Push LLVMContexts through the IntegerType APIs.
[oota-llvm.git] / lib / CodeGen / IntrinsicLowering.cpp
index 72c0f3b084aa3a8df5fe2bdd174f9c6061586aee..cf855f0e896a6d37018dc4d58d7bea7beb215d0e 100644 (file)
@@ -2,8 +2,8 @@
 //
 //                     The LLVM Compiler Infrastructure
 //
-// This file was developed by the LLVM research group and is distributed under
-// the University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 #include "llvm/Constants.h"
 #include "llvm/DerivedTypes.h"
 #include "llvm/Module.h"
-#include "llvm/Instructions.h"
 #include "llvm/Type.h"
 #include "llvm/CodeGen/IntrinsicLowering.h"
-#include "llvm/Support/Streams.h"
+#include "llvm/Support/IRBuilder.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Target/TargetData.h"
 #include "llvm/ADT/SmallVector.h"
 using namespace llvm;
@@ -33,6 +33,28 @@ static void EnsureFunctionExists(Module &M, const char *Name,
   M.getOrInsertFunction(Name, FunctionType::get(RetTy, ParamTys, false));
 }
 
+static void EnsureFPIntrinsicsExist(Module &M, Function *Fn,
+                                    const char *FName,
+                                    const char *DName, const char *LDName) {
+  // Insert definitions for all the floating point types.
+  switch((int)Fn->arg_begin()->getType()->getTypeID()) {
+  case Type::FloatTyID:
+    EnsureFunctionExists(M, FName, Fn->arg_begin(), Fn->arg_end(),
+                         Type::getFloatTy(M.getContext()));
+    break;
+  case Type::DoubleTyID:
+    EnsureFunctionExists(M, DName, Fn->arg_begin(), Fn->arg_end(),
+                         Type::getDoubleTy(M.getContext()));
+    break;
+  case Type::X86_FP80TyID:
+  case Type::FP128TyID:
+  case Type::PPC_FP128TyID:
+    EnsureFunctionExists(M, LDName, Fn->arg_begin(), Fn->arg_end(),
+                         Fn->arg_begin()->getType());
+    break;
+  }
+}
+
 /// ReplaceCallWith - This function is used when we want to lower an intrinsic
 /// call to a call of an external function.  This handles hard cases such as
 /// when there was already a prototype for the external function, and if that
@@ -40,156 +62,185 @@ static void EnsureFunctionExists(Module &M, const char *Name,
 template <class ArgIt>
 static CallInst *ReplaceCallWith(const char *NewFn, CallInst *CI,
                                  ArgIt ArgBegin, ArgIt ArgEnd,
-                                 const Type *RetTy, Constant *&FCache) {
-  if (!FCache) {
-    // If we haven't already looked up this function, check to see if the
-    // program already contains a function with this name.
-    Module *M = CI->getParent()->getParent()->getParent();
-    // Get or insert the definition now.
-    std::vector<const Type *> ParamTys;
-    for (ArgIt I = ArgBegin; I != ArgEnd; ++I)
-      ParamTys.push_back((*I)->getType());
-    FCache = M->getOrInsertFunction(NewFn,
-                                    FunctionType::get(RetTy, ParamTys, false));
-  }
-
-  SmallVector<Value*, 8> Operands(ArgBegin, ArgEnd);
-  CallInst *NewCI = new CallInst(FCache, &Operands[0], Operands.size(),
-                                 CI->getName(), CI);
+                                 const Type *RetTy) {
+  // If we haven't already looked up this function, check to see if the
+  // program already contains a function with this name.
+  Module *M = CI->getParent()->getParent()->getParent();
+  // Get or insert the definition now.
+  std::vector<const Type *> ParamTys;
+  for (ArgIt I = ArgBegin; I != ArgEnd; ++I)
+    ParamTys.push_back((*I)->getType());
+  Constant* FCache = M->getOrInsertFunction(NewFn,
+                                  FunctionType::get(RetTy, ParamTys, false));
+
+  IRBuilder<> Builder(CI->getParent(), CI);
+  SmallVector<Value *, 8> Args(ArgBegin, ArgEnd);
+  CallInst *NewCI = Builder.CreateCall(FCache, Args.begin(), Args.end());
+  NewCI->setName(CI->getName());
   if (!CI->use_empty())
     CI->replaceAllUsesWith(NewCI);
   return NewCI;
 }
 
 void IntrinsicLowering::AddPrototypes(Module &M) {
+  LLVMContext &Context = M.getContext();
   for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
     if (I->isDeclaration() && !I->use_empty())
       switch (I->getIntrinsicID()) {
       default: break;
       case Intrinsic::setjmp:
         EnsureFunctionExists(M, "setjmp", I->arg_begin(), I->arg_end(),
-                             Type::Int32Ty);
+                             Type::getInt32Ty(M.getContext()));
         break;
       case Intrinsic::longjmp:
         EnsureFunctionExists(M, "longjmp", I->arg_begin(), I->arg_end(),
-                             Type::VoidTy);
+                             Type::getVoidTy(M.getContext()));
         break;
       case Intrinsic::siglongjmp:
         EnsureFunctionExists(M, "abort", I->arg_end(), I->arg_end(),
-                             Type::VoidTy);
+                             Type::getVoidTy(M.getContext()));
+        break;
+      case Intrinsic::memcpy:
+        M.getOrInsertFunction("memcpy",
+          PointerType::getUnqual(Type::getInt8Ty(Context)),
+                              PointerType::getUnqual(Type::getInt8Ty(Context)), 
+                              PointerType::getUnqual(Type::getInt8Ty(Context)), 
+                              TD.getIntPtrType(Context), (Type *)0);
+        break;
+      case Intrinsic::memmove:
+        M.getOrInsertFunction("memmove",
+          PointerType::getUnqual(Type::getInt8Ty(Context)),
+                              PointerType::getUnqual(Type::getInt8Ty(Context)), 
+                              PointerType::getUnqual(Type::getInt8Ty(Context)), 
+                              TD.getIntPtrType(Context), (Type *)0);
+        break;
+      case Intrinsic::memset:
+        M.getOrInsertFunction("memset",
+          PointerType::getUnqual(Type::getInt8Ty(Context)),
+                              PointerType::getUnqual(Type::getInt8Ty(Context)), 
+                              Type::getInt32Ty(M.getContext()), 
+                              TD.getIntPtrType(Context), (Type *)0);
+        break;
+      case Intrinsic::sqrt:
+        EnsureFPIntrinsicsExist(M, I, "sqrtf", "sqrt", "sqrtl");
         break;
-      case Intrinsic::memcpy_i32:
-      case Intrinsic::memcpy_i64:
-        M.getOrInsertFunction("memcpy", PointerType::get(Type::Int8Ty),
-                              PointerType::get(Type::Int8Ty), 
-                              PointerType::get(Type::Int8Ty), 
-                              TD.getIntPtrType(), (Type *)0);
+      case Intrinsic::sin:
+        EnsureFPIntrinsicsExist(M, I, "sinf", "sin", "sinl");
         break;
-      case Intrinsic::memmove_i32:
-      case Intrinsic::memmove_i64:
-        M.getOrInsertFunction("memmove", PointerType::get(Type::Int8Ty),
-                              PointerType::get(Type::Int8Ty), 
-                              PointerType::get(Type::Int8Ty), 
-                              TD.getIntPtrType(), (Type *)0);
+      case Intrinsic::cos:
+        EnsureFPIntrinsicsExist(M, I, "cosf", "cos", "cosl");
         break;
-      case Intrinsic::memset_i32:
-      case Intrinsic::memset_i64:
-        M.getOrInsertFunction("memset", PointerType::get(Type::Int8Ty),
-                              PointerType::get(Type::Int8Ty), Type::Int32Ty, 
-                              TD.getIntPtrType(), (Type *)0);
+      case Intrinsic::pow:
+        EnsureFPIntrinsicsExist(M, I, "powf", "pow", "powl");
         break;
-      case Intrinsic::sqrt_f32:
-      case Intrinsic::sqrt_f64:
-        if(I->arg_begin()->getType() == Type::FloatTy)
-          EnsureFunctionExists(M, "sqrtf", I->arg_begin(), I->arg_end(),
-                               Type::FloatTy);
-        else
-          EnsureFunctionExists(M, "sqrt", I->arg_begin(), I->arg_end(),
-                               Type::DoubleTy);
+      case Intrinsic::log:
+        EnsureFPIntrinsicsExist(M, I, "logf", "log", "logl");
+        break;
+      case Intrinsic::log2:
+        EnsureFPIntrinsicsExist(M, I, "log2f", "log2", "log2l");
+        break;
+      case Intrinsic::log10:
+        EnsureFPIntrinsicsExist(M, I, "log10f", "log10", "log10l");
+        break;
+      case Intrinsic::exp:
+        EnsureFPIntrinsicsExist(M, I, "expf", "exp", "expl");
+        break;
+      case Intrinsic::exp2:
+        EnsureFPIntrinsicsExist(M, I, "exp2f", "exp2", "exp2l");
         break;
       }
 }
 
 /// LowerBSWAP - Emit the code to lower bswap of V before the specified
 /// instruction IP.
-static Value *LowerBSWAP(Value *V, Instruction *IP) {
+static Value *LowerBSWAP(LLVMContext &Context, Value *V, Instruction *IP) {
   assert(V->getType()->isInteger() && "Can't bswap a non-integer type!");
 
   unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
   
+  IRBuilder<> Builder(IP->getParent(), IP);
+
   switch(BitSize) {
-  default: assert(0 && "Unhandled type size of value to byteswap!");
+  default: llvm_unreachable("Unhandled type size of value to byteswap!");
   case 16: {
-    Value *Tmp1 = BinaryOperator::createShl(V,
-                                ConstantInt::get(V->getType(),8),"bswap.2",IP);
-    Value *Tmp2 = BinaryOperator::createLShr(V,
-                                ConstantInt::get(V->getType(),8),"bswap.1",IP);
-    V = BinaryOperator::createOr(Tmp1, Tmp2, "bswap.i16", IP);
+    Value *Tmp1 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
+                                    "bswap.2");
+    Value *Tmp2 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
+                                     "bswap.1");
+    V = Builder.CreateOr(Tmp1, Tmp2, "bswap.i16");
     break;
   }
   case 32: {
-    Value *Tmp4 = BinaryOperator::createShl(V,
-                              ConstantInt::get(V->getType(),24),"bswap.4", IP);
-    Value *Tmp3 = BinaryOperator::createShl(V,
-                              ConstantInt::get(V->getType(),8),"bswap.3",IP);
-    Value *Tmp2 = BinaryOperator::createLShr(V,
-                              ConstantInt::get(V->getType(),8),"bswap.2",IP);
-    Value *Tmp1 = BinaryOperator::createLShr(V,
-                              ConstantInt::get(V->getType(),24),"bswap.1", IP);
-    Tmp3 = BinaryOperator::createAnd(Tmp3, 
-                                     ConstantInt::get(Type::Int32Ty, 0xFF0000),
-                                     "bswap.and3", IP);
-    Tmp2 = BinaryOperator::createAnd(Tmp2, 
-                                     ConstantInt::get(Type::Int32Ty, 0xFF00),
-                                     "bswap.and2", IP);
-    Tmp4 = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.or1", IP);
-    Tmp2 = BinaryOperator::createOr(Tmp2, Tmp1, "bswap.or2", IP);
-    V = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.i32", IP);
+    Value *Tmp4 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 24),
+                                    "bswap.4");
+    Value *Tmp3 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
+                                    "bswap.3");
+    Value *Tmp2 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
+                                     "bswap.2");
+    Value *Tmp1 = Builder.CreateLShr(V,ConstantInt::get(V->getType(), 24),
+                                     "bswap.1");
+    Tmp3 = Builder.CreateAnd(Tmp3,
+                         ConstantInt::get(Type::getInt32Ty(Context), 0xFF0000),
+                             "bswap.and3");
+    Tmp2 = Builder.CreateAnd(Tmp2,
+                           ConstantInt::get(Type::getInt32Ty(Context), 0xFF00),
+                             "bswap.and2");
+    Tmp4 = Builder.CreateOr(Tmp4, Tmp3, "bswap.or1");
+    Tmp2 = Builder.CreateOr(Tmp2, Tmp1, "bswap.or2");
+    V = Builder.CreateOr(Tmp4, Tmp2, "bswap.i32");
     break;
   }
   case 64: {
-    Value *Tmp8 = BinaryOperator::createShl(V,
-                              ConstantInt::get(V->getType(),56),"bswap.8", IP);
-    Value *Tmp7 = BinaryOperator::createShl(V,
-                              ConstantInt::get(V->getType(),40),"bswap.7", IP);
-    Value *Tmp6 = BinaryOperator::createShl(V,
-                              ConstantInt::get(V->getType(),24),"bswap.6", IP);
-    Value *Tmp5 = BinaryOperator::createShl(V,
-                              ConstantInt::get(V->getType(),8),"bswap.5", IP);
-    Value* Tmp4 = BinaryOperator::createLShr(V,
-                              ConstantInt::get(V->getType(),8),"bswap.4", IP);
-    Value* Tmp3 = BinaryOperator::createLShr(V,
-                              ConstantInt::get(V->getType(),24),"bswap.3", IP);
-    Value* Tmp2 = BinaryOperator::createLShr(V,
-                              ConstantInt::get(V->getType(),40),"bswap.2", IP);
-    Value* Tmp1 = BinaryOperator::createLShr(V,
-                              ConstantInt::get(V->getType(),56),"bswap.1", IP);
-    Tmp7 = BinaryOperator::createAnd(Tmp7,
-                             ConstantInt::get(Type::Int64Ty, 
-                               0xFF000000000000ULL),
-                             "bswap.and7", IP);
-    Tmp6 = BinaryOperator::createAnd(Tmp6,
-                             ConstantInt::get(Type::Int64Ty, 0xFF0000000000ULL),
-                             "bswap.and6", IP);
-    Tmp5 = BinaryOperator::createAnd(Tmp5,
-                             ConstantInt::get(Type::Int64Ty, 0xFF00000000ULL),
-                             "bswap.and5", IP);
-    Tmp4 = BinaryOperator::createAnd(Tmp4,
-                             ConstantInt::get(Type::Int64Ty, 0xFF000000ULL),
-                             "bswap.and4", IP);
-    Tmp3 = BinaryOperator::createAnd(Tmp3,
-                             ConstantInt::get(Type::Int64Ty, 0xFF0000ULL),
-                             "bswap.and3", IP);
-    Tmp2 = BinaryOperator::createAnd(Tmp2,
-                             ConstantInt::get(Type::Int64Ty, 0xFF00ULL),
-                             "bswap.and2", IP);
-    Tmp8 = BinaryOperator::createOr(Tmp8, Tmp7, "bswap.or1", IP);
-    Tmp6 = BinaryOperator::createOr(Tmp6, Tmp5, "bswap.or2", IP);
-    Tmp4 = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.or3", IP);
-    Tmp2 = BinaryOperator::createOr(Tmp2, Tmp1, "bswap.or4", IP);
-    Tmp8 = BinaryOperator::createOr(Tmp8, Tmp6, "bswap.or5", IP);
-    Tmp4 = BinaryOperator::createOr(Tmp4, Tmp2, "bswap.or6", IP);
-    V = BinaryOperator::createOr(Tmp8, Tmp4, "bswap.i64", IP);
+    Value *Tmp8 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 56),
+                                    "bswap.8");
+    Value *Tmp7 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 40),
+                                    "bswap.7");
+    Value *Tmp6 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 24),
+                                    "bswap.6");
+    Value *Tmp5 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8),
+                                    "bswap.5");
+    Value* Tmp4 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8),
+                                     "bswap.4");
+    Value* Tmp3 = Builder.CreateLShr(V, 
+                                     ConstantInt::get(V->getType(), 24),
+                                     "bswap.3");
+    Value* Tmp2 = Builder.CreateLShr(V, 
+                                     ConstantInt::get(V->getType(), 40),
+                                     "bswap.2");
+    Value* Tmp1 = Builder.CreateLShr(V, 
+                                     ConstantInt::get(V->getType(), 56),
+                                     "bswap.1");
+    Tmp7 = Builder.CreateAnd(Tmp7,
+                             ConstantInt::get(Type::getInt64Ty(Context),
+                                              0xFF000000000000ULL),
+                             "bswap.and7");
+    Tmp6 = Builder.CreateAnd(Tmp6,
+                             ConstantInt::get(Type::getInt64Ty(Context),
+                                              0xFF0000000000ULL),
+                             "bswap.and6");
+    Tmp5 = Builder.CreateAnd(Tmp5,
+                        ConstantInt::get(Type::getInt64Ty(Context),
+                             0xFF00000000ULL),
+                             "bswap.and5");
+    Tmp4 = Builder.CreateAnd(Tmp4,
+                        ConstantInt::get(Type::getInt64Ty(Context),
+                             0xFF000000ULL),
+                             "bswap.and4");
+    Tmp3 = Builder.CreateAnd(Tmp3,
+                             ConstantInt::get(Type::getInt64Ty(Context),
+                             0xFF0000ULL),
+                             "bswap.and3");
+    Tmp2 = Builder.CreateAnd(Tmp2,
+                             ConstantInt::get(Type::getInt64Ty(Context),
+                             0xFF00ULL),
+                             "bswap.and2");
+    Tmp8 = Builder.CreateOr(Tmp8, Tmp7, "bswap.or1");
+    Tmp6 = Builder.CreateOr(Tmp6, Tmp5, "bswap.or2");
+    Tmp4 = Builder.CreateOr(Tmp4, Tmp3, "bswap.or3");
+    Tmp2 = Builder.CreateOr(Tmp2, Tmp1, "bswap.or4");
+    Tmp8 = Builder.CreateOr(Tmp8, Tmp6, "bswap.or5");
+    Tmp4 = Builder.CreateOr(Tmp4, Tmp2, "bswap.or6");
+    V = Builder.CreateOr(Tmp8, Tmp4, "bswap.i64");
     break;
   }
   }
@@ -198,7 +249,7 @@ static Value *LowerBSWAP(Value *V, Instruction *IP) {
 
 /// LowerCTPOP - Emit the code to lower ctpop of V before the specified
 /// instruction IP.
-static Value *LowerCTPOP(Value *V, Instruction *IP) {
+static Value *LowerCTPOP(LLVMContext &Context, Value *V, Instruction *IP) {
   assert(V->getType()->isInteger() && "Can't ctpop a non-integer type!");
 
   static const uint64_t MaskValues[6] = {
@@ -207,478 +258,143 @@ static Value *LowerCTPOP(Value *V, Instruction *IP) {
     0x0000FFFF0000FFFFULL, 0x00000000FFFFFFFFULL
   };
 
-  unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
+  IRBuilder<> Builder(IP->getParent(), IP);
 
-  for (unsigned i = 1, ct = 0; i != BitSize; i <<= 1, ++ct) {
-    Value *MaskCst = ConstantInt::get(V->getType(), MaskValues[ct]);
-    Value *LHS = BinaryOperator::createAnd(V, MaskCst, "cppop.and1", IP);
-    Value *VShift = BinaryOperator::createLShr(V,
-                      ConstantInt::get(V->getType(), i), "ctpop.sh", IP);
-    Value *RHS = BinaryOperator::createAnd(VShift, MaskCst, "cppop.and2", IP);
-    V = BinaryOperator::createAdd(LHS, RHS, "ctpop.step", IP);
+  unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
+  unsigned WordSize = (BitSize + 63) / 64;
+  Value *Count = ConstantInt::get(V->getType(), 0);
+
+  for (unsigned n = 0; n < WordSize; ++n) {
+    Value *PartValue = V;
+    for (unsigned i = 1, ct = 0; i < (BitSize>64 ? 64 : BitSize); 
+         i <<= 1, ++ct) {
+      Value *MaskCst = ConstantInt::get(V->getType(), MaskValues[ct]);
+      Value *LHS = Builder.CreateAnd(PartValue, MaskCst, "cppop.and1");
+      Value *VShift = Builder.CreateLShr(PartValue,
+                                        ConstantInt::get(V->getType(), i),
+                                         "ctpop.sh");
+      Value *RHS = Builder.CreateAnd(VShift, MaskCst, "cppop.and2");
+      PartValue = Builder.CreateAdd(LHS, RHS, "ctpop.step");
+    }
+    Count = Builder.CreateAdd(PartValue, Count, "ctpop.part");
+    if (BitSize > 64) {
+      V = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 64),
+                             "ctpop.part.sh");
+      BitSize -= 64;
+    }
   }
 
-  return CastInst::createIntegerCast(V, Type::Int32Ty, false, "ctpop", IP);
+  return Count;
 }
 
 /// LowerCTLZ - Emit the code to lower ctlz of V before the specified
 /// instruction IP.
-static Value *LowerCTLZ(Value *V, Instruction *IP) {
+static Value *LowerCTLZ(LLVMContext &Context, Value *V, Instruction *IP) {
+
+  IRBuilder<> Builder(IP->getParent(), IP);
 
   unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
-  for (unsigned i = 1; i != BitSize; i <<= 1) {
+  for (unsigned i = 1; i < BitSize; i <<= 1) {
     Value *ShVal = ConstantInt::get(V->getType(), i);
-    ShVal = BinaryOperator::createLShr(V, ShVal, "ctlz.sh", IP);
-    V = BinaryOperator::createOr(V, ShVal, "ctlz.step", IP);
+    ShVal = Builder.CreateLShr(V, ShVal, "ctlz.sh");
+    V = Builder.CreateOr(V, ShVal, "ctlz.step");
   }
 
-  V = BinaryOperator::createNot(V, "", IP);
-  return LowerCTPOP(V, IP);
+  V = Builder.CreateNot(V);
+  return LowerCTPOP(Context, V, IP);
 }
 
-/// Convert the llvm.part.select.iX.iY intrinsic. This intrinsic takes 
-/// three integer arguments. The first argument is the Value from which the
-/// bits will be selected. It may be of any bit width. The second and third
-/// arguments specify a range of bits to select with the second argument 
-/// specifying the low bit and the third argument specifying the high bit. Both
-/// must be type i32. The result is the corresponding selected bits from the
-/// Value in the same width as the Value (first argument). If the low bit index
-/// is higher than the high bit index then the inverse selection is done and 
-/// the bits are returned in inverse order. 
-/// @brief Lowering of llvm.part.select intrinsic.
-static Instruction *LowerPartSelect(CallInst *CI) {
-  // Make sure we're dealing with a part select intrinsic here
-  Function *F = CI->getCalledFunction();
-  const FunctionType *FT = F->getFunctionType();
-  if (!F->isDeclaration() || !FT->getReturnType()->isInteger() ||
-      FT->getNumParams() != 3 || !FT->getParamType(0)->isInteger() ||
-      !FT->getParamType(1)->isInteger() || !FT->getParamType(2)->isInteger())
-    return CI;
-
-  // Get the intrinsic implementation function by converting all the . to _
-  // in the intrinsic's function name and then reconstructing the function
-  // declaration.
-  std::string Name(F->getName());
-  for (unsigned i = 4; i < Name.length(); ++i)
-    if (Name[i] == '.')
-      Name[i] = '_';
-  Module* M = F->getParent();
-  F = cast<Function>(M->getOrInsertFunction(Name, FT));
-  F->setLinkage(GlobalValue::WeakLinkage);
-
-  // If we haven't defined the impl function yet, do so now
-  if (F->isDeclaration()) {
-
-    // Get the arguments to the function
-    Function::arg_iterator args = F->arg_begin();
-    Value* Val = args++; Val->setName("Val");
-    Value* Lo = args++; Lo->setName("Lo");
-    Value* Hi  = args++; Hi->setName("High");
-
-    // We want to select a range of bits here such that [Hi, Lo] is shifted
-    // down to the low bits. However, it is quite possible that Hi is smaller
-    // than Lo in which case the bits have to be reversed. 
-    
-    // Create the blocks we will need for the two cases (forward, reverse)
-    BasicBlock* CurBB   = new BasicBlock("entry", F);
-    BasicBlock *RevSize = new BasicBlock("revsize", CurBB->getParent());
-    BasicBlock *FwdSize = new BasicBlock("fwdsize", CurBB->getParent());
-    BasicBlock *Compute = new BasicBlock("compute", CurBB->getParent());
-    BasicBlock *Reverse = new BasicBlock("reverse", CurBB->getParent());
-    BasicBlock *RsltBlk = new BasicBlock("result",  CurBB->getParent());
-
-    // Cast Hi and Lo to the size of Val so the widths are all the same
-    if (Hi->getType() != Val->getType())
-      Hi = CastInst::createIntegerCast(Hi, Val->getType(), false, 
-                                         "tmp", CurBB);
-    if (Lo->getType() != Val->getType())
-      Lo = CastInst::createIntegerCast(Lo, Val->getType(), false, 
-                                          "tmp", CurBB);
-
-    // Compute a few things that both cases will need, up front.
-    Constant* Zero = ConstantInt::get(Val->getType(), 0);
-    Constant* One = ConstantInt::get(Val->getType(), 1);
-    Constant* AllOnes = ConstantInt::getAllOnesValue(Val->getType());
-
-    // Compare the Hi and Lo bit positions. This is used to determine 
-    // which case we have (forward or reverse)
-    ICmpInst *Cmp = new ICmpInst(ICmpInst::ICMP_ULT, Hi, Lo, "less",CurBB);
-    new BranchInst(RevSize, FwdSize, Cmp, CurBB);
-
-    // First, copmute the number of bits in the forward case.
-    Instruction* FBitSize = 
-      BinaryOperator::createSub(Hi, Lo,"fbits", FwdSize);
-    new BranchInst(Compute, FwdSize);
-
-    // Second, compute the number of bits in the reverse case.
-    Instruction* RBitSize = 
-      BinaryOperator::createSub(Lo, Hi, "rbits", RevSize);
-    new BranchInst(Compute, RevSize);
-
-    // Now, compute the bit range. Start by getting the bitsize and the shift
-    // amount (either Hi or Lo) from PHI nodes. Then we compute a mask for 
-    // the number of bits we want in the range. We shift the bits down to the 
-    // least significant bits, apply the mask to zero out unwanted high bits, 
-    // and we have computed the "forward" result. It may still need to be 
-    // reversed.
-
-    // Get the BitSize from one of the two subtractions
-    PHINode *BitSize = new PHINode(Val->getType(), "bits", Compute);
-    BitSize->reserveOperandSpace(2);
-    BitSize->addIncoming(FBitSize, FwdSize);
-    BitSize->addIncoming(RBitSize, RevSize);
-
-    // Get the ShiftAmount as the smaller of Hi/Lo
-    PHINode *ShiftAmt = new PHINode(Val->getType(), "shiftamt", Compute);
-    ShiftAmt->reserveOperandSpace(2);
-    ShiftAmt->addIncoming(Lo, FwdSize);
-    ShiftAmt->addIncoming(Hi, RevSize);
-
-    // Increment the bit size
-    Instruction *BitSizePlusOne = 
-      BinaryOperator::createAdd(BitSize, One, "bits", Compute);
-
-    // Create a Mask to zero out the high order bits.
-    Instruction* Mask = 
-      BinaryOperator::createShl(AllOnes, BitSizePlusOne, "mask", Compute);
-    Mask = BinaryOperator::createNot(Mask, "mask", Compute);
-
-    // Shift the bits down and apply the mask
-    Instruction* FRes = 
-      BinaryOperator::createLShr(Val, ShiftAmt, "fres", Compute);
-    FRes = BinaryOperator::createAnd(FRes, Mask, "fres", Compute);
-    new BranchInst(Reverse, RsltBlk, Cmp, Compute);
-
-    // In the Reverse block we have the mask already in FRes but we must reverse
-    // it by shifting FRes bits right and putting them in RRes by shifting them 
-    // in from left.
-
-    // First set up our loop counters
-    PHINode *Count = new PHINode(Val->getType(), "count", Reverse);
-    Count->reserveOperandSpace(2);
-    Count->addIncoming(BitSizePlusOne, Compute);
-
-    // Next, get the value that we are shifting.
-    PHINode *BitsToShift   = new PHINode(Val->getType(), "val", Reverse);
-    BitsToShift->reserveOperandSpace(2);
-    BitsToShift->addIncoming(FRes, Compute);
-
-    // Finally, get the result of the last computation
-    PHINode *RRes  = new PHINode(Val->getType(), "rres", Reverse);
-    RRes->reserveOperandSpace(2);
-    RRes->addIncoming(Zero, Compute);
-
-    // Decrement the counter
-    Instruction *Decr = BinaryOperator::createSub(Count, One, "decr", Reverse);
-    Count->addIncoming(Decr, Reverse);
-
-    // Compute the Bit that we want to move
-    Instruction *Bit = 
-      BinaryOperator::createAnd(BitsToShift, One, "bit", Reverse);
-
-    // Compute the new value for next iteration.
-    Instruction *NewVal = 
-      BinaryOperator::createLShr(BitsToShift, One, "rshift", Reverse);
-    BitsToShift->addIncoming(NewVal, Reverse);
-
-    // Shift the bit into the low bits of the result.
-    Instruction *NewRes = 
-      BinaryOperator::createShl(RRes, One, "lshift", Reverse);
-    NewRes = BinaryOperator::createOr(NewRes, Bit, "addbit", Reverse);
-    RRes->addIncoming(NewRes, Reverse);
-    
-    // Terminate loop if we've moved all the bits.
-    ICmpInst *Cond = 
-      new ICmpInst(ICmpInst::ICMP_EQ, Decr, Zero, "cond", Reverse);
-    new BranchInst(RsltBlk, Reverse, Cond, Reverse);
-
-    // Finally, in the result block, select one of the two results with a PHI
-    // node and return the result;
-    CurBB = RsltBlk;
-    PHINode *BitSelect = new PHINode(Val->getType(), "part_select", CurBB);
-    BitSelect->reserveOperandSpace(2);
-    BitSelect->addIncoming(FRes, Compute);
-    BitSelect->addIncoming(NewRes, Reverse);
-    new ReturnInst(BitSelect, CurBB);
-  }
-
-  // Return a call to the implementation function
-  Value *Args[3];
-  Args[0] = CI->getOperand(1);
-  Args[1] = CI->getOperand(2);
-  Args[2] = CI->getOperand(3);
-  return new CallInst(F, Args, 3, CI->getName(), CI);
-}
-
-/// Convert the llvm.part.set.iX.iY.iZ intrinsic. This intrinsic takes 
-/// four integer arguments (iAny %Value, iAny %Replacement, i32 %Low, i32 %High)
-/// The first two arguments can be any bit width. The result is the same width
-/// as %Value. The operation replaces bits between %Low and %High with the value
-/// in %Replacement. If %Replacement is not the same width, it is truncated or
-/// zero extended as appropriate to fit the bits being replaced. If %Low is
-/// greater than %High then the inverse set of bits are replaced.
-/// @brief Lowering of llvm.bit.part.set intrinsic.
-static Instruction *LowerPartSet(CallInst *CI) {
-  // Make sure we're dealing with a part select intrinsic here
-  Function *F = CI->getCalledFunction();
-  const FunctionType *FT = F->getFunctionType();
-  if (!F->isDeclaration() || !FT->getReturnType()->isInteger() ||
-      FT->getNumParams() != 4 || !FT->getParamType(0)->isInteger() ||
-      !FT->getParamType(1)->isInteger() || !FT->getParamType(2)->isInteger() ||
-      !FT->getParamType(3)->isInteger())
-    return CI;
-
-  // Get the intrinsic implementation function by converting all the . to _
-  // in the intrinsic's function name and then reconstructing the function
-  // declaration.
-  std::string Name(F->getName());
-  for (unsigned i = 4; i < Name.length(); ++i)
-    if (Name[i] == '.')
-      Name[i] = '_';
-  Module* M = F->getParent();
-  F = cast<Function>(M->getOrInsertFunction(Name, FT));
-  F->setLinkage(GlobalValue::WeakLinkage);
-
-  // If we haven't defined the impl function yet, do so now
-  if (F->isDeclaration()) {
-    // Note: the following code is based on code generated by llvm2cpp with 
-    // the following input. This is just *one* example of a generated function.
-    // The functions vary by bit width of result and first two arguments.
-    // The generated code has been changed to deal with any bit width not just
-    // the 32/64 bitwidths used in the above sample.
-    //
-    // define i64 @part_set(i64 %Val, i32 %Rep, i32 %Lo, i32 %Hi) {
-    // entry:
-    //   %is_forward = icmp ult i32 %Lo, %Hi
-    //   %Lo.pn = select i1 %is_forward, i32 %Hi, i32 %Lo
-    //   %Hi.pn = select i1 %is_forward, i32 %Lo, i32 %Hi
-    //   %iftmp.16.0 = sub i32 %Lo.pn, %Hi.pn
-    //   icmp ult i32 %iftmp.16.0, 32
-    //   br i1 %1, label %cond_true11, label %cond_next19
-    // cond_true11:
-    //   %tmp13 = sub i32 32, %iftmp.16.0
-    //   %tmp14 = lshr i32 -1, %tmp13
-    //   %tmp16 = and i32 %tmp14, %Rep
-    //   br label %cond_next19
-    // cond_next19:
-    //   %iftmp.17.0 = phi i32 [ %tmp16, %cond_true11 ], [ %Rep, %entry ]
-    //   %tmp2021 = zext i32 %iftmp.17.0 to i64
-    //   icmp ugt i32 %Lo, %Hi
-    //   br i1 %2, label %cond_next60, label %cond_true24
-    // cond_true24:
-    //   %tmp25.cast = zext i32 %Hi to i64
-    //   %tmp26 = lshr i64 -1, %tmp25.cast
-    //   %tmp27.cast = zext i32 %Lo to i64
-    //   %tmp28 = shl i64 %tmp26, %tmp27.cast
-    //   %tmp28not = xor i64 %tmp28, -1
-    //   %tmp31 = shl i64 %tmp2021, %tmp27.cast
-    //   %tmp34 = and i64 %tmp28not, %Val
-    //   %Val_addr.064 = or i64 %tmp31, %tmp34
-    //   ret i64 %Val_addr.064
-    // cond_next60:
-    //   %tmp39.cast = zext i32 %Lo to i64
-    //   %tmp40 = shl i64 -1, %tmp39.cast
-    //   %tmp41.cast = zext i32 %Hi to i64
-    //   %tmp42 = shl i64 -1, %tmp41.cast
-    //   %tmp45.demorgan = or i64 %tmp42, %tmp40
-    //   %tmp45 = xor i64 %tmp45.demorgan, -1
-    //   %tmp47 = and i64 %tmp45, %Val
-    //   %tmp50 = shl i64 %tmp2021, %tmp39.cast
-    //   %tmp52 = sub i32 32, %Hi
-    //   %tmp52.cast = zext i32 %tmp52 to i64
-    //   %tmp54 = lshr i64 %tmp2021, %tmp52.cast
-    //   %tmp57 = or i64 %tmp50, %tmp47
-    //   %Val_addr.0 = or i64 %tmp57, %tmp54
-    //   ret i64 %Val_addr.0
-    // }
-
-    // Get the arguments for the function.
-    Function::arg_iterator args = F->arg_begin();
-    Value* Val = args++; Val->setName("Val");
-    Value* Rep = args++; Rep->setName("Rep");
-    Value* Lo  = args++; Lo->setName("Lo");
-    Value* Hi  = args++; Hi->setName("Hi");
-
-    // Get some types we need
-    const IntegerType* ValTy = cast<IntegerType>(Val->getType());
-    const IntegerType* RepTy = cast<IntegerType>(Rep->getType());
-    uint32_t ValBits = ValTy->getBitWidth();
-    uint32_t RepBits = RepTy->getBitWidth();
-
-    // Constant Definitions
-    ConstantInt* RepBitWidth = ConstantInt::get(Type::Int32Ty, RepBits);
-    ConstantInt* RepMask = ConstantInt::getAllOnesValue(RepTy);
-    ConstantInt* ValMask = ConstantInt::getAllOnesValue(ValTy);
-
-    BasicBlock* entry = new BasicBlock("entry",F,0);
-    BasicBlock* large = new BasicBlock("large",F,0);
-    BasicBlock* small = new BasicBlock("small",F,0);
-    BasicBlock* forward = new BasicBlock("forward",F,0);
-    BasicBlock* reverse = new BasicBlock("reverse",F,0);
-
-    // Block entry (entry)
-    // First, get the number of bits that we're placing as an i32
-    ICmpInst* is_forward = 
-      new ICmpInst(ICmpInst::ICMP_ULT, Lo, Hi, "", entry);
-    SelectInst* Lo_pn = new SelectInst(is_forward, Hi, Lo, "", entry);
-    SelectInst* Hi_pn = new SelectInst(is_forward, Lo, Hi, "", entry);
-    BinaryOperator* NumBits = BinaryOperator::createSub(Lo_pn, Hi_pn, "",entry);
-    // Now, convert Lo and Hi to ValTy bit width
-    if (ValBits > 32) {
-      Hi = new ZExtInst(Hi, ValTy, "", entry);
-      Lo = new ZExtInst(Lo, ValTy, "", entry);
-    } else if (ValBits < 32) {
-      Hi = new TruncInst(Hi, ValTy, "", entry);
-      Lo = new TruncInst(Lo, ValTy, "", entry);
-    }
-    // Determine if the replacement bits are larger than the number of bits we
-    // are replacing and deal with it.
-    ICmpInst* is_large = 
-      new ICmpInst(ICmpInst::ICMP_ULT, NumBits, RepBitWidth, "", entry);
-    new BranchInst(large, small, is_large, entry);
-
-    // Block "large"
-    Instruction* MaskBits = 
-      BinaryOperator::createSub(RepBitWidth, NumBits, "", large);
-    MaskBits = CastInst::createIntegerCast(MaskBits, RepMask->getType(), 
-                                           false, "", large);
-    BinaryOperator* Mask1 = 
-      BinaryOperator::createLShr(RepMask, MaskBits, "", large);
-    BinaryOperator* Rep2 = BinaryOperator::createAnd(Mask1, Rep, "", large);
-    new BranchInst(small, large);
-
-    // Block "small"
-    PHINode* Rep3 = new PHINode(RepTy, "", small);
-    Rep3->reserveOperandSpace(2);
-    Rep3->addIncoming(Rep2, large);
-    Rep3->addIncoming(Rep, entry);
-    Value* Rep4 = Rep3;
-    if (ValBits > RepBits)
-      Rep4 = new ZExtInst(Rep3, ValTy, "", small);
-    else if (ValBits < RepBits)
-      Rep4 = new TruncInst(Rep3, ValTy, "", small);
-    ICmpInst* is_reverse = 
-      new ICmpInst(ICmpInst::ICMP_UGT, Lo, Hi, "", small);
-    new BranchInst(reverse, forward, is_reverse, small);
-
-    // Block "forward"
-    Value* t1    = BinaryOperator::createLShr(ValMask, Hi, "", forward);
-    Value* t2    = BinaryOperator::createShl(t1, Lo, "", forward);
-    Value* nott2 = BinaryOperator::createXor(t2, ValMask, "", forward);
-    Value* t3    = BinaryOperator::createShl(Rep4, Lo, "", forward);
-    Value* t4    = BinaryOperator::createAnd(nott2, Val, "", forward);
-    Value* FRslt = BinaryOperator::createOr(t3, t4, "", forward);
-    new ReturnInst(FRslt, forward);
-
-    // Block "reverse"
-    Value* t5    = BinaryOperator::createShl(ValMask, Lo, "", reverse);
-    Value* t6    = BinaryOperator::createShl(ValMask, Hi, "", reverse);
-    Value* t7    = BinaryOperator::createOr(t6, t5, "", reverse);
-    Value* t8    = BinaryOperator::createXor(t7, ValMask, "", reverse);
-    Value* t9    = BinaryOperator::createAnd(t8, Val, "", reverse);
-    Value* t10   = BinaryOperator::createShl(Rep4, Lo, "", reverse);
-    if (32 < ValBits)
-      RepBitWidth = 
-        cast<ConstantInt>(ConstantExpr::getZExt(RepBitWidth, ValTy));
-    else if (32 > ValBits)
-      RepBitWidth = 
-        cast<ConstantInt>(ConstantExpr::getTrunc(RepBitWidth, ValTy));
-    Value* t11   = BinaryOperator::createSub(RepBitWidth, Hi, "", reverse);
-    Value* t13   = BinaryOperator::createLShr(Rep4, t11, "",reverse);
-    Value* t14   = BinaryOperator::createOr(t10, t9, "", reverse);
-    Value* RRslt = BinaryOperator::createOr(t14, t13, "", reverse);
-    new ReturnInst(RRslt, reverse);
+static void ReplaceFPIntrinsicWithCall(CallInst *CI, const char *Fname,
+                                       const char *Dname,
+                                       const char *LDname) {
+  switch (CI->getOperand(1)->getType()->getTypeID()) {
+  default: llvm_unreachable("Invalid type in intrinsic");
+  case Type::FloatTyID:
+    ReplaceCallWith(Fname, CI, CI->op_begin() + 1, CI->op_end(),
+                  Type::getFloatTy(CI->getContext()));
+    break;
+  case Type::DoubleTyID:
+    ReplaceCallWith(Dname, CI, CI->op_begin() + 1, CI->op_end(),
+                  Type::getDoubleTy(CI->getContext()));
+    break;
+  case Type::X86_FP80TyID:
+  case Type::FP128TyID:
+  case Type::PPC_FP128TyID:
+    ReplaceCallWith(LDname, CI, CI->op_begin() + 1, CI->op_end(),
+                  CI->getOperand(1)->getType());
+    break;
   }
-
-  // Return a call to the implementation function
-  Value *Args[3];
-  Args[0] = CI->getOperand(1);
-  Args[1] = CI->getOperand(2);
-  Args[2] = CI->getOperand(3);
-  Args[3] = CI->getOperand(4);
-  return new CallInst(F, Args, 4, CI->getName(), CI);
 }
 
-
 void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
+  IRBuilder<> Builder(CI->getParent(), CI);
+  LLVMContext &Context = CI->getContext();
+
   Function *Callee = CI->getCalledFunction();
   assert(Callee && "Cannot lower an indirect call!");
 
   switch (Callee->getIntrinsicID()) {
   case Intrinsic::not_intrinsic:
-    cerr << "Cannot lower a call to a non-intrinsic function '"
-         << Callee->getName() << "'!\n";
-    abort();
+    llvm_report_error("Cannot lower a call to a non-intrinsic function '"+
+                      Callee->getName() + "'!");
   default:
-    cerr << "Error: Code generator does not support intrinsic function '"
-         << Callee->getName() << "'!\n";
-    abort();
+    llvm_report_error("Code generator does not support intrinsic function '"+
+                      Callee->getName()+"'!");
 
     // The setjmp/longjmp intrinsics should only exist in the code if it was
     // never optimized (ie, right out of the CFE), or if it has been hacked on
     // by the lowerinvoke pass.  In both cases, the right thing to do is to
     // convert the call to an explicit setjmp or longjmp call.
   case Intrinsic::setjmp: {
-    static Constant *SetjmpFCache = 0;
-    Value *V = ReplaceCallWith("setjmp", CI, CI->op_begin()+1, CI->op_end(),
-                               Type::Int32Ty, SetjmpFCache);
-    if (CI->getType() != Type::VoidTy)
+    Value *V = ReplaceCallWith("setjmp", CI, CI->op_begin() + 1, CI->op_end(),
+                               Type::getInt32Ty(Context));
+    if (CI->getType() != Type::getVoidTy(Context))
       CI->replaceAllUsesWith(V);
     break;
   }
   case Intrinsic::sigsetjmp:
-     if (CI->getType() != Type::VoidTy)
+     if (CI->getType() != Type::getVoidTy(Context))
        CI->replaceAllUsesWith(Constant::getNullValue(CI->getType()));
      break;
 
   case Intrinsic::longjmp: {
-    static Constant *LongjmpFCache = 0;
-    ReplaceCallWith("longjmp", CI, CI->op_begin()+1, CI->op_end(),
-                    Type::VoidTy, LongjmpFCache);
+    ReplaceCallWith("longjmp", CI, CI->op_begin() + 1, CI->op_end(),
+                    Type::getVoidTy(Context));
     break;
   }
 
   case Intrinsic::siglongjmp: {
     // Insert the call to abort
-    static Constant *AbortFCache = 0;
     ReplaceCallWith("abort", CI, CI->op_end(), CI->op_end(), 
-                    Type::VoidTy, AbortFCache);
+                    Type::getVoidTy(Context));
     break;
   }
   case Intrinsic::ctpop:
-    CI->replaceAllUsesWith(LowerCTPOP(CI->getOperand(1), CI));
+    CI->replaceAllUsesWith(LowerCTPOP(Context, CI->getOperand(1), CI));
     break;
 
   case Intrinsic::bswap:
-    CI->replaceAllUsesWith(LowerBSWAP(CI->getOperand(1), CI));
+    CI->replaceAllUsesWith(LowerBSWAP(Context, CI->getOperand(1), CI));
     break;
     
   case Intrinsic::ctlz:
-    CI->replaceAllUsesWith(LowerCTLZ(CI->getOperand(1), CI));
+    CI->replaceAllUsesWith(LowerCTLZ(Context, CI->getOperand(1), CI));
     break;
 
   case Intrinsic::cttz: {
     // cttz(x) -> ctpop(~X & (X-1))
     Value *Src = CI->getOperand(1);
-    Value *NotSrc = BinaryOperator::createNot(Src, Src->getName()+".not", CI);
-    Value *SrcM1  = ConstantInt::get(Src->getType(), 1);
-    SrcM1 = BinaryOperator::createSub(Src, SrcM1, "", CI);
-    Src = LowerCTPOP(BinaryOperator::createAnd(NotSrc, SrcM1, "", CI), CI);
+    Value *NotSrc = Builder.CreateNot(Src);
+    NotSrc->setName(Src->getName() + ".not");
+    Value *SrcM1 = ConstantInt::get(Src->getType(), 1);
+    SrcM1 = Builder.CreateSub(Src, SrcM1);
+    Src = LowerCTPOP(Context, Builder.CreateAnd(NotSrc, SrcM1), CI);
     CI->replaceAllUsesWith(Src);
     break;
   }
 
-  case Intrinsic::part_select:
-    CI->replaceAllUsesWith(LowerPartSelect(CI));
-    break;
-
-  case Intrinsic::part_set:
-    CI->replaceAllUsesWith(LowerPartSet(CI));
-    break;
-
   case Intrinsic::stacksave:
   case Intrinsic::stackrestore: {
-    static bool Warned = false;
     if (!Warned)
       cerr << "WARNING: this target does not support the llvm.stack"
            << (Callee->getIntrinsicID() == Intrinsic::stacksave ?
@@ -706,7 +422,7 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
   case Intrinsic::readcyclecounter: {
     cerr << "WARNING: this target does not support the llvm.readcyclecoun"
          << "ter intrinsic.  It is being lowered to a constant 0\n";
-    CI->replaceAllUsesWith(ConstantInt::get(Type::Int64Ty, 0));
+    CI->replaceAllUsesWith(ConstantInt::get(Type::getInt64Ty(Context), 0));
     break;
   }
 
@@ -715,81 +431,91 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
   case Intrinsic::dbg_region_end:
   case Intrinsic::dbg_func_start:
   case Intrinsic::dbg_declare:
+    break;    // Simply strip out debugging intrinsics
+
   case Intrinsic::eh_exception:
-  case Intrinsic::eh_selector:
-  case Intrinsic::eh_filter:
-    break;    // Simply strip out debugging and eh intrinsics
-
-  case Intrinsic::memcpy_i32:
-  case Intrinsic::memcpy_i64: {
-    static Constant *MemcpyFCache = 0;
-    Value *Size = CI->getOperand(3);
-    const Type *IntPtr = TD.getIntPtrType();
-    if (Size->getType()->getPrimitiveSizeInBits() <
-        IntPtr->getPrimitiveSizeInBits())
-      Size = new ZExtInst(Size, IntPtr, "", CI);
-    else if (Size->getType()->getPrimitiveSizeInBits() >
-             IntPtr->getPrimitiveSizeInBits())
-      Size = new TruncInst(Size, IntPtr, "", CI);
+  case Intrinsic::eh_selector_i32:
+  case Intrinsic::eh_selector_i64:
+    CI->replaceAllUsesWith(Constant::getNullValue(CI->getType()));
+    break;
+
+  case Intrinsic::eh_typeid_for_i32:
+  case Intrinsic::eh_typeid_for_i64:
+    // Return something different to eh_selector.
+    CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 1));
+    break;
+
+  case Intrinsic::var_annotation:
+    break;   // Strip out annotate intrinsic
+    
+  case Intrinsic::memcpy: {
+    const IntegerType *IntPtr = TD.getIntPtrType(Context);
+    Value *Size = Builder.CreateIntCast(CI->getOperand(3), IntPtr,
+                                        /* isSigned */ false);
     Value *Ops[3];
     Ops[0] = CI->getOperand(1);
     Ops[1] = CI->getOperand(2);
     Ops[2] = Size;
-    ReplaceCallWith("memcpy", CI, Ops, Ops+3, CI->getOperand(1)->getType(),
-                    MemcpyFCache);
+    ReplaceCallWith("memcpy", CI, Ops, Ops+3, CI->getOperand(1)->getType());
     break;
   }
-  case Intrinsic::memmove_i32: 
-  case Intrinsic::memmove_i64: {
-    static Constant *MemmoveFCache = 0;
-    Value *Size = CI->getOperand(3);
-    const Type *IntPtr = TD.getIntPtrType();
-    if (Size->getType()->getPrimitiveSizeInBits() <
-        IntPtr->getPrimitiveSizeInBits())
-      Size = new ZExtInst(Size, IntPtr, "", CI);
-    else if (Size->getType()->getPrimitiveSizeInBits() >
-             IntPtr->getPrimitiveSizeInBits())
-      Size = new TruncInst(Size, IntPtr, "", CI);
+  case Intrinsic::memmove: {
+    const IntegerType *IntPtr = TD.getIntPtrType(Context);
+    Value *Size = Builder.CreateIntCast(CI->getOperand(3), IntPtr,
+                                        /* isSigned */ false);
     Value *Ops[3];
     Ops[0] = CI->getOperand(1);
     Ops[1] = CI->getOperand(2);
     Ops[2] = Size;
-    ReplaceCallWith("memmove", CI, Ops, Ops+3, CI->getOperand(1)->getType(),
-                    MemmoveFCache);
+    ReplaceCallWith("memmove", CI, Ops, Ops+3, CI->getOperand(1)->getType());
     break;
   }
-  case Intrinsic::memset_i32:
-  case Intrinsic::memset_i64: {
-    static Constant *MemsetFCache = 0;
-    Value *Size = CI->getOperand(3);
-    const Type *IntPtr = TD.getIntPtrType();
-    if (Size->getType()->getPrimitiveSizeInBits() <
-        IntPtr->getPrimitiveSizeInBits())
-      Size = new ZExtInst(Size, IntPtr, "", CI);
-    else if (Size->getType()->getPrimitiveSizeInBits() >
-             IntPtr->getPrimitiveSizeInBits())
-      Size = new TruncInst(Size, IntPtr, "", CI);
+  case Intrinsic::memset: {
+    const IntegerType *IntPtr = TD.getIntPtrType(Context);
+    Value *Size = Builder.CreateIntCast(CI->getOperand(3), IntPtr,
+                                        /* isSigned */ false);
     Value *Ops[3];
     Ops[0] = CI->getOperand(1);
     // Extend the amount to i32.
-    Ops[1] = new ZExtInst(CI->getOperand(2), Type::Int32Ty, "", CI);
+    Ops[1] = Builder.CreateIntCast(CI->getOperand(2), Type::getInt32Ty(Context),
+                                   /* isSigned */ false);
     Ops[2] = Size;
-    ReplaceCallWith("memset", CI, Ops, Ops+3, CI->getOperand(1)->getType(),
-                    MemsetFCache);
+    ReplaceCallWith("memset", CI, Ops, Ops+3, CI->getOperand(1)->getType());
+    break;
+  }
+  case Intrinsic::sqrt: {
+    ReplaceFPIntrinsicWithCall(CI, "sqrtf", "sqrt", "sqrtl");
+    break;
+  }
+  case Intrinsic::log: {
+    ReplaceFPIntrinsicWithCall(CI, "logf", "log", "logl");
     break;
   }
-  case Intrinsic::sqrt_f32: {
-    static Constant *sqrtfFCache = 0;
-    ReplaceCallWith("sqrtf", CI, CI->op_begin()+1, CI->op_end(),
-                    Type::FloatTy, sqrtfFCache);
+  case Intrinsic::log2: {
+    ReplaceFPIntrinsicWithCall(CI, "log2f", "log2", "log2l");
     break;
   }
-  case Intrinsic::sqrt_f64: {
-    static Constant *sqrtFCache = 0;
-    ReplaceCallWith("sqrt", CI, CI->op_begin()+1, CI->op_end(),
-                    Type::DoubleTy, sqrtFCache);
+  case Intrinsic::log10: {
+    ReplaceFPIntrinsicWithCall(CI, "log10f", "log10", "log10l");
     break;
   }
+  case Intrinsic::exp: {
+    ReplaceFPIntrinsicWithCall(CI, "expf", "exp", "expl");
+    break;
+  }
+  case Intrinsic::exp2: {
+    ReplaceFPIntrinsicWithCall(CI, "exp2f", "exp2", "exp2l");
+    break;
+  }
+  case Intrinsic::pow: {
+    ReplaceFPIntrinsicWithCall(CI, "powf", "pow", "powl");
+    break;
+  }
+  case Intrinsic::flt_rounds:
+     // Lower to "round to the nearest"
+     if (CI->getType() != Type::getVoidTy(Context))
+       CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 1));
+     break;
   }
 
   assert(CI->use_empty() &&