignedMax & Mask;
    Known.Zero &= ~UnsignedMax & Mask;
  }
}
 
static bool isEphemeralValueOf(const Instruction *I, const Value *E) {
  SmallVector<const Value *, 16> WorkSet(1, I);
  SmallPtrSet<const Value *, 32> Visited;
  SmallPtrSet<const Value *, 16> EphValues;
 
  // The instruction defining an assumption's condition itself is always
  // considered ephemeral to that assumption (even if it has other
  // non-ephemeral users). See r246696's test case for an example.
  if (is_contained(I->operands(), E))
    return true;
 
  while (!WorkSet.empty()) {
    const Value *V = WorkSet.pop_back_val();
    if (!Visited.insert(V).second)
      continue;
 
    // If all uses of this value are ephemeral, then so is this value.
    if (llvm::all_of(V->users(), [&](const User *U) {
                                   return EphValues.count(U);
                                 })) {
      if (V == E)
        return true;
 
      if (V == I || (isa<Instruction>(V) &&
                     !cast<Instruction>(V)->mayHaveSideEffects() &&
                     !cast<Instruction>(V)->isTerminator())) {
       EphValues.insert(V);
       if (const User *U = dyn_cast<User>(V))
         append_range(WorkSet, U->operands());
      }
    }
  }
 
  return false;
}
 
// Is this an intrinsic that cannot be speculated but also cannot trap?
bool llvm::isAssumeLikeIntrinsic(const Instruction *I) {
  if (const IntrinsicInst *CI = dyn_cast<IntrinsicInst>(I))
    return CI->isAssumeLikeIntrinsic();
 
  return false;
}
 
bool llvm::isValidAssumeForContext(const Instruction *Inv,
                                   const Instruction *CxtI,
                                   const DominatorTree *DT,
                                   bool AllowEphemerals) {
  // There are two restrictions on the use of an assume:
  //  1. The assume must dominate the context (or the control flow must
  //     reach the assume whenever it reaches the context).
  //  2. The context must not be in the assume's set of ephemeral values
  //     (otherwise we will use the assume to prove that the condition
  //     feeding the assume is trivially true, thus causing the removal of
  //     the assume).
 
  if (Inv->getParent() == CxtI->getParent()) {
    // If Inv and CtxI are in the same block, check if the assume (Inv) is first
    // in the BB.
    if (Inv->comesBefore(CxtI))
      return true;
 
    // Don't let an assume affect itself - this would cause the problems
    // `isEphemeralValueOf` is trying to prevent, and it would also make
    // the loop below go out of bounds.
    if (!AllowEphemerals && Inv == CxtI)
      return false;
 
    // The context comes first, but they're both in the same block.
    // Make sure there is nothing in between that might interrupt
    // the control flow, not even CxtI itself.
    // We limit the scan distance between the assume and its context instruction
    // to avoid a compile-time explosion. This limit is chosen arbitrarily, so
    // it can be adjusted if needed (could be turned into a cl::opt).
    auto Range = make_range(CxtI->getIterator(), Inv->getIterator());
    if (!isGuaranteedToTransferExecutionToSuccessor(Range, 15))
      return false;
 
    return AllowEphemerals || !isEphemeralValueOf(Inv, CxtI);
  }
 
  // Inv and CxtI are in different blocks.
  if (DT) {
    if (DT->dominates(Inv, CxtI))
      return true;
  } else if (Inv->getParent() == CxtI->getParent()->getSinglePredecessor()) {
    // We don't have a DT, but this trivially dominates.
    return true;
  }
 
  return false;
}
 
// TODO: cmpExcludesZero misses many cases where `RHS` is non-constant but
// we still have enough information about `RHS` to conclude non-zero. For
// example Pred=EQ, RHS=isKnownNonZero. cmpExcludesZero is called in loops
// so the extra compile time may not be worth it, but possibly a second API
// should be created for use outside of loops.
static bool cmpExcludesZero(CmpInst::Predicate Pred, const Value *RHS) {
  // v u> y implies v != 0.
  if (Pred == ICmpInst::ICMP_UGT)
    return true;
 
  // Special-case v != 0 to also handle v != null.
  if (Pred == ICmpInst::ICMP_NE)
    return match(RHS, m_Zero());
 
  // All other predicates - rely on generic ConstantRange handling.
  const APInt *C;
  auto Zero = APInt::getZero(RHS->getType()->getScalarSizeInBits());
  if (match(RHS, m_APInt(C))) {
    ConstantRange TrueValues = ConstantRange::makeExactICmpRegion(Pred, *C);
    return !TrueValues.contains(Zero);
  }
 
  auto *VC = dyn_cast<ConstantDataVector>(RHS);
  if (VC == nullptr)
    return false;
 
  for (unsigned ElemIdx = 0, NElem = VC->getNumElements(); ElemIdx < NElem;
       ++ElemIdx) {
    ConstantRange TrueValues = ConstantRange::makeExactICmpRegion(
        Pred, VC->getElementAsAPInt(ElemIdx));
    if (TrueValues.contains(Zero))
      return false;
  }
  return true;
}
 
static bool isKnownNonZeroFromAssume(const Value *V, const SimplifyQuery &Q) {
  // Use of assumptions is context-sensitive. If we don't have a context, we
  // cannot use them!
  if (!Q.AC || !Q.CxtI)
    return false;
 
  for (AssumptionCache::ResultElem &Elem : Q.AC->assumptionsFor(V)) {
    if (!Elem.Assume)
      continue;
 
    AssumeInst *I = cast<AssumeInst>(Elem.Assume);
    assert(I->getFunction() == Q.CxtI->getFunction() &&
           "Got assumption for the wrong function!");
 
    if (Elem.Index != AssumptionCache::ExprResultIdx) {
      if (!V->getType()->isPointerTy())
        continue;
      if (RetainedKnowledge RK = getKnowledgeFromBundle(
              *I, I->bundle_op_info_begin()[Elem.Index])) {
        if (RK.WasOn == V &&
            (RK.AttrKind == Attribute::NonNull ||
             (RK.AttrKind == Attribute::Dereferenceable &&
              !NullPointerIsDefined(Q.CxtI->getFunction(),
                                    V->getType()->getPointerAddressSpace()))) &&
            isValidAssumeForContext(I, Q.CxtI, Q.DT))
          return true;
      }
      continue;
    }
 
    // Warning: This loop can end up being somewhat performance sensitive.
    // We're running this loop for once for each value queried resulting in a
    // runtime of ~O(#assumes * #values).
 
    Value *RHS;
    CmpInst::Predicate Pred;
    auto m_V = m_CombineOr(m_Specific(V), m_PtrToInt(m_Specific(V)));
    if (!match(I->getArgOperand(0), m_c_ICmp(Pred, m_V, m_Value(RHS))))
      return false;
 
    if (cmpExcludesZero(Pred, RHS) && isValidAssumeForContext(I, Q.CxtI, Q.DT))
      return true;
  }
 
  return false;
}
 
static void computeKnownBitsFromCmp(const Value *V, CmpInst::Predicate Pred,
                                    Value *LHS, Value *RHS, KnownBits &Known,
                                    const SimplifyQuery &Q) {
  if (RHS->getType()->isPointerTy()) {
    // Handle comparison of pointer to null explicitly, as it will not be
    // covered by the m_APInt() logic below.
    if (LHS == V && match(RHS, m_Zero())) {
      switch (Pred) {
      case ICmpInst::ICMP_EQ:
        Known.setAllZero();
        break;
      case ICmpInst::ICMP_SGE:
      case ICmpInst::ICMP_SGT:
        Known.makeNonNegative();
        break;
      case ICmpInst::ICMP_SLT:
        Known.makeNegative();
        break;
      default:
        break;
      }
    }
    return;
  }
 
  unsigned BitWidth = Known.getBitWidth();
  auto m_V =
      m_CombineOr(m_Specific(V), m_PtrToIntSameSize(Q.DL, m_Specific(V)));
 
  Value *Y;
  const APInt *Mask, *C;
  uint64_t ShAmt;
  switch (Pred) {
  case ICmpInst::ICMP_EQ:
    // assume(V = C)
    if (match(LHS, m_V) && match(RHS, m_APInt(C))) {
      Known = Known.unionWith(KnownBits::makeConstant(*C));
      // assume(V & Mask = C)
    } else if (match(LHS, m_c_And(m_V, m_Value(Y))) &&
               match(RHS, m_APInt(C))) {
      // For one bits in Mask, we can propagate bits from C to V.
      Known.One |= *C;
      if (match(Y, m_APInt(Mask)))
        Known.Zero |= ~*C & *Mask;
      // assume(V | Mask = C)
    } else if (match(LHS, m_c_Or(m_V, m_Value(Y))) && match(RHS, m_APInt(C))) {
      // For zero bits in Mask, we can propagate bits from C to V.
      Known.Zero |= ~*C;
      if (match(Y, m_APInt(Mask)))
        Known.One |= *C & ~*Mask;
      // assume(V ^ Mask = C)
    } else if (match(LHS, m_Xor(m_V, m_APInt(Mask))) &&
               match(RHS, m_APInt(C))) {
      // Equivalent to assume(V == Mask ^ C)
      Known = Known.unionWith(KnownBits::makeConstant(*C ^ *Mask));
      // assume(V << ShAmt = C)
    } else if (match(LHS, m_Shl(m_V, m_ConstantInt(ShAmt))) &&
               match(RHS, m_APInt(C)) && ShAmt < BitWidth) {
      // For those bits in C that are known, we can propagate them to known
      // bits in V shifted to the right by ShAmt.
      KnownBits RHSKnown = KnownBits::makeConstant(*C);
      RHSKnown.Zero.lshrInPlace(ShAmt);
      RHSKnown.One.lshrInPlace(ShAmt);
      Known = Known.unionWith(RHSKnown);
      // assume(V >> ShAmt = C)
    } else if (match(LHS, m_Shr(m_V, m_ConstantInt(ShAmt))) &&
               match(RHS, m_APInt(C)) && ShAmt < BitWidth) {
      KnownBits RHSKnown = KnownBits::makeConstant(*C);
      // For those bits in RHS that are known, we can propagate them to known
      // bits in V shifted to the right by C.
      Known.Zero |= RHSKnown.Zero << ShAmt;
      Known.One |= RHSKnown.One << ShAmt;
    }
    break;
  case ICmpInst::ICMP_NE: {
    // assume (V & B != 0) where B is a power of 2
    const APInt *BPow2;
    if (match(LHS, m_And(m_V, m_Power2(BPow2))) && match(RHS, m_Zero()))
      Known.One |= *BPow2;
    break;
  }
  default:
    if (match(RHS, m_APInt(C))) {
      const APInt *Offset = nullptr;
      if (match(LHS, m_CombineOr(m_V, m_AddLike(m_V, m_APInt(Offset))))) {
        ConstantRange LHSRange = ConstantRange::makeAllowedICmpRegion(Pred, *C);
        if (Offset)
          LHSRange = LHSRange.sub(*Offset);
        Known = Known.unionWith(LHSRange.toKnownBits());
      }
      if (Pred == ICmpInst::ICMP_UGT || Pred == ICmpInst::ICMP_UGE) {
        // X & Y u> C     -> X u> C && Y u> C
        // X nuw- Y u> C  -> X u> C
        if (match(LHS, m_c_And(m_V, m_Value())) ||
            match(LHS, m_NUWSub(m_V, m_Value())))
          Known.One.setHighBits(
              (*C + (Pred == ICmpInst::ICMP_UGT)).countLeadingOnes());
      }
      if (Pred == ICmpInst::ICMP_ULT || Pred == ICmpInst::ICMP_ULE) {
        // X | Y u< C    -> X u< C && Y u< C
        // X nuw+ Y u< C -> X u< C && Y u< C
        if (match(LHS, m_c_Or(m_V, m_Value())) ||
            match(LHS, m_c_NUWAdd(m_V, m_Value()))) {
          Known.Zero.setHighBits(
              (*C - (Pred == ICmpInst::ICMP_ULT)).countLeadingZeros());
        }
      }
    }
    break;
  }
}
 
static void computeKnownBitsFromICmpCond(const Value *V, ICmpInst *Cmp,
                                         KnownBits &Known,
                                         const SimplifyQuery &SQ, bool Invert) {
  ICmpInst::Predicate Pred =
      Invert ? Cmp->getInversePredicate() : Cmp->getPredicate();
  Value *LHS = Cmp->getOperand(0);
  Value *RHS = Cmp->getOperand(1);
 
  // Handle icmp pred (trunc V), C
  if (match(LHS, m_Trunc(m_Specific(V)))) {
    KnownBits DstKnown(LHS->getType()->getScalarSizeInBits());
    computeKnownBitsFromCmp(LHS, Pred, LHS, RHS, DstKnown, SQ);
    Known = Known.unionWith(DstKnown.anyext(Known.getBitWidth()));
    return;
  }
 
  computeKnownBitsFromCmp(V, Pred, LHS, RHS, Known, SQ);
}
 
static void computeKnownBitsFromCond(const Value *V, Value *Cond,
                                     KnownBits &Known, unsigned Depth,
                                     const SimplifyQuery &SQ, bool Invert) {
  Value *A, *B;
  if (Depth < MaxAnalysisRecursionDepth &&
      match(Cond, m_LogicalOp(m_Value(A), m_Value(B)))) {
    KnownBits Known2(Known.getBitWidth());
    KnownBits Known3(Known.getBitWidth());
    computeKnownBitsFromCond(V, A, Known2, Depth + 1, SQ, Invert);
    computeKnownBitsFromCond(V, B, Known3, Depth + 1, SQ, Invert);
    if (Invert ? match(Cond, m_LogicalOr(m_Value(), m_Value()))
               : match(Cond, m_LogicalAnd(m_Value(), m_Value())))
      Known2 = Known2.unionWith(Known3);
    else
      Known2 = Known2.intersectWith(Known3);
    Known = Known.unionWith(Known2);
  }
 
  if (auto *Cmp = dyn_cast<ICmpInst>(Cond))
    computeKnownBitsFromICmpCond(V, Cmp, Known, SQ, Invert);
}
 
void llvm::computeKnownBitsFromContext(const Value *V, KnownBits &Known,
                                       unsigned Depth, const SimplifyQuery &Q) {
  // Handle injected condition.
  if (Q.CC && Q.CC->AffectedValues.contains(V))
    computeKnownBitsFromCond(V, Q.CC->Cond, Known, Depth, Q, Q.CC->Invert);
 
  if (!Q.CxtI)
    return;
 
  if (Q.DC && Q.DT) {
    // Handle dominating conditions.
    for (BranchInst *BI : Q.DC->conditionsFor(V)) {
      BasicBlockEdge Edge0(BI->getParent(), BI->getSuccessor(0));
      if (Q.DT->dominates(Edge0, Q.CxtI->getParent()))
        computeKnownBitsFromCond(V, BI->getCondition(), Known, Depth, Q,
                                 /*Invert*/ false);
 
      BasicBlockEdge Edge1(BI->getParent(), BI->getSuccessor(1));
      if (Q.DT->dominates(Edge1, Q.CxtI->getParent()))
        computeKnownBitsFromCond(V, BI->getCondition(), Known, Depth, Q,
                                 /*Invert*/ true);
    }
 
    if (Known.hasConflict())
      Known.resetAll();
  }
 
  if (!Q.AC)
    return;
 
  unsigned BitWidth = Known.getBitWidth();
 
  // Note that the patterns below need to be kept in sync with the code
  // in AssumptionCache::updateAffectedValues.
 
  for (AssumptionCache::ResultElem &Elem : Q.AC->assumptionsFor(V)) {
    if (!Elem.Assume)
      continue;
 
    AssumeInst *I = cast<AssumeInst>(Elem.Assume);
    assert(I->getParent()->getParent() == Q.CxtI->getParent()->getParent() &&
           "Got assumption for the wrong function!");
 
    if (Elem.Index != AssumptionCache::ExprResultIdx) {
      if (!V->getType()->isPointerTy())
        continue;
      if (RetainedKnowledge RK = getKnowledgeFromBundle(
              *I, I->bundle_op_info_begin()[Elem.Index])) {
        if (RK.WasOn == V && RK.AttrKind == Attribute::Alignment &&
            isPowerOf2_64(RK.ArgValue) &&
            isValidAssumeForContext(I, Q.CxtI, Q.DT))
          Known.Zero.setLowBits(Log2_64(RK.ArgValue));
      }
      continue;
    }
 
    // Warning: This loop can end up being somewhat performance sensitive.
    // We're running this loop for once for each value queried resulting in a
    // runtime of ~O(#assumes * #values).
 
    Value *Arg = I->getArgOperand(0);
 
    if (Arg == V && isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
      assert(BitWidth == 1 && "assume operand is not i1?");
      (void)BitWidth;
      Known.setAllOnes();
      return;
    }
    if (match(Arg, m_Not(m_Specific(V))) &&
        isValidAssumeForContext(I, Q.CxtI, Q.DT)) {
      assert(BitWidth == 1 && "assume operand is not i1?");
      (void)BitWidth;
      Known.setAllZero();
      return;
    }
 
    // The remaining tests are all recursive, so bail out if we hit the limit.
    if (Depth == MaxAnalysisRecursionDepth)
      continue;
 
    ICmpInst *Cmp = dyn_cast<ICmpInst>(Arg);
    if (!Cmp)
      continue;
 
    if (!isValidAssumeForContext(I, Q.CxtI, Q.DT))
      continue;
 
    computeKnownBitsFromICmpCond(V, Cmp, Known, Q, /*Invert=*/false);
  }
 
  // Conflicting assumption: Undefined behavior will occur on this execution
  // path.
  if (Known.hasConflict())
    Known.resetAll();
}
 
/// Compute known bits from a shift operator, including those with a
/// non-constant shift amount. Known is the output of this function. Known2 is a
/// pre-allocated temporary with the same bit width as Known and on return
/// contains the known bit of the shift value source. KF is an
/// operator-specific function that, given the known-bits and a shift amount,
/// compute the implied known-bits of the shift operator's result respectively
/// for that shift amount. The results from calling KF are conservatively
/// combined for all permitted shift amounts.
static void computeKnownBitsFromShiftOperator(
    const Operator *I, const APInt &DemandedElts, KnownBits &Known,
    KnownBits &Known2, unsigned Depth, const SimplifyQuery &Q,
    function_ref<KnownBits(const KnownBits &, const KnownBits &, bool)> KF) {
  computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
  computeKnownBits(I->getOperand(1), DemandedElts, Known, Depth + 1, Q);
  // To limit compile-time impact, only query isKnownNonZero() if we know at
  // least something about the shift amount.
  bool ShAmtNonZero =
      Known.isNonZero() ||
      (Known.getMaxValue().ult(Known.getBitWidth()) &&
       isKnownNonZero(I->getOperand(1), DemandedElts, Q, Depth + 1));
  Known = KF(Known2, Known, ShAmtNonZero);
}
 
static KnownBits
getKnownBitsFromAndXorOr(const Operator *I, const APInt &DemandedElts,
                         const KnownBits &KnownLHS, const KnownBits &KnownRHS,
                         unsigned Depth, const SimplifyQuery &Q) {
  unsigned BitWidth = KnownLHS.getBitWidth();
  KnownBits KnownOut(BitWidth);
  bool IsAnd = false;
  bool HasKnownOne = !KnownLHS.One.isZero() || !KnownRHS.One.isZero();
  Value *X = nullptr, *Y = nullptr;
 
  switch (I->getOpcode()) {
  case Instruction::And:
    KnownOut = KnownLHS & KnownRHS;
    IsAnd = true;
    // and(x, -x) is common idioms that will clear all but lowest set
    // bit. If we have a single known bit in x, we can clear all bits
    // above it.
    // TODO: instcombine often reassociates independent `and` which can hide
    // this pattern. Try to match and(x, and(-x, y)) / and(and(x, y), -x).
    if (HasKnownOne && match(I, m_c_And(m_Value(X), m_Neg(m_Deferred(X))))) {
      // -(-x) == x so using whichever (LHS/RHS) gets us a better result.
      if (KnownLHS.countMaxTrailingZeros() <= KnownRHS.countMaxTrailingZeros())
        KnownOut = KnownLHS.blsi();
      else
        KnownOut = KnownRHS.blsi();
    }
    break;
  case Instruction::Or:
    KnownOut = KnownLHS | KnownRHS;
    break;
  case Instruction::Xor:
    KnownOut = KnownLHS ^ KnownRHS;
    // xor(x, x-1) is common idioms that will clear all but lowest set
    // bit. If we have a single known bit in x, we can clear all bits
    // above it.
    // TODO: xor(x, x-1) is often rewritting as xor(x, x-C) where C !=
    // -1 but for the purpose of demanded bits (xor(x, x-C) &
    // Demanded) == (xor(x, x-1) & Demanded). Extend the xor pattern
    // to use arbitrary C if xor(x, x-C) as the same as xor(x, x-1).
    if (HasKnownOne &&
        match(I, m_c_Xor(m_Value(X), m_Add(m_Deferred(X), m_AllOnes())))) {
      const KnownBits &XBits = I->getOperand(0) == X ? KnownLHS : KnownRHS;
      KnownOut = XBits.blsmsk();
    }
    break;
  default:
    llvm_unreachable("Invalid Op used in 'analyzeKnownBitsFromAndXorOr'");
  }
 
  // and(x, add (x, -1)) is a common idiom that always clears the low bit;
  // xor/or(x, add (x, -1)) is an idiom that will always set the low bit.
  // here we handle the more general case of adding any odd number by
  // matching the form and/xor/or(x, add(x, y)) where y is odd.
  // TODO: This could be generalized to clearing any bit set in y where the
  // following bit is known to be unset in y.
  if (!KnownOut.Zero[0] && !KnownOut.One[0] &&
      (match(I, m_c_BinOp(m_Value(X), m_c_Add(m_Deferred(X), m_Value(Y)))) ||
       match(I, m_c_BinOp(m_Value(X), m_Sub(m_Deferred(X), m_Value(Y)))) ||
       match(I, m_c_BinOp(m_Value(X), m_Sub(m_Value(Y), m_Deferred(X)))))) {
    KnownBits KnownY(BitWidth);
    computeKnownBits(Y, DemandedElts, KnownY, Depth + 1, Q);
    if (KnownY.countMinTrailingOnes() > 0) {
      if (IsAnd)
        KnownOut.Zero.setBit(0);
      else
        KnownOut.One.setBit(0);
    }
  }
  return KnownOut;
}
 
static KnownBits computeKnownBitsForHorizontalOperation(
    const Operator *I, const APInt &DemandedElts, unsigned Depth,
    const SimplifyQuery &Q,
    const function_ref<KnownBits(const KnownBits &, const KnownBits &)>
        KnownBitsFunc) {
  APInt DemandedEltsLHS, DemandedEltsRHS;
  getHorizDemandedEltsForFirstOperand(Q.DL.getTypeSizeInBits(I->getType()),
                                      DemandedElts, DemandedEltsLHS,
                                      DemandedEltsRHS);
 
  const auto ComputeForSingleOpFunc =
      [Depth, &Q, KnownBitsFunc](const Value *Op, APInt &DemandedEltsOp) {
        return KnownBitsFunc(
            computeKnownBits(Op, DemandedEltsOp, Depth + 1, Q),
            computeKnownBits(Op, DemandedEltsOp << 1, Depth + 1, Q));
      };
 
  if (DemandedEltsRHS.isZero())
    return ComputeForSingleOpFunc(I->getOperand(0), DemandedEltsLHS);
  if (DemandedEltsLHS.isZero())
    return ComputeForSingleOpFunc(I->getOperand(1), DemandedEltsRHS);
 
  return ComputeForSingleOpFunc(I->getOperand(0), DemandedEltsLHS)
      .intersectWith(ComputeForSingleOpFunc(I->getOperand(1), DemandedEltsRHS));
}
 
// Public so this can be used in `SimplifyDemandedUseBits`.
KnownBits llvm::analyzeKnownBitsFromAndXorOr(const Operator *I,
                                             const KnownBits &KnownLHS,
                                             const KnownBits &KnownRHS,
                                             unsigned Depth,
                                             const SimplifyQuery &SQ) {
  auto *FVTy = dyn_cast<FixedVectorType>(I->getType());
  APInt DemandedElts =
      FVTy ? APInt::getAllOnes(FVTy->getNumElements()) : APInt(1, 1);
 
  return getKnownBitsFromAndXorOr(I, DemandedElts, KnownLHS, KnownRHS, Depth,
                                  SQ);
}
 
ConstantRange llvm::getVScaleRange(const Function *F, unsigned BitWidth) {
  Attribute Attr = F->getFnAttribute(Attribute::VScaleRange);
  // Without vscale_range, we only know that vscale is non-zero.
  if (!Attr.isValid())
    return ConstantRange(APInt(BitWidth, 1), APInt::getZero(BitWidth));
 
  unsigned AttrMin = Attr.getVScaleRangeMin();
  // Minimum is larger than vscale width, result is always poison.
  if ((unsigned)llvm::bit_width(AttrMin) > BitWidth)
    return ConstantRange::getEmpty(BitWidth);
 
  APInt Min(BitWidth, AttrMin);
  std::optional<unsigned> AttrMax = Attr.getVScaleRangeMax();
  if (!AttrMax || (unsigned)llvm::bit_width(*AttrMax) > BitWidth)
    return ConstantRange(Min, APInt::getZero(BitWidth));
 
  return ConstantRange(Min, APInt(BitWidth, *AttrMax) + 1);
}
 
void llvm::adjustKnownBitsForSelectArm(KnownBits &Known, Value *Cond,
                                       Value *Arm, bool Invert, unsigned Depth,
                                       const SimplifyQuery &Q) {
  // If we have a constant arm, we are done.
  if (Known.isConstant())
    return;
 
  // See what condition implies about the bits of the select arm.
  KnownBits CondRes(Known.getBitWidth());
  computeKnownBitsFromCond(Arm, Cond, CondRes, Depth + 1, Q, Invert);
  // If we don't get any information from the condition, no reason to
  // proceed.
  if (CondRes.isUnknown())
    return;
 
  // We can have conflict if the condition is dead. I.e if we have
  // (x | 64) < 32 ? (x | 64) : y
  // we will have conflict at bit 6 from the condition/the `or`.
  // In that case just return. Its not particularly important
  // what we do, as this select is going to be simplified soon.
  CondRes = CondRes.unionWith(Known);
  if (CondRes.hasConflict())
    return;
 
  // Finally make sure the information we found is valid. This is relatively
  // expensive so it's left for the very end.
  if (!isGuaranteedNotToBeUndef(Arm, Q.AC, Q.CxtI, Q.DT, Depth + 1))
    return;
 
  // Finally, we know we get information from the condition and its valid,
  // so return it.
  Known = CondRes;
}
 
static void computeKnownBitsFromOperator(const Operator *I,
                                         const APInt &DemandedElts,
                                         KnownBits &Known, unsigned Depth,
                                         const SimplifyQuery &Q) {
  unsigned BitWidth = Known.getBitWidth();
 
  KnownBits Known2(BitWidth);
  switch (I->getOpcode()) {
  default: break;
  case Instruction::Load:
    if (MDNode *MD =
            Q.IIQ.getMetadata(cast<LoadInst>(I), LLVMContext::MD_range))
      computeKnownBitsFromRangeMetadata(*MD, Known);
    break;
  case Instruction::And:
    computeKnownBits(I->getOperand(1), DemandedElts, Known, Depth + 1, Q);
    computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
 
    Known = getKnownBitsFromAndXorOr(I, DemandedElts, Known2, Known, Depth, Q);
    break;
  case Instruction::Or:
    computeKnownBits(I->getOperand(1), DemandedElts, Known, Depth + 1, Q);
    computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
 
    Known = getKnownBitsFromAndXorOr(I, DemandedElts, Known2, Known, Depth, Q);
    break;
  case Instruction::Xor:
    computeKnownBits(I->getOperand(1), DemandedElts, Known, Depth + 1, Q);
    computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
 
    Known = getKnownBitsFromAndXorOr(I, DemandedElts, Known2, Known, Depth, Q);
    break;
  case Instruction::Mul: {
    bool NSW = Q.IIQ.hasNoSignedWrap(cast<OverflowingBinaryOperator>(I));
    computeKnownBitsMul(I->getOperand(0), I->getOperand(1), NSW, DemandedElts,
                        Known, Known2, Depth, Q);
    break;
  }
  case Instruction::UDiv: {
    computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
    computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
    Known =
        KnownBits::udiv(Known, Known2, Q.IIQ.isExact(cast<BinaryOperator>(I)));
    break;
  }
  case Instruction::SDiv: {
    computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
    computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
    Known =
        KnownBits::sdiv(Known, Known2, Q.IIQ.isExact(cast<BinaryOperator>(I)));
    break;
  }
  case Instruction::Select: {
    auto ComputeForArm = [&](Value *Arm, bool Invert) {
      KnownBits Res(Known.getBitWidth());
      computeKnownBits(Arm, DemandedElts, Res, Depth + 1, Q);
      adjustKnownBitsForSelectArm(Res, I->getOperand(0), Arm, Invert, Depth, Q);
      return Res;
    };
    // Only known if known in both the LHS and RHS.
    Known =
        ComputeForArm(I->getOperand(1), /*Invert=*/false)
            .intersectWith(ComputeForArm(I->getOperand(2), /*Invert=*/true));
    break;
  }
  case Instruction::FPTrunc:
  case Instruction::FPExt:
  case Instruction::FPToUI:
  case Instruction::FPToSI:
  case Instruction::SIToFP:
  case Instruction::UIToFP:
    break; // Can't work with floating point.
  case Instruction::PtrToInt:
  case Instruction::IntToPtr:
    // Fall through and handle them the same as zext/trunc.
    [[fallthrough]];
  case Instruction::ZExt:
  case Instruction::Trunc: {
    Type *SrcTy = I->getOperand(0)->getType();
 
    unsigned SrcBitWidth;
    // Note that we handle pointer operands here because of inttoptr/ptrtoint
    // which fall through here.
    Type *ScalarTy = SrcTy->getScalarType();
    SrcBitWidth = ScalarTy->isPointerTy() ?
      Q.DL.getPointerTypeSizeInBits(ScalarTy) :
      Q.DL.getTypeSizeInBits(ScalarTy);
 
    assert(SrcBitWidth && "SrcBitWidth can't be zero");
    Known = Known.anyextOrTrunc(SrcBitWidth);
    computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
    if (auto *Inst = dyn_cast<PossiblyNonNegInst>(I);
        Inst && Inst->hasNonNeg() && !Known.isNegative())
      Known.makeNonNegative();
    Known = Known.zextOrTrunc(BitWidth);
    break;
  }
  case Instruction::BitCast: {
    Type *SrcTy = I->getOperand(0)->getType();
    if (SrcTy->isIntOrPtrTy() &&
        // TODO: For now, not handling conversions like:
        // (bitcast i64 %x to <2 x i32>)
        !I->getType()->isVectorTy()) {
      computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
      break;
    }
 
    const Value *V;
    // Handle bitcast from floating point to integer.
    if (match(I, m_ElementWiseBitCast(m_Value(V))) &&
        V->getType()->isFPOrFPVectorTy()) {
      Type *FPType = V->getType()->getScalarType();
      KnownFPClass Result =
          computeKnownFPClass(V, DemandedElts, fcAllFlags, Depth + 1, Q);
      FPClassTest FPClasses = Result.KnownFPClasses;
 
      // TODO: Treat it as zero/poison if the use of I is unreachable.
      if (FPClasses == fcNone)
        break;
 
      if (Result.isKnownNever(fcNormal | fcSubnormal | fcNan)) {
        Known.Zero.setAllBits();
        Known.One.setAllBits();
 
        if (FPClasses & fcInf)
          Known = Known.intersectWith(KnownBits::makeConstant(
              APFloat::getInf(FPType->getFltSemantics()).bitcastToAPInt()));
 
        if (FPClasses & fcZero)
          Known = Known.intersectWith(KnownBits::makeConstant(
              APInt::getZero(FPType->getScalarSizeInBits())));
 
        Known.Zero.clearSignBit();
        Known.One.clearSignBit();
      }
 
      if (Result.SignBit) {
        if (*Result.SignBit)
          Known.makeNegative();
        else
          Known.makeNonNegative();
      }
 
      break;
    }
 
    // Handle cast from vector integer type to scalar or vector integer.
    auto *SrcVecTy = dyn_cast<FixedVectorType>(SrcTy);
    if (!SrcVecTy || !SrcVecTy->getElementType()->isIntegerTy() ||
        !I->getType()->isIntOrIntVectorTy() ||
        isa<ScalableVectorType>(I->getType()))
      break;
 
    // Look through a cast from narrow vector elements to wider type.
    // Examples: v4i32 -> v2i64, v3i8 -> v24
    unsigned SubBitWidth = SrcVecTy->getScalarSizeInBits();
    if (BitWidth % SubBitWidth == 0) {
      // Known bits are automatically intersected across demanded elements of a
      // vector. So for example, if a bit is computed as known zero, it must be
      // zero across all demanded elements of the vector.
      //
      // For this bitcast, each demanded element of the output is sub-divided
      // across a set of smaller vector elements in the source vector. To get
      // the known bits for an entire element of the output, compute the known
      // bits for each sub-element sequentially. This is done by shifting the
      // one-set-bit demanded elements parameter across the sub-elements for
      // consecutive calls to computeKnownBits. We are using the demanded
      // elements parameter as a mask operator.
      //
      // The known bits of each sub-element are then inserted into place
      // (dependent on endian) to form the full result of known bits.
      unsigned NumElts = DemandedElts.getBitWidth();
      unsigned SubScale = BitWidth / SubBitWidth;
      APInt SubDemandedElts = APInt::getZero(NumElts * SubScale);
      for (unsigned i = 0; i != NumElts; ++i) {
        if (DemandedElts[i])
          SubDemandedElts.setBit(i * SubScale);
      }
 
      KnownBits KnownSrc(SubBitWidth);
      for (unsigned i = 0; i != SubScale; ++i) {
        computeKnownBits(I->getOperand(0), SubDemandedElts.shl(i), KnownSrc,
                         Depth + 1, Q);
        unsigned ShiftElt = Q.DL.isLittleEndian() ? i : SubScale - 1 - i;
        Known.insertBits(KnownSrc, ShiftElt * SubBitWidth);
      }
    }
    break;
  }
  case Instruction::SExt: {
    // Compute the bits in the result that are not present in the input.
    unsigned SrcBitWidth = I->getOperand(0)->getType()->getScalarSizeInBits();
 
    Known = Known.trunc(SrcBitWidth);
    computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
    // If the sign bit of the input is known set or clear, then we know the
    // top bits of the result.
    Known = Known.sext(BitWidth);
    break;
  }
  case Instruction::Shl: {
    bool NUW = Q.IIQ.hasNoUnsignedWrap(cast<OverflowingBinaryOperator>(I));
    bool NSW = Q.IIQ.hasNoSignedWrap(cast<OverflowingBinaryOperator>(I));
    auto KF = [NUW, NSW](const KnownBits &KnownVal, const KnownBits &KnownAmt,
                         bool ShAmtNonZero) {
      return KnownBits::shl(KnownVal, KnownAmt, NUW, NSW, ShAmtNonZero);
    };
    computeKnownBitsFromShiftOperator(I, DemandedElts, Known, Known2, Depth, Q,
                                      KF);
    // Trailing zeros of a right-shifted constant never decrease.
    const APInt *C;
    if (match(I->getOperand(0), m_APInt(C)))
      Known.Zero.setLowBits(C->countr_zero());
    break;
  }
  case Instruction::LShr: {
    bool Exact = Q.IIQ.isExact(cast<BinaryOperator>(I));
    auto KF = [Exact](const KnownBits &KnownVal, const KnownBits &KnownAmt,
                      bool ShAmtNonZero) {
      return KnownBits::lshr(KnownVal, KnownAmt, ShAmtNonZero, Exact);
    };
    computeKnownBitsFromShiftOperator(I, DemandedElts, Known, Known2, Depth, Q,
                                      KF);
    // Leading zeros of a left-shifted constant never decrease.
    const APInt *C;
    if (match(I->getOperand(0), m_APInt(C)))
      Known.Zero.setHighBits(C->countl_zero());
    break;
  }
  case Instruction::AShr: {
    bool Exact = Q.IIQ.isExact(cast<BinaryOperator>(I));
    auto KF = [Exact](const KnownBits &KnownVal, const KnownBits &KnownAmt,
                      bool ShAmtNonZero) {
      return KnownBits::ashr(KnownVal, KnownAmt, ShAmtNonZero, Exact);
    };
    computeKnownBitsFromShiftOperator(I, DemandedElts, Known, Known2, Depth, Q,
                                      KF);
    break;
  }
  case Instruction::Sub: {
    bool NSW = Q.IIQ.hasNoSignedWrap(cast<OverflowingBinaryOperator>(I));
    bool NUW = Q.IIQ.hasNoUnsignedWrap(cast<OverflowingBinaryOperator>(I));
    computeKnownBitsAddSub(false, I->getOperand(0), I->getOperand(1), NSW, NUW,
                           DemandedElts, Known, Known2, Depth, Q);
    break;
  }
  case Instruction::Add: {
    bool NSW = Q.IIQ.hasNoSignedWrap(cast<OverflowingBinaryOperator>(I));
    bool NUW = Q.IIQ.hasNoUnsignedWrap(cast<OverflowingBinaryOperator>(I));
    computeKnownBitsAddSub(true, I->getOperand(0), I->getOperand(1), NSW, NUW,
                           DemandedElts, Known, Known2, Depth, Q);
    break;
  }
  case Instruction::SRem:
    computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
    computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
    Known = KnownBits::srem(Known, Known2);
    break;
 
  case Instruction::URem:
    computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
    computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
    Known = KnownBits::urem(Known, Known2);
    break;
  case Instruction::Alloca:
    Known.Zero.setLowBits(Log2(cast<AllocaInst>(I)->getAlign()));
    break;
  case Instruction::GetElementPtr: {
    // Analyze all of the subscripts of this getelementptr instruction
    // to determine if we can prove known low zero bits.
    computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
    // Accumulate the constant indices in a separate variable
    // to minimize the number of calls to computeForAddSub.
    APInt AccConstIndices(BitWidth, 0, /*IsSigned*/ true);
 
    gep_type_iterator GTI = gep_type_begin(I);
    for (unsigned i = 1, e = I->getNumOperands(); i != e; ++i, ++GTI) {
      // TrailZ can only become smaller, short-circuit if we hit zero.
      if (Known.isUnknown())
        break;
 
      Value *Index = I->getOperand(i);
 
      // Handle case when index is zero.
      Constant *CIndex = dyn_cast<Constant>(Index);
      if (CIndex && CIndex->isZeroValue())
        continue;
 
      if (StructType *STy = GTI.getStructTypeOrNull()) {
        // Handle struct member offset arithmetic.
 
        assert(CIndex &&
               "Access to structure field must be known at compile time");
 
        if (CIndex->getType()->isVectorTy())
          Index = CIndex->getSplatValue();
 
        unsigned Idx = cast<ConstantInt>(Index)->getZExtValue();
        const StructLayout *SL = Q.DL.getStructLayout(STy);
        uint64_t Offset = SL->getElementOffset(Idx);
        AccConstIndices += Offset;
        continue;
      }
 
      // Handle array index arithmetic.
      Type *IndexedTy = GTI.getIndexedType();
      if (!IndexedTy->isSized()) {
        Known.resetAll();
        break;
      }
 
      unsigned IndexBitWidth = Index->getType()->getScalarSizeInBits();
      KnownBits IndexBits(IndexBitWidth);
      computeKnownBits(Index, IndexBits, Depth + 1, Q);
      TypeSize IndexTypeSize = GTI.getSequentialElementStride(Q.DL);
      uint64_t TypeSizeInBytes = IndexTypeSize.getKnownMinValue();
      KnownBits ScalingFactor(IndexBitWidth);
      // Multiply by current sizeof type.
      // &A[i] == A + i * sizeof(*A[i]).
      if (IndexTypeSize.isScalable()) {
        // For scalable types the only thing we know about sizeof is
        // that this is a multiple of the minimum size.
        ScalingFactor.Zero.setLowBits(llvm::countr_zero(TypeSizeInBytes));
      } else if (IndexBits.isConstant()) {
        APInt IndexConst = IndexBits.getConstant();
        APInt ScalingFactor(IndexBitWidth, TypeSizeInBytes);
        IndexConst *= ScalingFactor;
        AccConstIndices += IndexConst.sextOrTrunc(BitWidth);
        continue;
      } else {
        ScalingFactor =
            KnownBits::makeConstant(APInt(IndexBitWidth, TypeSizeInBytes));
      }
      IndexBits = KnownBits::mul(IndexBits, ScalingFactor);
 
      // If the offsets have a different width from the pointer, according
      // to the language reference we need to sign-extend or truncate them
      // to the width of the pointer.
      IndexBits = IndexBits.sextOrTrunc(BitWidth);
 
      // Note that inbounds does *not* guarantee nsw for the addition, as only
      // the offset is signed, while the base address is unsigned.
      Known = KnownBits::add(Known, IndexBits);
    }
    if (!Known.isUnknown() && !AccConstIndices.isZero()) {
      KnownBits Index = KnownBits::makeConstant(AccConstIndices);
      Known = KnownBits::add(Known, Index);
    }
    break;
  }
  case Instruction::PHI: {
    const PHINode *P = cast<PHINode>(I);
    BinaryOperator *BO = nullptr;
    Value *R = nullptr, *L = nullptr;
    if (matchSimpleRecurrence(P, BO, R, L)) {
      // Handle the case of a simple two-predecessor recurrence PHI.
      // There's a lot more that could theoretically be done here, but
      // this is sufficient to catch some interesting cases.
      unsigned Opcode = BO->getOpcode();
 
      // If this is a shift recurrence, we know the bits being shifted in.
      // We can combine that with information about the start value of the
      // recurrence to conclude facts about the result.
      if ((Opcode == Instruction::LShr || Opcode == Instruction::AShr ||
           Opcode == Instruction::Shl) &&
          BO->getOperand(0) == I) {
 
        // We have matched a recurrence of the form:
        // %iv = [R, %entry], [%iv.next, %backedge]
        // %iv.next = shift_op %iv, L
 
        // Recurse with the phi context to avoid concern about whether facts
        // inferred hold at original context instruction.  TODO: It may be
        // correct to use the original context.  IF warranted, explore and
        // add sufficient tests to cover.
        SimplifyQuery RecQ = Q.getWithoutCondContext();
        RecQ.CxtI = P;
        computeKnownBits(R, DemandedElts, Known2, Depth + 1, RecQ);
        switch (Opcode) {
        case Instruction::Shl:
          // A shl recurrence will only increase the tailing zeros
          Known.Zero.setLowBits(Known2.countMinTrailingZeros());
          break;
        case Instruction::LShr:
          // A lshr recurrence will preserve the leading zeros of the
          // start value
          Known.Zero.setHighBits(Known2.countMinLeadingZeros());
          break;
        case Instruction::AShr:
          // An ashr recurrence will extend the initial sign bit
          Known.Zero.setHighBits(Known2.countMinLeadingZeros());
          Known.One.setHighBits(Known2.countMinLeadingOnes());
          break;
        };
      }
 
      // Check for operations that have the property that if
      // both their operands have low zero bits, the result
      // will have low zero bits.
      if (Opcode == Instruction::Add ||
          Opcode == Instruction::Sub ||
          Opcode == Instruction::And ||
          Opcode == Instruction::Or ||
          Opcode == Instruction::Mul) {
        // Change the context instruction to the "edge" that flows into the
        // phi. This is important because that is where the value is actually
        // "evaluated" even though it is used later somewhere else. (see also
        // D69571).
        SimplifyQuery RecQ = Q.getWithoutCondContext();
 
        unsigned OpNum = P->getOperand(0) == R ? 0 : 1;
        Instruction *RInst = P->getIncomingBlock(OpNum)->getTerminator();
        Instruction *LInst = P->getIncomingBlock(1 - OpNum)->getTerminator();
 
        // Ok, we have a PHI of the form L op= R. Check for low
        // zero bits.
        RecQ.CxtI = RInst;
        computeKnownBits(R, DemandedElts, Known2, Depth + 1, RecQ);
 
        // We need to take the minimum number of known bits
        KnownBits Known3(BitWidth);
        RecQ.CxtI = LInst;
        computeKnownBits(L, DemandedElts, Known3, Depth + 1, RecQ);
 
        Known.Zero.setLowBits(std::min(Known2.countMinTrailingZeros(),
                                       Known3.countMinTrailingZeros()));
 
        auto *OverflowOp = dyn_cast<OverflowingBinaryOperator>(BO);
        if (OverflowOp && Q.IIQ.hasNoSignedWrap(OverflowOp)) {
          // If initial value of recurrence is nonnegative, and we are adding
          // a nonnegative number with nsw, the result can only be nonnegative
          // or poison value regardless of the number of times we execute the
          // add in phi recurrence. If initial value is negative and we are
          // adding a negative number with nsw, the result can only be
          // negative or poison value. Similar arguments apply to sub and mul.
          //
          // (add non-negative, non-negative) --> non-negative
          // (add negative, negative) --> negative
          if (Opcode == Instruction::Add) {
            if (Known2.isNonNegative() && Known3.isNonNegative())
              Known.makeNonNegative();
            else if (Known2.isNegative() && Known3.isNegative())
              Known.makeNegative();
          }
 
          // (sub nsw non-negative, negative) --> non-negative
          // (sub nsw negative, non-negative) --> negative
          else if (Opcode == Instruction::Sub && BO->getOperand(0) == I) {
            if (Known2.isNonNegative() && Known3.isNegative())
              Known.makeNonNegative();
            else if (Known2.isNegative() && Known3.isNonNegative())
              Known.makeNegative();
          }
 
          // (mul nsw non-negative, non-negative) --> non-negative
          else if (Opcode == Instruction::Mul && Known2.isNonNegative() &&
                   Known3.isNonNegative())
            Known.makeNonNegative();
        }
 
        break;
      }
    }
 
    // Unreachable blocks may have zero-operand PHI nodes.
    if (P->getNumIncomingValues() == 0)
      break;
 
    // Otherwise take the unions of the known bit sets of the operands,
    // taking conservative care to avoid excessive recursion.
    if (Depth < MaxAnalysisRecursionDepth - 1 && Known.isUnknown()) {
      // Skip if every incoming value references to ourself.
      if (isa_and_nonnull<UndefValue>(P->hasConstantValue()))
        break;
 
      Known.Zero.setAllBits();
      Known.One.setAllBits();
      for (unsigned u = 0, e = P->getNumIncomingValues(); u < e; ++u) {
        Value *IncValue = P->getIncomingValue(u);
        // Skip direct self references.
        if (IncValue == P) continue;
 
        // Change the context instruction to the "edge" that flows into the
        // phi. This is important because that is where the value is actually
        // "evaluated" even though it is used later somewhere else. (see also
        // D69571).
        SimplifyQuery RecQ = Q.getWithoutCondContext();
        RecQ.CxtI = P->getIncomingBlock(u)->getTerminator();
 
        Known2 = KnownBits(BitWidth);
 
        // Recurse, but cap the recursion to one level, because we don't
        // want to waste time spinning around in loops.
        // TODO: See if we can base recursion limiter on number of incoming phi
        // edges so we don't overly clamp analysis.
        computeKnownBits(IncValue, DemandedElts, Known2,
                         MaxAnalysisRecursionDepth - 1, RecQ);
 
        // See if we can further use a conditional branch into the phi
        // to help us determine the range of the value.
        if (!Known2.isConstant()) {
          ICmpInst::Predicate Pred;
          const APInt *RHSC;
          BasicBlock *TrueSucc, *FalseSucc;
          // TODO: Use RHS Value and compute range from its known bits.
          if (match(RecQ.CxtI,
                    m_Br(m_c_ICmp(Pred, m_Specific(IncValue), m_APInt(RHSC)),
                         m_BasicBlock(TrueSucc), m_BasicBlock(FalseSucc)))) {
            // Check for cases of duplicate successors.
            if ((TrueSucc == P->getParent()) != (FalseSucc == P->getParent())) {
              // If we're using the false successor, invert the predicate.
              if (FalseSucc == P->getParent())
                Pred = CmpInst::getInversePredicate(Pred);
              // Get the knownbits implied by the incoming phi condition.
              auto CR = ConstantRange::makeExactICmpRegion(Pred, *RHSC);
              KnownBits KnownUnion = Known2.unionWith(CR.toKnownBits());
              // We can have conflicts here if we are analyzing deadcode (its
              // impossible for us reach this BB based the icmp).
              if (KnownUnion.hasConflict()) {
                // No reason to continue analyzing in a known dead region, so
                // just resetAll and break. This will cause us to also exit the
                // outer loop.
                Known.resetAll();
                break;
              }
              Known2 = KnownUnion;
            }
          }
        }
 
        Known = Known.intersectWith(Known2);
        // If all bits have been ruled out, there's no need to check
        // more operands.
        if (Known.isUnknown())
          break;
      }
    }
    break;
  }
  case Instruction::Call:
  case Instruction::Invoke: {
    // If range metadata is attached to this call, set known bits from that,
    // and then intersect with known bits based on other properties of the
    // function.
    if (MDNode *MD =
            Q.IIQ.getMetadata(cast<Instruction>(I), LLVMContext::MD_range))
      computeKnownBitsFromRangeMetadata(*MD, Known);
 
    const auto *CB = cast<CallBase>(I);
 
    if (std::optional<ConstantRange> Range = CB->getRange())
      Known = Known.unionWith(Range->toKnownBits());
 
    if (const Value *RV = CB->getReturnedArgOperand()) {
      if (RV->getType() == I->getType()) {
        computeKnownBits(RV, Known2, Depth + 1, Q);
        Known = Known.unionWith(Known2);
        // If the function doesn't return properly for all input values
        // (e.g. unreachable exits) then there might be conflicts between the
        // argument value and the range metadata. Simply discard the known bits
        // in case of conflicts.
        if (Known.hasConflict())
          Known.resetAll();
      }
    }
    if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
      switch (II->getIntrinsicID()) {
      default:
        break;
      case Intrinsic::abs: {
        computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
        bool IntMinIsPoison = match(II->getArgOperand(1), m_One());
        Known = Known2.abs(IntMinIsPoison);
        break;
      }
      case Intrinsic::bitreverse:
        computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
        Known.Zero |= Known2.Zero.reverseBits();
        Known.One |= Known2.One.reverseBits();
        break;
      case Intrinsic::bswap:
        computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
        Known.Zero |= Known2.Zero.byteSwap();
        Known.One |= Known2.One.byteSwap();
        break;
      case Intrinsic::ctlz: {
        computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
        // If we have a known 1, its position is our upper bound.
        unsigned PossibleLZ = Known2.countMaxLeadingZeros();
        // If this call is poison for 0 input, the result will be less than 2^n.
        if (II->getArgOperand(1) == ConstantInt::getTrue(II->getContext()))
          PossibleLZ = std::min(PossibleLZ, BitWidth - 1);
        unsigned LowBits = llvm::bit_width(PossibleLZ);
        Known.Zero.setBitsFrom(LowBits);
        break;
      }
      case Intrinsic::cttz: {
        computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
        // If we have a known 1, its position is our upper bound.
        unsigned PossibleTZ = Known2.countMaxTrailingZeros();
        // If this call is poison for 0 input, the result will be less than 2^n.
        if (II->getArgOperand(1) == ConstantInt::getTrue(II->getContext()))
          PossibleTZ = std::min(PossibleTZ, BitWidth - 1);
        unsigned LowBits = llvm::bit_width(PossibleTZ);
        Known.Zero.setBitsFrom(LowBits);
        break;
      }
      case Intrinsic::ctpop: {
        computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
        // We can bound the space the count needs.  Also, bits known to be zero
        // can't contribute to the population.
        unsigned BitsPossiblySet = Known2.countMaxPopulation();
        unsigned LowBits = llvm::bit_width(BitsPossiblySet);
        Known.Zero.setBitsFrom(LowBits);
        // TODO: we could bound KnownOne using the lower bound on the number
        // of bits which might be set provided by popcnt KnownOne2.
        break;
      }
      case Intrinsic::fshr:
      case Intrinsic::fshl: {
        const APInt *SA;
        if (!match(I->getOperand(2), m_APInt(SA)))
          break;
 
        // Normalize to funnel shift left.
        uint64_t ShiftAmt = SA->urem(BitWidth);
        if (II->getIntrinsicID() == Intrinsic::fshr)
          ShiftAmt = BitWidth - ShiftAmt;
 
        KnownBits Known3(BitWidth);
        computeKnownBits(I->getOperand(0), DemandedElts, Known2, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known3, Depth + 1, Q);
 
        Known.Zero =
            Known2.Zero.shl(ShiftAmt) | Known3.Zero.lshr(BitWidth - ShiftAmt);
        Known.One =
            Known2.One.shl(ShiftAmt) | Known3.One.lshr(BitWidth - ShiftAmt);
        break;
      }
      case Intrinsic::uadd_sat:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::uadd_sat(Known, Known2);
        break;
      case Intrinsic::usub_sat:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::usub_sat(Known, Known2);
        break;
      case Intrinsic::sadd_sat:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::sadd_sat(Known, Known2);
        break;
      case Intrinsic::ssub_sat:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::ssub_sat(Known, Known2);
        break;
        // Vec reverse preserves bits from input vec.
      case Intrinsic::vector_reverse:
        computeKnownBits(I->getOperand(0), DemandedElts.reverseBits(), Known,
                         Depth + 1, Q);
        break;
        // for min/max/and/or reduce, any bit common to each element in the
        // input vec is set in the output.
      case Intrinsic::vector_reduce_and:
      case Intrinsic::vector_reduce_or:
      case Intrinsic::vector_reduce_umax:
      case Intrinsic::vector_reduce_umin:
      case Intrinsic::vector_reduce_smax:
      case Intrinsic::vector_reduce_smin:
        computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
        break;
      case Intrinsic::vector_reduce_xor: {
        computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
        // The zeros common to all vecs are zero in the output.
        // If the number of elements is odd, then the common ones remain. If the
        // number of elements is even, then the common ones becomes zeros.
        auto *VecTy = cast<VectorType>(I->getOperand(0)->getType());
        // Even, so the ones become zeros.
        bool EvenCnt = VecTy->getElementCount().isKnownEven();
        if (EvenCnt)
          Known.Zero |= Known.One;
        // Maybe even element count so need to clear ones.
        if (VecTy->isScalableTy() || EvenCnt)
          Known.One.clearAllBits();
        break;
      }
      case Intrinsic::umin:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::umin(Known, Known2);
        break;
      case Intrinsic::umax:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::umax(Known, Known2);
        break;
      case Intrinsic::smin:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::smin(Known, Known2);
        break;
      case Intrinsic::smax:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::smax(Known, Known2);
        break;
      case Intrinsic::ptrmask: {
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
 
        const Value *Mask = I->getOperand(1);
        Known2 = KnownBits(Mask->getType()->getScalarSizeInBits());
        computeKnownBits(Mask, DemandedElts, Known2, Depth + 1, Q);
        // TODO: 1-extend would be more precise.
        Known &= Known2.anyextOrTrunc(BitWidth);
        break;
      }
      case Intrinsic::x86_sse2_pmulh_w:
      case Intrinsic::x86_avx2_pmulh_w:
      case Intrinsic::x86_avx512_pmulh_w_512:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::mulhs(Known, Known2);
        break;
      case Intrinsic::x86_sse2_pmulhu_w:
      case Intrinsic::x86_avx2_pmulhu_w:
      case Intrinsic::x86_avx512_pmulhu_w_512:
        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
        Known = KnownBits::mulhu(Known, Known2);
        break;
      case Intrinsic::x86_sse42_crc32_64_64:
        Known.Zero.setBitsFrom(32);
        break;
      case Intrinsic::x86_ssse3_phadd_d_128:
      case Intrinsic::x86_ssse3_phadd_w_128:
      case Intrinsic::x86_avx2_phadd_d:
      case Intrinsic::x86_avx2_phadd_w: {
        Known = computeKnownBitsForHorizontalOperation(
            I, DemandedElts, Depth, Q,
            [](const KnownBits &KnownLHS, const KnownBits &KnownRHS) {
              return KnownBits::add(KnownLHS, KnownRHS);
            });
        break;
      }
      case Intrinsic::x86_ssse3_phadd_sw_128:
      case Intrinsic::x86_avx2_phadd_sw: {
        Known = computeKnownBitsForHorizontalOperation(I, DemandedElts, Depth,
                                                       Q, KnownBits::sadd_sat);
        break;
      }
      case Intrinsic::x86_ssse3_phsub_d_128:
      case Intrinsic::x86_ssse3_phsub_w_128:
      case Intrinsic::x86_avx2_phsub_d:
      case Intrinsic::x86_avx2_phsub_w: {
        Known = computeKnownBitsForHorizontalOperation(
            I, DemandedElts, Depth, Q,
            [](const KnownBits &KnownLHS, const KnownBits &KnownRHS) {
              return KnownBits::sub(KnownLHS, KnownRHS);
            });
        break;
      }
      case Intrinsic::x86_ssse3_phsub_sw_128:
      case Intrinsic::x86_avx2_phsub_sw: {
        Known = computeKnownBitsForHorizontalOperation(I, DemandedElts, Depth,
                                                       Q, KnownBits::ssub_sat);
        break;
      }
      case Intrinsic::riscv_vsetvli:
      case Intrinsic::riscv_vsetvlimax: {
        bool HasAVL = II->getIntrinsicID() == Intrinsic::riscv_vsetvli;
        const ConstantRange Range = getVScaleRange(II->getFunction(), BitWidth);
        uint64_t SEW = RISCVVType::decodeVSEW(
            cast<ConstantInt>(II->getArgOperand(HasAVL))->getZExtValue());
        RISCVII::VLMUL VLMUL = static_cast<RISCVII::VLMUL>(
            cast<ConstantInt>(II->getArgOperand(1 + HasAVL))->getZExtValue());
        uint64_t MaxVLEN =
            Range.getUnsignedMax().getZExtValue() * RISCV::RVVBitsPerBlock;
        uint64_t MaxVL = MaxVLEN / RISCVVType::getSEWLMULRatio(SEW, VLMUL);
 
        // Result of vsetvli must be not larger than AVL.
        if (HasAVL)
          if (auto *CI = dyn_cast<ConstantInt>(II->getArgOperand(0)))
            MaxVL = std::min(MaxVL, CI->getZExtValue());
 
        unsigned KnownZeroFirstBit = Log2_32(MaxVL) + 1;
        if (BitWidth > KnownZeroFirstBit)
          Known.Zero.setBitsFrom(KnownZeroFirstBit);
        break;
      }
      case Intrinsic::vscale: {
        if (!II->getParent() || !II->getFunction())
          break;
 
        Known = getVScaleRange(II->getFunction(), BitWidth).toKnownBits();
        break;
      }
      }
    }
    break;
  }
  case Instruction::ShuffleVector: {
    auto *Shuf = dyn_cast<ShuffleVectorInst>(I);
    // FIXME: Do we need to handle ConstantExpr involving shufflevectors?
    if (!Shuf) {
      Known.resetAll();
      return;
    }
    // For undef elements, we don't know anything about the common state of
    // the shuffle result.
    APInt DemandedLHS, DemandedRHS;
    if (!getShuffleDemandedElts(Shuf, DemandedElts, DemandedLHS, DemandedRHS)) {
      Known.resetAll();
      return;
    }
    Known.One.setAllBits();
    Known.Zero.setAllBits();
    if (!!DemandedLHS) {
      const Value *LHS = Shuf->getOperand(0);
      computeKnownBits(LHS, DemandedLHS, Known, Depth + 1, Q);
      // If we don't know any bits, early out.
      if (Known.isUnknown())
        break;
    }
    if (!!DemandedRHS) {
      const Value *RHS = Shuf->getOperand(1);
      computeKnownBits(RHS, DemandedRHS, Known2, Depth + 1, Q);
      Known = Known.intersectWith(Known2);
    }
    break;
  }
  case Instruction::InsertElement: {
    if (isa<ScalableVectorType>(I->getType())) {
      Known.resetAll();
      return;
    }
    const Value *Vec = I->getOperand(0);
    const Value *Elt = I->getOperand(1);
    auto *CIdx = dyn_cast<ConstantInt>(I->getOperand(2));
    unsigned NumElts = DemandedElts.getBitWidth();
    APInt DemandedVecElts = DemandedElts;
    bool NeedsElt = true;
    // If we know the index we are inserting too, clear it from Vec check.
    if (CIdx && CIdx->getValue().ult(NumElts)) {
      DemandedVecElts.clearBit(CIdx->getZExtValue());
      NeedsElt = DemandedElts[CIdx->getZExtValue()];
    }
 
    Known.One.setAllBits();
    Known.Zero.setAllBits();
    if (NeedsElt) {
      computeKnownBits(Elt, Known, Depth + 1, Q);
      // If we don't know any bits, early out.
      if (Known.isUnknown())
        break;
    }
 
    if (!DemandedVecElts.isZero()) {
      computeKnownBits(Vec, DemandedVecElts, Known2, Depth + 1, Q);
      Known = Known.intersectWith(Known2);
    }
    break;
  }
  case Instruction::ExtractElement: {
    // Look through extract element. If the index is non-constant or
    // out-of-range demand all elements, otherwise just the extracted element.
    const Value *Vec = I->getOperand(0);
    const Value *Idx = I->getOperand(1);
    auto *CIdx = dyn_cast<ConstantInt>(Idx);
    if (isa<ScalableVectorType>(Vec->getType())) {
      // FIXME: there's probably *something* we can do with scalable vectors
      Known.resetAll();
      break;
    }
    unsigned NumElts = cast<FixedVectorType>(Vec->getType())->getNumElements();
    APInt DemandedVecElts = APInt::getAllOnes(NumElts);
    if (CIdx && CIdx->getValue().ult(NumElts))
      DemandedVecElts = APInt::getOneBitSet(NumElts, CIdx->getZExtValue());
    computeKnownBits(Vec, DemandedVecElts, Known, Depth + 1, Q);
    break;
  }
  case Instruction::ExtractValue:
    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I->getOperand(0))) {
      const ExtractValueInst *EVI = cast<ExtractValueInst>(I);
      if (EVI->getNumIndices() != 1) break;
      if (EVI->getIndices()[0] == 0) {
        switch (II->getIntrinsicID()) {
        default: break;
        case Intrinsic::uadd_with_overflow:
        case Intrinsic::sadd_with_overflow:
          computeKnownBitsAddSub(
              true, II->getArgOperand(0), II->getArgOperand(1), /*NSW=*/false,
              /* NUW=*/false, DemandedElts, Known, Known2, Depth, Q);
          break;
        case Intrinsic::usub_with_overflow:
        case Intrinsic::ssub_with_overflow:
          computeKnownBitsAddSub(
              false, II->getArgOperand(0), II->getArgOperand(1), /*NSW=*/false,
              /* NUW=*/false, DemandedElts, Known, Known2, Depth, Q);
          break;
        case Intrinsic::umul_with_overflow:
        case Intrinsic::smul_with_overflow:
          computeKnownBitsMul(II->getArgOperand(0), II->getArgOperand(1), false,
                              DemandedElts, Known, Known2, Depth, Q);
          break;
        }
      }
    }
    break;
  case Instruction::Freeze:
    if (isGuaranteedNotToBePoison(I->getOperand(0), Q.AC, Q.CxtI, Q.DT,
                                  Depth + 1))
      computeKnownBits(I->getOperand(0), Known, Depth + 1, Q);
    break;
  }
}
 
/// Determine which bits of V are known to be either zero or one and return
/// them.
KnownBits llvm::computeKnownBits(const Value *V, const APInt &DemandedElts,
                                 unsigned Depth, const SimplifyQuery &Q) {
  KnownBits Known(getBitWidth(V->getType(), Q.DL));
  ::computeKnownBits(V, DemandedElts, Known, Depth, Q);
  return Known;
}
 
/// Determine which bits of V are known to be either zero or one and return
/// them.
KnownBits llvm::computeKnownBits(const Value *V, unsigned Depth,
                                 const SimplifyQuery &Q) {
  KnownBits Known(getBitWidth(V->getType(), Q.DL));
  computeKnownBits(V, Known, Depth, Q);
  return Known;
}
 
/// Determine which bits of V are known to be either zero or one and return
/// them in the Known bit set.
///
/// NOTE: we cannot consider 'undef' to be "IsZero" here.  The problem is that
/// we cannot optimize based on the assumption that it is zero without changing
/// it to be an explicit zero.  If we don't change it to zero, other code could
/// optimized based on the contradictory assumption that it is non-zero.
/// Because instcombine aggressively folds operations with undef args anyway,
/// this won't lose us code quality.
///
/// This function is defined on values with integer type, values with pointer
/// type, and vectors of integers.  In the case
/// where V is a vector, known zero, and known one values are the
/// same width as the vector element, and the bit is set only if it is true
/// for all of the demanded elements in the vector specified by DemandedElts.
void computeKnownBits(const Value *V, const APInt &DemandedElts,
                      KnownBits &Known, unsigned Depth,
                      const SimplifyQuery &Q) {
  if (!DemandedElts) {
    // No demanded elts, better to assume we don't know anything.
    Known.resetAll();
    return;
  }
 
  assert(V && "No Value?");
  assert(Depth <= MaxAnalysisRecursionDepth && "Limit Search Depth");
 
#ifndef NDEBUG
  Type *Ty = V->getType();
  unsigned BitWidth = Known.getBitWidth();
 
  assert((Ty->isIntOrIntVectorTy(BitWidth) || Ty->isPtrOrPtrVectorTy()) &&
         "Not integer or pointer type!");
 
  if (auto *FVTy = dyn_cast<FixedVectorType>(Ty)) {
    assert(
        FVTy->getNumElements() == DemandedElts.getBitWidth() &&
        "DemandedElt width should equal the fixed vector number of elements");
  } else {
    assert(DemandedElts == APInt(1, 1) &&
           "DemandedElt width should be 1 for scalars or scalable vectors");
  }
 
  Type *ScalarTy = Ty->getScalarType();
  if (ScalarTy->isPointerTy()) {
    assert(BitWidth == Q.DL.getPointerTypeSizeInBits(ScalarTy) &&
           "V and Known should have same BitWidth");
  } else {
    assert(BitWidth == Q.DL.getTypeSizeInBits(ScalarTy) &&
           "V and Known should have same BitWidth");
  }
#endif
 
  const APInt *C;
  if (match(V, m_APInt(C))) {
    // We know all of the bits for a scalar constant or a splat vector constant!
    Known = KnownBits::makeConstant(*C);
    return;
  }
  // Null and aggregate-zero are all-zeros.
  if (isa<ConstantPointerNull>(V) || isa<ConstantAggregateZero>(V)) {
    Known.setAllZero();
    return;
  }
  // Handle a constant vector by taking the intersection of the known bits of
  // each element.
  if (const ConstantDataVector *CDV = dyn_cast<ConstantDataVector>(V)) {
    assert(!isa<ScalableVectorType>(V->getType()));
    // We know that CDV must be a vector of integers. Take the intersection of
    // each element.
    Known.Zero.setAllBits(); Known.One.setAllBits();
    for (unsigned i = 0, e = CDV->getNumElements(); i != e; ++i) {
      if (!DemandedElts[i])
        continue;
      APInt Elt = CDV->getElementAsAPInt(i);
      Known.Zero &= ~Elt;
      Known.One &= Elt;
    }
    if (Known.hasConflict())
      Known.resetAll();
    return;
  }
 
  if (const auto *CV = dyn_cast<ConstantVector>(V)) {
    assert(!isa<ScalableVectorType>(V->getType()));
    // We know that CV must be a vector of integers. Take the intersection of
    // each element.
    Known.Zero.setAllBits(); Known.One.setAllBits();
    for (unsigned i = 0, e = CV->getNumOperands(); i != e; ++i) {
      if (!DemandedElts[i])
        continue;
      Constant *Element = CV->getAggregateElement(i);
      if (isa<PoisonValue>(Element))
        continue;
      auto *ElementCI = dyn_cast_or_null<ConstantInt>(Element);
      if (!ElementCI) {
        Known.resetAll();
        return;
      }
      const APInt &Elt = ElementCI->getValue();
      Known.Zero &= ~Elt;
      Known.One &= Elt;
    }
    if (Known.hasConflict())
      Known.resetAll();
    return;
  }
 
  // Start out not knowing anything.
  Known.resetAll();
 
  // We can't imply anything about undefs.
  if (isa<UndefValue>(V))
    return;
 
  // There's no point in looking through other users of ConstantData for
  // assumptions.  Confirm that we've handled them all.
  assert(!isa<ConstantData>(V) && "Unhandled constant data!");
 
  if (const auto *A = dyn_cast<Argument>(V))
    if (std::optional<ConstantRange> Range = A->getRange())
      Known = Range->toKnownBits();
 
  // All recursive calls that increase depth must come after this.
  if (Depth == MaxAnalysisRecursionDepth)
    return;
 
  // A weak GlobalAlias is totally unknown. A non-weak GlobalAlias has
  // the bits of its aliasee.
  if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
    if (!GA->isInterposable())
      computeKnownBits(GA->getAliasee(), Known, Depth + 1, Q);
    return;
  }
 
  if (const Operator *I = dyn_cast<Operator>(V))
    computeKnownBitsFromOperator(I, DemandedElts, Known, Depth, Q);
  else if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
    if (std::optional<ConstantRange> CR = GV->getAbsoluteSymbolRange())
      Known = CR->toKnownBits();
  }
 
  // Aligned pointers have trailing zeros - refine Known.Zero set
  if (isa<PointerType>(V->getType())) {
    Align Alignment = V->getPointerAlignment(Q.DL);
    Known.Zero.setLowBits(Log2(Alignment));
  }
 
  // computeKnownBitsFromContext strictly refines Known.
  // Therefore, we run them after computeKnownBitsFromOperator.
 
  // Check whether we can determine known bits from context such as assumes.
  computeKnownBitsFromContext(V, Known, Depth, Q);
}
 
/// Try to detect a recurrence that the value of the induction variable is
/// always a power of two (or zero).
static bool isPowerOfTwoRecurrence(const PHINode *PN, bool OrZero,
                                   unsigned Depth, SimplifyQuery &Q) {
  BinaryOperator *BO = nullptr;
  Value *Start = nullptr, *Step = nullptr;
  if (!matchSimpleRecurrence(PN, BO, Start, Step))
    return false;
 
  // Initial value must be a power of two.
  for (const Use &U : PN->operands()) {
    if (U.get() == Start) {
      // Initial value comes from a different BB, need to adjust context
      // instruction for analysis.
      Q.CxtI = PN->getIncomingBlock(U)->getTerminator();
      if (!isKnownToBeAPowerOfTwo(Start, OrZero, Depth, Q))
        return false;
    }
  }
 
  // Except for Mul, the induction variable must be on the left side of the
  // increment expression, otherwise its value can be arbitrary.
  if (BO->getOpcode() != Instruction::Mul && BO->getOperand(1) != Step)
    return false;
 
  Q.CxtI = BO->getParent()->getTerminator();
  switch (BO->getOpcode()) {
  case Instruction::Mul:
    // Power of two is closed under multiplication.
    return (OrZero || Q.IIQ.hasNoUnsignedWrap(BO) ||
            Q.IIQ.hasNoSignedWrap(BO)) &&
           isKnownToBeAPowerOfTwo(Step, OrZero, Depth, Q);
  case Instruction::SDiv:
    // Start value must not be signmask for signed division, so simply being a
    // power of two is not sufficient, and it has to be a constant.
    if (!match(Start, m_Power2()) || match(Start, m_SignMask()))
      return false;
    [[fallthrough]];
  case Instruction::UDiv:
    // Divisor must be a power of two.
    // If OrZero is false, cannot guarantee induction variable is non-zero after
    // division, same for Shr, unless it is exact division.
    return (OrZero || Q.IIQ.isExact(BO)) &&
           isKnownToBeAPowerOfTwo(Step, false, Depth, Q);
  case Instruction::Shl:
    return OrZero || Q.IIQ.hasNoUnsignedWrap(BO) || Q.IIQ.hasNoSignedWrap(BO);
  case Instruction::AShr:
    if (!match(Start, m_Power2()) || match(Start, m_SignMask()))
      return false;
    [[fallthrough]];
  case Instruction::LShr:
    return OrZero || Q.IIQ.isExact(BO);
  default:
    return false;
  }
}
 
/// Return true if the given value is known to have exactly one
/// bit set when defined. For vectors return true if every element is known to
/// be a power of two when defined. Supports values with integer or pointer
/// types and vectors of integers.
bool isKnownToBeAPowerOfTwo(const Value *V, bool OrZero, unsigned Depth,
                            const SimplifyQuery &Q) {
  assert(Depth <= MaxAnalysisRecursionDepth && "Limit Search Depth");
 
  if (isa<Constant>(V))
    return OrZero ? match(V, m_Power2OrZero()) : match(V, m_Power2());
 
  // i1 is by definition a power of 2 or zero.
  if (OrZero && V->getType()->getScalarSizeInBits() == 1)
    return true;
 
  auto *I = dyn_cast<Instruction>(V);
  if (!I)
    return false;
 
  if (Q.CxtI && match(V, m_VScale())) {
    const Function *F = Q.CxtI->getFunction();
    // The vscale_range indicates vscale is a power-of-two.
    return F->hasFnAttribute(Attribute::VScaleRange);
  }
 
  // 1 << X is clearly a power of two if the one is not shifted off the end.  If
  // it is shifted off the end then the result is undefined.
  if (match(I, m_Shl(m_One(), m_Value())))
    return true;
 
  // (signmask) >>l X is clearly a power of two if the one is not shifted off
  // the bottom.  If it is shifted off the bottom then the result is undefined.
  if (match(I, m_LShr(m_SignMask(), m_Value())))
    return true;
 
  // The remaining tests are all recursive, so bail out if we hit the limit.
  if (Depth++ == MaxAnalysisRecursionDepth)
    return false;
 
  switch (I->getOpcode()) {
  case Instruction::ZExt:
    return isKnownToBeAPowerOfTwo(I->getOperand(0), OrZero, Depth, Q);
  case Instruction::Trunc:
    return OrZero && isKnownToBeAPowerOfTwo(I->getOperand(0), OrZero, Depth, Q);
  case Instruction::Shl:
    if (OrZero || Q.IIQ.hasNoUnsignedWrap(I) || Q.IIQ.hasNoSignedWrap(I))
      return isKnownToBeAPowerOfTwo(I->getOperand(0), OrZero, Depth, Q);
    return false;
  case Instruction::LShr:
    if (OrZero || Q.IIQ.isExact(cast<BinaryOperator>(I)))
      return isKnownToBeAPowerOfTwo(I->getOperand(0), OrZero, Depth, Q);
    return false;
  case Instruction::UDiv:
    if (Q.IIQ.isExact(cast<BinaryOperator>(I)))
      return isKnownToBeAPowerOfTwo(I->getOperand(0), OrZero, Depth, Q);
    return false;
  case Instruction::Mul:
    return isKnownToBeAPowerOfTwo(I->getOperand(1), OrZero, Depth, Q) &&
           isKnownToBeAPowerOfTwo(I->getOperand(0), OrZero, Depth, Q) &&
           (OrZero || isKnownNonZero(I, Q, Depth));
  case Instruction::And:
    // A power of two and'd with anything is a power of two or zero.
    if (OrZero &&
        (isKnownToBeAPowerOfTwo(I->getOperand(1), /*OrZero*/ true, Depth, Q) ||
         isKnownToBeAPowerOfTwo(I->getOperand(0), /*OrZero*/ true, Depth, Q)))
      return true;
    // X & (-X) is always a power of two or zero.
    if (match(I->getOperand(0), m_Neg(m_Specific(I->getOperand(1)))) ||
        match(I->getOperand(1), m_Neg(m_Specific(I->getOperand(0)))))
      return OrZero || isKnownNonZero(I->getOperand(0), Q, Depth);
    return false;
  case Instruction::Add: {
    // Adding a power-of-two or zero to the same power-of-two or zero yields
    // either the original power-of-two, a larger power-of-two or zero.
    const OverflowingBinaryOperator *VOBO = cast<OverflowingBinaryOperator>(V);
    if (OrZero || Q.IIQ.hasNoUnsignedWrap(VOBO) ||
        Q.IIQ.hasNoSignedWrap(VOBO)) {
      if (match(I->getOperand(0),
                m_c_And(m_Specific(I->getOperand(1)), m_Value())) &&
          isKnownToBeAPowerOfTwo(I->getOperand(1), OrZero, Depth, Q))
        return true;
      if (match(I->getOperand(1),
                m_c_And(m_Specific(I->getOperand(0)), m_Value())) &&
          isKnownToBeAPowerOfTwo(I->getOperand(0), OrZero, Depth, Q))
        return true;
 
      unsigned BitWidth = V->getType()->getScalarSizeInBits();
      KnownBits LHSBits(BitWidth);
      computeKnownBits(I->getOperand(0), LHSBits, Depth, Q);
 
      KnownBits RHSBits(BitWidth);
      computeKnownBits(I->getOperand(1), RHSBits, Depth, Q);
      // If i8 V is a power of two or zero:
      //  ZeroBits: 1 1 1 0 1 1 1 1
      // ~ZeroBits: 0 0 0 1 0 0 0 0
      if ((~(LHSBits.Zero & RHSBits.Zero)).isPowerOf2())
        // If OrZero isn't set, we cannot give back a zero result.
        // Make sure either the LHS or RHS has a bit set.
        if (OrZero || RHSBits.One.getBoolValue() || LHSBits.One.getBoolValue())
          return true;
    }
 
    // LShr(UINT_MAX, Y) + 1 is a power of two (if add is nuw) or zero.
    if (OrZero || Q.IIQ.hasNoUnsignedWrap(VOBO))
      if (match(I, m_Add(m_LShr(m_AllOnes(), m_Value()), m_One())))
        return true;
    return false;
  }
  case Instruction::Select:
    return isKnownToBeAPowerOfTwo(I->getOperand(1), OrZero, Depth, Q) &&
           isKnownToBeAPowerOfTwo(I->getOperand(2), OrZero, Depth, Q);
  case Instruction::PHI: {
    // A PHI node is power of two if all incoming values are power of two, or if
    // it is an induction variable where in each step its value is a power of
    // two.
    auto *PN = cast<PHINode>(I);
    SimplifyQuery RecQ = Q.getWithoutCondContext();
 
    // Check if it is an induction variable and always power of two.
    if (isPowerOfTwoRecurrence(PN, OrZero, Depth, RecQ))
      return true;
 
    // Recursively check all incoming values. Limit recursion to 2 levels, so
    // that search complexity is limited to number of operands^2.
    unsigned NewDepth = std::max(Depth, MaxAnalysisRecursionDepth - 1);
    return llvm::all_of(PN->operands(), [&](const Use &U) {
      // Value is power of 2 if it is coming from PHI node itself by induction.
      if (U.get() == PN)
        return true;
 
      // Change the context instruction to the incoming block where it is
      // evaluated.
      RecQ.CxtI = PN->getIncomingBlock(U)->getTerminator();
      return isKnownToBeAPowerOfTwo(U.get(), OrZero, NewDepth, RecQ);
    });
  }
  case Instruction::Invoke:
  case Instruction::Call: {
    if (auto *II = dyn_cast<IntrinsicInst>(I)) {
      switch (II->getIntrinsicID()) {
      case Intrinsic::umax:
      case Intrinsic::smax:
      case Intrinsic::umin:
      case Intrinsic::smin:
        return isKnownToBeAPowerOfTwo(II->getArgOperand(1), OrZero, Depth, Q) &&
               isKnownToBeAPowerOfTwo(II->getArgOperand(0), OrZero, Depth, Q);
      // bswap/bitreverse just move around bits, but don't change any 1s/0s
      // thus dont change pow2/non-pow2 status.
      case Intrinsic::bitreverse:
      case Intrinsic::bswap:
        return isKnownToBeAPowerOfTwo(II->getArgOperand(0), OrZero, Depth, Q);
      case Intrinsic::fshr:
      case Intrinsic::fshl:
        // If Op0 == Op1, this is a rotate. is_pow2(rotate(x, y)) == is_pow2(x)
        if (II->getArgOperand(0) == II->getArgOperand(1))
          return isKnownToBeAPowerOfTwo(II->getArgOperand(0), OrZero, Depth, Q);
        break;
      default:
        break;
      }
    }
    return false;
  }
  default:
    return false;
  }
}
 
/// Test whether a GEP's result is known to be non-null.
///
/// Uses properties inherent in a GEP to try to determine whether it is known
/// to be non-null.
///
/// Currently this routine does not support vector GEPs.
static bool isGEPKnownNonNull(const GEPOperator *GEP, unsigned Depth,
                              const SimplifyQuery &Q) {
  const Function *F = nullptr;
  if (const Instruction *I = dyn_cast<Instruction>(GEP))
    F = I->getFunction();
 
  // If the gep is nuw or inbounds with invalid null pointer, then the GEP
  // may be null iff the base pointer is null and the offset is zero.
  if (!GEP->hasNoUnsignedWrap() &&
      !(GEP->isInBounds() &&
        !NullPointerIsDefined(F, GEP->getPointerAddressSpace())))
    return false;
 
  // FIXME: Support vector-GEPs.
  assert(GEP->getType()->isPointerTy() && "We only support plain pointer GEP");
 
  // If the base pointer is non-null, we cannot walk to a null address with an
  // inbounds GEP in address space zero.
  if (isKnownNonZero(GEP->getPointerOperand(), Q, Depth))
    return true;
 
  // Walk the GEP operands and see if any operand introduces a non-zero offset.
  // If so, then the GEP cannot produce a null pointer, as doing so would
  // inherently violate the inbounds contract within address space zero.
  for (gep_type_iterator GTI = gep_type_begin(GEP), GTE = gep_type_end(GEP);
       GTI != GTE; ++GTI) {
    // Struct types are easy -- they must always be indexed by a constant.
    if (StructType *STy = GTI.getStructTypeOrNull()) {
      ConstantInt *OpC = cast<ConstantInt>(GTI.getOperand());
      unsigned ElementIdx = OpC->getZExtValue();
      const StructLayout *SL = Q.DL.getStructLayout(STy);
      uint64_t ElementOffset = SL->getElementOffset(ElementIdx);
      if (ElementOffset > 0)
        return true;
      continue;
    }
 
    // If we have a zero-sized type, the index doesn't matter. Keep looping.
    if (GTI.getSequentialElementStride(Q.DL).isZero())
      continue;
 
    // Fast path the constant operand case both for efficiency and so we don't
    // increment Depth when just zipping down an all-constant GEP.
    if (ConstantInt *OpC = dyn_cast<ConstantInt>(GTI.getOperand())) {
      if (!OpC->isZero())
        return true;
      continue;
    }
 
    // We post-increment Depth here because while isKnownNonZero increments it
    // as well, when we pop back up that increment won't persist. We don't want
    // to recurse 10k times just because we have 10k GEP operands. We don't
    // bail completely out because we want to handle constant GEPs regardless
    // of depth.
    if (Depth++ >= MaxAnalysisRecursionDepth)
      continue;
 
    if (isKnownNonZero(GTI.getOperand(), Q, Depth))
      return true;
  }
 
  return false;
}
 
static bool isKnownNonNullFromDominatingCondition(const Value *V,
                                                  const Instruction *CtxI,
                                                  const DominatorTree *DT) {
  assert(!isa<Constant>(V) && "Called for constant?");
 
  if (!CtxI || !DT)
    return false;
 
  unsigned NumUsesExplored = 0;
  for (const auto *U : V->users()) {
    // Avoid massive lists
    if (NumUsesExplored >= DomConditionsMaxUses)
      break;
    NumUsesExplored++;
 
    // If the value is used as an argument to a call or invoke, then argument
    // attributes may provide an answer about null-ness.
    if (const auto *CB = dyn_cast<CallBase>(U))
      if (auto *CalledFunc = CB->getCalledFunction())
        for (const Argument &Arg : CalledFunc->args())
          if (CB->getArgOperand(Arg.getArgNo()) == V &&
              Arg.hasNonNullAttr(/* AllowUndefOrPoison */ false) &&
              DT->dominates(CB, CtxI))
            return true;
 
    // If the value is used as a load/store, then the pointer must be non null.
    if (V == getLoadStorePointerOperand(U)) {
      const Instruction *I = cast<Instruction>(U);
      if (!NullPointerIsDefined(I->getFunction(),
                                V->getType()->getPointerAddressSpace()) &&
          DT->dominates(I, CtxI))
        return true;
    }
 
    if ((match(U, m_IDiv(m_Value(), m_Specific(V))) ||
         match(U, m_IRem(m_Value(), m_Specific(V)))) &&
        isValidAssumeForContext(cast<Instruction>(U), CtxI, DT))
      return true;
 
    // Consider only compare instructions uniquely controlling a branch
    Value *RHS;
    CmpInst::Predicate Pred;
    if (!match(U, m_c_ICmp(Pred, m_Specific(V), m_Value(RHS))))
      continue;
 
    bool NonNullIfTrue;
    if (cmpExcludesZero(Pred, RHS))
      NonNullIfTrue = true;
    else if (cmpExcludesZero(CmpInst::getInversePredicate(Pred), RHS))
      NonNullIfTrue = false;
    else
      continue;
 
    SmallVector<const User *, 4> WorkList;
    SmallPtrSet<const User *, 4> Visited;
    for (const auto *CmpU : U->users()) {
      assert(WorkList.empty() && "Should be!");
      if (Visited.insert(CmpU).second)
        WorkList.push_back(CmpU);
 
      while (!WorkList.empty()) {
        auto *Curr = WorkList.pop_back_val();
 
        // If a user is an AND, add all its users to the work list. We only
        // propagate "pred != null" condition through AND because it is only
        // correct to assume that all conditions of AND are met in true branch.
        // TODO: Support similar logic of OR and EQ predicate?
        if (NonNullIfTrue)
          if (match(Curr, m_LogicalAnd(m_Value(), m_Value()))) {
            for (const auto *CurrU : Curr->users())
              if (Visited.insert(CurrU).second)
                WorkList.push_back(CurrU);
            continue;
          }
 
        if (const BranchInst *BI = dyn_cast<BranchInst>(Curr)) {
          assert(BI->isConditional() && "uses a comparison!");
 
          BasicBlock *NonNullSuccessor =
              BI->getSuccessor(NonNullIfTrue ? 0 : 1);
          BasicBlockEdge Edge(BI->getParent(), NonNullSuccessor);
          if (Edge.isSingleEdge() && DT->dominates(Edge, CtxI->getParent()))
            return true;
        } else if (NonNullIfTrue && isGuard(Curr) &&
                   DT->dominates(cast<Instruction>(Curr), CtxI)) {
          return true;
        }
      }
    }
  }
 
  return false;
}
 
/// Does the 'Range' metadata (which must be a valid MD_range operand list)
/// ensure that the value it's attached to is never Value?  'RangeType' is
/// is the type of the value described by the range.
static bool rangeMetadataExcludesValue(const MDNode* Ranges, const APInt& Value) {
  const unsigned NumRanges = Ranges->getNumOperands() / 2;
  assert(NumRanges >= 1);
  for (unsigned i = 0; i < NumRanges; ++i) {
    ConstantInt *Lower =
        mdconst::extract<ConstantInt>(Ranges->getOperand(2 * i + 0));
    ConstantInt *Upper =
        mdconst::extract<ConstantInt>(Ranges->getOperand(2 * i + 1));
    ConstantRange Range(Lower->getValue(), Upper->getValue());
    if (Range.contains(Value))
      return false;
  }
  return true;
}
 
/// Try to detect a recurrence that monotonically increases/decreases from a
/// non-zero starting value. These are common as induction variables.
static bool isNonZeroRecurrence(const PHINode *PN) {
  BinaryOperator *BO = nullptr;
  Value *Start = nullptr, *Step = nullptr;
  const APInt *StartC, *StepC;
  if (!matchSimpleRecurrence(PN, BO, Start, Step) ||
      !match(Start, m_APInt(StartC)) || StartC->isZero())
    return false;
 
  switch (BO->getOpcode()) {
  case Instruction::Add:
    // Starting from non-zero and stepping away from zero can never wrap back
    // to zero.
    return BO->hasNoUnsignedWrap() ||
           (BO->hasNoSignedWrap() && match(Step, m_APInt(StepC)) &&
            StartC->isNegative() == StepC->isNegative());
  case Instruction::Mul:
    return (BO->hasNoUnsignedWrap() || BO->hasNoSignedWrap()) &&
           match(Step, m_APInt(StepC)) && !StepC->isZero();
  case Instruction::Shl:
    return BO->hasNoUnsignedWrap() || BO->hasNoSignedWrap();
  case Instruction::AShr:
  case Instruction::LShr:
    return BO->isExact();
  default:
    return false;
  }
}
 
static bool matchOpWithOpEqZero(Value *Op0, Value *Op1) {
  return match(Op0, m_ZExtOrSExt(m_SpecificICmp(ICmpInst::ICMP_EQ,
                                                m_Specific(Op1), m_Zero()))) ||
         match(Op1, m_ZExtOrSExt(m_SpecificICmp(ICmpInst::ICMP_EQ,
                                                m_Specific(Op0), m_Zero())));
}
 
static bool isNonZeroAdd(const APInt &DemandedElts, unsigned Depth,
                         const SimplifyQuery &Q, unsigned BitWidth, Value *X,
                         Value *Y, bool NSW, bool NUW) {
  // (X + (X != 0)) is non zero
  if (matchOpWithOpEqZero(X, Y))
    return true;
 
  if (NUW)
    return isKnownNonZero(Y, DemandedElts, Q, Depth) ||
           isKnownNonZero(X, DemandedElts, Q, Depth);
 
  KnownBits XKnown = computeKnownBits(X, DemandedElts, Depth, Q);
  KnownBits YKnown = computeKnownBits(Y, DemandedElts, Depth, Q);
 
  // If X and Y are both non-negative (as signed values) then their sum is not
  // zero unless both X and Y are zero.
  if (XKnown.isNonNegative() && YKnown.isNonNegative())
    if (isKnownNonZero(Y, DemandedElts, Q, Depth) ||
        isKnownNonZero(X, DemandedElts, Q, Depth))
      return true;
 
  // If X and Y are both negative (as signed values) then their sum is not
  // zero unless both X and Y equal INT_MIN.
  if (XKnown.isNegative() && YKnown.isNegative()) {
    APInt Mask = APInt::getSignedMaxValue(BitWidth);
    // The sign bit of X is set.  If some other bit is set then X is not equal
    // to INT_MIN.
    if (XKnown.One.intersects(Mask))
      return true;
    // The sign bit of Y is set.  If some other bit is set then Y is not equal
    // to INT_MIN.
    if (YKnown.One.intersects(Mask))
      return true;
  }
 
  // The sum of a non-negative number and a power of two is not zero.
  if (XKnown.isNonNegative() &&
      isKnownToBeAPowerOfTwo(Y, /*OrZero*/ false, Depth, Q))
    return true;
  if (YKnown.isNonNegative() &&
      isKnownToBeAPowerOfTwo(X, /*OrZero*/ false, Depth, Q))
    return true;
 
  return KnownBits::add(XKnown, YKnown, NSW, NUW).isNonZero();
}
 
static bool isNonZeroSub(const APInt &DemandedElts, unsigned Depth,
                         const SimplifyQuery &Q, unsigned BitWidth, Value *X,
                         Value *Y) {
  // (X - (X != 0)) is non zero
  // ((X != 0) - X) is non zero
  if (matchOpWithOpEqZero(X, Y))
    return true;
 
  // TODO: Move this case into isKnownNonEqual().
  if (auto *C = dyn_cast<Constant>(X))
    if (C->isNullValue() && isKnownNonZero(Y, DemandedElts, Q, Depth))
      return true;
 
  return ::isKnownNonEqual(X, Y, DemandedElts, Depth, Q);
}
 
static bool isNonZeroMul(const APInt &DemandedElts, unsigned Depth,
                         const SimplifyQuery &Q, unsigned BitWidth, Value *X,
                         Value *Y, bool NSW, bool NUW) {
  // If X and Y are non-zero then so is X * Y as long as the multiplication
  // does not overflow.
  if (NSW || NUW)
    return isKnownNonZero(X, DemandedElts, Q, Depth) &&
           isKnownNonZero(Y, DemandedElts, Q, Depth);
 
  // If either X or Y is odd, then if the other is non-zero the result can't
  // be zero.
  KnownBits XKnown = computeKnownBits(X, DemandedElts, Depth, Q);
  if (XKnown.One[0])
    return isKnownNonZero(Y, DemandedElts, Q, Depth);
 
  KnownBits YKnown = computeKnownBits(Y, DemandedElts, Depth, Q);
  if (YKnown.One[0])
    return XKnown.isNonZero() || isKnownNonZero(X, DemandedElts, Q, Depth);
 
  // If there exists any subset of X (sX) and subset of Y (sY) s.t sX * sY is
  // non-zero, then X * Y is non-zero. We can find sX and sY by just taking
  // the lowest known One of X and Y. If they are non-zero, the result
  // must be non-zero. We can check if LSB(X) * LSB(Y) != 0 by doing
  // X.CountLeadingZeros + Y.CountLeadingZeros < BitWidth.
  return (XKnown.countMaxTrailingZeros() + YKnown.countMaxTrailingZeros()) <
         BitWidth;
}
 
static bool isNonZeroShift(const Operator *I, const APInt &DemandedElts,
                           unsigned Depth, const SimplifyQuery &Q,
                           const KnownBits &KnownVal) {
  auto ShiftOp = [&](const APInt &Lhs, const APInt &Rhs) {
    switch (I->getOpcode()) {
    case Instruction::Shl:
      return Lhs.shl(Rhs);
    case Instruction::LShr:
      return Lhs.lshr(Rhs);
    case Instruction::AShr:
      return Lhs.ashr(Rhs);
    default:
      llvm_unreachable("Unknown Shift Opcode");
    }
  };
 
  auto InvShiftOp = [&](const APInt &Lhs, const APInt &Rhs) {
    switch (I->getOpcode()) {
    case Instruction::Shl:
      return Lhs.lshr(Rhs);
    case Instruction::LShr:
    case Instruction::AShr:
      return Lhs.shl(Rhs);
    default:
      llvm_unreachable("Unknown Shift Opcode");
    }
  };
 
  if (KnownVal.isUnknown())
    return false;
 
  KnownBits KnownCnt =
      computeKnownBits(I->getOperand(1), DemandedElts, Depth, Q);
  APInt MaxShift = KnownCnt.getMaxValue();
  unsigned NumBits = KnownVal.getBitWidth();
  if (MaxShift.uge(NumBits))
    return false;
 
  if (!ShiftOp(KnownVal.One, MaxShift).isZero())
    return true;
 
  // If all of the bits shifted out are known to be zero, and Val is known
  // non-zero then at least one non-zero bit must remain.
  if (InvShiftOp(KnownVal.Zero, NumBits - MaxShift)
          .eq(InvShiftOp(APInt::getAllOnes(NumBits), NumBits - MaxShift)) &&
      isKnownNonZero(I->getOperand(0), DemandedElts, Q, Depth))
    return true;
 
  return false;
}
 
static bool isKnownNonZeroFromOperator(const Operator *I,
                                       const APInt &DemandedElts,
                                       unsigned Depth, const SimplifyQuery &Q) {
  unsigned BitWidth = getBitWidth(I->getType()->getScalarType(), Q.DL);
  switch (I->getOpcode()) {
  case Instruction::Alloca:
    // Alloca never returns null, malloc might.
    return I->getType()->getPointerAddressSpace() == 0;
  case Instruction::GetElementPtr:
    if (I->getType()->isPointerTy())
      return isGEPKnownNonNull(cast<GEPOperator>(I), Depth, Q);
    break;
  case Instruction::BitCast: {
    // We need to be a bit careful here. We can only peek through the bitcast
    // if the scalar size of elements in the operand are smaller than and a
    // multiple of the size they are casting too. Take three cases:
    //
    // 1) Unsafe:
    //        bitcast <2 x i16> %NonZero to <4 x i8>
    //
    //    %NonZero can have 2 non-zero i16 elements, but isKnownNonZero on a
    //    <4 x i8> requires that all 4 i8 elements be non-zero which isn't
    //    guranteed (imagine just sign bit set in the 2 i16 elements).
    //
    // 2) Unsafe:
    //        bitcast <4 x i3> %NonZero to <3 x i4>
    //
    //    Even though the scalar size of the src (`i3`) is smaller than the
    //    scalar size of the dst `i4`, because `i3` is not a multiple of `i4`
    //    its possible for the `3 x i4` elements to be zero because there are
    //    some elements in the destination that don't contain any full src
    //    element.
    //
    // 3) Safe:
    //        bitcast <4 x i8> %NonZero to <2 x i16>
    //
    //    This is always safe as non-zero in the 4 i8 elements implies
    //    non-zero in the combination of any two adjacent ones. Since i8 is a
    //    multiple of i16, each i16 is guranteed to have 2 full i8 elements.
    //    This all implies the 2 i16 elements are non-zero.
    Type *FromTy = I->getOperand(0)->getType();
    if ((FromTy->isIntOrIntVectorTy() || FromTy->isPtrOrPtrVectorTy()) &&
        (BitWidth % getBitWidth(FromTy->getScalarType(), Q.DL)) == 0)
      return isKnownNonZero(I->getOperand(0), Q, Depth);
  } break;
  case Instruction::IntToPtr:
    // Note that we have to take special care to avoid looking through
    // truncating casts, e.g., int2ptr/ptr2int with appropriate sizes, as well
    // as casts that can alter the value, e.g., AddrSpaceCasts.
    if (!isa<ScalableVectorType>(I->getType()) &&
        Q.DL.getTypeSizeInBits(I->getOperand(0)->getType()).getFixedValue() <=
            Q.DL.getTypeSizeInBits(I->getType()).getFixedValue())
      return isKnownNonZero(I->getOperand(0), DemandedElts, Q, Depth);
    break;
  case Instruction::PtrToInt:
    // Similar to int2ptr above, we can look through ptr2int here if the cast
    // is a no-op or an extend and not a truncate.
    if (!isa<ScalableVectorType>(I->getType()) &&
        Q.DL.getTypeSizeInBits(I->getOperand(0)->getType()).getFixedValue() <=
            Q.DL.getTypeSizeInBits(I->getType()).getFixedValue())
      return isKnownNonZero(I->getOperand(0), DemandedElts, Q, Depth);
    break;
  case Instruction::Trunc:
    // nuw/nsw trunc preserves zero/non-zero status of input.
    if (auto *TI = dyn_cast<TruncInst>(I))
      if (TI->hasNoSignedWrap() || TI->hasNoUnsignedWrap())
        return isKnownNonZero(TI->getOperand(0), DemandedElts, Q, Depth);
    break;
 
  case Instruction::Sub:
    return isNonZeroSub(DemandedElts, Depth, Q, BitWidth, I->getOperand(0),
                        I->getOperand(1));
  case Instruction::Xor:
    // (X ^ (X != 0)) is non zero
    if (matchOpWithOpEqZero(I->getOperand(0), I->getOperand(1)))
      return true;
    break;
  case Instruction::Or:
    // (X | (X != 0)) is non zero
    if (matchOpWithOpEqZero(I->getOperand(0), I->getOperand(1)))
      return true;
    // X | Y != 0 if X != 0 or Y != 0.
    return isKnownNonZero(I->getOperand(1), DemandedElts, Q, Depth) ||
           isKnownNonZero(I->getOperand(0), DemandedElts, Q, Depth);
  case Instruction::SExt:
  case Instruction::ZExt:
    // ext X != 0 if X != 0.
    return isKnownNonZero(I->getOperand(0), DemandedElts, Q, Depth);
 
  case Instruction::Shl: {
    // shl nsw/nuw can't remove any non-zero bits.
    const OverflowingBinaryOperator *BO = cast<OverflowingBinaryOperator>(I);
    if (Q.IIQ.hasNoUnsignedWrap(BO) || Q.IIQ.hasNoSignedWrap(BO))
      return isKnownNonZero(I->getOperand(0), DemandedElts, Q, Depth);
 
    // shl X, Y != 0 if X is odd.  Note that the value of the shift is undefined
    // if the lowest bit is shifted off the end.
    KnownBits Known(BitWidth);
    computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth, Q);
    if (Known.One[0])
      return true;
 
    return isNonZeroShift(I, DemandedElts, Depth, Q, Known);
  }
  case Instruction::LShr:
  case Instruction::AShr: {
    // shr exact can only shift out zero bits.
    const PossiblyExactOperator *BO = cast<PossiblyExactOperator>(I);
    if (BO->isExact())
      return isKnownNonZero(I->getOperand(0), DemandedElts, Q, Depth);
 
    // shr X, Y != 0 if X is negative.  Note that the value of the shift is not
    // defined if the sign bit is shifted off the end.
    KnownBits Known =
        computeKnownBits(I->getOperand(0), DemandedElts, Depth, Q);
    if (Known.isNegative())
      return true;
 
    return isNonZeroShift(I, DemandedElts, Depth, Q, Known);
  }
  case Instruction::UDiv:
  case Instruction::SDiv: {
    // X / Y
    // div exact can only produce a zero if the dividend is zero.
    if (cast<PossiblyExactOperator>(I)->isExact())
      return isKnownNonZero(I->getOperand(0), DemandedElts, Q, Depth);
 
    KnownBits XKnown =
        computeKnownBits(I->getOperand(0), DemandedElts, Depth, Q);
    // If X is fully unknown we won't be able to figure anything out so don't
    // both computing knownbits for Y.
    if (XKnown.isUnknown())
      return false;
 
    KnownBits YKnown =
        computeKnownBits(I->getOperand(1), DemandedElts, Depth, Q);
    if (I->getOpcode() == Instruction::SDiv) {
      // For signed division need to compare abs value of the operands.
      XKnown = XKnown.abs(/*IntMinIsPoison*/ false);
      YKnown = YKnown.abs(/*IntMinIsPoison*/ false);
    }
    // If X u>= Y then div is non zero (0/0 is UB).
    std::optional<bool> XUgeY = KnownBits::uge(XKnown, YKnown);
    // If X is total unknown or X u< Y we won't be able to prove non-zero
    // with compute known bits so just return early.
    return XUgeY && *XUgeY;
  }
  case Instruction::Add: {
    // X + Y.
 
    // If Add has nuw wrap flag, then if either X or Y is non-zero the result is
    // non-zero.
    auto *BO = cast<OverflowingBinaryOperator>(I);
    return isNonZeroAdd(DemandedElts, Depth, Q, BitWidth, I->getOperand(0),
                        I->getOperand(1), Q.IIQ.hasNoSignedWrap(BO),
                        Q.IIQ.hasNoUnsignedWrap(BO));
  }
  case Instruction::Mul: {
    const OverflowingBinaryOperator *BO = cast<OverflowingBinaryOperator>(I);
    return isNonZeroMul(DemandedElts, Depth, Q, BitWidth, I->getOperand(0),
                        I->getOperand(1), Q.IIQ.hasNoSignedWrap(BO),
                        Q.IIQ.hasNoUnsignedWrap(BO));
  }
  case Instruction::Select: {
    // (C ? X : Y) != 0 if X != 0 and Y != 0.
 
    // First check if the arm is non-zero using `isKnownNonZero`. If that fails,
    // then see if the select condition implies the arm is non-zero. For example
    // (X != 0 ? X : Y), we know the true arm is non-zero as the `X` "return" is
    // dominated by `X != 0`.
    auto SelectArmIsNonZero = [&](bool IsTrueArm) {
      Value *Op;
      Op = IsTrueArm ? I->getOperand(1) : I->getOperand(2);
      // Op is trivially non-zero.
      if (isKnownNonZero(Op, DemandedElts, Q, Depth))
        return true;
 
      // The condition of the select dominates the true/false arm. Check if the
      // condition implies that a given arm is non-zero.
      Value *X;
      CmpInst::Predicate Pred;
      if (!match(I->getOperand(0), m_c_ICmp(Pred, m_Specific(Op), m_Value(X))))
        return false;
 
      if (!IsTrueArm)
        Pred = ICmpInst::getInversePredicate(Pred);
 
      return cmpExcludesZero(Pred, X);
    };
 
    if (SelectArmIsNonZero(/* IsTrueArm */ true) &&
        SelectArmIsNonZero(/* IsTrueArm */ false))
      return true;
    break;
  }
  case Instruction::PHI: {
    auto *PN = cast<PHINode>(I);
    if (Q.IIQ.UseInstrInfo && isNonZeroRecurrence(PN))
      return true;
 
    // Check if all incoming values are non-zero using recursion.
    SimplifyQuery RecQ = Q.getWithoutCondContext();
    unsigned NewDepth = std::max(Depth, MaxAnalysisRecursionDepth - 1);
    return llvm::all_of(PN->operands(), [&](const Use &U) {
      if (U.get() == PN)
        return true;
      RecQ.CxtI = PN->getIncomingBlock(U)->getTerminator();
      // Check if the branch on the phi excludes zero.
      ICmpInst::Predicate Pred;
      Value *X;
      BasicBlock *TrueSucc, *FalseSucc;
      if (match(RecQ.CxtI,
                m_Br(m_c_ICmp(Pred, m_Specific(U.get()), m_Value(X)),
                     m_BasicBlock(TrueSucc), m_BasicBlock(FalseSucc)))) {
        // Check for cases of duplicate successors.
        if ((TrueSucc == PN->getParent()) != (FalseSucc == PN->getParent())) {
          // If we're using the false successor, invert the predicate.
          if (FalseSucc == PN->getParent())
            Pred = CmpInst::getInversePredicate(Pred);
          if (cmpExcludesZero(Pred, X))
            return true;
        }
      }
      // Finally recurse on the edge and check it directly.
      return isKnownNonZero(U.get(), DemandedElts, RecQ, NewDepth);
    });
  }
  case Instruction::InsertElement: {
    if (isa<ScalableVectorType>(I->getType()))
      break;
 
    const Value *Vec = I->getOperand(0);
    const Value *Elt = I->getOperand(1);
    auto *CIdx = dyn_cast<ConstantInt>(I->getOperand(2));
 
    unsigned NumElts = DemandedElts.getBitWidth();
    APInt DemandedVecElts = DemandedElts;
    bool SkipElt = false;
    // If we know the index we are inserting too, clear it from Vec check.
    if (CIdx && CIdx->getValue().ult(NumElts)) {
      DemandedVecElts.clearBit(CIdx->getZExtValue());
      SkipElt = !DemandedElts[CIdx->getZExtValue()];
    }
 
    // Result is zero if Elt is non-zero and rest of the demanded elts in Vec
    // are non-zero.
    return (SkipElt || isKnownNonZero(Elt, Q, Depth)) &&
           (DemandedVecElts.isZero() ||
            isKnownNonZero(Vec, DemandedVecElts, Q, Depth));
  }
  case Instruction::ExtractElement:
    if (const auto *EEI = dyn_cast<ExtractElementInst>(I)) {
      const Value *Vec = EEI->getVectorOperand();
      const Value *Idx = EEI->getIndexOperand();
      auto *CIdx = dyn_cast<ConstantInt>(Idx);
      if (auto *VecTy = dyn_cast<FixedVectorType>(Vec->getType())) {
        unsigned NumElts = VecTy->getNumElements();
        APInt DemandedVecElts = APInt::getAllOnes(NumElts);
        if (CIdx && CIdx->getValue().ult(NumElts))
          DemandedVecElts = APInt::getOneBitSet(NumElts, CIdx->getZExtValue());
        return isKnownNonZero(Vec, DemandedVecElts, Q, Depth);
      }
    }
    break;
  case Instruction::ShuffleVector: {
    auto *Shuf = dyn_cast<ShuffleVectorInst>(I);
    if (!Shuf)
      break;
    APInt DemandedLHS, DemandedRHS;
    // For undef elements, we don't know anything about the common state of
    // the shuffle result.
    if (!getShuffleDemandedElts(Shuf, DemandedElts, DemandedLHS, DemandedRHS))
      break;
    // If demanded elements for both vecs are non-zero, the shuffle is non-zero.
    return (DemandedRHS.isZero() ||
            isKnownNonZero(Shuf->getOperand(1), DemandedRHS, Q, Depth)) &&
           (DemandedLHS.isZero() ||
            isKnownNonZero(Shuf->getOperand(0), DemandedLHS, Q, Depth));
  }
  case Instruction::Freeze:
    return isKnownNonZero(I->getOperand(0), Q, Depth) &&
           isGuaranteedNotToBePoison(I->getOperand(0), Q.AC, Q.CxtI, Q.DT,
                                     Depth);
  case Instruction::Load: {
    auto *LI = cast<LoadInst>(I);
    // A Load tagged with nonnull or dereferenceable with null pointer undefined
    // is never null.
    if (auto *PtrT = dyn_cast<PointerType>(I->getType())) {
      if (Q.IIQ.getMetadata(LI, LLVMContext::MD_nonnull) ||
          (Q.IIQ.getMetadata(LI, LLVMContext::MD_dereferenceable) &&
           !NullPointerIsDefined(LI->getFunction(), PtrT->getAddressSpace())))
        return true;
    } else if (MDNode *Ranges = Q.IIQ.getMetadata(LI, LLVMContext::MD_range)) {
      return rangeMetadataExcludesValue(Ranges, APInt::getZero(BitWidth));
    }
 
    // No need to fall through to computeKnownBits as range metadata is already
    // handled in isKnownNonZero.
    return false;
  }
  case Instruction::ExtractValue: {
    const WithOverflowInst *WO;
    if (match(I, m_ExtractValue<0>(m_WithOverflowInst(WO)))) {
      switch (WO->getBinaryOp()) {
      default:
        break;
      case Instruction::Add:
        return isNonZeroAdd(DemandedElts, Depth, Q, BitWidth,
                            WO->getArgOperand(0), WO->getArgOperand(1),
                            /*NSW=*/false,
                            /*NUW=*/false);
      case Instruction::Sub:
        return isNonZeroSub(DemandedElts, Depth, Q, BitWidth,
                            WO->getArgOperand(0), WO->getArgOperand(1));
      case Instruction::Mul:
        return isNonZeroMul(DemandedElts, Depth, Q, BitWidth,
                            WO->getArgOperand(0), WO->getArgOperand(1),
                            /*NSW=*/false, /*NUW=*/false);
        break;
      }
    }
    break;
  }
  case Instruction::Call:
  case Instruction::Invoke: {
    const auto *Call = cast<CallBase>(I);
    if (I->getType()->isPointerTy()) {
      if (Call->isReturnNonNull())
        return true;
      if (const auto *RP = getArgumentAliasingToReturnedPointer(Call, true))
        return isKnownNonZero(RP, Q, Depth);
    } else {
      if (MDNode *Ranges = Q.IIQ.getMetadata(Call, LLVMContext::MD_range))
        return rangeMetadataExcludesValue(Ranges, APInt::getZero(BitWidth));
      if (std::optional<ConstantRange> Range = Call->getRange()) {
        const APInt ZeroValue(Range->getBitWidth(), 0);
        if (!Range->contains(ZeroValue))
          return true;
      }
      if (const Value *RV = Call->getReturnedArgOperand())
        if (RV->getType() == I->getType() && isKnownNonZero(RV, Q, Depth))
          return true;
    }
 
    if (auto *II = dyn_cast<IntrinsicInst>(I)) {
      switch (II->getIntrinsicID()) {
      case Intrinsic::sshl_sat:
      case Intrinsic::ushl_sat:
      case Intrinsic::abs:
      case Intrinsic::bitreverse:
      case Intrinsic::bswap:
      case Intrinsic::ctpop:
        return isKnownNonZero(II->getArgOperand(0), DemandedElts, Q, Depth);
        // NB: We don't do usub_sat here as in any case we can prove its
        // non-zero, we will fold it to `sub nuw` in InstCombine.
      case Intrinsic::ssub_sat:
        return isNonZeroSub(DemandedElts, Depth, Q, BitWidth,
                            II->getArgOperand(0), II->getArgOperand(1));
      case Intrinsic::sadd_sat:
        return isNonZeroAdd(DemandedElts, Depth, Q, BitWidth,
                            II->getArgOperand(0), II->getArgOperand(1),
                            /*NSW=*/true, /* NUW=*/false);
        // Vec reverse preserves zero/non-zero status from input vec.
      case Intrinsic::vector_reverse:
        return isKnownNonZero(II->getArgOperand(0), DemandedElts.reverseBits(),
                              Q, Depth);
        // umin/smin/smax/smin/or of all non-zero elements is always non-zero.
      case Intrinsic::vector_reduce_or:
      case Intrinsic::vector_reduce_umax:
      case Intrinsic::vector_reduce_umin:
      case Intrinsic::vector_reduce_smax:
      case Intrinsic::vector_reduce_smin:
        return isKnownNonZero(II->getArgOperand(0), Q, Depth);
      case Intrinsic::umax:
      case Intrinsic::uadd_sat:
        // umax(X, (X != 0)) is non zero
        // X +usat (X != 0) is non zero
        if (matchOpWithOpEqZero(II->getArgOperand(0), II->getArgOperand(1)))
          return true;
 
        return isKnownNonZero(II->getArgOperand(1), DemandedElts, Q, Depth) ||
               isKnownNonZero(II->getArgOperand(0), DemandedElts, Q, Depth);
      case Intrinsic::smax: {
        // If either arg is strictly positive the result is non-zero. Otherwise
        // the result is non-zero if both ops are non-zero.
        auto IsNonZero = [&](Value *Op, std::optional<bool> &OpNonZero,
                             const KnownBits &OpKnown) {
          if (!OpNonZero.has_value())
            OpNonZero = OpKnown.isNonZero() ||
                        isKnownNonZero(Op, DemandedElts, Q, Depth);
          return *OpNonZero;
        };
        // Avoid re-computing isKnownNonZero.
        std::optional<bool> Op0NonZero, Op1NonZero;
        KnownBits Op1Known =
            computeKnownBits(II->getArgOperand(1), DemandedElts, Depth, Q);
        if (Op1Known.isNonNegative() &&
            IsNonZero(II->getArgOperand(1), Op1NonZero, Op1Known))
          return true;
        KnownBits Op0Known =
            computeKnownBits(II->getArgOperand(0), DemandedElts, Depth, Q);
        if (Op0Known.isNonNegative() &&
            IsNonZero(II->getArgOperand(0), Op0NonZero, Op0Known))
          return true;
        return IsNonZero(II->getArgOperand(1), Op1NonZero, Op1Known) &&
               IsNonZero(II->getArgOperand(0), Op0NonZero, Op0Known);
      }
      case Intrinsic::smin: {
        // If either arg is negative the result is non-zero. Otherwise
        // the result is non-zero if both ops are non-zero.
        KnownBits Op1Known =
            computeKnownBits(II->getArgOperand(1), DemandedElts, Depth, Q);
        if (Op1Known.isNegative())
          return true;
        KnownBits Op0Known =
            computeKnownBits(II->getArgOperand(0), DemandedElts, Depth, Q);
        if (Op0Known.isNegative())
          return true;
 
        if (Op1Known.isNonZero() && Op0Known.isNonZero())
          return true;
      }
        [[fallthrough]];
      case Intrinsic::umin:
        return isKnownNonZero(II->getArgOperand(0), DemandedElts, Q, Depth) &&
               isKnownNonZero(II->getArgOperand(1), DemandedElts, Q, Depth);
      case Intrinsic::cttz:
        return computeKnownBits(II->getArgOperand(0), DemandedElts, Depth, Q)
            .Zero[0];
      case Intrinsic::ctlz:
        return computeKnownBits(II->getArgOperand(0), DemandedElts, Depth, Q)
            .isNonNegative();
      case Intrinsic::fshr:
      case Intrinsic::fshl:
        // If Op0 == Op1, this is a rotate. rotate(x, y) != 0 iff x != 0.
        if (II->getArgOperand(0) == II->getArgOperand(1))
          return isKnownNonZero(II->getArgOperand(0), DemandedElts, Q, Depth);
        break;
      case Intrinsic::vscale:
        return true;
      case Intrinsic::experimental_get_vector_length:
        return isKnownNonZero(I->getOperand(0), Q, Depth);
      default:
        break;
      }
      break;
    }
 
    return false;
  }
  }
 
  KnownBits Known(BitWidth);
  computeKnownBits(I, DemandedElts, Known, Depth, Q);
  return Known.One != 0;
}
 
/// Return true if the given value is known to be non-zero when defined. For
/// vectors, return true if every demanded element is known to be non-zero when
/// defined. For pointers, if the context instruction and dominator tree are
/// specified, perform context-sensitive analysis and return true if the
/// pointer couldn't possibly be null at the specified instruction.
/// Supports values with integer or pointer type and vectors of integers.
bool isKnownNonZero(const Value *V, const APInt &DemandedElts,
                    const SimplifyQuery &Q, unsigned Depth) {
  Type *Ty = V->getType();
 
#ifndef NDEBUG
  assert(Depth <= MaxAnalysisRecursionDepth && "Limit Search Depth");
 
  if (auto *FVTy = dyn_cast<FixedVectorType>(Ty)) {
    assert(
        FVTy->getNumElements() == DemandedElts.getBitWidth() &&
        "DemandedElt width should equal the fixed vector number of elements");
  } else {
    assert(DemandedElts == APInt(1, 1) &&
           "DemandedElt width should be 1 for scalars");
  }
#endif
 
  if (auto *C = dyn_cast<Constant>(V)) {
    if (C->isNullValue())
      return false;
    if (isa<ConstantInt>(C))
      // Must be non-zero due to null test above.
      return true;
 
    // For constant vectors, check that all elements are poison or known
    // non-zero to determine that the whole vector is known non-zero.
    if (auto *VecTy = dyn_cast<FixedVectorType>(Ty)) {
      for (unsigned i = 0, e = VecTy->getNumElements(); i != e; ++i) {
        if (!DemandedElts[i])
          continue;
        Constant *Elt = C->getAggregateElement(i);
        if (!Elt || Elt->isNullValue())
          return false;
        if (!isa<PoisonValue>(Elt) && !isa<ConstantInt>(Elt))
          return false;
      }
      return true;
    }
 
    // Constant ptrauth can be null, iff the base pointer can be.
    if (auto *CPA = dyn_cast<ConstantPtrAuth>(V))
      return isKnownNonZero(CPA->getPointer(), DemandedElts, Q, Depth);
 
    // A global variable in address space 0 is non null unless extern weak
    // or an absolute symbol reference. Other address spaces may have null as a
    // valid address for a global, so we can't assume anything.
    if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
      if (!GV->isAbsoluteSymbolRef() && !GV->hasExternalWeakLinkage() &&
          GV->getType()->getAddressSpace() == 0)
        return true;
    }
 
    // For constant expressions, fall through to the Operator code below.
    if (!isa<ConstantExpr>(V))
      return false;
  }
 
  if (const auto *A = dyn_cast<Argument>(V))
    if (std::optional<ConstantRange> Range = A->getRange()) {
      const APInt ZeroValue(Range->getBitWidth(), 0);
      if (!Range->contains(ZeroValue))
        return true;
    }
 
  if (!isa<Constant>(V) && isKnownNonZeroFromAssume(V, Q))
    return true;
 
  // Some of the tests below are recursive, so bail out if we hit the limit.
  if (Depth++ >= MaxAnalysisRecursionDepth)
    return false;
 
  // Check for pointer simplifications.
 
  if (PointerType *PtrTy = dyn_cast<PointerType>(Ty)) {
    // A byval, inalloca may not be null in a non-default addres space. A
    // nonnull argument is assumed never 0.
    if (const Argument *A = dyn_cast<Argument>(V)) {
      if (((A->hasPassPointeeByValueCopyAttr() &&
            !NullPointerIsDefined(A->getParent(), PtrTy->getAddressSpace())) ||
           A->hasNonNullAttr()))
        return true;
    }
  }
 
  if (const auto *I = dyn_cast<Operator>(V))
    if (isKnownNonZeroFromOperator(I, DemandedElts, Depth, Q))
      return true;
 
  if (!isa<Constant>(V) &&
      isKnownNonNullFromDominatingCondition(V, Q.CxtI, Q.DT))
    return true;
 
  return false;
}
 
bool llvm::isKnownNonZero(const Value *V, const SimplifyQuery &Q,
                          unsigned Depth) {
  auto *FVTy = dyn_cast<FixedVectorType>(V->getType());
  APInt DemandedElts =
      FVTy ? APInt::getAllOnes(FVTy->getNumElements()) : APInt(1, 1);
  return ::isKnownNonZero(V, DemandedElts, Q, Depth);
}
 
/// If the pair of operators are the same invertible function, return the
/// the operands of the function corresponding to each input. Otherwise,
/// return std::nullopt.  An invertible function is one that is 1-to-1 and maps
/// every input value to exactly one output value.  This is equivalent to
/// saying that Op1 and Op2 are equal exactly when the specified pair of
/// operands are equal, (except that Op1 and Op2 may be poison more often.)
static std::optional<std::pair<Value*, Value*>>
getInvertibleOperands(const Operator *Op1,
                      const Operator *Op2) {
  if (Op1->getOpcode() != Op2->getOpcode())
    return std::nullopt;
 
  auto getOperands = [&](unsigned OpNum) -> auto {
    return std::make_pair(Op1->getOperand(OpNum), Op2->getOperand(OpNum));
  };
 
  switch (Op1->getOpcode()) {
  default:
    break;
  case Instruction::Or:
    if (!cast<PossiblyDisjointInst>(Op1)->isDisjoint() ||
        !cast<PossiblyDisjointInst>(Op2)->isDisjoint())
      break;
    [[fallthrough]];
  case Instruction::Xor:
  case Instruction::Add: {
    Value *Other;
    if (match(Op2, m_c_BinOp(m_Specific(Op1->getOperand(0)), m_Value(Other))))
      return std::make_pair(Op1->getOperand(1), Other);
    if (match(Op2, m_c_BinOp(m_Specific(Op1->getOperand(1)), m_Value(Other))))
      return std::make_pair(Op1->getOperand(0), Other);
    break;
  }
  case Instruction::Sub:
    if (Op1->getOperand(0) == Op2->getOperand(0))
      return getOperands(1);
    if (Op1->getOperand(1) == Op2->getOperand(1))
      return getOperands(0);
    break;
  case Instruction::Mul: {
    // invertible if A * B == (A * B) mod 2^N where A, and B are integers
    // and N is the bitwdith.  The nsw case is non-obvious, but proven by
    // alive2: https://alive2.llvm.org/ce/z/Z6D5qK
    auto *OBO1 = cast<OverflowingBinaryOperator>(Op1);
    auto *OBO2 = cast<OverflowingBinaryOperator>(Op2);
    if ((!OBO1->hasNoUnsignedWrap() || !OBO2->hasNoUnsignedWrap()) &&
        (!OBO1->hasNoSignedWrap() || !OBO2->hasNoSignedWrap()))
      break;
 
    // Assume operand order has been canonicalized
    if (Op1->getOperand(1) == Op2->getOperand(1) &&
        isa<ConstantInt>(Op1->getOperand(1)) &&
        !cast<ConstantInt>(Op1->getOperand(1))->isZero())
      return getOperands(0);
    break;
  }
  case Instruction::Shl: {
    // Same as multiplies, with the difference that we don't need to check
    // for a non-zero multiply. Shifts always multiply by non-zero.
    auto *OBO1 = cast<OverflowingBinaryOperator>(Op1);
    auto *OBO2 = cast<OverflowingBinaryOperator>(Op2);
    if ((!OBO1->hasNoUnsignedWrap() || !OBO2->hasNoUnsignedWrap()) &&
        (!OBO1->hasNoSignedWrap() || !OBO2->hasNoSignedWrap()))
      break;
 
    if (Op1->getOperand(1) == Op2->getOperand(1))
      return getOperands(0);
    break;
  }
  case Instruction::AShr:
  case Instruction::LShr: {
    auto *PEO1 = cast<PossiblyExactOperator>(Op1);
    auto *PEO2 = cast<PossiblyExactOperator>(Op2);
    if (!PEO1->isExact() || !PEO2->isExact())
      break;
 
    if (Op1->getOperand(1) == Op2->getOperand(1))
      return getOperands(0);
    break;
  }
  case Instruction::SExt:
  case Instruction::ZExt:
    if (Op1->getOperand(0)->getType() == Op2->getOperand(0)->getType())
      return getOperands(0);
    break;
  case Instruction::PHI: {
    const PHINode *PN1 = cast<PHINode>(Op1);
    const PHINode *PN2 = cast<PHINode>(Op2);
 
    // If PN1 and PN2 are both recurrences, can we prove the entire recurrences
    // are a single invertible function of the start values? Note that repeated
    // application of an invertible function is also invertible
    BinaryOperator *BO1 = nullptr;
    Value *Start1 = nullptr, *Step1 = nullptr;
    BinaryOperator *BO2 = nullptr;
    Value *Start2 = nullptr, *Step2 = nullptr;
    if (PN1->getParent() != PN2->getParent() ||
        !matchSimpleRecurrence(PN1, BO1, Start1, Step1) ||
        !matchSimpleRecurrence(PN2, BO2, Start2, Step2))
      break;
 
    auto Values = getInvertibleOperands(cast<Operator>(BO1),
                                        cast<Operator>(BO2));
    if (!Values)
       break;
 
    // We have to be careful of mutually defined recurrences here.  Ex:
    // * X_i = X_(i-1) OP Y_(i-1), and Y_i = X_(i-1) OP V
    // * X_i = Y_i = X_(i-1) OP Y_(i-1)
    // The invertibility of these is complicated, and not worth reasoning
    // about (yet?).
    if (Values->first != PN1 || Values->second != PN2)
      break;
 
    return std::make_pair(Start1, Start2);
  }
  }
  return std::nullopt;
}
 
/// Return true if V1 == (binop V2, X), where X is known non-zero.
/// Only handle a small subset of binops where (binop V2, X) with non-zero X
/// implies V2 != V1.
static bool isModifyingBinopOfNonZero(const Value *V1, const Value *V2,
                                      const APInt &DemandedElts, unsigned Depth,
                                      const SimplifyQuery &Q) {
  const BinaryOperator *BO = dyn_cast<BinaryOperator>(V1);
  if (!BO)
    return false;
  switch (BO->getOpcode()) {
  default:
    break;
  case Instruction::Or:
    if (!cast<PossiblyDisjointInst>(V1)->isDisjoint())
      break;
    [[fallthrough]];
  case Instruction::Xor:
  case Instruction::Add:
    Value *Op = nullptr;
    if (V2 == BO->getOperand(0))
      Op = BO->getOperand(1);
    else if (V2 == BO->getOperand(1))
      Op = BO->getOperand(0);
    else
      return false;
    return isKnownNonZero(Op, DemandedElts, Q, Depth + 1);
  }
  return false;
}
 
/// Return true if V2 == V1 * C, where V1 is known non-zero, C is not 0/1 and
/// the multiplication is nuw or nsw.
static bool isNonEqualMul(const Value *V1, const Value *V2,
                          const APInt &DemandedElts, unsigned Depth,
                          const SimplifyQuery &Q) {
  if (auto *OBO = dyn_cast<OverflowingBinaryOperator>(V2)) {
    const APInt *C;
    return match(OBO, m_Mul(m_Specific(V1), m_APInt(C))) &&
           (OBO->hasNoUnsignedWrap() || OBO->hasNoSignedWrap()) &&
           !C->isZero() && !C->isOne() &&
           isKnownNonZero(V1, DemandedElts, Q, Depth + 1);
  }
  return false;
}
 
/// Return true if V2 == V1 << C, where V1 is known non-zero, C is not 0 and
/// the shift is nuw or nsw.
static bool isNonEqualShl(const Value *V1, const Value *V2,
                          const APInt &DemandedElts, unsigned Depth,
                          const SimplifyQuery &Q) {
  if (auto *OBO = dyn_cast<OverflowingBinaryOperator>(V2)) {
    const APInt *C;
    return match(OBO, m_Shl(m_Specific(V1), m_APInt(C))) &&
           (OBO->hasNoUnsignedWrap() || OBO->hasNoSignedWrap()) &&
           !C->isZero() && isKnownNonZero(V1, DemandedElts, Q, Depth + 1);
  }
  return false;
}
 
static bool isNonEqualPHIs(const PHINode *PN1, const PHINode *PN2,
                           const APInt &DemandedElts, unsigned Depth,
                           const SimplifyQuery &Q) {
  // Check two PHIs are in same block.
  if (PN1->getParent() != PN2->getParent())
    return false;
 
  SmallPtrSet<const BasicBlock *, 8> VisitedBBs;
  bool UsedFullRecursion = false;
  for (const BasicBlock *IncomBB : PN1->blocks()) {
    if (!VisitedBBs.insert(IncomBB).second)
      continue; // Don't reprocess blocks that we have dealt with already.
    const Value *IV1 = PN1->getIncomingValueForBlock(IncomBB);
    const Value *IV2 = PN2->getIncomingValueForBlock(IncomBB);
    const APInt *C1, *C2;
    if (match(IV1, m_APInt(C1)) && match(IV2, m_APInt(C2)) && *C1 != *C2)
      continue;
 
    // Only one pair of phi operands is allowed for full recursion.
    if (UsedFullRecursion)
      return false;
 
    SimplifyQuery RecQ = Q.getWithoutCondContext();
    RecQ.CxtI = IncomBB->getTerminator();
    if (!isKnownNonEqual(IV1, IV2, DemandedElts, Depth + 1, RecQ))
      return false;
    UsedFullRecursion = true;
  }
  return true;
}
 
static bool isNonEqualSelect(const Value *V1, const Value *V2,
                             const APInt &DemandedElts, unsigned Depth,
                             const SimplifyQuery &Q) {
  const SelectInst *SI1 = dyn_cast<SelectInst>(V1);
  if (!SI1)
    return false;
 
  if (const SelectInst *SI2 = dyn_cast<SelectInst>(V2)) {
    const Value *Cond1 = SI1->getCondition();
    const Value *Cond2 = SI2->getCondition();
    if (Cond1 == Cond2)
      return isKnownNonEqual(SI1->getTrueValue(), SI2->getTrueValue(),
                             DemandedElts, Depth + 1, Q) &&
             isKnownNonEqual(SI1->getFalseValue(), SI2->getFalseValue(),
                             DemandedElts, Depth + 1, Q);
  }
  return isKnownNonEqual(SI1->getTrueValue(), V2, DemandedElts, Depth + 1, Q) &&
         isKnownNonEqual(SI1->getFalseValue(), V2, DemandedElts, Depth + 1, Q);
}
 
// Check to see if A is both a GEP and is the incoming value for a PHI in the
// loop, and B is either a ptr or another GEP. If the PHI has 2 incoming values,
// one of them being the recursive GEP A and the other a ptr at same base and at
// the same/higher offset than B we are only incrementing the pointer further in
// loop if offset of recursive GEP is greater than 0.
static bool isNonEqualPointersWithRecursiveGEP(const Value *A, const Value *B,
                                               const SimplifyQuery &Q) {
  if (!A->getType()->isPointerTy() || !B->getType()->isPointerTy())
    return false;
 
  auto *GEPA = dyn_cast<GEPOperator>(A);
  if (!GEPA || GEPA->getNumIndices() != 1 || !isa<Constant>(GEPA->idx_begin()))
    return false;
 
  // Handle 2 incoming PHI values with one being a recursive GEP.
  auto *PN = dyn_cast<PHINode>(GEPA->getPointerOperand());
  if (!PN || PN->getNumIncomingValues() != 2)
    return false;
 
  // Search for the recursive GEP as an incoming operand, and record that as
  // Step.
  Value *Start = nullptr;
  Value *Step = const_cast<Value *>(A);
  if (PN->getIncomingValue(0) == Step)
    Start = PN->getIncomingValue(1);
  else if (PN->getIncomingValue(1) == Step)
    Start = PN->getIncomingValue(0);
  else
    return false;
 
  // Other incoming node base should match the B base.
  // StartOffset >= OffsetB && StepOffset > 0?
  // StartOffset <= OffsetB && StepOffset < 0?
  // Is non-equal if above are true.
  // We use stripAndAccumulateInBoundsConstantOffsets to restrict the
  // optimisation to inbounds GEPs only.
  unsigned IndexWidth = Q.DL.getIndexTypeSizeInBits(Start->getType());
  APInt StartOffset(IndexWidth, 0);
  Start = Start->stripAndAccumulateInBoundsConstantOffsets(Q.DL, StartOffset);
  APInt StepOffset(IndexWidth, 0);
  Step = Step->stripAndAccumulateInBoundsConstantOffsets(Q.DL, StepOffset);
 
  // Check if Base Pointer of Step matches the PHI.
  if (Step != PN)
    return false;
  APInt OffsetB(IndexWidth, 0);
  B = B->stripAndAccumulateInBoundsConstantOffsets(Q.DL, OffsetB);
  return Start == B &&
         ((StartOffset.sge(OffsetB) && StepOffset.isStrictlyPositive()) ||
          (StartOffset.sle(OffsetB) && StepOffset.isNegative()));
}
 
/// Return true if it is known that V1 != V2.
static bool isKnownNonEqual(const Value *V1, const Value *V2,
                            const APInt &DemandedElts, unsigned Depth,
                            const SimplifyQuery &Q) {
  if (V1 == V2)
    return false;
  if (V1->getType() != V2->getType())
    // We can't look through casts yet.
    return false;
 
  if (Depth >= MaxAnalysisRecursionDepth)
    return false;
 
  // See if we can recurse through (exactly one of) our operands.  This
  // requires our operation be 1-to-1 and map every input value to exactly
  // one output value.  Such an operation is invertible.
  auto *O1 = dyn_cast<Operator>(V1);
  auto *O2 = dyn_cast<Operator>(V2);
  if (O1 && O2 && O1->getOpcode() == O2->getOpcode()) {
    if (auto Values = getInvertibleOperands(O1, O2))
      return isKnownNonEqual(Values->first, Values->second, DemandedElts,
                             Depth + 1, Q);
 
    if (const PHINode *PN1 = dyn_cast<PHINode>(V1)) {
      const PHINode *PN2 = cast<PHINode>(V2);
      // FIXME: This is missing a generalization to handle the case where one is
      // a PHI and another one isn't.
      if (isNonEqualPHIs(PN1, PN2, DemandedElts, Depth, Q))
        return true;
    };
  }
 
  if (isModifyingBinopOfNonZero(V1, V2, DemandedElts, Depth, Q) ||
      isModifyingBinopOfNonZero(V2, V1, DemandedElts, Depth, Q))
    return true;
 
  if (isNonEqualMul(V1, V2, DemandedElts, Depth, Q) ||
      isNonEqualMul(V2, V1, DemandedElts, Depth, Q))
    return true;
 
  if (isNonEqualShl(V1, V2, DemandedElts, Depth, Q) ||
      isNonEqualShl(V2, V1, DemandedElts, Depth, Q))
    return true;
 
  if (V1->getType()->isIntOrIntVectorTy()) {
    // Are any known bits in V1 contradictory to known bits in V2? If V1
    // has a known zero where V2 has a known one, they must not be equal.
    KnownBits Known1 = computeKnownBits(V1, DemandedElts, Depth, Q);
    if (!Known1.isUnknown()) {
      KnownBits Known2 = computeKnownBits(V2, DemandedElts, Depth, Q);
      if (Known1.Zero.intersects(Known2.One) ||
          Known2.Zero.intersects(Known1.One))
        return true;
    }
  }
 
  if (isNonEqualSelect(V1, V2, DemandedElts, Depth, Q) ||
      isNonEqualSelect(V2, V1, DemandedElts, Depth, Q))
    return true;
 
  if (isNonEqualPointersWithRecursiveGEP(V1, V2, Q) ||
      isNonEqualPointersWithRecursiveGEP(V2, V1, Q))
    return true;
 
  Value *A, *B;
  // PtrToInts are NonEqual if their Ptrs are NonEqual.
  // Check PtrToInt type matches the pointer size.
  if (match(V1, m_PtrToIntSameSize(Q.DL, m_Value(A))) &&
      match(V2, m_PtrToIntSameSize(Q.DL, m_Value(B))))
    return isKnownNonEqual(A, B, DemandedElts, Depth + 1, Q);
 
  return false;
}
 
// Match a signed min+max clamp pattern like smax(smin(In, CHigh), CLow).
// Returns the input and lower/upper bounds.
static bool isSignedMinMaxClamp(const Value *Select, const Value *&In,
                                const APInt *&CLow, const APInt *&CHigh) {
  assert(isa<Operator>(Select) &&
         cast<Operator>(Select)->getOpcode() == Instruction::Select &&
         "Input should be a Select!");
 
  const Value *LHS = nullptr, *RHS = nullptr;
  SelectPatternFlavor SPF = matchSelectPattern(Select, LHS, RHS).Flavor;
  if (SPF != SPF_SMAX && SPF != SPF_SMIN)
    return false;
 
  if (!match(RHS, m_APInt(CLow)))
    return false;
 
  const Value *LHS2 = nullptr, *RHS2 = nullptr;
  SelectPatternFlavor SPF2 = matchSelectPattern(LHS, LHS2, RHS2).Flavor;
  if (getInverseMinMaxFlavor(SPF) != SPF2)
    return false;
 
  if (!match(RHS2, m_APInt(CHigh)))
    return false;
 
  if (SPF == SPF_SMIN)
    std::swap(CLow, CHigh);
 
  In = LHS2;
  return CLow->sle(*CHigh);
}
 
static bool isSignedMinMaxIntrinsicClamp(const IntrinsicInst *II,
                                         const APInt *&CLow,
                                         const APInt *&CHigh) {
  assert((II->getIntrinsicID() == Intrinsic::smin ||
          II->getIntrinsicID() == Intrinsic::smax) && "Must be smin/smax");
 
  Intrinsic::ID InverseID = getInverseMinMaxIntrinsic(II->getIntrinsicID());
  auto *InnerII = dyn_cast<IntrinsicInst>(II->getArgOperand(0));
  if (!InnerII || InnerII->getIntrinsicID() != InverseID ||
      !match(II->getArgOperand(1), m_APInt(CLow)) ||
      !match(InnerII->getArgOperand(1), m_APInt(CHigh)))
    return false;
 
  if (II->getIntrinsicID() == Intrinsic::smin)
    std::swap(CLow, CHigh);
  return CLow->sle(*CHigh);
}
 
/// For vector constants, loop over the elements and find the constant with the
/// minimum number of sign bits. Return 0 if the value is not a vector constant
/// or if any element was not analyzed; otherwise, return the count for the
/// element with the minimum number of sign bits.
static unsigned computeNumSignBitsVectorConstant(const Value *V,
                                                 const APInt &DemandedElts,
                                                 unsigned TyBits) {
  const auto *CV = dyn_cast<Constant>(V);
  if (!CV || !isa<FixedVectorType>(CV->getType()))
    return 0;
 
  unsigned MinSignBits = TyBits;
  unsigned NumElts = cast<FixedVectorType>(CV->getType())->getNumElements();
  for (unsigned i = 0; i != NumElts; ++i) {
    if (!DemandedElts[i])
      continue;
    // If we find a non-ConstantInt, bail out.
    auto *Elt = dyn_cast_or_null<ConstantInt>(CV->getAggregateElement(i));
    if (!Elt)
      return 0;
 
    MinSignBits = std::min(MinSignBits, Elt->getValue().getNumSignBits());
  }
 
  return MinSignBits;
}
 
static unsigned ComputeNumSignBitsImpl(const Value *V,
                                       const APInt &DemandedElts,
                                       unsigned Depth, const SimplifyQuery &Q);
 
static unsigned ComputeNumSignBits(const Value *V, const APInt &DemandedElts,
                                   unsigned Depth, const SimplifyQuery &Q) {
  unsigned Result = ComputeNumSignBitsImpl(V, DemandedElts, Depth, Q);
  assert(Result > 0 && "At least one sign bit needs to be present!");
  return Result;
}
 
/// Return the number of times the sign bit of the register is replicated into
/// the other bits. We know that at least 1 bit is always equal to the sign bit
/// (itself), but other cases can give us information. For example, immediately
/// after an "ashr X, 2", we know that the top 3 bits are all equal to each
/// other, so we return 3. For vectors, return the number of sign bits for the
/// vector element with the minimum number of known sign bits of the demanded
/// elements in the vector specified by DemandedElts.
static unsigned ComputeNumSignBitsImpl(const Value *V,
                                       const APInt &DemandedElts,
                                       unsigned Depth, const SimplifyQuery &Q) {
  Type *Ty = V->getType();
#ifndef NDEBUG
  assert(Depth <= MaxAnalysisRecursionDepth && "Limit Search Depth");
 
  if (auto *FVTy = dyn_cast<FixedVectorType>(Ty)) {
    assert(
        FVTy->getNumElements() == DemandedElts.getBitWidth() &&
        "DemandedElt width should equal the fixed vector number of elements");
  } else {
    assert(DemandedElts == APInt(1, 1) &&
           "DemandedElt width should be 1 for scalars");
  }
#endif
 
  // We return the minimum number of sign bits that are guaranteed to be present
  // in V, so for undef we have to conservatively return 1.  We don't have the
  // same behavior for poison though -- that's a FIXME today.
 
  Type *ScalarTy = Ty->getScalarType();
  unsigned TyBits = ScalarTy->isPointerTy() ?
    Q.DL.getPointerTypeSizeInBits(ScalarTy) :
    Q.DL.getTypeSizeInBits(ScalarTy);
 
  unsigned Tmp, Tmp2;
  unsigned FirstAnswer = 1;
 
  // Note that ConstantInt is handled by the general computeKnownBits case
  // below.
 
  if (Depth == MaxAnalysisRecursionDepth)
    return 1;
 
  if (auto *U = dyn_cast<Operator>(V)) {
    switch (Operator::getOpcode(V)) {
    default: break;
    case Instruction::SExt:
      Tmp = TyBits - U->getOperand(0)->getType()->getScalarSizeInBits();
      return ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q) +
             Tmp;
 
    case Instruction::SDiv: {
      const APInt *Denominator;
      // sdiv X, C -> adds log(C) sign bits.
      if (match(U->getOperand(1), m_APInt(Denominator))) {
 
        // Ignore non-positive denominator.
        if (!Denominator->isStrictlyPositive())
          break;
 
        // Calculate the incoming numerator bits.
        unsigned NumBits =
            ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q);
 
        // Add floor(log(C)) bits to the numerator bits.
        return std::min(TyBits, NumBits + Denominator->logBase2());
      }
      break;
    }
 
    case Instruction::SRem: {
      Tmp = ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q);
 
      const APInt *Denominator;
      // srem X, C -> we know that the result is within [-C+1,C) when C is a
      // positive constant.  This let us put a lower bound on the number of sign
      // bits.
      if (match(U->getOperand(1), m_APInt(Denominator))) {
 
        // Ignore non-positive denominator.
        if (Denominator->isStrictlyPositive()) {
          // Calculate the leading sign bit constraints by examining the
          // denominator.  Given that the denominator is positive, there are two
          // cases:
          //
          //  1. The numerator is positive. The result range is [0,C) and
          //     [0,C) u< (1 << ceilLogBase2(C)).
          //
          //  2. The numerator is negative. Then the result range is (-C,0] and
          //     integers in (-C,0] are either 0 or >u (-1 << ceilLogBase2(C)).
          //
          // Thus a lower bound on the number of sign bits is `TyBits -
          // ceilLogBase2(C)`.
 
          unsigned ResBits = TyBits - Denominator->ceilLogBase2();
          Tmp = std::max(Tmp, ResBits);
        }
      }
      return Tmp;
    }
 
    case Instruction::AShr: {
      Tmp = ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q);
      // ashr X, C   -> adds C sign bits.  Vectors too.
      const APInt *ShAmt;
      if (match(U->getOperand(1), m_APInt(ShAmt))) {
        if (ShAmt->uge(TyBits))
          break; // Bad shift.
        unsigned ShAmtLimited = ShAmt->getZExtValue();
        Tmp += ShAmtLimited;
        if (Tmp > TyBits) Tmp = TyBits;
      }
      return Tmp;
    }
    case Instruction::Shl: {
      const APInt *ShAmt;
      Value *X = nullptr;
      if (match(U->getOperand(1), m_APInt(ShAmt))) {
        // shl destroys sign bits.
        if (ShAmt->uge(TyBits))
          break; // Bad shift.
        // We can look through a zext (more or less treating it as a sext) if
        // all extended bits are shifted out.
        if (match(U->getOperand(0), m_ZExt(m_Value(X))) &&
            ShAmt->uge(TyBits - X->getType()->getScalarSizeInBits())) {
          Tmp = ComputeNumSignBits(X, DemandedElts, Depth + 1, Q);
          Tmp += TyBits - X->getType()->getScalarSizeInBits();
        } else
          Tmp =
              ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q);
        if (ShAmt->uge(Tmp))
          break; // Shifted all sign bits out.
        Tmp2 = ShAmt->getZExtValue();
        return Tmp - Tmp2;
      }
      break;
    }
    case Instruction::And:
    case Instruction::Or:
    case Instruction::Xor: // NOT is handled here.
      // Logical binary ops preserve the number of sign bits at the worst.
      Tmp = ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q);
      if (Tmp != 1) {
        Tmp2 = ComputeNumSignBits(U->getOperand(1), DemandedElts, Depth + 1, Q);
        FirstAnswer = std::min(Tmp, Tmp2);
        // We computed what we know about the sign bits as our first
        // answer. Now proceed to the generic code that uses
        // computeKnownBits, and pick whichever answer is better.
      }
      break;
 
    case Instruction::Select: {
      // If we have a clamp pattern, we know that the number of sign bits will
      // be the minimum of the clamp min/max range.
      const Value *X;
      const APInt *CLow, *CHigh;
      if (isSignedMinMaxClamp(U, X, CLow, CHigh))
        return std::min(CLow->getNumSignBits(), CHigh->getNumSignBits());
 
      Tmp = ComputeNumSignBits(U->getOperand(1), DemandedElts, Depth + 1, Q);
      if (Tmp == 1)
        break;
      Tmp2 = ComputeNumSignBits(U->getOperand(2), DemandedElts, Depth + 1, Q);
      return std::min(Tmp, Tmp2);
    }
 
    case Instruction::Add:
      // Add can have at most one carry bit.  Thus we know that the output
      // is, at worst, one more bit than the inputs.
      Tmp = ComputeNumSignBits(U->getOperand(0), Depth + 1, Q);
      if (Tmp == 1) break;
 
      // Special case decrementing a value (ADD X, -1):
      if (const auto *CRHS = dyn_cast<Constant>(U->getOperand(1)))
        if (CRHS->isAllOnesValue()) {
          KnownBits Known(TyBits);
          computeKnownBits(U->getOperand(0), DemandedElts, Known, Depth + 1, Q);
 
          // If the input is known to be 0 or 1, the output is 0/-1, which is
          // all sign bits set.
          if ((Known.Zero | 1).isAllOnes())
            return TyBits;
 
          // If we are subtracting one from a positive number, there is no carry
          // out of the result.
          if (Known.isNonNegative())
            return Tmp;
        }
 
      Tmp2 = ComputeNumSignBits(U->getOperand(1), DemandedElts, Depth + 1, Q);
      if (Tmp2 == 1)
        break;
      return std::min(Tmp, Tmp2) - 1;
 
    case Instruction::Sub:
      Tmp2 = ComputeNumSignBits(U->getOperand(1), DemandedElts, Depth + 1, Q);
      if (Tmp2 == 1)
        break;
 
      // Handle NEG.
      if (const auto *CLHS = dyn_cast<Constant>(U->getOperand(0)))
        if (CLHS->isNullValue()) {
          KnownBits Known(TyBits);
          computeKnownBits(U->getOperand(1), DemandedElts, Known, Depth + 1, Q);
          // If the input is known to be 0 or 1, the output is 0/-1, which is
          // all sign bits set.
          if ((Known.Zero | 1).isAllOnes())
            return TyBits;
 
          // If the input is known to be positive (the sign bit is known clear),
          // the output of the NEG has the same number of sign bits as the
          // input.
          if (Known.isNonNegative())
            return Tmp2;
 
          // Otherwise, we treat this like a SUB.
        }
 
      // Sub can have at most one carry bit.  Thus we know that the output
      // is, at worst, one more bit than the inputs.
      Tmp = ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q);
      if (Tmp == 1)
        break;
      return std::min(Tmp, Tmp2) - 1;
 
    case Instruction::Mul: {
      // The output of the Mul can be at most twice the valid bits in the
      // inputs.
      unsigned SignBitsOp0 =
          ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q);
      if (SignBitsOp0 == 1)
        break;
      unsigned SignBitsOp1 =
          ComputeNumSignBits(U->getOperand(1), DemandedElts, Depth + 1, Q);
      if (SignBitsOp1 == 1)
        break;
      unsigned OutValidBits =
          (TyBits - SignBitsOp0 + 1) + (TyBits - SignBitsOp1 + 1);
      return OutValidBits > TyBits ? 1 : TyBits - OutValidBits + 1;
    }
 
    case Instruction::PHI: {
      const PHINode *PN = cast<PHINode>(U);
      unsigned NumIncomingValues = PN->getNumIncomingValues();
      // Don't analyze large in-degree PHIs.
      if (NumIncomingValues > 4) break;
      // Unreachable blocks may have zero-operand PHI nodes.
      if (NumIncomingValues == 0) break;
 
      // Take the minimum of all incoming values.  This can't infinitely loop
      // because of our depth threshold.
      SimplifyQuery RecQ = Q.getWithoutCondContext();
      Tmp = TyBits;
      for (unsigned i = 0, e = NumIncomingValues; i != e; ++i) {
        if (Tmp == 1) return Tmp;
        RecQ.CxtI = PN->getIncomingBlock(i)->getTerminator();
        Tmp = std::min(Tmp, ComputeNumSignBits(PN->getIncomingValue(i),
                                               DemandedElts, Depth + 1, RecQ));
      }
      return Tmp;
    }
 
    case Instruction::Trunc: {
      // If the input contained enough sign bits that some remain after the
      // truncation, then we can make use of that. Otherwise we don't know
      // anything.
      Tmp = ComputeNumSignBits(U->getOperand(0), Depth + 1, Q);
      unsigned OperandTyBits = U->getOperand(0)->getType()->getScalarSizeInBits();
      if (Tmp > (OperandTyBits - TyBits))
        return Tmp - (OperandTyBits - TyBits);
 
      return 1;
    }
 
    case Instruction::ExtractElement:
      // Look through extract element. At the moment we keep this simple and
      // skip tracking the specific element. But at least we might find
      // information valid for all elements of the vector (for example if vector
      // is sign extended, shifted, etc).
      return ComputeNumSignBits(U->getOperand(0), Depth + 1, Q);
 
    case Instruction::ShuffleVector: {
      // Collect the minimum number of sign bits that are shared by every vector
      // element referenced by the shuffle.
      auto *Shuf = dyn_cast<ShuffleVectorInst>(U);
      if (!Shuf) {
        // FIXME: Add support for shufflevector constant expressions.
        return 1;
      }
      APInt DemandedLHS, DemandedRHS;
      // For undef elements, we don't know anything about the common state of
      // the shuffle result.
      if (!getShuffleDemandedElts(Shuf, DemandedElts, DemandedLHS, DemandedRHS))
        return 1;
      Tmp = std::numeric_limits<unsigned>::max();
      if (!!DemandedLHS) {
        const Value *LHS = Shuf->getOperand(0);
        Tmp = ComputeNumSignBits(LHS, DemandedLHS, Depth + 1, Q);
      }
      // If we don't know anything, early out and try computeKnownBits
      // fall-back.
      if (Tmp == 1)
        break;
      if (!!DemandedRHS) {
        const Value *RHS = Shuf->getOperand(1);
        Tmp2 = ComputeNumSignBits(RHS, DemandedRHS, Depth + 1, Q);
        Tmp = std::min(Tmp, Tmp2);
      }
      // If we don't know anything, early out and try computeKnownBits
      // fall-back.
      if (Tmp == 1)
        break;
      assert(Tmp <= TyBits && "Failed to determine minimum sign bits");
      return Tmp;
    }
    case Instruction::Call: {
      if (const auto *II = dyn_cast<IntrinsicInst>(U)) {
        switch (II->getIntrinsicID()) {
        default:
          break;
        case Intrinsic::abs:
          Tmp =
              ComputeNumSignBits(U->getOperand(0), DemandedElts, Depth + 1, Q);
          if (Tmp == 1)
            break;
 
          // Absolute value reduces number of sign bits by at most 1.
          return Tmp - 1;
        case Intrinsic::smin:
        case Intrinsic::smax: {
          const APInt *CLow, *CHigh;
          if (isSignedMinMaxIntrinsicClamp(II, CLow, CHigh))
            return std::min(CLow->getNumSignBits(), CHigh->getNumSignBits());
        }
        }
      }
    }
    }
  }
 
  // Finally, if we can prove that the top bits of the result are 0's or 1's,
  // use this information.
 
  // If we can examine all elements of a vector constant successfully, we're
  // done (we can't do any better than that). If not, keep trying.
  if (unsigned VecSignBits =
          computeNumSignBitsVectorConstant(V, DemandedElts, TyBits))
    return VecSignBits;
 
  KnownBits Known(TyBits);
  computeKnownBits(V, DemandedElts, Known, Depth, Q);
 
  // If we know that the sign bit is either zero or one, determine the number of
  // identical bits in the top of the input value.
  return std::max(FirstAnswer, Known.countMinSignBits());
}
 
Intrinsic::ID llvm::getIntrinsicForCallSite(const CallBase &CB,
                                            const TargetLibraryInfo *TLI) {
  const Function *F = CB.getCalledFunction();
  if (!F)
    return Intrinsic::not_intrinsic;
 
  if (F->isIntrinsic())
    return F->getIntrinsicID();
 
  // We are going to infer semantics of a library function based on mapping it
  // to an LLVM intrinsic. Check that the library function is available from
  // this callbase and in this environment.
  LibFunc Func;
  if (F->hasLocalLinkage() || !TLI || !TLI->getLibFunc(CB, Func) ||
      !CB.onlyReadsMemory())
    return Intrinsic::not_intrinsic;
 
  switch (Func) {
  default:
    break;
  case LibFunc_sin:
  case LibFunc_sinf:
  case LibFunc_sinl:
    return Intrinsic::sin;
  case LibFunc_cos:
  case LibFunc_cosf:
  case LibFunc_cosl:
    return Intrinsic::cos;
  case LibFunc_tan:
  case LibFunc_tanf:
  case LibFunc_tanl:
    return Intrinsic::tan;
  case LibFunc_exp:
  case LibFunc_expf:
  case LibFunc_expl:
    return Intrinsic::exp;
  case LibFunc_exp2:
  case LibFunc_exp2f:
  case LibFunc_exp2l:
    return Intrinsic::exp2;
  case LibFunc_log:
  case LibFunc_logf:
  case LibFunc_logl:
    return Intrinsic::log;
  case LibFunc_log10:
  case LibFunc_log10f:
  case LibFunc_log10l:
    return Intrinsic::log10;
  case LibFunc_log2:
  case LibFunc_log2f:
  case LibFunc_log2l:
    return Intrinsic::log2;
  case LibFunc_fabs:
  case LibFunc_fabsf:
  case LibFunc_fabsl:
    return Intrinsic::fabs;
  case LibFunc_fmin:
  case LibFunc_fminf:
  case LibFunc_fminl:
    return Intrinsic::minnum;
  case LibFunc_fmax:
  case LibFunc_fmaxf:
  case LibFunc_fmaxl:
    return Intrinsic::maxnum;
  case LibFunc_copysign:
  case LibFunc_copysignf:
  case LibFunc_copysignl:
    return Intrinsic::copysign;
  case LibFunc_floor:
  case LibFunc_floorf:
  case LibFunc_floorl:
    return Intrinsic::floor;
  case LibFunc_ceil:
  case LibFunc_ceilf:
  case LibFunc_ceill:
    return Intrinsic::ceil;
  case LibFunc_trunc:
  case LibFunc_truncf:
  case LibFunc_truncl:
    return Intrinsic::trunc;
  case LibFunc_rint:
  case LibFunc_rintf:
  case LibFunc_rintl:
    return Intrinsic::rint;
  case LibFunc_nearbyint:
  case LibFunc_nearbyintf:
  case LibFunc_nearbyintl:
    return Intrinsic::nearbyint;
  case LibFunc_round:
  case LibFunc_roundf:
  case LibFunc_roundl:
    return Intrinsic::round;
  case LibFunc_roundeven:
  case LibFunc_roundevenf:
  case LibFunc_roundevenl:
    return Intrinsic::roundeven;
  case LibFunc_pow:
  case LibFunc_powf:
  case LibFunc_powl:
    return Intrinsic::pow;
  case LibFunc_sqrt:
  case LibFunc_sqrtf:
  case LibFunc_sqrtl:
    return Intrinsic::sqrt;
  }
 
  return Intrinsic::not_intrinsic;
}
 
/// Return true if it's possible to assume IEEE treatment of input denormals in
/// \p F for \p Val.
static bool inputDenormalIsIEEE(const Function &F, const Type *Ty) {
  Ty = Ty->getScalarType();
  return F.getDenormalMode(Ty->getFltSemantics()).Input == DenormalMode::IEEE;
}
 
static bool inputDenormalIsIEEEOrPosZero(const Function &F, const Type *Ty) {
  Ty = Ty->getScalarType();
  DenormalMode Mode = F.getDenormalMode(Ty->getFltSemantics());
  return Mode.Input == DenormalMode::IEEE ||
         Mode.Input == DenormalMode::PositiveZero;
}
 
static bool outputDenormalIsIEEEOrPosZero(const Function &F, const Type *Ty) {
  Ty = Ty->getScalarType();
  DenormalMode Mode = F.getDenormalMode(Ty->getFltSemantics());
  return Mode.Output == DenormalMode::IEEE ||
         Mode.Output == DenormalMode::PositiveZero;
}
 
bool KnownFPClass::isKnownNeverLogicalZero(const Function &F, Type *Ty) const {
  return isKnownNeverZero() &&
         (isKnownNeverSubnormal() || inputDenormalIsIEEE(F, Ty));
}
 
bool KnownFPClass::isKnownNeverLogicalNegZero(const Function &F,
                                              Type *Ty) const {
  return isKnownNeverNegZero() &&
         (isKnownNeverNegSubnormal() || inputDenormalIsIEEEOrPosZero(F, Ty));
}
 
bool KnownFPClass::isKnownNeverLogicalPosZero(const Function &F,
                                              Type *Ty) const {
  if (!isKnownNeverPosZero())
    return false;
 
  // If we know there are no denormals, nothing can be flushed to zero.
  if (isKnownNeverSubnormal())
    return true;
 
  DenormalMode Mode = F.getDenormalMode(Ty->getScalarType()->getFltSemantics());
  switch (Mode.Input) {
  case DenormalMode::IEEE:
    return true;
  case DenormalMode::PreserveSign:
    // Negative subnormal won't flush to +0
    return isKnownNeverPosSubnormal();
  case DenormalMode::PositiveZero:
  default:
    // Both positive and negative subnormal could flush to +0
    return false;
  }
 
  llvm_unreachable("covered switch over denormal mode");
}
 
void KnownFPClass::propagateDenormal(const KnownFPClass &Src, const Function &F,
                                     Type *Ty) {
  KnownFPClasses = Src.KnownFPClasses;
  // If we aren't assuming the source can't be a zero, we don't have to check if
  // a denormal input could be flushed.
  if (!Src.isKnownNeverPosZero() && !Src.isKnownNeverNegZero())
    return;
 
  // If we know the input can't be a denormal, it can't be flushed to 0.
  if (Src.isKnownNeverSubnormal())
    return;
 
  DenormalMode Mode = F.getDenormalMode(Ty->getScalarType()->getFltSemantics());
 
  if (!Src.isKnownNeverPosSubnormal() && Mode != DenormalMode::getIEEE())
    KnownFPClasses |= fcPosZero;
 
  if (!Src.isKnownNeverNegSubnormal() && Mode != DenormalMode::getIEEE()) {
    if (Mode != DenormalMode::getPositiveZero())
      KnownFPClasses |= fcNegZero;
 
    if (Mode.Input == DenormalMode::PositiveZero ||
        Mode.Output == DenormalMode::PositiveZero ||
        Mode.Input == DenormalMode::Dynamic ||
        Mode.Output == DenormalMode::Dynamic)
      KnownFPClasses |= fcPosZero;
  }
}
 
void KnownFPClass::propagateCanonicalizingSrc(const KnownFPClass &Src,
                                              const Function &F, Type *Ty) {
  propagateDenormal(Src, F, Ty);
  propagateNaN(Src, /*PreserveSign=*/true);
}
 
/// Given an exploded icmp instruction, return true if the comparison only
/// checks the sign bit. If it only checks the sign bit, set TrueIfSigned if
/// the result of the comparison is true when the input value is signed.
bool llvm::isSignBitCheck(ICmpInst::Predicate Pred, const APInt &RHS,
                          bool &TrueIfSigned) {
  switch (Pred) {
  case ICmpInst::ICMP_SLT: // True if LHS s< 0
    TrueIfSigned = true;
    return RHS.isZero();
  case ICmpInst::ICMP_SLE: // True if LHS s<= -1
    TrueIfSigned = true;
    return RHS.isAllOnes();
  case ICmpInst::ICMP_SGT: // True if LHS s> -1
    TrueIfSigned = false;
    return RHS.isAllOnes();
  case ICmpInst::ICMP_SGE: // True if LHS s>= 0
    TrueIfSigned = false;
    return RHS.isZero();
  case ICmpInst::ICMP_UGT:
    // True if LHS u> RHS and RHS == sign-bit-mask - 1
    TrueIfSigned = true;
    return RHS.isMaxSignedValue();
  case ICmpInst::ICMP_UGE:
    // True if LHS u>= RHS and RHS == sign-bit-mask (2^7, 2^15, 2^31, etc)
    TrueIfSigned = true;
    return RHS.isMinSignedValue();
  case ICmpInst::ICMP_ULT:
    // True if LHS u< RHS and RHS == sign-bit-mask (2^7, 2^15, 2^31, etc)
    TrueIfSigned = false;
    return RHS.isMinSignedValue();
  case ICmpInst::ICMP_ULE:
    // True if LHS u<= RHS and RHS == sign-bit-mask - 1
    TrueIfSigned = false;
    return RHS.isMaxSignedValue();
  default:
    return false;
  }
}
 
/// Returns a pair of values, which if passed to llvm.is.fpclass, returns the
/// same result as an fcmp with the given operands.
std::pair<Value *, FPClassTest> llvm::fcmpToClassTest(FCmpInst::Predicate Pred,
                                                      const Function &F,
                                                      Value *LHS, Value *RHS,
                                                      bool LookThroughSrc) {
  const APFloat *ConstRHS;
  if (!match(RHS, m_APFloatAllowPoison(ConstRHS)))
    return {nullptr, fcAllFlags};
 
  return fcmpToClassTest(Pred, F, LHS, ConstRHS, LookThroughSrc);
}
 
std::pair<Value *, FPClassTest>
llvm::fcmpToClassTest(FCmpInst::Predicate Pred, const Function &F, Value *LHS,
                      const APFloat *ConstRHS, bool LookThroughSrc) {
 
  auto [Src, ClassIfTrue, ClassIfFalse] =
      fcmpImpliesClass(Pred, F, LHS, *ConstRHS, LookThroughSrc);
  if (Src && ClassIfTrue == ~ClassIfFalse)
    return {Src, ClassIfTrue};
  return {nullptr, fcAllFlags};
}
 
/// Return the return value for fcmpImpliesClass for a compare that produces an
/// exact class test.
static std::tuple<Value *, FPClassTest, FPClassTest> exactClass(Value *V,
                                                                FPClassTest M) {
  return {V, M, ~M};
}
 
std::tuple<Value *, FPClassTest, FPClassTest>
llvm::fcmpImpliesClass(CmpInst::Predicate Pred, const Function &F, Value *LHS,
                       FPClassTest RHSClass, bool LookThroughSrc) {
  assert(RHSClass != fcNone);
  Value *Src = LHS;
 
  if (Pred == FCmpInst::FCMP_TRUE)
    return exactClass(Src, fcAllFlags);
 
  if (Pred == FCmpInst::FCMP_FALSE)
    return exactClass(Src, fcNone);
 
  const FPClassTest OrigClass = RHSClass;
 
  const bool IsNegativeRHS = (RHSClass & fcNegative) == RHSClass;
  const bool IsPositiveRHS = (RHSClass & fcPositive) == RHSClass;
  const bool IsNaN = (RHSClass & ~fcNan) == fcNone;
 
  if (IsNaN) {
    // fcmp o__ x, nan -> false
    // fcmp u__ x, nan -> true
    return exactClass(Src, CmpInst::isOrdered(Pred) ? fcNone : fcAllFlags);
  }
 
  // fcmp ord x, zero|normal|subnormal|inf -> ~fcNan
  if (Pred == FCmpInst::FCMP_ORD)
    return exactClass(Src, ~fcNan);
 
  // fcmp uno x, zero|normal|subnormal|inf -> fcNan
  if (Pred == FCmpInst::FCMP_UNO)
    return exactClass(Src, fcNan);
 
  const bool IsFabs = LookThroughSrc && match(LHS, m_FAbs(m_Value(Src)));
  if (IsFabs)
    RHSClass = llvm::inverse_fabs(RHSClass);
 
  const bool IsZero = (OrigClass & fcZero) == OrigClass;
  if (IsZero) {
    assert(Pred != FCmpInst::FCMP_ORD && Pred != FCmpInst::FCMP_UNO);
    // Compares with fcNone are only exactly equal to fcZero if input denormals
    // are not flushed.
    // TODO: Handle DAZ by expanding masks to cover subnormal cases.
    if (!inputDenormalIsIEEE(F, LHS->getType()))
      return {nullptr, fcAllFlags, fcAllFlags};
 
    switch (Pred) {
    case FCmpInst::FCMP_OEQ: // Match x == 0.0
      return exactClass(Src, fcZero);
    case FCmpInst::FCMP_UEQ: // Match isnan(x) || (x == 0.0)
      return exactClass(Src, fcZero | fcNan);
    case FCmpInst::FCMP_UNE: // Match (x != 0.0)
      return exactClass(Src, ~fcZero);
    case FCmpInst::FCMP_ONE: // Match !isnan(x) && x != 0.0
      return exactClass(Src, ~fcNan & ~fcZero);
    case FCmpInst::FCMP_ORD:
      // Canonical form of ord/uno is with a zero. We could also handle
      // non-canonical other non-NaN constants or LHS == RHS.
      return exactClass(Src, ~fcNan);
    case FCmpInst::FCMP_UNO:
      return exactClass(Src, fcNan);
    case FCmpInst::FCMP_OGT: // x > 0
      return exactClass(Src, fcPosSubnormal | fcPosNormal | fcPosInf);
    case FCmpInst::FCMP_UGT: // isnan(x) || x > 0
      return exactClass(Src, fcPosSubnormal | fcPosNormal | fcPosInf | fcNan);
    case FCmpInst::FCMP_OGE: // x >= 0