llvm/lib/Transforms/Scalar/SeparateConstOffsetFromGEP.cpp

//===-- SeparateConstOffsetFromGEP.cpp - ------------------------*- C++ -*-===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// Loop unrolling may create many similar GEPs for array accesses.
// e.g., a 2-level loop
//
// float a[32][32]; // global variable
//
// for (int i = 0; i < 2; ++i) {
//   for (int j = 0; j < 2; ++j) {
//     ...
//     ... = a[x + i][y + j];
//     ...
//   }
// }
//
// will probably be unrolled to:
//
// gep %a, 0, %x, %y; load
// gep %a, 0, %x, %y + 1; load
// gep %a, 0, %x + 1, %y; load
// gep %a, 0, %x + 1, %y + 1; load
//
// LLVM's GVN does not use partial redundancy elimination yet, and is thus
// unable to reuse (gep %a, 0, %x, %y). As a result, this misoptimization incurs
// significant slowdown in targets with limited addressing modes. For instance,
// because the PTX target does not support the reg+reg addressing mode, the
// NVPTX backend emits PTX code that literally computes the pointer address of
// each GEP, wasting tons of registers. It emits the following PTX for the
// first load and similar PTX for other loads.
//
// mov.u32         %r1, %x;
// mov.u32         %r2, %y;
// mul.wide.u32    %rl2, %r1, 128;
// mov.u64         %rl3, a;
// add.s64         %rl4, %rl3, %rl2;
// mul.wide.u32    %rl5, %r2, 4;
// add.s64         %rl6, %rl4, %rl5;
// ld.global.f32   %f1, [%rl6];
//
// To reduce the register pressure, the optimization implemented in this file
// merges the common part of a group of GEPs, so we can compute each pointer
// address by adding a simple offset to the common part, saving many registers.
//
// It works by splitting each GEP into a variadic base and a constant offset.
// The variadic base can be computed once and reused by multiple GEPs, and the
// constant offsets can be nicely folded into the reg+immediate addressing mode
// (supported by most targets) without using any extra register.
//
// For instance, we transform the four GEPs and four loads in the above example
// into:
//
// base = gep a, 0, x, y
// load base
// laod base + 1  * sizeof(float)
// load base + 32 * sizeof(float)
// load base + 33 * sizeof(float)
//
// Given the transformed IR, a backend that supports the reg+immediate
// addressing mode can easily fold the pointer arithmetics into the loads. For
// example, the NVPTX backend can easily fold the pointer arithmetics into the
// ld.global.f32 instructions, and the resultant PTX uses much fewer registers.
//
// mov.u32         %r1, %tid.x;
// mov.u32         %r2, %tid.y;
// mul.wide.u32    %rl2, %r1, 128;
// mov.u64         %rl3, a;
// add.s64         %rl4, %rl3, %rl2;
// mul.wide.u32    %rl5, %r2, 4;
// add.s64         %rl6, %rl4, %rl5;
// ld.global.f32   %f1, [%rl6]; // so far the same as unoptimized PTX
// ld.global.f32   %f2, [%rl6+4]; // much better
// ld.global.f32   %f3, [%rl6+128]; // much better
// ld.global.f32   %f4, [%rl6+132]; // much better
//
//===----------------------------------------------------------------------===//

#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Operator.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Scalar.h"

using namespace llvm;

static cl::opt<bool> DisableSeparateConstOffsetFromGEP(
    "disable-separate-const-offset-from-gep", cl::init(false),
    cl::desc("Do not separate the constant offset from a GEP instruction"),
    cl::Hidden);

namespace {

/// \brief A helper class for separating a constant offset from a GEP index.
///
/// In real programs, a GEP index may be more complicated than a simple addition
/// of something and a constant integer which can be trivially splitted. For
/// example, to split ((a << 3) | 5) + b, we need to search deeper for the
/// constant offset, so that we can separate the index to (a << 3) + b and 5.
///
/// Therefore, this class looks into the expression that computes a given GEP
/// index, and tries to find a constant integer that can be hoisted to the
/// outermost level of the expression as an addition. Not every constant in an
/// expression can jump out. e.g., we cannot transform (b * (a + 5)) to (b * a +
/// 5); nor can we transform (3 * (a + 5)) to (3 * a + 5), however in this case,
/// -instcombine probably already optimized (3 * (a + 5)) to (3 * a + 15).
class ConstantOffsetExtractor {
 public:
  /// Extracts a constant offset from the given GEP index. It outputs the
  /// numeric value of the extracted constant offset (0 if failed), and a
  /// new index representing the remainder (equal to the original index minus
  /// the constant offset).
  /// \p Idx    The given GEP index
  /// \p NewIdx The new index to replace (output)
  /// \p DL     The datalayout of the module
  /// \p GEP    The given GEP
  static int64_t Extract(Value *Idx, Value *&NewIdx, const DataLayout *DL,
                         GetElementPtrInst *GEP);
  /// Looks for a constant offset without extracting it. The meaning of the
  /// arguments and the return value are the same as Extract.
  static int64_t Find(Value *Idx, const DataLayout *DL, GetElementPtrInst *GEP);

 private:
  ConstantOffsetExtractor(const DataLayout *Layout, Instruction *InsertionPt)
      : DL(Layout), IP(InsertionPt) {}
  /// Searches the expression that computes V for a non-zero constant C s.t.
  /// V can be reassociated into the form V' + C. If the searching is
  /// successful, returns C and update UserChain as a def-use chain from C to V;
  /// otherwise, UserChain is empty.
  ///
  /// \p V            The given expression
  /// \p SignExtended Whether V will be sign-extended in the computation of the
  ///                 GEP index
  /// \p ZeroExtended Whether V will be zero-extended in the computation of the
  ///                 GEP index
  /// \p NonNegative  Whether V is guaranteed to be non-negative. For example,
  ///                 an index of an inbounds GEP is guaranteed to be
  ///                 non-negative. Levaraging this, we can better split
  ///                 inbounds GEPs.
  APInt find(Value *V, bool SignExtended, bool ZeroExtended, bool NonNegative);
  /// A helper function to look into both operands of a binary operator.
  APInt findInEitherOperand(BinaryOperator *BO, bool SignExtended,
                            bool ZeroExtended);
  /// After finding the constant offset C from the GEP index I, we build a new
  /// index I' s.t. I' + C = I. This function builds and returns the new
  /// index I' according to UserChain produced by function "find".
  ///
  /// The building conceptually takes two steps:
  /// 1) iteratively distribute s/zext towards the leaves of the expression tree
  /// that computes I
  /// 2) reassociate the expression tree to the form I' + C.
  ///
  /// For example, to extract the 5 from sext(a + (b + 5)), we first distribute
  /// sext to a, b and 5 so that we have
  ///   sext(a) + (sext(b) + 5).
  /// Then, we reassociate it to
  ///   (sext(a) + sext(b)) + 5.
  /// Given this form, we know I' is sext(a) + sext(b).
  Value *rebuildWithoutConstOffset();
  /// After the first step of rebuilding the GEP index without the constant
  /// offset, distribute s/zext to the operands of all operators in UserChain.
  /// e.g., zext(sext(a + (b + 5)) (assuming no overflow) =>
  /// zext(sext(a)) + (zext(sext(b)) + zext(sext(5))).
  ///
  /// The function also updates UserChain to point to new subexpressions after
  /// distributing s/zext. e.g., the old UserChain of the above example is
  /// 5 -> b + 5 -> a + (b + 5) -> sext(...) -> zext(sext(...)),
  /// and the new UserChain is
  /// zext(sext(5)) -> zext(sext(b)) + zext(sext(5)) ->
  ///   zext(sext(a)) + (zext(sext(b)) + zext(sext(5))
  ///
  /// \p ChainIndex The index to UserChain. ChainIndex is initially
  ///               UserChain.size() - 1, and is decremented during
  ///               the recursion.
  Value *distributeExtsAndCloneChain(unsigned ChainIndex);
  /// Reassociates the GEP index to the form I' + C and returns I'.
  Value *removeConstOffset(unsigned ChainIndex);
  /// A helper function to apply ExtInsts, a list of s/zext, to value V.
  /// e.g., if ExtInsts = [sext i32 to i64, zext i16 to i32], this function
  /// returns "sext i32 (zext i16 V to i32) to i64".
  Value *applyExts(Value *V);

  /// Returns true if LHS and RHS have no bits in common, i.e., LHS | RHS == 0.
  bool NoCommonBits(Value *LHS, Value *RHS) const;
  /// Computes which bits are known to be one or zero.
  /// \p KnownOne Mask of all bits that are known to be one.
  /// \p KnownZero Mask of all bits that are known to be zero.
  void ComputeKnownBits(Value *V, APInt &KnownOne, APInt &KnownZero) const;
  /// A helper function that returns whether we can trace into the operands
  /// of binary operator BO for a constant offset.
  ///
  /// \p SignExtended Whether BO is surrounded by sext
  /// \p ZeroExtended Whether BO is surrounded by zext
  /// \p NonNegative Whether BO is known to be non-negative, e.g., an in-bound
  ///                array index.
  bool CanTraceInto(bool SignExtended, bool ZeroExtended, BinaryOperator *BO,
                    bool NonNegative);

  /// The path from the constant offset to the old GEP index. e.g., if the GEP
  /// index is "a * b + (c + 5)". After running function find, UserChain[0] will
  /// be the constant 5, UserChain[1] will be the subexpression "c + 5", and
  /// UserChain[2] will be the entire expression "a * b + (c + 5)".
  ///
  /// This path helps to rebuild the new GEP index.
  SmallVector<User *, 8> UserChain;
  /// A data structure used in rebuildWithoutConstOffset. Contains all
  /// sext/zext instructions along UserChain.
  SmallVector<CastInst *, 16> ExtInsts;
  /// The data layout of the module. Used in ComputeKnownBits.
  const DataLayout *DL;
  Instruction *IP;  /// Insertion position of cloned instructions.
};

/// \brief A pass that tries to split every GEP in the function into a variadic
/// base and a constant offset. It is a FunctionPass because searching for the
/// constant offset may inspect other basic blocks.
class SeparateConstOffsetFromGEP : public FunctionPass {
 public:
  static char ID;
  SeparateConstOffsetFromGEP() : FunctionPass(ID) {
    initializeSeparateConstOffsetFromGEPPass(*PassRegistry::getPassRegistry());
  }

  void getAnalysisUsage(AnalysisUsage &AU) const override {
    AU.addRequired<DataLayoutPass>();
    AU.addRequired<TargetTransformInfo>();
  }
  bool runOnFunction(Function &F) override;

 private:
  /// Tries to split the given GEP into a variadic base and a constant offset,
  /// and returns true if the splitting succeeds.
  bool splitGEP(GetElementPtrInst *GEP);
  /// Finds the constant offset within each index, and accumulates them. This
  /// function only inspects the GEP without changing it. The output
  /// NeedsExtraction indicates whether we can extract a non-zero constant
  /// offset from any index.
  int64_t accumulateByteOffset(GetElementPtrInst *GEP, const DataLayout *DL,
                               bool &NeedsExtraction);
};
}  // anonymous namespace

char SeparateConstOffsetFromGEP::ID = 0;
INITIALIZE_PASS_BEGIN(
    SeparateConstOffsetFromGEP, "separate-const-offset-from-gep",
    "Split GEPs to a variadic base and a constant offset for better CSE", false,
    false)
INITIALIZE_AG_DEPENDENCY(TargetTransformInfo)
INITIALIZE_PASS_DEPENDENCY(DataLayoutPass)
INITIALIZE_PASS_END(
    SeparateConstOffsetFromGEP, "separate-const-offset-from-gep",
    "Split GEPs to a variadic base and a constant offset for better CSE", false,
    false)

FunctionPass *llvm::createSeparateConstOffsetFromGEPPass() {
  return new SeparateConstOffsetFromGEP();
}

bool ConstantOffsetExtractor::CanTraceInto(bool SignExtended,
                                            bool ZeroExtended,
                                            BinaryOperator *BO,
                                            bool NonNegative) {
  // We only consider ADD, SUB and OR, because a non-zero constant found in
  // expressions composed of these operations can be easily hoisted as a
  // constant offset by reassociation.
  if (BO->getOpcode() != Instruction::Add &&
      BO->getOpcode() != Instruction::Sub &&
      BO->getOpcode() != Instruction::Or) {
    return false;
  }

  Value *LHS = BO->getOperand(0), *RHS = BO->getOperand(1);
  // Do not trace into "or" unless it is equivalent to "add". If LHS and RHS
  // don't have common bits, (LHS | RHS) is equivalent to (LHS + RHS).
  if (BO->getOpcode() == Instruction::Or && !NoCommonBits(LHS, RHS))
    return false;

  // In addition, tracing into BO requires that its surrounding s/zext (if
  // any) is distributable to both operands.
  //
  // Suppose BO = A op B.
  //  SignExtended | ZeroExtended | Distributable?
  // --------------+--------------+----------------------------------
  //       0       |      0       | true because no s/zext exists
  //       0       |      1       | zext(BO) == zext(A) op zext(B)
  //       1       |      0       | sext(BO) == sext(A) op sext(B)
  //       1       |      1       | zext(sext(BO)) ==
  //               |              |     zext(sext(A)) op zext(sext(B))
  if (BO->getOpcode() == Instruction::Add && NonNegative) {
    // If a + b >= 0 and (a >= 0 or b >= 0), then
    //   s/zext(a + b) = s/zext(a) + s/zext(b)
    // even if the addition is not marked nsw.
    //
    // Leveraging this invarient, we can trace into an sext'ed inbound GEP
    // index if the constant offset is non-negative.
    //
    // Verified in @sext_add in split-gep.ll.
    if (ConstantInt *ConstLHS = dyn_cast<ConstantInt>(LHS)) {
      if (!ConstLHS->isNegative())
        return true;
    }
    if (ConstantInt *ConstRHS = dyn_cast<ConstantInt>(RHS)) {
      if (!ConstRHS->isNegative())
        return true;
    }
  }

  // sext (add/sub nsw A, B) == add/sub nsw (sext A), (sext B)
  // zext (add/sub nuw A, B) == add/sub nuw (zext A), (zext B)
  if (BO->getOpcode() == Instruction::Add ||
      BO->getOpcode() == Instruction::Sub) {
    if (SignExtended && !BO->hasNoSignedWrap())
      return false;
    if (ZeroExtended && !BO->hasNoUnsignedWrap())
      return false;
  }

  return true;
}

APInt ConstantOffsetExtractor::findInEitherOperand(BinaryOperator *BO,
                                                   bool SignExtended,
                                                   bool ZeroExtended) {
  // BO being non-negative does not shed light on whether its operands are
  // non-negative. Clear the NonNegative flag here.
  APInt ConstantOffset = find(BO->getOperand(0), SignExtended, ZeroExtended,
                              /* NonNegative */ false);
  // If we found a constant offset in the left operand, stop and return that.
  // This shortcut might cause us to miss opportunities of combining the
  // constant offsets in both operands, e.g., (a + 4) + (b + 5) => (a + b) + 9.
  // However, such cases are probably already handled by -instcombine,
  // given this pass runs after the standard optimizations.
  if (ConstantOffset != 0) return ConstantOffset;
  ConstantOffset = find(BO->getOperand(1), SignExtended, ZeroExtended,
                        /* NonNegative */ false);
  // If U is a sub operator, negate the constant offset found in the right
  // operand.
  if (BO->getOpcode() == Instruction::Sub)
    ConstantOffset = -ConstantOffset;
  return ConstantOffset;
}

APInt ConstantOffsetExtractor::find(Value *V, bool SignExtended,
                                    bool ZeroExtended, bool NonNegative) {
  // TODO(jingyue): We could trace into integer/pointer casts, such as
  // inttoptr, ptrtoint, bitcast, and addrspacecast. We choose to handle only
  // integers because it gives good enough results for our benchmarks.
  unsigned BitWidth = cast<IntegerType>(V->getType())->getBitWidth();

  // We cannot do much with Values that are not a User, such as an Argument.
  User *U = dyn_cast<User>(V);
  if (U == nullptr) return APInt(BitWidth, 0);

  APInt ConstantOffset(BitWidth, 0);
  if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
    // Hooray, we found it!
    ConstantOffset = CI->getValue();
  } else if (BinaryOperator *BO = dyn_cast<BinaryOperator>(V)) {
    // Trace into subexpressions for more hoisting opportunities.
    if (CanTraceInto(SignExtended, ZeroExtended, BO, NonNegative)) {
      ConstantOffset = findInEitherOperand(BO, SignExtended, ZeroExtended);
    }
  } else if (isa<SExtInst>(V)) {
    ConstantOffset = find(U->getOperand(0), /* SignExtended */ true,
                          ZeroExtended, NonNegative).sext(BitWidth);
  } else if (isa<ZExtInst>(V)) {
    // As an optimization, we can clear the SignExtended flag because
    // sext(zext(a)) = zext(a). Verified in @sext_zext in split-gep.ll.
    //
    // Clear the NonNegative flag, because zext(a) >= 0 does not imply a >= 0.
    // TODO: if zext(a) < 2 ^ (bitwidth(a) - 1), we can prove a >= 0.
    ConstantOffset =
        find(U->getOperand(0), /* SignExtended */ false,
             /* ZeroExtended */ true, /* NonNegative */ false).zext(BitWidth);
  }

  // If we found a non-zero constant offset, add it to the path for
  // rebuildWithoutConstOffset. Zero is a valid constant offset, but doesn't
  // help this optimization.
  if (ConstantOffset != 0)
    UserChain.push_back(U);
  return ConstantOffset;
}

Value *ConstantOffsetExtractor::applyExts(Value *V) {
  Value *Current = V;
  // ExtInsts is built in the use-def order. Therefore, we apply them to V
  // in the reversed order.
  for (auto I = ExtInsts.rbegin(), E = ExtInsts.rend(); I != E; ++I) {
    if (Constant *C = dyn_cast<Constant>(Current)) {
      // If Current is a constant, apply s/zext using ConstantExpr::getCast.
      // ConstantExpr::getCast emits a ConstantInt if C is a ConstantInt.
      Current = ConstantExpr::getCast((*I)->getOpcode(), C, (*I)->getType());
    } else {
      Instruction *Ext = (*I)->clone();
      Ext->setOperand(0, Current);
      Ext->insertBefore(IP);
      Current = Ext;
    }
  }
  return Current;
}

Value *ConstantOffsetExtractor::rebuildWithoutConstOffset() {
  distributeExtsAndCloneChain(UserChain.size() - 1);
  // Remove all nullptrs (used to be s/zext) from UserChain.
  unsigned NewSize = 0;
  for (auto I = UserChain.begin(), E = UserChain.end(); I != E; ++I) {
    if (*I != nullptr) {
      UserChain[NewSize] = *I;
      NewSize++;
    }
  }
  UserChain.resize(NewSize);
  return removeConstOffset(UserChain.size() - 1);
}

Value *
ConstantOffsetExtractor::distributeExtsAndCloneChain(unsigned ChainIndex) {
  User *U = UserChain[ChainIndex];
  if (ChainIndex == 0) {
    assert(isa<ConstantInt>(U));
    // If U is a ConstantInt, applyExts will return a ConstantInt as well.
    return UserChain[ChainIndex] = cast<ConstantInt>(applyExts(U));
  }

  if (CastInst *Cast = dyn_cast<CastInst>(U)) {
    assert((isa<SExtInst>(Cast) || isa<ZExtInst>(Cast)) &&
           "We only traced into two types of CastInst: sext and zext");
    ExtInsts.push_back(Cast);
    UserChain[ChainIndex] = nullptr;
    return distributeExtsAndCloneChain(ChainIndex - 1);
  }

  // Function find only trace into BinaryOperator and CastInst.
  BinaryOperator *BO = cast<BinaryOperator>(U);
  // OpNo = which operand of BO is UserChain[ChainIndex - 1]
  unsigned OpNo = (BO->getOperand(0) == UserChain[ChainIndex - 1] ? 0 : 1);
  Value *TheOther = applyExts(BO->getOperand(1 - OpNo));
  Value *NextInChain = distributeExtsAndCloneChain(ChainIndex - 1);

  BinaryOperator *NewBO = nullptr;
  if (OpNo == 0) {
    NewBO = BinaryOperator::Create(BO->getOpcode(), NextInChain, TheOther,
                                   BO->getName(), IP);
  } else {
    NewBO = BinaryOperator::Create(BO->getOpcode(), TheOther, NextInChain,
                                   BO->getName(), IP);
  }
  return UserChain[ChainIndex] = NewBO;
}

Value *ConstantOffsetExtractor::removeConstOffset(unsigned ChainIndex) {
  if (ChainIndex == 0) {
    assert(isa<ConstantInt>(UserChain[ChainIndex]));
    return ConstantInt::getNullValue(UserChain[ChainIndex]->getType());
  }

  BinaryOperator *BO = cast<BinaryOperator>(UserChain[ChainIndex]);
  unsigned OpNo = (BO->getOperand(0) == UserChain[ChainIndex - 1] ? 0 : 1);
  assert(BO->getOperand(OpNo) == UserChain[ChainIndex - 1]);
  Value *NextInChain = removeConstOffset(ChainIndex - 1);
  Value *TheOther = BO->getOperand(1 - OpNo);

  // If NextInChain is 0 and not the LHS of a sub, we can simplify the
  // sub-expression to be just TheOther.
  if (ConstantInt *CI = dyn_cast<ConstantInt>(NextInChain)) {
    if (CI->isZero() && !(BO->getOpcode() == Instruction::Sub && OpNo == 0))
      return TheOther;
  }

  if (BO->getOpcode() == Instruction::Or) {
    // Rebuild "or" as "add", because "or" may be invalid for the new
    // epxression.
    //
    // For instance, given
    //   a | (b + 5) where a and b + 5 have no common bits,
    // we can extract 5 as the constant offset.
    //
    // However, reusing the "or" in the new index would give us
    //   (a | b) + 5
    // which does not equal a | (b + 5).
    //
    // Replacing the "or" with "add" is fine, because
    //   a | (b + 5) = a + (b + 5) = (a + b) + 5
    return BinaryOperator::CreateAdd(BO->getOperand(0), BO->getOperand(1),
                                     BO->getName(), IP);
  }

  // We can reuse BO in this case, because the new expression shares the same
  // instruction type and BO is used at most once.
  assert(BO->getNumUses() <= 1 &&
         "distributeExtsAndCloneChain clones each BinaryOperator in "
         "UserChain, so no one should be used more than "
         "once");
  BO->setOperand(OpNo, NextInChain);
  BO->setHasNoSignedWrap(false);
  BO->setHasNoUnsignedWrap(false);
  // Make sure it appears after all instructions we've inserted so far.
  BO->moveBefore(IP);
  return BO;
}

int64_t ConstantOffsetExtractor::Extract(Value *Idx, Value *&NewIdx,
                                         const DataLayout *DL,
                                         GetElementPtrInst *GEP) {
  ConstantOffsetExtractor Extractor(DL, GEP);
  // Find a non-zero constant offset first.
  APInt ConstantOffset =
      Extractor.find(Idx, /* SignExtended */ false, /* ZeroExtended */ false,
                     GEP->isInBounds());
  if (ConstantOffset != 0) {
    // Separates the constant offset from the GEP index.
    NewIdx = Extractor.rebuildWithoutConstOffset();
  }
  return ConstantOffset.getSExtValue();
}

int64_t ConstantOffsetExtractor::Find(Value *Idx, const DataLayout *DL,
      GetElementPtrInst *GEP) {
  // If Idx is an index of an inbound GEP, Idx is guaranteed to be non-negative.
  return ConstantOffsetExtractor(DL, GEP)
      .find(Idx, /* SignExtended */ false, /* ZeroExtended */ false,
            GEP->isInBounds())
      .getSExtValue();
}

void ConstantOffsetExtractor::ComputeKnownBits(Value *V, APInt &KnownOne,
                                               APInt &KnownZero) const {
  IntegerType *IT = cast<IntegerType>(V->getType());
  KnownOne = APInt(IT->getBitWidth(), 0);
  KnownZero = APInt(IT->getBitWidth(), 0);
  llvm::computeKnownBits(V, KnownZero, KnownOne, DL, 0);
}

bool ConstantOffsetExtractor::NoCommonBits(Value *LHS, Value *RHS) const {
  assert(LHS->getType() == RHS->getType() &&
         "LHS and RHS should have the same type");
  APInt LHSKnownOne, LHSKnownZero, RHSKnownOne, RHSKnownZero;
  ComputeKnownBits(LHS, LHSKnownOne, LHSKnownZero);
  ComputeKnownBits(RHS, RHSKnownOne, RHSKnownZero);
  return (LHSKnownZero | RHSKnownZero).isAllOnesValue();
}

int64_t SeparateConstOffsetFromGEP::accumulateByteOffset(
    GetElementPtrInst *GEP, const DataLayout *DL, bool &NeedsExtraction) {
  NeedsExtraction = false;
  int64_t AccumulativeByteOffset = 0;
  gep_type_iterator GTI = gep_type_begin(*GEP);
  for (unsigned I = 1, E = GEP->getNumOperands(); I != E; ++I, ++GTI) {
    if (isa<SequentialType>(*GTI)) {
      // Tries to extract a constant offset from this GEP index.
      int64_t ConstantOffset =
          ConstantOffsetExtractor::Find(GEP->getOperand(I), DL, GEP);
      if (ConstantOffset != 0) {
        NeedsExtraction = true;
        // A GEP may have multiple indices.  We accumulate the extracted
        // constant offset to a byte offset, and later offset the remainder of
        // the original GEP with this byte offset.
        AccumulativeByteOffset +=
            ConstantOffset * DL->getTypeAllocSize(GTI.getIndexedType());
      }
    }
  }
  return AccumulativeByteOffset;
}

bool SeparateConstOffsetFromGEP::splitGEP(GetElementPtrInst *GEP) {
  // Skip vector GEPs.
  if (GEP->getType()->isVectorTy())
    return false;

  // The backend can already nicely handle the case where all indices are
  // constant.
  if (GEP->hasAllConstantIndices())
    return false;

  bool Changed = false;
  // Canonicalize array indices to pointer-size integers. This helps to simplify
  // the logic of splitting a GEP. For example, if a + b is a pointer-size
  // integer, we have
  //   gep base, a + b = gep (gep base, a), b
  // However, this equality may not hold if the size of a + b is smaller than
  // the pointer size, because LLVM conceptually sign-extends GEP indices to
  // pointer size before computing the address
  // (http://llvm.org/docs/LangRef.html#id181).
  //
  // This canonicalization is very likely already done in clang and instcombine.
  // Therefore, the program will probably remain the same.
  //
  // Verified in @i32_add in split-gep.ll
  const DataLayout *DL = &getAnalysis<DataLayoutPass>().getDataLayout();
  Type *IntPtrTy = DL->getIntPtrType(GEP->getType());
  gep_type_iterator GTI = gep_type_begin(*GEP);
  for (User::op_iterator I = GEP->op_begin() + 1, E = GEP->op_end();
       I != E; ++I, ++GTI) {
    if (isa<SequentialType>(*GTI)) {
      if ((*I)->getType() != IntPtrTy) {
        *I = CastInst::CreateIntegerCast(*I, IntPtrTy, true, "idxprom", GEP);
        Changed = true;
      }
    }
  }

  bool NeedsExtraction;
  int64_t AccumulativeByteOffset =
      accumulateByteOffset(GEP, DL, NeedsExtraction);

  if (!NeedsExtraction)
    return Changed;
  // Before really splitting the GEP, check whether the backend supports the
  // addressing mode we are about to produce. If no, this splitting probably
  // won't be beneficial.
  TargetTransformInfo &TTI = getAnalysis<TargetTransformInfo>();
  if (!TTI.isLegalAddressingMode(GEP->getType()->getElementType(),
                                 /*BaseGV=*/nullptr, AccumulativeByteOffset,
                                 /*HasBaseReg=*/true, /*Scale=*/0)) {
    return Changed;
  }

  // Remove the constant offset in each GEP index. The resultant GEP computes
  // the variadic base.
  GTI = gep_type_begin(*GEP);
  for (unsigned I = 1, E = GEP->getNumOperands(); I != E; ++I, ++GTI) {
    if (isa<SequentialType>(*GTI)) {
      Value *NewIdx = nullptr;
      // Tries to extract a constant offset from this GEP index.
      int64_t ConstantOffset =
          ConstantOffsetExtractor::Extract(GEP->getOperand(I), NewIdx, DL, GEP);
      if (ConstantOffset != 0) {
        assert(NewIdx != nullptr &&
               "ConstantOffset != 0 implies NewIdx is set");
        GEP->setOperand(I, NewIdx);
      }
    }
  }
  // Clear the inbounds attribute because the new index may be off-bound.
  // e.g.,
  //
  // b = add i64 a, 5
  // addr = gep inbounds float* p, i64 b
  //
  // is transformed to:
  //
  // addr2 = gep float* p, i64 a
  // addr = gep float* addr2, i64 5
  //
  // If a is -4, although the old index b is in bounds, the new index a is
  // off-bound. http://llvm.org/docs/LangRef.html#id181 says "if the
  // inbounds keyword is not present, the offsets are added to the base
  // address with silently-wrapping two's complement arithmetic".
  // Therefore, the final code will be a semantically equivalent.
  //
  // TODO(jingyue): do some range analysis to keep as many inbounds as
  // possible. GEPs with inbounds are more friendly to alias analysis.
  GEP->setIsInBounds(false);

  // Offsets the base with the accumulative byte offset.
  //
  //   %gep                        ; the base
  //   ... %gep ...
  //
  // => add the offset
  //
  //   %gep2                       ; clone of %gep
  //   %new.gep = gep %gep2, <offset / sizeof(*%gep)>
  //   %gep                        ; will be removed
  //   ... %gep ...
  //
  // => replace all uses of %gep with %new.gep and remove %gep
  //
  //   %gep2                       ; clone of %gep
  //   %new.gep = gep %gep2, <offset / sizeof(*%gep)>
  //   ... %new.gep ...
  //
  // If AccumulativeByteOffset is not a multiple of sizeof(*%gep), we emit an
  // uglygep (http://llvm.org/docs/GetElementPtr.html#what-s-an-uglygep):
  // bitcast %gep2 to i8*, add the offset, and bitcast the result back to the
  // type of %gep.
  //
  //   %gep2                       ; clone of %gep
  //   %0       = bitcast %gep2 to i8*
  //   %uglygep = gep %0, <offset>
  //   %new.gep = bitcast %uglygep to <type of %gep>
  //   ... %new.gep ...
  Instruction *NewGEP = GEP->clone();
  NewGEP->insertBefore(GEP);

  uint64_t ElementTypeSizeOfGEP =
      DL->getTypeAllocSize(GEP->getType()->getElementType());
  if (AccumulativeByteOffset % ElementTypeSizeOfGEP == 0) {
    // Very likely. As long as %gep is natually aligned, the byte offset we
    // extracted should be a multiple of sizeof(*%gep).
    // Per ANSI C standard, signed / unsigned = unsigned. Therefore, we
    // cast ElementTypeSizeOfGEP to signed.
    int64_t Index =
        AccumulativeByteOffset / static_cast<int64_t>(ElementTypeSizeOfGEP);
    NewGEP = GetElementPtrInst::Create(
        NewGEP, ConstantInt::get(IntPtrTy, Index, true), GEP->getName(), GEP);
  } else {
    // Unlikely but possible. For example,
    // #pragma pack(1)
    // struct S {
    //   int a[3];
    //   int64 b[8];
    // };
    // #pragma pack()
    //
    // Suppose the gep before extraction is &s[i + 1].b[j + 3]. After
    // extraction, it becomes &s[i].b[j] and AccumulativeByteOffset is
    // sizeof(S) + 3 * sizeof(int64) = 100, which is not a multiple of
    // sizeof(int64).
    //
    // Emit an uglygep in this case.
    Type *I8PtrTy = Type::getInt8PtrTy(GEP->getContext(),
                                       GEP->getPointerAddressSpace());
    NewGEP = new BitCastInst(NewGEP, I8PtrTy, "", GEP);
    NewGEP = GetElementPtrInst::Create(
        NewGEP, ConstantInt::get(IntPtrTy, AccumulativeByteOffset, true),
        "uglygep", GEP);
    if (GEP->getType() != I8PtrTy)
      NewGEP = new BitCastInst(NewGEP, GEP->getType(), GEP->getName(), GEP);
  }

  GEP->replaceAllUsesWith(NewGEP);
  GEP->eraseFromParent();

  return true;
}

bool SeparateConstOffsetFromGEP::runOnFunction(Function &F) {
  if (DisableSeparateConstOffsetFromGEP)
    return false;

  bool Changed = false;
  for (Function::iterator B = F.begin(), BE = F.end(); B != BE; ++B) {
    for (BasicBlock::iterator I = B->begin(), IE = B->end(); I != IE; ) {
      if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I++)) {
        Changed |= splitGEP(GEP);
      }
      // No need to split GEP ConstantExprs because all its indices are constant
      // already.
    }
  }
  return Changed;
}