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). For example, a call to the function "exit(3)" that
// occurs within the main() function can be transformed into a simple "return 3"
// instruction. Any optimization that takes this 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/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/ADT/SmallPtrSet.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Config/config.h"
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 VISIBILITY_HIDDEN LibCallOptimization {
protected:
  Function *Caller;
  const TargetData *TD;
  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, IRBuilder<> &B) {
    Caller = CI->getParent()->getParent();
    this->TD = TD;
    if (CI->getCalledFunction())
      Context = &CI->getCalledFunction()->getContext();
    return CallOptimizer(CI->getCalledFunction(), CI, B);
  }

  /// CastToCStr - Return V if it is an i8*, otherwise cast it to i8*.
  Value *CastToCStr(Value *V, IRBuilder<> &B);

  /// EmitStrLen - Emit a call to the strlen function to the builder, for the
  /// specified pointer.  Ptr is required to be some pointer type, and the
  /// return value has 'intptr_t' type.
  Value *EmitStrLen(Value *Ptr, IRBuilder<> &B);
  
  /// EmitMemCpy - Emit a call to the memcpy function to the builder.  This
  /// always expects that the size has type 'intptr_t' and Dst/Src are pointers.
  Value *EmitMemCpy(Value *Dst, Value *Src, Value *Len, 
                    unsigned Align, IRBuilder<> &B);
  
  /// EmitMemChr - Emit a call to the memchr function.  This assumes that Ptr is
  /// a pointer, Val is an i32 value, and Len is an 'intptr_t' value.
  Value *EmitMemChr(Value *Ptr, Value *Val, Value *Len, IRBuilder<> &B);

  /// EmitMemCmp - Emit a call to the memcmp function.
  Value *EmitMemCmp(Value *Ptr1, Value *Ptr2, Value *Len, IRBuilder<> &B);

  /// EmitMemSet - Emit a call to the memset function
  Value *EmitMemSet(Value *Dst, Value *Val, Value *Len, IRBuilder<> &B);

  /// EmitUnaryFloatFnCall - Emit a call to the unary function named 'Name' (e.g.
  /// 'floor').  This function is known to take a single of type matching 'Op'
  /// and returns one value with the same type.  If 'Op' is a long double, 'l'
  /// is added as the suffix of name, if 'Op' is a float, we add a 'f' suffix.
  Value *EmitUnaryFloatFnCall(Value *Op, const char *Name, IRBuilder<> &B);
  
  /// EmitPutChar - Emit a call to the putchar function.  This assumes that Char
  /// is an integer.
  void EmitPutChar(Value *Char, IRBuilder<> &B);
  
  /// EmitPutS - Emit a call to the puts function.  This assumes that Str is
  /// some pointer.
  void EmitPutS(Value *Str, IRBuilder<> &B);
    
  /// EmitFPutC - Emit a call to the fputc function.  This assumes that Char is
  /// an i32, and File is a pointer to FILE.
  void EmitFPutC(Value *Char, Value *File, IRBuilder<> &B);
  
  /// EmitFPutS - Emit a call to the puts function.  Str is required to be a
  /// pointer and File is a pointer to FILE.
  void EmitFPutS(Value *Str, Value *File, IRBuilder<> &B);
  
  /// EmitFWrite - Emit a call to the fwrite function.  This assumes that Ptr is
  /// a pointer, Size is an 'intptr_t', and File is a pointer to FILE.
  void EmitFWrite(Value *Ptr, Value *Size, Value *File, IRBuilder<> &B);
  
};
} // End anonymous namespace.

/// CastToCStr - Return V if it is an i8*, otherwise cast it to i8*.
Value *LibCallOptimization::CastToCStr(Value *V, IRBuilder<> &B) {
  return
        B.CreateBitCast(V, PointerType::getUnqual(Type::getInt8Ty(*Context)), "cstr");
}

/// EmitStrLen - Emit a call to the strlen function to the builder, for the
/// specified pointer.  This always returns an integer value of size intptr_t.
Value *LibCallOptimization::EmitStrLen(Value *Ptr, IRBuilder<> &B) {
  Module *M = Caller->getParent();
  AttributeWithIndex AWI[2];
  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
  AWI[1] = AttributeWithIndex::get(~0u, Attribute::ReadOnly |
                                   Attribute::NoUnwind);

  Constant *StrLen =M->getOrInsertFunction("strlen", AttrListPtr::get(AWI, 2),
                                           TD->getIntPtrType(*Context),
                                    PointerType::getUnqual(Type::getInt8Ty(*Context)),
                                           NULL);
  CallInst *CI = B.CreateCall(StrLen, CastToCStr(Ptr, B), "strlen");
  if (const Function *F = dyn_cast<Function>(StrLen->stripPointerCasts()))
    CI->setCallingConv(F->getCallingConv());

  return CI;
}

/// EmitMemCpy - Emit a call to the memcpy function to the builder.  This always
/// expects that the size has type 'intptr_t' and Dst/Src are pointers.
Value *LibCallOptimization::EmitMemCpy(Value *Dst, Value *Src, Value *Len,
                                       unsigned Align, IRBuilder<> &B) {
  Module *M = Caller->getParent();
  Intrinsic::ID IID = Intrinsic::memcpy;
  const Type *Tys[1];
  Tys[0] = Len->getType();
  Value *MemCpy = Intrinsic::getDeclaration(M, IID, Tys, 1);
  return B.CreateCall4(MemCpy, CastToCStr(Dst, B), CastToCStr(Src, B), Len,
                       ConstantInt::get(Type::getInt32Ty(*Context), Align));
}

/// EmitMemChr - Emit a call to the memchr function.  This assumes that Ptr is
/// a pointer, Val is an i32 value, and Len is an 'intptr_t' value.
Value *LibCallOptimization::EmitMemChr(Value *Ptr, Value *Val,
                                       Value *Len, IRBuilder<> &B) {
  Module *M = Caller->getParent();
  AttributeWithIndex AWI;
  AWI = AttributeWithIndex::get(~0u, Attribute::ReadOnly | Attribute::NoUnwind);

  Value *MemChr = M->getOrInsertFunction("memchr", AttrListPtr::get(&AWI, 1),
                                    PointerType::getUnqual(Type::getInt8Ty(*Context)),
                                    PointerType::getUnqual(Type::getInt8Ty(*Context)),
                                         Type::getInt32Ty(*Context), TD->getIntPtrType(*Context),
                                         NULL);
  CallInst *CI = B.CreateCall3(MemChr, CastToCStr(Ptr, B), Val, Len, "memchr");

  if (const Function *F = dyn_cast<Function>(MemChr->stripPointerCasts()))
    CI->setCallingConv(F->getCallingConv());

  return CI;
}

/// EmitMemCmp - Emit a call to the memcmp function.
Value *LibCallOptimization::EmitMemCmp(Value *Ptr1, Value *Ptr2,
                                       Value *Len, IRBuilder<> &B) {
  Module *M = Caller->getParent();
  AttributeWithIndex AWI[3];
  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
  AWI[1] = AttributeWithIndex::get(2, Attribute::NoCapture);
  AWI[2] = AttributeWithIndex::get(~0u, Attribute::ReadOnly |
                                   Attribute::NoUnwind);

  Value *MemCmp = M->getOrInsertFunction("memcmp", AttrListPtr::get(AWI, 3),
                                         Type::getInt32Ty(*Context),
                                    PointerType::getUnqual(Type::getInt8Ty(*Context)),
                                    PointerType::getUnqual(Type::getInt8Ty(*Context)),
                                         TD->getIntPtrType(*Context), NULL);
  CallInst *CI = B.CreateCall3(MemCmp, CastToCStr(Ptr1, B), CastToCStr(Ptr2, B),
                               Len, "memcmp");

  if (const Function *F = dyn_cast<Function>(MemCmp->stripPointerCasts()))
    CI->setCallingConv(F->getCallingConv());

  return CI;
}

/// EmitMemSet - Emit a call to the memset function
Value *LibCallOptimization::EmitMemSet(Value *Dst, Value *Val,
                                       Value *Len, IRBuilder<> &B) {
 Module *M = Caller->getParent();
 Intrinsic::ID IID = Intrinsic::memset;
 const Type *Tys[1];
 Tys[0] = Len->getType();
 Value *MemSet = Intrinsic::getDeclaration(M, IID, Tys, 1);
 Value *Align = ConstantInt::get(Type::getInt32Ty(*Context), 1);
 return B.CreateCall4(MemSet, CastToCStr(Dst, B), Val, Len, Align);
}

/// EmitUnaryFloatFnCall - Emit a call to the unary function named 'Name' (e.g.
/// 'floor').  This function is known to take a single of type matching 'Op' and
/// returns one value with the same type.  If 'Op' is a long double, 'l' is
/// added as the suffix of name, if 'Op' is a float, we add a 'f' suffix.
Value *LibCallOptimization::EmitUnaryFloatFnCall(Value *Op, const char *Name,
                                                 IRBuilder<> &B) {
  char NameBuffer[20];
  if (Op->getType() != Type::getDoubleTy(*Context)) {
    // If we need to add a suffix, copy into NameBuffer.
    unsigned NameLen = strlen(Name);
    assert(NameLen < sizeof(NameBuffer)-2);
    memcpy(NameBuffer, Name, NameLen);
    if (Op->getType() == Type::getFloatTy(*Context))
      NameBuffer[NameLen] = 'f';  // floorf
    else
      NameBuffer[NameLen] = 'l';  // floorl
    NameBuffer[NameLen+1] = 0;
    Name = NameBuffer;
  }

  Module *M = Caller->getParent();
  Value *Callee = M->getOrInsertFunction(Name, Op->getType(),
                                         Op->getType(), NULL);
  CallInst *CI = B.CreateCall(Callee, Op, Name);

  if (const Function *F = dyn_cast<Function>(Callee->stripPointerCasts()))
    CI->setCallingConv(F->getCallingConv());

  return CI;
}

/// EmitPutChar - Emit a call to the putchar function.  This assumes that Char
/// is an integer.
void LibCallOptimization::EmitPutChar(Value *Char, IRBuilder<> &B) {
  Module *M = Caller->getParent();
  Value *PutChar = M->getOrInsertFunction("putchar", Type::getInt32Ty(*Context),
                                          Type::getInt32Ty(*Context), NULL);
  CallInst *CI = B.CreateCall(PutChar,
                              B.CreateIntCast(Char, Type::getInt32Ty(*Context), "chari"),
                              "putchar");

  if (const Function *F = dyn_cast<Function>(PutChar->stripPointerCasts()))
    CI->setCallingConv(F->getCallingConv());
}

/// EmitPutS - Emit a call to the puts function.  This assumes that Str is
/// some pointer.
void LibCallOptimization::EmitPutS(Value *Str, IRBuilder<> &B) {
  Module *M = Caller->getParent();
  AttributeWithIndex AWI[2];
  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
  AWI[1] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);

  Value *PutS = M->getOrInsertFunction("puts", AttrListPtr::get(AWI, 2),
                                       Type::getInt32Ty(*Context),
                                    PointerType::getUnqual(Type::getInt8Ty(*Context)),
                                       NULL);
  CallInst *CI = B.CreateCall(PutS, CastToCStr(Str, B), "puts");
  if (const Function *F = dyn_cast<Function>(PutS->stripPointerCasts()))
    CI->setCallingConv(F->getCallingConv());

}

/// EmitFPutC - Emit a call to the fputc function.  This assumes that Char is
/// an integer and File is a pointer to FILE.
void LibCallOptimization::EmitFPutC(Value *Char, Value *File, IRBuilder<> &B) {
  Module *M = Caller->getParent();
  AttributeWithIndex AWI[2];
  AWI[0] = AttributeWithIndex::get(2, Attribute::NoCapture);
  AWI[1] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
  Constant *F;
  if (isa<PointerType>(File->getType()))
    F = M->getOrInsertFunction("fputc", AttrListPtr::get(AWI, 2), Type::getInt32Ty(*Context),
                               Type::getInt32Ty(*Context), File->getType(), NULL);
  else
    F = M->getOrInsertFunction("fputc", Type::getInt32Ty(*Context), Type::getInt32Ty(*Context),
                               File->getType(), NULL);
  Char = B.CreateIntCast(Char, Type::getInt32Ty(*Context), "chari");
  CallInst *CI = B.CreateCall2(F, Char, File, "fputc");

  if (const Function *Fn = dyn_cast<Function>(F->stripPointerCasts()))
    CI->setCallingConv(Fn->getCallingConv());
}

/// EmitFPutS - Emit a call to the puts function.  Str is required to be a
/// pointer and File is a pointer to FILE.
void LibCallOptimization::EmitFPutS(Value *Str, Value *File, IRBuilder<> &B) {
  Module *M = Caller->getParent();
  AttributeWithIndex AWI[3];
  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
  AWI[1] = AttributeWithIndex::get(2, Attribute::NoCapture);
  AWI[2] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
  Constant *F;
  if (isa<PointerType>(File->getType()))
    F = M->getOrInsertFunction("fputs", AttrListPtr::get(AWI, 3), Type::getInt32Ty(*Context),
                               PointerType::getUnqual(Type::getInt8Ty(*Context)),
                               File->getType(), NULL);
  else
    F = M->getOrInsertFunction("fputs", Type::getInt32Ty(*Context),
                               PointerType::getUnqual(Type::getInt8Ty(*Context)),
                               File->getType(), NULL);
  CallInst *CI = B.CreateCall2(F, CastToCStr(Str, B), File, "fputs");

  if (const Function *Fn = dyn_cast<Function>(F->stripPointerCasts()))
    CI->setCallingConv(Fn->getCallingConv());
}

/// EmitFWrite - Emit a call to the fwrite function.  This assumes that Ptr is
/// a pointer, Size is an 'intptr_t', and File is a pointer to FILE.
void LibCallOptimization::EmitFWrite(Value *Ptr, Value *Size, Value *File,
                                     IRBuilder<> &B) {
  Module *M = Caller->getParent();
  AttributeWithIndex AWI[3];
  AWI[0] = AttributeWithIndex::get(1, Attribute::NoCapture);
  AWI[1] = AttributeWithIndex::get(4, Attribute::NoCapture);
  AWI[2] = AttributeWithIndex::get(~0u, Attribute::NoUnwind);
  Constant *F;
  if (isa<PointerType>(File->getType()))
    F = M->getOrInsertFunction("fwrite", AttrListPtr::get(AWI, 3),
                               TD->getIntPtrType(*Context),
                               PointerType::getUnqual(Type::getInt8Ty(*Context)),
                               TD->getIntPtrType(*Context), TD->getIntPtrType(*Context),
                               File->getType(), NULL);
  else
    F = M->getOrInsertFunction("fwrite", TD->getIntPtrType(*Context),
                               PointerType::getUnqual(Type::getInt8Ty(*Context)),
                               TD->getIntPtrType(*Context), TD->getIntPtrType(*Context),
                               File->getType(), NULL);
  CallInst *CI = B.CreateCall4(F, CastToCStr(Ptr, B), Size,
                        ConstantInt::get(TD->getIntPtrType(*Context), 1), File);

  if (const Function *Fn = dyn_cast<Function>(F->stripPointerCasts()))
    CI->setCallingConv(Fn->getCallingConv());
}

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

/// GetStringLengthH - If we can compute the length of the string pointed to by
/// the specified pointer, return 'len+1'.  If we can't, return 0.
static uint64_t GetStringLengthH(Value *V, SmallPtrSet<PHINode*, 32> &PHIs) {
  // Look through noop bitcast instructions.
  if (BitCastInst *BCI = dyn_cast<BitCastInst>(V))
    return GetStringLengthH(BCI->getOperand(0), PHIs);
  
  // If this is a PHI node, there are two cases: either we have already seen it
  // or we haven't.
  if (PHINode *PN = dyn_cast<PHINode>(V)) {
    if (!PHIs.insert(PN))
      return ~0ULL;  // already in the set.
    
    // If it was new, see if all the input strings are the same length.
    uint64_t LenSoFar = ~0ULL;
    for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
      uint64_t Len = GetStringLengthH(PN->getIncomingValue(i), PHIs);
      if (Len == 0) return 0; // Unknown length -> unknown.
      
      if (Len == ~0ULL) continue;
      
      if (Len != LenSoFar && LenSoFar != ~0ULL)
        return 0;    // Disagree -> unknown.
      LenSoFar = Len;
    }
    
    // Success, all agree.
    return LenSoFar;
  }
  
  // strlen(select(c,x,y)) -> strlen(x) ^ strlen(y)
  if (SelectInst *SI = dyn_cast<SelectInst>(V)) {
    uint64_t Len1 = GetStringLengthH(SI->getTrueValue(), PHIs);
    if (Len1 == 0) return 0;
    uint64_t Len2 = GetStringLengthH(SI->getFalseValue(), PHIs);
    if (Len2 == 0) return 0;
    if (Len1 == ~0ULL) return Len2;
    if (Len2 == ~0ULL) return Len1;
    if (Len1 != Len2) return 0;
    return Len1;
  }
  
  // If the value is not a GEP instruction nor a constant expression with a
  // GEP instruction, then return unknown.
  User *GEP = 0;
  if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(V)) {
    GEP = GEPI;
  } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
    if (CE->getOpcode() != Instruction::GetElementPtr)
      return 0;
    GEP = CE;
  } else {
    return 0;
  }
  
  // Make sure the GEP has exactly three arguments.
  if (GEP->getNumOperands() != 3)
    return 0;
  
  // Check to make sure that the first operand of the GEP is an integer and
  // has value 0 so that we are sure we're indexing into the initializer.
  if (ConstantInt *Idx = dyn_cast<ConstantInt>(GEP->getOperand(1))) {
    if (!Idx->isZero())
      return 0;
  } else
    return 0;
  
  // If the second index isn't a ConstantInt, then this is a variable index
  // into the array.  If this occurs, we can't say anything meaningful about
  // the string.
  uint64_t StartIdx = 0;
  if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(2)))
    StartIdx = CI->getZExtValue();
  else
    return 0;
  
  // The GEP instruction, constant or instruction, must reference a global
  // variable that is a constant and is initialized. The referenced constant
  // initializer is the array that we'll use for optimization.
  GlobalVariable* GV = dyn_cast<GlobalVariable>(GEP->getOperand(0));
  if (!GV || !GV->isConstant() || !GV->hasInitializer() ||
      GV->mayBeOverridden())
    return 0;
  Constant *GlobalInit = GV->getInitializer();
  
  // Handle the ConstantAggregateZero case, which is a degenerate case. The
  // initializer is constant zero so the length of the string must be zero.
  if (isa<ConstantAggregateZero>(GlobalInit))
    return 1;  // Len = 0 offset by 1.
  
  // Must be a Constant Array
  ConstantArray *Array = dyn_cast<ConstantArray>(GlobalInit);
  if (!Array ||
      Array->getType()->getElementType() != Type::getInt8Ty(V->getContext()))
    return false;
  
  // Get the number of elements in the array
  uint64_t NumElts = Array->getType()->getNumElements();
  
  // Traverse the constant array from StartIdx (derived above) which is
  // the place the GEP refers to in the array.
  for (unsigned i = StartIdx; i != NumElts; ++i) {
    Constant *Elt = Array->getOperand(i);
    ConstantInt *CI = dyn_cast<ConstantInt>(Elt);
    if (!CI) // This array isn't suitable, non-int initializer.
      return 0;
    if (CI->isZero())
      return i-StartIdx+1; // We found end of string, success!
  }
  
  return 0; // The array isn't null terminated, conservatively return 'unknown'.
}

/// GetStringLength - If we can compute the length of the string pointed to by
/// the specified pointer, return 'len+1'.  If we can't, return 0.
static uint64_t GetStringLength(Value *V) {
  if (!isa<PointerType>(V->getType())) return 0;
  
  SmallPtrSet<PHINode*, 32> PHIs;
  uint64_t Len = GetStringLengthH(V, PHIs);
  // If Len is ~0ULL, we had an infinite phi cycle: this is dead code, so return
  // an empty string as a length.
  return Len == ~0ULL ? 1 : Len;
}

/// 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;
}

//===----------------------------------------------------------------------===//
// Miscellaneous LibCall Optimizations
//===----------------------------------------------------------------------===//

namespace {
//===---------------------------------------===//
// 'exit' Optimizations

/// ExitOpt - int main() { exit(4); } --> int main() { return 4; }
struct VISIBILITY_HIDDEN ExitOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify we have a reasonable prototype for exit.
    if (Callee->arg_size() == 0 || !CI->use_empty())
      return 0;

    // Verify the caller is main, and that the result type of main matches the
    // argument type of exit.
    if (Caller->getName() != "main" || !Caller->hasExternalLinkage() ||
        Caller->getReturnType() != CI->getOperand(1)->getType())
      return 0;

    TerminatorInst *OldTI = CI->getParent()->getTerminator();

    // Drop all successor phi node entries.
    for (unsigned i = 0, e = OldTI->getNumSuccessors(); i != e; ++i)
      OldTI->getSuccessor(i)->removePredecessor(CI->getParent());

    // Remove all instructions after the exit.
    BasicBlock::iterator Dead = CI, E = OldTI; ++Dead;
    while (Dead != E) {
      BasicBlock::iterator Next = next(Dead);
      if (Dead->getType() != Type::getVoidTy(*Context))
        Dead->replaceAllUsesWith(UndefValue::get(Dead->getType()));
      Dead->eraseFromParent();
      Dead = Next;
    }

    // Insert a return instruction.
    OldTI->eraseFromParent();
    B.SetInsertPoint(B.GetInsertBlock());
    B.CreateRet(CI->getOperand(1));

    return CI;
  }
};

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

//===---------------------------------------===//
// 'strcat' Optimizations

struct VISIBILITY_HIDDEN StrCatOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strcat" function prototype.
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        FT->getReturnType() != PointerType::getUnqual(Type::getInt8Ty(*Context)) ||
        FT->getParamType(0) != FT->getReturnType() ||
        FT->getParamType(1) != FT->getReturnType())
      return 0;
    
    // Extract some information from the instruction
    Value *Dst = CI->getOperand(1);
    Value *Src = CI->getOperand(2);
    
    // 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);
    
    // 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.
    EmitMemCpy(CpyDst, Src,
               ConstantInt::get(TD->getIntPtrType(*Context), Len+1), 1, B);
  }
};

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

struct VISIBILITY_HIDDEN StrNCatOpt : public StrCatOpt {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strncat" function prototype.
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 ||
        FT->getReturnType() != PointerType::getUnqual(Type::getInt8Ty(*Context)) ||
        FT->getParamType(0) != FT->getReturnType() ||
        FT->getParamType(1) != FT->getReturnType() ||
        !isa<IntegerType>(FT->getParamType(2)))
      return 0;

    // Extract some information from the instruction
    Value *Dst = CI->getOperand(1);
    Value *Src = CI->getOperand(2);
    uint64_t Len;

    // We don't do anything if length is not constant
    if (ConstantInt *LengthArg = dyn_cast<ConstantInt>(CI->getOperand(3)))
      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 VISIBILITY_HIDDEN StrChrOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strchr" function prototype.
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 ||
        FT->getReturnType() != PointerType::getUnqual(Type::getInt8Ty(*Context)) ||
        FT->getParamType(0) != FT->getReturnType())
      return 0;
    
    Value *SrcStr = CI->getOperand(1);
    
    // 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->getOperand(2));
    if (CharC == 0) {
      // These optimizations require TargetData.
      if (!TD) return 0;

      uint64_t Len = GetStringLength(SrcStr);
      if (Len == 0 || FT->getParamType(1) != Type::getInt32Ty(*Context)) // memchr needs i32.
        return 0;
      
      return EmitMemChr(SrcStr, CI->getOperand(2), // include nul.
                        ConstantInt::get(TD->getIntPtrType(*Context), Len), B);
    }

    // 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';
    char CharValue = CharC->getSExtValue();
    
    // Compute the offset.
    uint64_t i = 0;
    while (1) {
      if (i == Str.size())    // Didn't find the char.  strchr returns null.
        return Constant::getNullValue(CI->getType());
      // Did we find our match?
      if (Str[i] == CharValue)
        break;
      ++i;
    }
    
    // strchr(s+n,c)  -> gep(s+n+i,c)
    Value *Idx = ConstantInt::get(Type::getInt64Ty(*Context), i);
    return B.CreateGEP(SrcStr, Idx, "strchr");
  }
};

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

struct VISIBILITY_HIDDEN StrCmpOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strcmp" function prototype.
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 || FT->getReturnType() != Type::getInt32Ty(*Context) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        FT->getParamType(0) != PointerType::getUnqual(Type::getInt8Ty(*Context)))
      return 0;
    
    Value *Str1P = CI->getOperand(1), *Str2P = CI->getOperand(2);
    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);
    
    if (HasStr1 && Str1.empty()) // strcmp("", x) -> *x
      return 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(x, y)  -> cnst  (if both x and y are constant strings)
    if (HasStr1 && HasStr2)
      return ConstantInt::get(CI->getType(), 
                                     strcmp(Str1.c_str(),Str2.c_str()));

    // 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);
    }

    return 0;
  }
};

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

struct VISIBILITY_HIDDEN StrNCmpOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strncmp" function prototype.
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 || FT->getReturnType() != Type::getInt32Ty(*Context) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        FT->getParamType(0) != PointerType::getUnqual(Type::getInt8Ty(*Context)) ||
        !isa<IntegerType>(FT->getParamType(2)))
      return 0;
    
    Value *Str1P = CI->getOperand(1), *Str2P = CI->getOperand(2);
    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->getOperand(3)))
      Length = LengthArg->getZExtValue();
    else
      return 0;
    
    if (Length == 0) // strncmp(x,y,0)   -> 0
      return ConstantInt::get(CI->getType(), 0);
    
    std::string Str1, Str2;
    bool HasStr1 = GetConstantStringInfo(Str1P, Str1);
    bool HasStr2 = GetConstantStringInfo(Str2P, Str2);
    
    if (HasStr1 && Str1.empty())  // strncmp("", x, n) -> *x
      return 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());
    
    // strncmp(x, y)  -> cnst  (if both x and y are constant strings)
    if (HasStr1 && HasStr2)
      return ConstantInt::get(CI->getType(),
                              strncmp(Str1.c_str(), Str2.c_str(), Length));
    return 0;
  }
};


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

struct VISIBILITY_HIDDEN StrCpyOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    // Verify the "strcpy" function prototype.
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 2 || FT->getReturnType() != FT->getParamType(0) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        FT->getParamType(0) != PointerType::getUnqual(Type::getInt8Ty(*Context)))
      return 0;
    
    Value *Dst = CI->getOperand(1), *Src = CI->getOperand(2);
    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.
    EmitMemCpy(Dst, Src,
               ConstantInt::get(TD->getIntPtrType(*Context), Len), 1, B);
    return Dst;
  }
};

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

struct VISIBILITY_HIDDEN StrNCpyOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 || FT->getReturnType() != FT->getParamType(0) ||
        FT->getParamType(0) != FT->getParamType(1) ||
        FT->getParamType(0) != PointerType::getUnqual(Type::getInt8Ty(*Context)) ||
        !isa<IntegerType>(FT->getParamType(2)))
      return 0;

    Value *Dst = CI->getOperand(1);
    Value *Src = CI->getOperand(2);
    Value *LenOp = CI->getOperand(3);

    // 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)
      EmitMemSet(Dst, ConstantInt::get(Type::getInt8Ty(*Context), '\0'), LenOp, B);
      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]
    EmitMemCpy(Dst, Src,
               ConstantInt::get(TD->getIntPtrType(*Context), Len), 1, B);

    return Dst;
  }
};

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

struct VISIBILITY_HIDDEN StrLenOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 1 ||
        FT->getParamType(0) != PointerType::getUnqual(Type::getInt8Ty(*Context)) ||
        !isa<IntegerType>(FT->getReturnType()))
      return 0;
    
    Value *Src = CI->getOperand(1);

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

    // Handle strlen(p) != 0.
    if (!IsOnlyUsedInZeroEqualityComparison(CI)) return 0;

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

//===---------------------------------------===//
// 'strto*' Optimizations

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

    Value *EndPtr = CI->getOperand(2);
    if (isa<ConstantPointerNull>(EndPtr)) {
      CI->setOnlyReadsMemory();
      CI->addAttribute(1, Attribute::NoCapture);
    }

    return 0;
  }
};


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

struct VISIBILITY_HIDDEN MemCmpOpt : public LibCallOptimization {
  virtual Value *CallOptimizer(Function *Callee, CallInst *CI, IRBuilder<> &B) {
    const FunctionType *FT = Callee->getFunctionType();
    if (FT->getNumParams() != 3 || !isa<PointerType>(FT->getParamType(0)) ||
        !isa<PointerType>(FT->getParamType(1)) ||
        FT->getReturnType() != Type::getInt32Ty(*Context))
      return 0;

    Value *LHS = CI->getOperand(1), *RHS = CI->getOperand(2);

    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->getOperand(3));
    if (!LenC) return 0;
    uint64_t Len = LenC->getZExtValue();

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

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

    // memcmp(S1,S2,2) != 0 -> (*(short*)LHS ^ *(short*)RHS)  != 0
    // memcmp(S1,S2,4) != 0 -> (*(int*)LHS ^ *(int*)RHS)  != 0
    if ((Len == 2 || Len == 4) && IsOnlyUsedInZeroEqualityComparison(CI)) {
      const Type *PTy = PointerType::getUnqual(Len == 2 ?
                       Type::getInt16Ty(*Context) : Type::getInt32Ty(*Context));
      LHS = B.CreateBitCast(LHS, PTy, "tmp");
      RHS = B.CreateBitCast(RHS, PTy, "tmp");
      LoadInst *LHSV = B.CreateLoad(LHS, "lhsv");
      LoadInst *RHSV = B.CreateLoad(RHS, "rhsv");
      LHSV->setAlignment(1); RHSV->setAlignment(1);  // Unaligned loads.
      return B.CreateZExt(B.CreateXor(LHSV, RHSV, "shortdiff"), CI->getType());
    }

    return 0;
  }
};

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

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

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

    // memcpy(x, y, n) -> llvm.memcpy(x, y, n, 1)
    EmitMemCpy(CI->getOperand(1), CI->getOperand(2), CI->getOperand(3), 1, B);
    return CI->getOperand(1);
  }
};

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

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