SimplifyLibCalls.cpp

//===- SimplifyLibCalls.cpp - Optimize specific well-known library calls --===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements a simple pass that applies a variety of small
// optimizations for calls to specific well-known function calls (e.g. runtime
// library functions).   Any optimization that takes the very simple form
// "replace call to library function with simpler code that provides the same
// result" belongs in this file.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "simplify-libcalls"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Transforms/Utils/BuildLibCalls.h"
#include "llvm/Intrinsics.h"
#include "llvm/LLVMContext.h"
#include "llvm/Module.h"
#include "llvm/Pass.h"
#include "llvm/Support/IRBuilder.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Config/config.h"            // FIXME: Shouldn't depend on host!
using namespace llvm;

STATISTIC(NumSimplified, "Number of library calls simplified");
STATISTIC(NumAnnotated, "Number of attributes added to library functions");

//===----------------------------------------------------------------------===//
// Optimizer Base Class
//===----------------------------------------------------------------------===//

/// This class is the abstract base class for the set of optimizations that
/// corresponds to one library call.
namespace {
class LibCallOptimization {
protected:
  Function *Caller;
  const TargetData *TD;
  const TargetLibraryInfo *TLI;
  LLVMContext* Context;
public:
  LibCallOptimization() { }
  virtual ~LibCallOptimization() {}

  /// CallOptimizer - This pure virtual method is implemented by base classes to
  /// do various optimizations.  If this returns null then no transformation was
  /// performed.  If it returns CI, then it transformed the call and CI is to be
  /// deleted.  If it returns something else, replace CI with the new value and
  /// delete CI.
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B)
    =0;

  Value *OptimizeCall(CallInst *CI, const TargetData *TD,
                      const TargetLibraryInfo *TLI, IRBuilder<> &B) {
    Caller = CI->getParent()->getParent();
    this->TD = TD;
    this->TLI = TLI;
    if (CI->getCalledFunction())
      Context = &CI->getCalledFunction()->getContext();

    // We never change the calling convention.
    if (CI->getCallingConv() != llvm::CallingConv::C)
      return NULL;

    return CallOptimizer(CI->getCalledFunction(), CI, B);
  }
};
} // End anonymous namespace.


//===----------------------------------------------------------------------===//
// Helper Functions
//===----------------------------------------------------------------------===//

/// IsOnlyUsedInZeroEqualityComparison - Return true if it only matters that the
/// value is equal or not-equal to zero.
static bool IsOnlyUsedInZeroEqualityComparison(Value *V) {
  for (Value::use_iterator UI = V->use_begin(), E = V->use_end();
       UI != E; ++UI) {
    if (ICmpInst *IC = dyn_cast<ICmpInst>(*UI))
      if (IC->isEquality())
        if (Constant *C = dyn_cast<Constant>(IC->getOperand(1)))
          if (C->isNullValue())
            continue;
    // Unknown instruction.
    return false;
  }
  return true;
}
 
static bool CallHasFloatingPointArgument(const CallInst *CI) {
  for (CallInst::const_op_iterator it = CI->op_begin(), e = CI->op_end();
       it != e; ++it) {
    if ((*it)->getType()->isFloatingPointTy())
      return true;
  }
  return false;
}

/// IsOnlyUsedInEqualityComparison - Return true if it is only used in equality
/// comparisons with With.
static bool IsOnlyUsedInEqualityComparison(Value *V, Value *With) {
  for (Value::use_iterator UI = V->use_begin(), E = V->use_end();
       UI != E; ++UI) {
    if (ICmpInst *IC = dyn_cast<ICmpInst>(*UI))
      if (IC->isEquality() && IC->getOperand(1) == With)
        continue;
    // Unknown instruction.
    return false;
  }
  return true;
}

//===----------------------------------------------------------------------===//
// String and Memory LibCall Optimizations
//===----------------------------------------------------------------------===//

//===---------------------------------------===//
// 'strcat' Optimizations
namespace {
struct StrCatOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strcat" function prototype.
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        FT->getReturnType() != B.getInt8PtrTy() ||
        FT->getParamType(0) != FT->getReturnType() ||
        FT->getParamType(1) != FT->getReturnType())
      return 0;

    // Extract some information from the instruction
    Value *Dst = CI->getArgOperand(0);
    Value *Src = CI->getArgOperand(1);

    // See if we can get the length of the input string.
    uint64_t Len = GetStringLength(Src);
    if (Len == 0) return 0;
    --Len;  // Unbias length.

    // Handle the simple, do-nothing case: strcat(x, "") -> x
    if (Len == 0)
      return Dst;

    // These optimizations require TargetData.
    if (!TD) return 0;

    EmitStrLenMemCpy(Src, Dst, Len, B);
    return Dst;
  }

  void EmitStrLenMemCpy(Value *Src, Value *Dst, uint64_t Len, IRBuilder<> &B) {
    // We need to find the end of the destination string.  That's where the
    // memory is to be moved to. We just generate a call to strlen.
    Value *DstLen = EmitStrLen(Dst, B, TD);

    // Now that we have the destination's length, we must index into the
    // destination's pointer to get the actual memcpy destination (end of
    // the string .. we're concatenating).
    Value *CpyDst = B.CreateGEP(Dst, DstLen, "endptr");

    // We have enough information to now generate the memcpy call to do the
    // concatenation for us.  Make a memcpy to copy the nul byte with align = 1.
    B.CreateMemCpy(CpyDst, Src,
                   ConstantInt::get(TD->getIntPtrType(*Context), Len + 1), 1);
  }
};

//===---------------------------------------===//
// 'strncat' Optimizations

struct StrNCatOpt : public StrCatOpt {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strncat" function prototype.
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 ||
        FT->getReturnType() != B.getInt8PtrTy() ||
        FT->getParamType(0) != FT->getReturnType() ||
        FT->getParamType(1) != FT->getReturnType() ||
        !FT->getParamType(2)->isIntegerTy())
      return 0;

    // Extract some information from the instruction
    Value *Dst = CI->getArgOperand(0);
    Value *Src = CI->getArgOperand(1);
    uint64_t Len;

    // We don't do anything if length is not constant
    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getArgOperand(2)))
      Len = LengthArg->getZExtValue();
    else
      return 0;

    // See if we can get the length of the input string.
    uint64_t SrcLen = GetStringLength(Src);
    if (SrcLen == 0) return 0;
    --SrcLen;  // Unbias length.

    // Handle the simple, do-nothing cases:
    // strncat(x, "", c) -> x
    // strncat(x,  c, 0) -> x
    if (SrcLen == 0 || Len == 0) return Dst;

    // These optimizations require TargetData.
    if (!TD) return 0;

    // We don't optimize this case
    if (Len < SrcLen) return 0;

    // strncat(x, s, c) -> strcat(x, s)
    // s is constant so the strcat can be optimized further
    EmitStrLenMemCpy(Src, Dst, SrcLen, B);
    return Dst;
  }
};

//===---------------------------------------===//
// 'strchr' Optimizations

struct StrChrOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strchr" function prototype.
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        FT->getReturnType() != B.getInt8PtrTy() ||
        FT->getParamType(0) != FT->getReturnType() ||
        !FT->getParamType(1)->isIntegerTy(32))
      return 0;

    Value *SrcStr = CI->getArgOperand(0);

    // If the second operand is non-constant, see if we can compute the length
    // of the input string and turn this into memchr.
    ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));
    if (CharC == 0) {
      // These optimizations require TargetData.
      if (!TD) return 0;

      uint64_t Len = GetStringLength(SrcStr);
      if (Len == 0 || !FT->getParamType(1)->isIntegerTy(32))// memchr needs i32.
        return 0;

      return EmitMemChr(SrcStr, CI->getArgOperand(1), // include nul.
                        ConstantInt::get(TD->getIntPtrType(*Context), Len),
                        B, TD);
    }

    // Otherwise, the character is a constant, see if the first argument is
    // a string literal.  If so, we can constant fold.
    std::string Str;
    if (!GetConstantStringInfo(SrcStr, Str))
      return 0;

    // strchr can find the nul character.
    Str += '\0';

    // Compute the offset.
    size_t I = Str.find(CharC->getSExtValue());
    if (I == std::string::npos) // Didn't find the char.  strchr returns null.
      return Constant::getNullValue(CI->getType());

    // strchr(s+n,c)  -> gep(s+n+i,c)
    return B.CreateGEP(SrcStr, B.getInt64(I), "strchr");
  }
};

//===---------------------------------------===//
// 'strrchr' Optimizations

struct StrRChrOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strrchr" function prototype.
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        FT->getReturnType() != B.getInt8PtrTy() ||
        FT->getParamType(0) != FT->getReturnType() ||
        !FT->getParamType(1)->isIntegerTy(32))
      return 0;

    Value *SrcStr = CI->getArgOperand(0);
    ConstantInt *CharC = dyn_cast<ConstantInt>(CI->getArgOperand(1));

    // Cannot fold anything if we're not looking for a constant.
    if (!CharC)
      return 0;

    std::string Str;
    if (!GetConstantStringInfo(SrcStr, Str)) {
      // strrchr(s, 0) -> strchr(s, 0)
      if (TD && CharC->isZero())
        return EmitStrChr(SrcStr, '\0', B, TD);
      return 0;
    }

    // strrchr can find the nul character.
    Str += '\0';

    // Compute the offset.
    size_t I = Str.rfind(CharC->getSExtValue());
    if (I == std::string::npos) // Didn't find the char. Return null.
      return Constant::getNullValue(CI->getType());

    // strrchr(s+n,c) -> gep(s+n+i,c)
    return B.CreateGEP(SrcStr, B.getInt64(I), "strrchr");
  }
};

//===---------------------------------------===//
// 'strcmp' Optimizations

struct StrCmpOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strcmp" function prototype.
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        !FT->getReturnType()->isIntegerTy(32) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        FT->getParamType(0) != B.getInt8PtrTy())
      return 0;

    Value *Str1P = CI->getArgOperand(0), *Str2P = CI->getArgOperand(1);
    if (Str1P == Str2P)      // strcmp(x,x)  -> 0
      return ConstantInt::get(CI->getType(), 0);

    std::string Str1, Str2;
    bool HasStr1 = GetConstantStringInfo(Str1P, Str1);
    bool HasStr2 = GetConstantStringInfo(Str2P, Str2);

    // strcmp(x, y)  -> cnst  (if both x and y are constant strings)
    if (HasStr1 && HasStr2)
      return ConstantInt::get(CI->getType(),
                              StringRef(Str1).compare(Str2));

    if (HasStr1 && Str1.empty()) // strcmp("", x) -> -*x
      return B.CreateNeg(B.CreateZExt(B.CreateLoad(Str2P, "strcmpload"),
                                      CI->getType()));

    if (HasStr2 && Str2.empty()) // strcmp(x,"") -> *x
      return B.CreateZExt(B.CreateLoad(Str1P, "strcmpload"), CI->getType());

    // strcmp(P, "x") -> memcmp(P, "x", 2)
    uint64_t Len1 = GetStringLength(Str1P);
    uint64_t Len2 = GetStringLength(Str2P);
    if (Len1 && Len2) {
      // These optimizations require TargetData.
      if (!TD) return 0;

      return EmitMemCmp(Str1P, Str2P,
                        ConstantInt::get(TD->getIntPtrType(*Context),
                        std::min(Len1, Len2)), B, TD);
    }

    return 0;
  }
};

//===---------------------------------------===//
// 'strncmp' Optimizations

struct StrNCmpOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strncmp" function prototype.
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 ||
        !FT->getReturnType()->isIntegerTy(32) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        FT->getParamType(0) != B.getInt8PtrTy() ||
        !FT->getParamType(2)->isIntegerTy())
      return 0;

    Value *Str1P = CI->getArgOperand(0), *Str2P = CI->getArgOperand(1);
    if (Str1P == Str2P)      // strncmp(x,x,n)  -> 0
      return ConstantInt::get(CI->getType(), 0);

    // Get the length argument if it is constant.
    uint64_t Length;
    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getArgOperand(2)))
      Length = LengthArg->getZExtValue();
    else
      return 0;

    if (Length == 0) // strncmp(x,y,0)   -> 0
      return ConstantInt::get(CI->getType(), 0);

    if (TD && Length == 1) // strncmp(x,y,1) -> memcmp(x,y,1)
      return EmitMemCmp(Str1P, Str2P, CI->getArgOperand(2), B, TD);

    std::string Str1, Str2;
    bool HasStr1 = GetConstantStringInfo(Str1P, Str1);
    bool HasStr2 = GetConstantStringInfo(Str2P, Str2);

    // strncmp(x, y)  -> cnst  (if both x and y are constant strings)
    if (HasStr1 && HasStr2) {
      StringRef SubStr1 = StringRef(Str1).substr(0, Length);
      StringRef SubStr2 = StringRef(Str2).substr(0, Length);
      return ConstantInt::get(CI->getType(), SubStr1.compare(SubStr2));
    }

    if (HasStr1 && Str1.empty())  // strncmp("", x, n) -> -*x
      return B.CreateNeg(B.CreateZExt(B.CreateLoad(Str2P, "strcmpload"),
                                      CI->getType()));

    if (HasStr2 && Str2.empty())  // strncmp(x, "", n) -> *x
      return B.CreateZExt(B.CreateLoad(Str1P, "strcmpload"), CI->getType());

    return 0;
  }
};


//===---------------------------------------===//
// 'strcpy' Optimizations

struct StrCpyOpt : public LibCallOptimization {
  bool OptChkCall;  // True if it's optimizing a __strcpy_chk libcall.

  StrCpyOpt(bool c) : OptChkCall(c) {}

  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strcpy" function prototype.
    unsigned NumParams = OptChkCall ? 3 : 2;
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != NumParams ||
        FT->getReturnType() != FT->getParamType(0) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        FT->getParamType(0) != B.getInt8PtrTy())
      return 0;

    Value *Dst = CI->getArgOperand(0), *Src = CI->getArgOperand(1);
    if (Dst == Src)      // strcpy(x,x)  -> x
      return Src;

    // These optimizations require TargetData.
    if (!TD) return 0;

    // See if we can get the length of the input string.
    uint64_t Len = GetStringLength(Src);
    if (Len == 0) return 0;

    // We have enough information to now generate the memcpy call to do the
    // concatenation for us.  Make a memcpy to copy the nul byte with align = 1.
    if (OptChkCall)
      EmitMemCpyChk(Dst, Src,
                    ConstantInt::get(TD->getIntPtrType(*Context), Len),
                    CI->getArgOperand(2), B, TD);
    else
      B.CreateMemCpy(Dst, Src,
                     ConstantInt::get(TD->getIntPtrType(*Context), Len), 1);
    return Dst;
  }
};

//===---------------------------------------===//
// 'strncpy' Optimizations

struct StrNCpyOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        FT->getParamType(0) != B.getInt8PtrTy() ||
        !FT->getParamType(2)->isIntegerTy())
      return 0;

    Value *Dst = CI->getArgOperand(0);
    Value *Src = CI->getArgOperand(1);
    Value *LenOp = CI->getArgOperand(2);

    // See if we can get the length of the input string.
    uint64_t SrcLen = GetStringLength(Src);
    if (SrcLen == 0) return 0;
    --SrcLen;

    if (SrcLen == 0) {
      // strncpy(x, "", y) -> memset(x, '\0', y, 1)
      B.CreateMemSet(Dst, B.getInt8('\0'), LenOp, 1);
      return Dst;
    }

    uint64_t Len;
    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(LenOp))
      Len = LengthArg->getZExtValue();
    else
      return 0;

    if (Len == 0) return Dst; // strncpy(x, y, 0) -> x

    // These optimizations require TargetData.
    if (!TD) return 0;

    // Let strncpy handle the zero padding
    if (Len > SrcLen+1) return 0;

    // strncpy(x, s, c) -> memcpy(x, s, c, 1) [s and c are constant]
    B.CreateMemCpy(Dst, Src,
                   ConstantInt::get(TD->getIntPtrType(*Context), Len), 1);

    return Dst;
  }
};

//===---------------------------------------===//
// 'strlen' Optimizations

struct StrLenOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 1 ||
        FT->getParamType(0) != B.getInt8PtrTy() ||
        !FT->getReturnType()->isIntegerTy())
      return 0;

    Value *Src = CI->getArgOperand(0);

    // Constant folding: strlen("xyz") -> 3
    if (uint64_t Len = GetStringLength(Src))
      return ConstantInt::get(CI->getType(), Len-1);

    // strlen(x) != 0 --> *x != 0
    // strlen(x) == 0 --> *x == 0
    if (IsOnlyUsedInZeroEqualityComparison(CI))
      return B.CreateZExt(B.CreateLoad(Src, "strlenfirst"), CI->getType());
    return 0;
  }
};


//===---------------------------------------===//
// 'strpbrk' Optimizations

struct StrPBrkOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        FT->getParamType(0) != B.getInt8PtrTy() ||
        FT->getParamType(1) != FT->getParamType(0) ||
        FT->getReturnType() != FT->getParamType(0))
      return 0;

    std::string S1, S2;
    bool HasS1 = GetConstantStringInfo(CI->getArgOperand(0), S1);
    bool HasS2 = GetConstantStringInfo(CI->getArgOperand(1), S2);

    // strpbrk(s, "") -> NULL
    // strpbrk("", s) -> NULL
    if ((HasS1 && S1.empty()) || (HasS2 && S2.empty()))
      return Constant::getNullValue(CI->getType());

    // Constant folding.
    if (HasS1 && HasS2) {
      size_t I = S1.find_first_of(S2);
      if (I == std::string::npos) // No match.
        return Constant::getNullValue(CI->getType());

      return B.CreateGEP(CI->getArgOperand(0), B.getInt64(I), "strpbrk");
    }

    // strpbrk(s, "a") -> strchr(s, 'a')
    if (TD && HasS2 && S2.size() == 1)
      return EmitStrChr(CI->getArgOperand(0), S2[0], B, TD);

    return 0;
  }
};

//===---------------------------------------===//
// 'strto*' Optimizations.  This handles strtol, strtod, strtof, strtoul, etc.

struct StrToOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if ((FT->getNumParams() != 2 && FT->getNumParams() != 3) ||
        !FT->getParamType(0)->isPointerTy() ||
        !FT->getParamType(1)->isPointerTy())
      return 0;

    Value *EndPtr = CI->getArgOperand(1);
    if (isa<ConstantPointerNull>(EndPtr)) {
      // With a null EndPtr, this function won't capture the main argument.
      // It would be readonly too, except that it still may write to errno.
      CI->addAttribute(1, Attribute::NoCapture);
    }

    return 0;
  }
};

//===---------------------------------------===//
// 'strspn' Optimizations

struct StrSpnOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        FT->getParamType(0) != B.getInt8PtrTy() ||
        FT->getParamType(1) != FT->getParamType(0) ||
        !FT->getReturnType()->isIntegerTy())
      return 0;

    std::string S1, S2;
    bool HasS1 = GetConstantStringInfo(CI->getArgOperand(0), S1);
    bool HasS2 = GetConstantStringInfo(CI->getArgOperand(1), S2);

    // strspn(s, "") -> 0
    // strspn("", s) -> 0
    if ((HasS1 && S1.empty()) || (HasS2 && S2.empty()))
      return Constant::getNullValue(CI->getType());

    // Constant folding.
    if (HasS1 && HasS2)
      return ConstantInt::get(CI->getType(), strspn(S1.c_str(), S2.c_str()));

    return 0;
  }
};

//===---------------------------------------===//
// 'strcspn' Optimizations

struct StrCSpnOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        FT->getParamType(0) != B.getInt8PtrTy() ||
        FT->getParamType(1) != FT->getParamType(0) ||
        !FT->getReturnType()->isIntegerTy())
      return 0;

    std::string S1, S2;
    bool HasS1 = GetConstantStringInfo(CI->getArgOperand(0), S1);
    bool HasS2 = GetConstantStringInfo(CI->getArgOperand(1), S2);

    // strcspn("", s) -> 0
    if (HasS1 && S1.empty())
      return Constant::getNullValue(CI->getType());

    // Constant folding.
    if (HasS1 && HasS2)
      return ConstantInt::get(CI->getType(), strcspn(S1.c_str(), S2.c_str()));

    // strcspn(s, "") -> strlen(s)
    if (TD && HasS2 && S2.empty())
      return EmitStrLen(CI->getArgOperand(0), B, TD);

    return 0;
  }
};

//===---------------------------------------===//
// 'strstr' Optimizations

struct StrStrOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        !FT->getParamType(0)->isPointerTy() ||
        !FT->getParamType(1)->isPointerTy() ||
        !FT->getReturnType()->isPointerTy())
      return 0;

    // fold strstr(x, x) -> x.
    if (CI->getArgOperand(0) == CI->getArgOperand(1))
      return B.CreateBitCast(CI->getArgOperand(0), CI->getType());

    // fold strstr(a, b) == a -> strncmp(a, b, strlen(b)) == 0
    if (TD && IsOnlyUsedInEqualityComparison(CI, CI->getArgOperand(0))) {
      Value *StrLen = EmitStrLen(CI->getArgOperand(1), B, TD);
      Value *StrNCmp = EmitStrNCmp(CI->getArgOperand(0), CI->getArgOperand(1),
                                   StrLen, B, TD);
      for (Value::use_iterator UI = CI->use_begin(), UE = CI->use_end();
           UI != UE; ) {
        ICmpInst *Old = cast<ICmpInst>(*UI++);
        Value *Cmp = B.CreateICmp(Old->getPredicate(), StrNCmp,
                                  ConstantInt::getNullValue(StrNCmp->getType()),
                                  "cmp");
        Old->replaceAllUsesWith(Cmp);
        Old->eraseFromParent();
      }
      return CI;
    }

    // See if either input string is a constant string.
    std::string SearchStr, ToFindStr;
    bool HasStr1 = GetConstantStringInfo(CI->getArgOperand(0), SearchStr);
    bool HasStr2 = GetConstantStringInfo(CI->getArgOperand(1), ToFindStr);

    // fold strstr(x, "") -> x.
    if (HasStr2 && ToFindStr.empty())
      return B.CreateBitCast(CI->getArgOperand(0), CI->getType());

    // If both strings are known, constant fold it.
    if (HasStr1 && HasStr2) {
      std::string::size_type Offset = SearchStr.find(ToFindStr);

      if (Offset == std::string::npos) // strstr("foo", "bar") -> null
        return Constant::getNullValue(CI->getType());

      // strstr("abcd", "bc") -> gep((char*)"abcd", 1)
      Value *Result = CastToCStr(CI->getArgOperand(0), B);
      Result = B.CreateConstInBoundsGEP1_64(Result, Offset, "strstr");
      return B.CreateBitCast(Result, CI->getType());
    }

    // fold strstr(x, "y") -> strchr(x, 'y').
    if (HasStr2 && ToFindStr.size() == 1)
      return B.CreateBitCast(EmitStrChr(CI->getArgOperand(0),
                             ToFindStr[0], B, TD), CI->getType());
    return 0;
  }
};


//===---------------------------------------===//
// 'memcmp' Optimizations

struct MemCmpOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 || !FT->getParamType(0)->isPointerTy() ||
        !FT->getParamType(1)->isPointerTy() ||
        !FT->getReturnType()->isIntegerTy(32))
      return 0;

    Value *LHS = CI->getArgOperand(0), *RHS = CI->getArgOperand(1);

    if (LHS == RHS)  // memcmp(s,s,x) -> 0
      return Constant::getNullValue(CI->getType());

    // Make sure we have a constant length.
    ConstantInt *LenC = dyn_cast<ConstantInt>(CI->getArgOperand(2));
    if (!LenC) return 0;
    uint64_t Len = LenC->getZExtValue();

    if (Len == 0) // memcmp(s1,s2,0) -> 0
      return Constant::getNullValue(CI->getType());

    // memcmp(S1,S2,1) -> *(unsigned char*)LHS - *(unsigned char*)RHS
    if (Len == 1) {
      Value *LHSV = B.CreateZExt(B.CreateLoad(CastToCStr(LHS, B), "lhsc"),
                                 CI->getType(), "lhsv");
      Value *RHSV = B.CreateZExt(B.CreateLoad(CastToCStr(RHS, B), "rhsc"),
                                 CI->getType(), "rhsv");
      return B.CreateSub(LHSV, RHSV, "chardiff");
    }

    // Constant folding: memcmp(x, y, l) -> cnst (all arguments are constant)
    std::string LHSStr, RHSStr;
    if (GetConstantStringInfo(LHS, LHSStr) &&
        GetConstantStringInfo(RHS, RHSStr)) {
      // Make sure we're not reading out-of-bounds memory.
      if (Len > LHSStr.length() || Len > RHSStr.length())
        return 0;
      uint64_t Ret = memcmp(LHSStr.data(), RHSStr.data(), Len);
      return ConstantInt::get(CI->getType(), Ret);
    }

    return 0;
  }
};

//===---------------------------------------===//
// 'memcpy' Optimizations

struct MemCpyOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // These optimizations require TargetData.
    if (!TD) return 0;

    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
        !FT->getParamType(0)->isPointerTy() ||
        !FT->getParamType(1)->isPointerTy() ||
        FT->getParamType(2) != TD->getIntPtrType(*Context))
      return 0;

    // memcpy(x, y, n) -> llvm.memcpy(x, y, n, 1)
    B.CreateMemCpy(CI->getArgOperand(0), CI->getArgOperand(1),
                   CI->getArgOperand(2), 1);
    return CI->getArgOperand(0);
  }
};

//===---------------------------------------===//
// 'memmove' Optimizations

struct MemMoveOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // These optimizations require TargetData.
    if (!TD) return 0;

    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
        !FT->getParamType(0)->isPointerTy() ||
        !FT->getParamType(1)->isPointerTy() ||
        FT->getParamType(2) != TD->getIntPtrType(*Context))
      return 0;

    // memmove(x, y, n) -> llvm.memmove(x, y, n, 1)
    B.CreateMemMove(CI->getArgOperand(0), CI->getArgOperand(1),
                    CI->getArgOperand(2), 1);
    return CI->getArgOperand(0);
  }
};

//===---------------------------------------===//
// 'memset' Optimizations

struct MemSetOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // These optimizations require TargetData.
    if (!TD) return 0;

    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
        !FT->getParamType(0)->isPointerTy() ||
        !FT->getParamType(1)->isIntegerTy() ||
        FT->getParamType(2) != TD->getIntPtrType(*Context))
      return 0;

    // memset(p, v, n) -> llvm.memset(p, v, n, 1)
    Value *Val = B.CreateIntCast(CI->getArgOperand(1), B.getInt8Ty(), false);
    B.CreateMemSet(CI->getArgOperand(0), Val, CI->getArgOperand(2), 1);
    return CI->getArgOperand(0);
  }
};

//===----------------------------------------------------------------------===//
// Math Library Optimizations
//===----------------------------------------------------------------------===//

//===---------------------------------------===//
// 'pow*' Optimizations

struct PowOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    // Just make sure this has 2 arguments of the same FP type, which match the
    // result type.
    if (FT->getNumParams() != 2 || FT->getReturnType() != FT->getParamType(0) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        !FT->getParamType(0)->isFloatingPointTy())
      return 0;

    Value *Op1 = CI->getArgOperand(0), *Op2 = CI->getArgOperand(1);
    if (ConstantFP *Op1C = dyn_cast<ConstantFP>(Op1)) {
      if (Op1C->isExactlyValue(1.0))  // pow(1.0, x) -> 1.0
        return Op1C;
      if (Op1C->isExactlyValue(2.0))  // pow(2.0, x) -> exp2(x)
        return EmitUnaryFloatFnCall(Op2, "exp2", B, Callee->getAttributes());
    }

    ConstantFP *Op2C = dyn_cast<ConstantFP>(Op2);
    if (Op2C == 0) return 0;

    if (Op2C->getValueAPF().isZero())  // pow(x, 0.0) -> 1.0
      return ConstantFP::get(CI->getType(), 1.0);

    if (Op2C->isExactlyValue(0.5)) {
      // Expand pow(x, 0.5) to (x == -infinity ? +infinity : fabs(sqrt(x))).
      // This is faster than calling pow, and still handles negative zero
      // and negative infinite correctly.
      // TODO: In fast-math mode, this could be just sqrt(x).
      // TODO: In finite-only mode, this could be just fabs(sqrt(x)).
      Value *Inf = ConstantFP::getInfinity(CI->getType());
      Value *NegInf = ConstantFP::getInfinity(CI->getType(), true);
      Value *Sqrt = EmitUnaryFloatFnCall(Op1, "sqrt", B,
                                         Callee->getAttributes());
      Value *FAbs = EmitUnaryFloatFnCall(Sqrt, "fabs", B,
                                         Callee->getAttributes());
      Value *FCmp = B.CreateFCmpOEQ(Op1, NegInf);
      Value *Sel = B.CreateSelect(FCmp, Inf, FAbs);
      return Sel;
    }

    if (Op2C->isExactlyValue(1.0))  // pow(x, 1.0) -> x
      return Op1;
    if (Op2C->isExactlyValue(2.0))  // pow(x, 2.0) -> x*x
      return B.CreateFMul(Op1, Op1, "pow2");
    if (Op2C->isExactlyValue(-1.0)) // pow(x, -1.0) -> 1.0/x
      return B.CreateFDiv(ConstantFP::get(CI->getType(), 1.0),
                          Op1, "powrecip");
    return 0;
  }
};

//===---------------------------------------===//
// 'exp2' Optimizations

struct Exp2Opt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    // Just make sure this has 1 argument of FP type, which matches the
    // result type.
    if (FT->getNumParams() != 1 || FT->getReturnType() != FT->getParamType(0) ||
        !FT->getParamType(0)->isFloatingPointTy())
      return 0;

    Value *Op = CI->getArgOperand(0);
    // Turn exp2(sitofp(x)) -> ldexp(1.0, sext(x))  if sizeof(x) <= 32
    // Turn exp2(uitofp(x)) -> ldexp(1.0, zext(x))  if sizeof(x) < 32
    Value *LdExpArg = 0;
    if (SIToFPInst *OpC = dyn_cast<SIToFPInst>(Op)) {
      if (OpC->getOperand(0)->getType()->getPrimitiveSizeInBits() <= 32)
        LdExpArg = B.CreateSExt(OpC->getOperand(0), B.getInt32Ty());
    } else if (UIToFPInst *OpC = dyn_cast<UIToFPInst>(Op)) {
      if (OpC->getOperand(0)->getType()->getPrimitiveSizeInBits() < 32)
        LdExpArg = B.CreateZExt(OpC->getOperand(0), B.getInt32Ty());
    }

    if (LdExpArg) {
      const char *Name;
      if (Op->getType()->isFloatTy())
        Name = "ldexpf";
      else if (Op->getType()->isDoubleTy())
        Name = "ldexp";
      else
        Name = "ldexpl";

      Constant *One = ConstantFP::get(*Context, APFloat(1.0f));
      if (!Op->getType()->isFloatTy())
        One = ConstantExpr::getFPExtend(One, Op->getType());

      Module *M = Caller->getParent();
      Value *Callee = M->getOrInsertFunction(Name, Op->getType(),
                                             Op->getType(),
                                             B.getInt32Ty(), NULL);
      CallInst *CI = B.CreateCall2(Callee, One, LdExpArg);
      if (const Function *F = dyn_cast<Function>(Callee->stripPointerCasts()))
        CI->setCallingConv(F->getCallingConv());

      return CI;
    }
    return 0;
  }
};

//===---------------------------------------===//
// Double -> Float Shrinking Optimizations for Unary Functions like 'floor'

struct UnaryDoubleFPOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 1 || !FT->getReturnType()->isDoubleTy() ||
        !FT->getParamType(0)->isDoubleTy())
      return 0;

    // If this is something like 'floor((double)floatval)', convert to floorf.
    FPExtInst *Cast = dyn_cast<FPExtInst>(CI->getArgOperand(0));
    if (Cast == 0 || !Cast->getOperand(0)->getType()->isFloatTy())
      return 0;

    // floor((double)floatval) -> (double)floorf(floatval)
    Value *V = Cast->getOperand(0);
    V = EmitUnaryFloatFnCall(V, Callee->getName(), B, Callee->getAttributes());
    return B.CreateFPExt(V, B.getDoubleTy());
  }
};

//===----------------------------------------------------------------------===//
// Integer Optimizations
//===----------------------------------------------------------------------===//

//===---------------------------------------===//
// 'ffs*' Optimizations

struct FFSOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    FunctionType *FT = Callee->getFunctionType();
    // Just make sure this has 2 arguments of the same FP type, which match the
    // result type.
    if (FT->getNumParams() != 1 ||
        !FT->getReturnType()->isIntegerTy(32) ||
        !FT->getParamType(0)->isIntegerTy())
      return 0;

    Value *Op = CI->getArgOperand(0);

    // Constant fold.
    if (ConstantInt *CI = dyn_cast<ConstantInt>(Op)) {
      if (CI->getValue() == 0)  // ffs(0) -> 0.
        return Constant::getNullValue(CI->getType());
      // ffs(c) -> cttz(c)+1
      return B.getInt32(CI->getValue().countTrailingZeros() + 1);
    }

    // ffs(x) -> x != 0 ? (i32)llvm.cttz(x)+1 : 0
    Type *ArgType = Op->getType();
    Value *F = Intrinsic::getDeclaration(Callee->getParent(),