The ARC language-specific optimizer. Credit to Dan Gohman.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@133108 91177308-0d34-0410-b5e6-96231b3b80d8

28 files changed, 6947 insertions, 0 deletions

diff --git a/include/llvm/Analysis/Passes.h b/include/llvm/Analysis/Passes.h
index 0eff75fe2f..a22bd12dec 100644
--- a/include/llvm/Analysis/Passes.h
+++ b/include/llvm/Analysis/Passes.h
@@ -88,6 +88,13 @@ namespace llvm {
 
   //===--------------------------------------------------------------------===//
   //
+  // createObjCARCAliasAnalysisPass - This pass implements ObjC-ARC-based
+  // alias analysis.
+  //
+  ImmutablePass *createObjCARCAliasAnalysisPass();
+
+  //===--------------------------------------------------------------------===//
+  //
   // createProfileLoaderPass - This pass loads information from a profile dump
   // file.
   //
diff --git a/include/llvm/InitializePasses.h b/include/llvm/InitializePasses.h
index 372eaba74a..de58252117 100644
--- a/include/llvm/InitializePasses.h
+++ b/include/llvm/InitializePasses.h
@@ -160,6 +160,10 @@ void initializeModuleDebugInfoPrinterPass(PassRegistry&);
 void initializeNoAAPass(PassRegistry&);
 void initializeNoProfileInfoPass(PassRegistry&);
 void initializeNoPathProfileInfoPass(PassRegistry&);
+void initializeObjCARCAliasAnalysisPass(PassRegistry&);
+void initializeObjCARCExpandPass(PassRegistry&);
+void initializeObjCARCContractPass(PassRegistry&);
+void initializeObjCARCOptPass(PassRegistry&);
 void initializeOptimalEdgeProfilerPass(PassRegistry&);
 void initializeOptimizePHIsPass(PassRegistry&);
 void initializePEIPass(PassRegistry&);
diff --git a/include/llvm/LinkAllPasses.h b/include/llvm/LinkAllPasses.h
index c02e1615de..c2ea8efb9a 100644
--- a/include/llvm/LinkAllPasses.h
+++ b/include/llvm/LinkAllPasses.h
@@ -97,6 +97,10 @@ namespace {
       (void) llvm::createLowerSwitchPass();
       (void) llvm::createNoAAPass();
       (void) llvm::createNoProfileInfoPass();
+      (void) llvm::createObjCARCAliasAnalysisPass();
+      (void) llvm::createObjCARCExpandPass();
+      (void) llvm::createObjCARCContractPass();
+      (void) llvm::createObjCARCOptPass();
       (void) llvm::createProfileEstimatorPass();
       (void) llvm::createProfileVerifierPass();
       (void) llvm::createPathProfileVerifierPass();
diff --git a/include/llvm/Support/PassManagerBuilder.h b/include/llvm/Support/PassManagerBuilder.h
index 513bb887a0..31624db813 100644
--- a/include/llvm/Support/PassManagerBuilder.h
+++ b/include/llvm/Support/PassManagerBuilder.h
@@ -188,6 +188,7 @@ public:
       MPM.add(createArgumentPromotionPass());   // Scalarize uninlined fn args
     
     // Start of function pass.
+    MPM.add(createObjCARCExpandPass());         // Canonicalize ObjC ARC code.
     // Break up aggregate allocas, using SSAUpdater.
     MPM.add(createScalarReplAggregatesPass(-1, false));
     MPM.add(createEarlyCSEPass());              // Catch trivial redundancies
@@ -223,6 +224,7 @@ public:
     MPM.add(createJumpThreadingPass());         // Thread jumps
     MPM.add(createCorrelatedValuePropagationPass());
     MPM.add(createDeadStoreEliminationPass());  // Delete dead stores
+    MPM.add(createObjCARCOptPass());            // Objective-C ARC optimizations.
     MPM.add(createAggressiveDCEPass());         // Delete dead instructions
     MPM.add(createCFGSimplificationPass());     // Merge & remove BBs
     MPM.add(createInstructionCombiningPass());  // Clean up after everything.
diff --git a/include/llvm/Transforms/Scalar.h b/include/llvm/Transforms/Scalar.h
index de46a8d984..e830435751 100644
--- a/include/llvm/Transforms/Scalar.h
+++ b/include/llvm/Transforms/Scalar.h
@@ -338,6 +338,24 @@ Pass *createCorrelatedValuePropagationPass();
 
 //===----------------------------------------------------------------------===//
 //
+// ObjCARCExpand - ObjC ARC preliminary simplifications.
+//
+Pass *createObjCARCExpandPass();
+
+//===----------------------------------------------------------------------===//
+//
+// ObjCARCContract - Late ObjC ARC cleanups.
+//
+Pass *createObjCARCContractPass();
+
+//===----------------------------------------------------------------------===//
+//
+// ObjCARCOpt - ObjC ARC optimization.
+//
+Pass *createObjCARCOptPass();
+
+//===----------------------------------------------------------------------===//
+//
 // InstructionSimplifier - Remove redundant instructions.
 //
 FunctionPass *createInstructionSimplifierPass();
diff --git a/lib/CodeGen/LLVMTargetMachine.cpp b/lib/CodeGen/LLVMTargetMachine.cpp
index 589d0a9a67..b98fbed695 100644
--- a/lib/CodeGen/LLVMTargetMachine.cpp
+++ b/lib/CodeGen/LLVMTargetMachine.cpp
@@ -303,6 +303,10 @@ bool LLVMTargetMachine::addCommonCodeGenPasses(PassManagerBase &PM,
   if (!DisableVerify)
     PM.add(createVerifierPass());
 
+  // Simplify ObjC ARC code. This is done late because it makes re-optimization
+  // difficult.
+  PM.add(createObjCARCContractPass());
+
   // Run loop strength reduction before anything else.
   if (OptLevel != CodeGenOpt::None && !DisableLSR) {
     PM.add(createLoopStrengthReducePass(getTargetLowering()));
diff --git a/lib/Transforms/Scalar/CMakeLists.txt b/lib/Transforms/Scalar/CMakeLists.txt
index fcf914f8ba..c223da60e0 100644
--- a/lib/Transforms/Scalar/CMakeLists.txt
+++ b/lib/Transforms/Scalar/CMakeLists.txt
@@ -20,6 +20,7 @@ add_llvm_library(LLVMScalarOpts
   LoopUnswitch.cpp
   LowerAtomic.cpp
   MemCpyOptimizer.cpp
+  ObjCARC.cpp
   Reassociate.cpp
   Reg2Mem.cpp
   SCCP.cpp
diff --git a/lib/Transforms/Scalar/ObjCARC.cpp b/lib/Transforms/Scalar/ObjCARC.cpp
new file mode 100644
index 0000000000..e65e2853c2
--- /dev/null
+++ b/lib/Transforms/Scalar/ObjCARC.cpp
@@ -0,0 +1,3520 @@
+//===- ObjCARC.cpp - ObjC ARC Optimization --------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines ObjC ARC optimizations. ARC stands for
+// Automatic Reference Counting and is a system for managing reference counts
+// for objects in Objective C.
+//
+// The optimizations performed include elimination of redundant, partially
+// redundant, and inconsequential reference count operations, elimination of
+// redundant weak pointer operations, pattern-matching and replacement of
+// low-level operations into higher-level operations, and numerous minor
+// simplifications.
+//
+// This file also defines a simple ARC-aware AliasAnalysis.
+//
+// WARNING: This file knows about certain library functions. It recognizes them
+// by name, and hardwires knowedge of their semantics.
+//
+// WARNING: This file knows about how certain Objective-C library functions are
+// used. Naive LLVM IR transformations which would otherwise be
+// behavior-preserving may break these assumptions.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "objc-arc"
+#include "llvm/Function.h"
+#include "llvm/Intrinsics.h"
+#include "llvm/GlobalVariable.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Transforms/Utils/Local.h"
+#include "llvm/Support/CallSite.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/STLExtras.h"
+using namespace llvm;
+
+// A handy option to enable/disable all optimizations in this file.
+static cl::opt<bool> EnableARCOpts("enable-objc-arc-opts", cl::init(true));
+
+//===----------------------------------------------------------------------===//
+// Misc. Utilities
+//===----------------------------------------------------------------------===//
+
+namespace {
+  /// MapVector - An associative container with fast insertion-order
+  /// (deterministic) iteration over its elements. Plus the special
+  /// blot operation.
+  template<class KeyT, class ValueT>
+  class MapVector {
+    /// Map - Map keys to indices in Vector.
+    typedef DenseMap<KeyT, size_t> MapTy;
+    MapTy Map;
+
+    /// Vector - Keys and values.
+    typedef std::vector<std::pair<KeyT, ValueT> > VectorTy;
+    VectorTy Vector;
+
+  public:
+    typedef typename VectorTy::iterator iterator;
+    typedef typename VectorTy::const_iterator const_iterator;
+    iterator begin() { return Vector.begin(); }
+    iterator end() { return Vector.end(); }
+    const_iterator begin() const { return Vector.begin(); }
+    const_iterator end() const { return Vector.end(); }
+
+#ifdef XDEBUG
+    ~MapVector() {
+      assert(Vector.size() >= Map.size()); // May differ due to blotting.
+      for (typename MapTy::const_iterator I = Map.begin(), E = Map.end();
+           I != E; ++I) {
+        assert(I->second < Vector.size());
+        assert(Vector[I->second].first == I->first);
+      }
+      for (typename VectorTy::const_iterator I = Vector.begin(),
+           E = Vector.end(); I != E; ++I)
+        assert(!I->first ||
+               (Map.count(I->first) &&
+                Map[I->first] == size_t(I - Vector.begin())));
+    }
+#endif
+
+    ValueT &operator[](KeyT Arg) {
+      std::pair<typename MapTy::iterator, bool> Pair =
+        Map.insert(std::make_pair(Arg, size_t(0)));
+      if (Pair.second) {
+        Pair.first->second = Vector.size();
+        Vector.push_back(std::make_pair(Arg, ValueT()));
+        return Vector.back().second;
+      }
+      return Vector[Pair.first->second].second;
+    }
+
+    std::pair<iterator, bool>
+    insert(const std::pair<KeyT, ValueT> &InsertPair) {
+      std::pair<typename MapTy::iterator, bool> Pair =
+        Map.insert(std::make_pair(InsertPair.first, size_t(0)));
+      if (Pair.second) {
+        Pair.first->second = Vector.size();
+        Vector.push_back(InsertPair);
+        return std::make_pair(llvm::prior(Vector.end()), true);
+      }
+      return std::make_pair(Vector.begin() + Pair.first->second, false);
+    }
+
+    const_iterator find(KeyT Key) const {
+      typename MapTy::const_iterator It = Map.find(Key);
+      if (It == Map.end()) return Vector.end();
+      return Vector.begin() + It->second;
+    }
+
+    /// blot - This is similar to erase, but instead of removing the element
+    /// from the vector, it just zeros out the key in the vector. This leaves
+    /// iterators intact, but clients must be prepared for zeroed-out keys when
+    /// iterating.
+    void blot(KeyT Key) {
+      typename MapTy::iterator It = Map.find(Key);
+      if (It == Map.end()) return;
+      Vector[It->second].first = KeyT();
+      Map.erase(It);
+    }
+
+    void clear() {
+      Map.clear();
+      Vector.clear();
+    }
+  };
+}
+
+//===----------------------------------------------------------------------===//
+// ARC Utilities.
+//===----------------------------------------------------------------------===//
+
+namespace {
+  /// InstructionClass - A simple classification for instructions.
+  enum InstructionClass {
+    IC_Retain,              ///< objc_retain
+    IC_RetainRV,            ///< objc_retainAutoreleasedReturnValue
+    IC_RetainBlock,         ///< objc_retainBlock
+    IC_Release,             ///< objc_release
+    IC_Autorelease,         ///< objc_autorelease
+    IC_AutoreleaseRV,       ///< objc_autoreleaseReturnValue
+    IC_AutoreleasepoolPush, ///< objc_autoreleasePoolPush
+    IC_AutoreleasepoolPop,  ///< objc_autoreleasePoolPop
+    IC_NoopCast,            ///< objc_retainedObject, etc.
+    IC_FusedRetainAutorelease, ///< objc_retainAutorelease
+    IC_FusedRetainAutoreleaseRV, ///< objc_retainAutoreleaseReturnValue
+    IC_LoadWeakRetained,    ///< objc_loadWeakRetained (primitive)
+    IC_StoreWeak,           ///< objc_storeWeak (primitive)
+    IC_InitWeak,            ///< objc_initWeak (derived)
+    IC_LoadWeak,            ///< objc_loadWeak (derived)
+    IC_MoveWeak,            ///< objc_moveWeak (derived)
+    IC_CopyWeak,            ///< objc_copyWeak (derived)
+    IC_DestroyWeak,         ///< objc_destroyWeak (derived)
+    IC_CallOrUser,          ///< could call objc_release and/or "use" pointers
+    IC_Call,                ///< could call objc_release
+    IC_User,                ///< could "use" a pointer
+    IC_None                 ///< anything else
+  };
+}
+
+/// IsPotentialUse - Test whether the given value is possible a
+/// reference-counted pointer.
+static bool IsPotentialUse(const Value *Op) {
+  // Pointers to static or stack storage are not reference-counted pointers.
+  if (isa<Constant>(Op) || isa<AllocaInst>(Op))
+    return false;
+  // Special arguments are not reference-counted.
+  if (const Argument *Arg = dyn_cast<Argument>(Op))
+    if (Arg->hasByValAttr() ||
+        Arg->hasNestAttr() ||
+        Arg->hasStructRetAttr())
+      return false;
+  // Only consider values with pointer types, and not function pointers.
+  const PointerType *Ty = dyn_cast<PointerType>(Op->getType());
+  if (!Ty || isa<FunctionType>(Ty->getElementType()))
+    return false;
+  // Conservatively assume anything else is a potential use.
+  return true;
+}
+
+/// GetCallSiteClass - Helper for GetInstructionClass. Determines what kind
+/// of construct CS is.
+static InstructionClass GetCallSiteClass(ImmutableCallSite CS) {
+  for (ImmutableCallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end();
+       I != E; ++I)
+    if (IsPotentialUse(*I))
+      return CS.onlyReadsMemory() ? IC_User : IC_CallOrUser;
+
+  return CS.onlyReadsMemory() ? IC_None : IC_Call;
+}
+
+/// GetFunctionClass - Determine if F is one of the special known Functions.
+/// If it isn't, return IC_CallOrUser.
+static InstructionClass GetFunctionClass(const Function *F) {
+  Function::const_arg_iterator AI = F->arg_begin(), AE = F->arg_end();
+
+  // No arguments.
+  if (AI == AE)
+    return StringSwitch<InstructionClass>(F->getName())
+      .Case("objc_autoreleasePoolPush",  IC_AutoreleasepoolPush)
+      .Default(IC_CallOrUser);
+
+  // One argument.
+  const Argument *A0 = AI++;
+  if (AI == AE)
+    // Argument is a pointer.
+    if (const PointerType *PTy = dyn_cast<PointerType>(A0->getType())) {
+      const Type *ETy = PTy->getElementType();
+      // Argument is i8*.
+      if (ETy->isIntegerTy(8))
+        return StringSwitch<InstructionClass>(F->getName())
+          .Case("objc_retain",                IC_Retain)
+          .Case("objc_retainAutoreleasedReturnValue", IC_RetainRV)
+          .Case("objc_retainBlock",           IC_RetainBlock)
+          .Case("objc_release",               IC_Release)
+          .Case("objc_autorelease",           IC_Autorelease)
+          .Case("objc_autoreleaseReturnValue", IC_AutoreleaseRV)
+          .Case("objc_autoreleasePoolPop",    IC_AutoreleasepoolPop)
+          .Case("objc_retainedObject",        IC_NoopCast)
+          .Case("objc_unretainedObject",      IC_NoopCast)
+          .Case("objc_unretainedPointer",     IC_NoopCast)
+          .Case("objc_retain_autorelease",    IC_FusedRetainAutorelease)
+          .Case("objc_retainAutorelease",     IC_FusedRetainAutorelease)
+          .Case("objc_retainAutoreleaseReturnValue",IC_FusedRetainAutoreleaseRV)
+          .Default(IC_CallOrUser);
+
+      // Argument is i8**
+      if (const PointerType *Pte = dyn_cast<PointerType>(ETy))
+        if (Pte->getElementType()->isIntegerTy(8))
+          return StringSwitch<InstructionClass>(F->getName())
+            .Case("objc_loadWeakRetained",      IC_LoadWeakRetained)
+            .Case("objc_loadWeak",              IC_LoadWeak)
+            .Case("objc_destroyWeak",           IC_DestroyWeak)
+            .Default(IC_CallOrUser);
+    }
+
+  // Two arguments, first is i8**.
+  const Argument *A1 = AI++;
+  if (AI == AE)
+    if (const PointerType *PTy = dyn_cast<PointerType>(A0->getType()))
+      if (const PointerType *Pte = dyn_cast<PointerType>(PTy->getElementType()))
+        if (Pte->getElementType()->isIntegerTy(8))
+          if (const PointerType *PTy1 = dyn_cast<PointerType>(A1->getType())) {
+            const Type *ETy1 = PTy1->getElementType();
+            // Second argument is i8*
+            if (ETy1->isIntegerTy(8))
+              return StringSwitch<InstructionClass>(F->getName())
+                     .Case("objc_storeWeak",             IC_StoreWeak)
+                     .Case("objc_initWeak",              IC_InitWeak)
+                     .Default(IC_CallOrUser);
+            // Second argument is i8**.
+            if (const PointerType *Pte1 = dyn_cast<PointerType>(ETy1))
+              if (Pte1->getElementType()->isIntegerTy(8))
+                return StringSwitch<InstructionClass>(F->getName())
+                       .Case("objc_moveWeak",              IC_MoveWeak)
+                       .Case("objc_copyWeak",              IC_CopyWeak)
+                       .Default(IC_CallOrUser);
+          }
+
+  // Anything else.
+  return IC_CallOrUser;
+}
+
+/// GetInstructionClass - Determine what kind of construct V is.
+static InstructionClass GetInstructionClass(const Value *V) {
+  if (const Instruction *I = dyn_cast<Instruction>(V)) {
+    // Any instruction other than bitcast and gep with a pointer operand have a
+    // use of an objc pointer. Bitcasts, GEPs, Selects, PHIs transfer a pointer
+    // to a subsequent use, rather than using it themselves, in this sense.
+    // As a short cut, several other opcodes are known to have no pointer
+    // operands of interest. And ret is never followed by a release, so it's
+    // not interesting to examine.
+    switch (I->getOpcode()) {
+    case Instruction::Call: {
+      const CallInst *CI = cast<CallInst>(I);
+      // Check for calls to special functions.
+      if (const Function *F = CI->getCalledFunction()) {
+        InstructionClass Class = GetFunctionClass(F);
+        if (Class != IC_CallOrUser)
+          return Class;
+
+        // None of the intrinsic functions do objc_release. For intrinsics, the
+        // only question is whether or not they may be users.
+        switch (F->getIntrinsicID()) {
+        case 0: break;
+        case Intrinsic::bswap: case Intrinsic::ctpop:
+        case Intrinsic::ctlz: case Intrinsic::cttz:
+        case Intrinsic::returnaddress: case Intrinsic::frameaddress:
+        case Intrinsic::stacksave: case Intrinsic::stackrestore:
+        case Intrinsic::vastart: case Intrinsic::vacopy: case Intrinsic::vaend:
+        // Don't let dbg info affect our results.
+        case Intrinsic::dbg_declare: case Intrinsic::dbg_value:
+          // Short cut: Some intrinsics obviously don't use ObjC pointers.
+          return IC_None;
+        default:
+          for (Function::const_arg_iterator AI = F->arg_begin(),
+               AE = F->arg_end(); AI != AE; ++AI)
+            if (IsPotentialUse(AI))
+              return IC_User;
+          return IC_None;
+        }
+      }
+      return GetCallSiteClass(CI);
+    }
+    case Instruction::Invoke:
+      return GetCallSiteClass(cast<InvokeInst>(I));
+    case Instruction::BitCast:
+    case Instruction::GetElementPtr:
+    case Instruction::Select: case Instruction::PHI:
+    case Instruction::Ret: case Instruction::Br:
+    case Instruction::Switch: case Instruction::IndirectBr:
+    case Instruction::Alloca: case Instruction::VAArg:
+    case Instruction::Add: case Instruction::FAdd:
+    case Instruction::Sub: case Instruction::FSub:
+    case Instruction::Mul: case Instruction::FMul:
+    case Instruction::SDiv: case Instruction::UDiv: case Instruction::FDiv:
+    case Instruction::SRem: case Instruction::URem: case Instruction::FRem:
+    case Instruction::Shl: case Instruction::LShr: case Instruction::AShr:
+    case Instruction::And: case Instruction::Or: case Instruction::Xor:
+    case Instruction::SExt: case Instruction::ZExt: case Instruction::Trunc:
+    case Instruction::IntToPtr: case Instruction::FCmp:
+    case Instruction::FPTrunc: case Instruction::FPExt:
+    case Instruction::FPToUI: case Instruction::FPToSI:
+    case Instruction::UIToFP: case Instruction::SIToFP:
+    case Instruction::InsertElement: case Instruction::ExtractElement:
+    case Instruction::ShuffleVector:
+    case Instruction::ExtractValue:
+      break;
+    case Instruction::ICmp:
+      // Comparing a pointer with null, or any other constant, isn't an
+      // interesting use, because we don't care what the pointer points to, or
+      // about the values of any other dynamic reference-counted pointers.
+      if (IsPotentialUse(I->getOperand(1)))
+        return IC_User;
+      break;
+    default:
+      // For anything else, check all the operands.
+      for (User::const_op_iterator OI = I->op_begin(), OE = I->op_end();
+           OI != OE; ++OI)
+        if (IsPotentialUse(*OI))
+          return IC_User;
+    }
+  }
+
+  // Otherwise, it's totally inert for ARC purposes.
+  return IC_None;
+}
+
+/// GetBasicInstructionClass - Determine what kind of construct V is. This is
+/// similar to GetInstructionClass except that it only detects objc runtine
+/// calls. This allows it to be faster.
+static InstructionClass GetBasicInstructionClass(const Value *V) {
+  if (const CallInst *CI = dyn_cast<CallInst>(V)) {
+    if (const Function *F = CI->getCalledFunction())
+      return GetFunctionClass(F);
+    // Otherwise, be conservative.
+    return IC_CallOrUser;
+  }
+
+  // Otherwise, be conservative.
+  return IC_User;
+}
+
+/// IsRetain - Test if the the given class is objc_retain or
+/// equivalent.
+static bool IsRetain(InstructionClass Class) {
+  return Class == IC_Retain ||
+         Class == IC_RetainRV;
+}
+
+/// IsAutorelease - Test if the the given class is objc_autorelease or
+/// equivalent.
+static bool IsAutorelease(InstructionClass Class) {
+  return Class == IC_Autorelease ||
+         Class == IC_AutoreleaseRV;
+}
+
+/// IsForwarding - Test if the given class represents instructions which return
+/// their argument verbatim.
+static bool IsForwarding(InstructionClass Class) {
+  // objc_retainBlock technically doesn't always return its argument
+  // verbatim, but it doesn't matter for our purposes here.
+  return Class == IC_Retain ||
+         Class == IC_RetainRV ||
+         Class == IC_Autorelease ||
+         Class == IC_AutoreleaseRV ||
+         Class == IC_RetainBlock ||
+         Class == IC_NoopCast;
+}
+
+/// IsNoopOnNull - Test if the given class represents instructions which do
+/// nothing if passed a null pointer.
+static bool IsNoopOnNull(InstructionClass Class) {
+  return Class == IC_Retain ||
+         Class == IC_RetainRV ||
+         Class == IC_Release ||
+         Class == IC_Autorelease ||
+         Class == IC_AutoreleaseRV ||
+         Class == IC_RetainBlock;
+}
+
+/// IsAlwaysTail - Test if the given class represents instructions which are
+/// always safe to mark with the "tail" keyword.
+static bool IsAlwaysTail(InstructionClass Class) {
+  // IC_RetainBlock may be given a stack argument.
+  return Class == IC_Retain ||
+         Class == IC_RetainRV ||
+         Class == IC_Autorelease ||
+         Class == IC_AutoreleaseRV;
+}
+
+/// IsNoThrow - Test if the given class represents instructions which are always
+/// safe to mark with the nounwind attribute..
+static bool IsNoThrow(InstructionClass Class) {
+  return Class == IC_Retain ||
+         Class == IC_RetainRV ||
+         Class == IC_RetainBlock ||
+         Class == IC_Release ||
+         Class == IC_Autorelease ||
+         Class == IC_AutoreleaseRV ||
+         Class == IC_AutoreleasepoolPush ||
+         Class == IC_AutoreleasepoolPop;
+}
+
+/// EraseInstruction - Erase the given instruction. ObjC calls return their
+/// argument verbatim, so if it's such a call and the return value has users,
+/// replace them with the argument value.
+static void EraseInstruction(Instruction *CI) {
+  Value *OldArg = cast<CallInst>(CI)->getArgOperand(0);
+
+  bool Unused = CI->use_empty();
+
+  if (!Unused) {
+    // Replace the return value with the argument.
+    assert(IsForwarding(GetBasicInstructionClass(CI)) &&
+           "Can't delete non-forwarding instruction with users!");
+    CI->replaceAllUsesWith(OldArg);
+  }
+
+  CI->eraseFromParent();
+
+  if (Unused)
+    RecursivelyDeleteTriviallyDeadInstructions(OldArg);
+}
+
+/// GetUnderlyingObjCPtr - This is a wrapper around getUnderlyingObject which
+/// also knows how to look through objc_retain and objc_autorelease calls, which
+/// we know to return their argument verbatim.
+static const Value *GetUnderlyingObjCPtr(const Value *V) {
+  for (;;) {
+    V = GetUnderlyingObject(V);
+    if (!IsForwarding(GetBasicInstructionClass(V)))
+      break;
+    V = cast<CallInst>(V)->getArgOperand(0);
+  }
+
+  return V;
+}
+
+/// StripPointerCastsAndObjCCalls - This is a wrapper around
+/// Value::stripPointerCasts which also knows how to look through objc_retain
+/// and objc_autorelease calls, which we know to return their argument verbatim.
+static const Value *StripPointerCastsAndObjCCalls(const Value *V) {
+  for (;;) {
+    V = V->stripPointerCasts();
+    if (!IsForwarding(GetBasicInstructionClass(V)))
+      break;
+    V = cast<CallInst>(V)->getArgOperand(0);
+  }
+  return V;
+}
+
+/// StripPointerCastsAndObjCCalls - This is a wrapper around
+/// Value::stripPointerCasts which also knows how to look through objc_retain
+/// and objc_autorelease calls, which we know to return their argument verbatim.
+static Value *StripPointerCastsAndObjCCalls(Value *V) {
+  for (;;) {
+    V = V->stripPointerCasts();
+    if (!IsForwarding(GetBasicInstructionClass(V)))
+      break;
+    V = cast<CallInst>(V)->getArgOperand(0);
+  }
+  return V;
+}
+
+/// GetObjCArg - Assuming the given instruction is one of the special calls such
+/// as objc_retain or objc_release, return the argument value, stripped of no-op
+/// casts and forwarding calls.
+static Value *GetObjCArg(Value *Inst) {
+  return StripPointerCastsAndObjCCalls(cast<CallInst>(Inst)->getArgOperand(0));
+}
+
+/// IsObjCIdentifiedObject - This is similar to AliasAnalysis'
+/// isObjCIdentifiedObject, except that it uses special knowledge of
+/// ObjC conventions...
+static bool IsObjCIdentifiedObject(const Value *V) {
+  // Assume that call results and arguments have their own "provenance".
+  // Constants (including GlobalVariables) and Allocas are never
+  // reference-counted.
+  if (isa<CallInst>(V) || isa<InvokeInst>(V) ||
+      isa<Argument>(V) || isa<Constant>(V) ||
+      isa<AllocaInst>(V))
+    return true;
+
+  if (const LoadInst *LI = dyn_cast<LoadInst>(V)) {
+    const Value *Pointer =
+      StripPointerCastsAndObjCCalls(LI->getPointerOperand());
+    if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(Pointer)) {
+      StringRef Name = GV->getName();
+      // These special variables are known to hold values which are not
+      // reference-counted pointers.
+      if (Name.startswith("\01L_OBJC_SELECTOR_REFERENCES_") ||
+          Name.startswith("\01L_OBJC_CLASSLIST_REFERENCES_") ||
+          Name.startswith("\01L_OBJC_CLASSLIST_SUP_REFS_$_") ||
+          Name.startswith("\01L_OBJC_METH_VAR_NAME_") ||
+          Name.startswith("\01l_objc_msgSend_fixup_"))
+        return true;
+    }
+  }
+
+  return false;
+}
+
+/// FindSingleUseIdentifiedObject - This is similar to
+/// StripPointerCastsAndObjCCalls but it stops as soon as it finds a value
+/// with multiple uses.
+static const Value *FindSingleUseIdentifiedObject(const Value *Arg) {
+  if (Arg->hasOneUse()) {
+    if (const BitCastInst *BC = dyn_cast<BitCastInst>(Arg))
+      return FindSingleUseIdentifiedObject(BC->getOperand(0));
+    if (const GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Arg))
+      if (GEP->hasAllZeroIndices())
+        return FindSingleUseIdentifiedObject(GEP->getPointerOperand());
+    if (IsForwarding(GetBasicInstructionClass(Arg)))
+      return FindSingleUseIdentifiedObject(
+               cast<CallInst>(Arg)->getArgOperand(0));
+    if (!IsObjCIdentifiedObject(Arg))
+      return 0;
+    return Arg;
+  }
+
+  // If we found an identifiable object but it has multiple uses, but they
+  // are trivial uses, we can still consider this to be a single-use
+  // value.
+  if (IsObjCIdentifiedObject(Arg)) {
+    for (Value::const_use_iterator UI = Arg->use_begin(), UE = Arg->use_end();
+         UI != UE; ++UI) {
+      const User *U = *UI;
+      if (!U->use_empty() || StripPointerCastsAndObjCCalls(U) != Arg)
+         return 0;
+    }
+
+    return Arg;
+  }
+
+  return 0;
+}
+
+//===----------------------------------------------------------------------===//
+// ARC AliasAnalysis.
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Pass.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/Passes.h"
+
+namespace {
+  /// ObjCARCAliasAnalysis - This is a simple alias analysis
+  /// implementation that uses knowledge of ARC constructs to answer queries.
+  ///
+  /// TODO: This class could be generalized to know about other ObjC-specific
+  /// tricks. Such as knowing that ivars in the non-fragile ABI are non-aliasing
+  /// even though their offsets are dynamic.
+  class ObjCARCAliasAnalysis : public ImmutablePass,
+                               public AliasAnalysis {
+  public:
+    static char ID; // Class identification, replacement for typeinfo
+    ObjCARCAliasAnalysis() : ImmutablePass(ID) {
+      initializeObjCARCAliasAnalysisPass(*PassRegistry::getPassRegistry());
+    }
+
+  private:
+    virtual void initializePass() {
+      InitializeAliasAnalysis(this);
+    }
+
+    /// getAdjustedAnalysisPointer - This method is used when a pass implements
+    /// an analysis interface through multiple inheritance.  If needed, it
+    /// should override this to adjust the this pointer as needed for the
+    /// specified pass info.
+    virtual void *getAdjustedAnalysisPointer(const void *PI) {
+      if (PI == &AliasAnalysis::ID)
+        return (AliasAnalysis*)this;
+      return this;
+    }
+
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+    virtual AliasResult alias(const Location &LocA, const Location &LocB);
+    virtual bool pointsToConstantMemory(const Location &Loc, bool OrLocal);
+    virtual ModRefBehavior getModRefBehavior(ImmutableCallSite CS);
+    virtual ModRefBehavior getModRefBehavior(const Function *F);
+    virtual ModRefResult getModRefInfo(ImmutableCallSite CS,
+                                       const Location &Loc);
+    virtual ModRefResult getModRefInfo(ImmutableCallSite CS1,
+                                       ImmutableCallSite CS2);
+  };
+}  // End of anonymous namespace
+
+// Register this pass...
+char ObjCARCAliasAnalysis::ID = 0;
+INITIALIZE_AG_PASS(ObjCARCAliasAnalysis, AliasAnalysis, "objc-arc-aa",
+                   "ObjC-ARC-Based Alias Analysis", false, true, false)
+
+ImmutablePass *llvm::createObjCARCAliasAnalysisPass() {
+  return new ObjCARCAliasAnalysis();
+}
+
+void
+ObjCARCAliasAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.setPreservesAll();
+  AliasAnalysis::getAnalysisUsage(AU);
+}
+
+AliasAnalysis::AliasResult
+ObjCARCAliasAnalysis::alias(const Location &LocA, const Location &LocB) {
+  if (!EnableARCOpts)
+    return AliasAnalysis::alias(LocA, LocB);
+
+  // First, strip off no-ops, including ObjC-specific no-ops, and try making a
+  // precise alias query.
+  const Value *SA = StripPointerCastsAndObjCCalls(LocA.Ptr);
+  const Value *SB = StripPointerCastsAndObjCCalls(LocB.Ptr);
+  AliasResult Result =
+    AliasAnalysis::alias(Location(SA, LocA.Size, LocA.TBAATag),
+                         Location(SB, LocB.Size, LocB.TBAATag));
+  if (Result != MayAlias)
+    return Result;
+
+  // If that failed, climb to the underlying object, including climbing through
+  // ObjC-specific no-ops, and try making an imprecise alias query.
+  const Value *UA = GetUnderlyingObjCPtr(SA);
+  const Value *UB = GetUnderlyingObjCPtr(SB);
+  if (UA != SA || UB != SB) {
+    Result = AliasAnalysis::alias(Location(UA), Location(UB));
+    // We can't use MustAlias or PartialAlias results here because
+    // GetUnderlyingObjCPtr may return an offsetted pointer value.
+    if (Result == NoAlias)
+      return NoAlias;
+  }
+
+  // If that failed, fail. We don't need to chain here, since that's covered
+  // by the earlier precise query.
+  return MayAlias;
+}
+
+bool
+ObjCARCAliasAnalysis::pointsToConstantMemory(const Location &Loc,
+                                             bool OrLocal) {
+  if (!EnableARCOpts)
+    return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal);
+
+  // First, strip off no-ops, including ObjC-specific no-ops, and try making
+  // a precise alias query.
+  const Value *S = StripPointerCastsAndObjCCalls(Loc.Ptr);
+  if (AliasAnalysis::pointsToConstantMemory(Location(S, Loc.Size, Loc.TBAATag),
+                                            OrLocal))
+    return true;
+
+  // If that failed, climb to the underlying object, including climbing through
+  // ObjC-specific no-ops, and try making an imprecise alias query.
+  const Value *U = GetUnderlyingObjCPtr(S);
+  if (U != S)
+    return AliasAnalysis::pointsToConstantMemory(Location(U), OrLocal);
+
+  // If that failed, fail. We don't need to chain here, since that's covered
+  // by the earlier precise query.
+  return false;
+}
+
+AliasAnalysis::ModRefBehavior
+ObjCARCAliasAnalysis::getModRefBehavior(ImmutableCallSite CS) {
+  // We have nothing to do. Just chain to the next AliasAnalysis.
+  return AliasAnalysis::getModRefBehavior(CS);
+}
+
+AliasAnalysis::ModRefBehavior
+ObjCARCAliasAnalysis::getModRefBehavior(const Function *F) {
+  if (!EnableARCOpts)
+    return AliasAnalysis::getModRefBehavior(F);
+
+  switch (GetFunctionClass(F)) {
+  case IC_NoopCast:
+    return DoesNotAccessMemory;
+  default:
+    break;
+  }
+
+  return AliasAnalysis::getModRefBehavior(F);
+}
+
+AliasAnalysis::ModRefResult
+ObjCARCAliasAnalysis::getModRefInfo(ImmutableCallSite CS, const Location &Loc) {
+  if (!EnableARCOpts)
+    return AliasAnalysis::getModRefInfo(CS, Loc);
+
+  switch (GetBasicInstructionClass(CS.getInstruction())) {
+  case IC_Retain:
+  case IC_RetainRV:
+  case IC_RetainBlock:
+  case IC_Autorelease:
+  case IC_AutoreleaseRV:
+  case IC_NoopCast:
+  case IC_AutoreleasepoolPush:
+  case IC_FusedRetainAutorelease:
+  case IC_FusedRetainAutoreleaseRV:
+    // These functions don't access any memory visible to the compiler.
+    return NoModRef;
+  default:
+    break;
+  }
+
+  return AliasAnalysis::getModRefInfo(CS, Loc);
+}
+
+AliasAnalysis::ModRefResult
+ObjCARCAliasAnalysis::getModRefInfo(ImmutableCallSite CS1,
+                                    ImmutableCallSite CS2) {
+  // TODO: Theoretically we could check for dependencies between objc_* calls
+  // and OnlyAccessesArgumentPointees calls or other well-behaved calls.
+  return AliasAnalysis::getModRefInfo(CS1, CS2);
+}
+
+//===----------------------------------------------------------------------===//
+// ARC expansion.
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Support/InstIterator.h"
+#include "llvm/Transforms/Scalar.h"
+
+namespace {
+  /// ObjCARCExpand - Early ARC transformations.
+  class ObjCARCExpand : public FunctionPass {
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+    virtual bool runOnFunction(Function &F);
+
+  public:
+    static char ID;
+    ObjCARCExpand() : FunctionPass(ID) {
+      initializeObjCARCExpandPass(*PassRegistry::getPassRegistry());
+    }
+  };
+}
+
+char ObjCARCExpand::ID = 0;
+INITIALIZE_PASS(ObjCARCExpand,
+                "objc-arc-expand", "ObjC ARC expansion", false, false)
+
+Pass *llvm::createObjCARCExpandPass() {
+  return new ObjCARCExpand();
+}
+
+void ObjCARCExpand::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.setPreservesCFG();
+}
+
+bool ObjCARCExpand::runOnFunction(Function &F) {
+  if (!EnableARCOpts)
+    return false;
+
+  bool Changed = false;
+
+  for (inst_iterator I = inst_begin(&F), E = inst_end(&F); I != E; ++I) {
+    Instruction *Inst = &*I;
+
+    switch (GetBasicInstructionClass(Inst)) {
+    case IC_Retain:
+    case IC_RetainRV:
+    case IC_Autorelease:
+    case IC_AutoreleaseRV:
+    case IC_FusedRetainAutorelease:
+    case IC_FusedRetainAutoreleaseRV:
+      // These calls return their argument verbatim, as a low-level
+      // optimization. However, this makes high-level optimizations
+      // harder. Undo any uses of this optimization that the front-end
+      // emitted here. We'll redo them in a later pass.
+      Changed = true;
+      Inst->replaceAllUsesWith(cast<CallInst>(Inst)->getArgOperand(0));
+      break;
+    default:
+      break;
+    }
+  }
+
+  return Changed;
+}
+
+//===----------------------------------------------------------------------===//
+// ARC optimization.
+//===----------------------------------------------------------------------===//
+
+// TODO: On code like this:
+//
+// objc_retain(%x)
+// stuff_that_cannot_release()
+// objc_autorelease(%x)
+// stuff_that_cannot_release()
+// objc_retain(%x)
+// stuff_that_cannot_release()
+// objc_autorelease(%x)
+//
+// The second retain and autorelease can be deleted.
+
+// TODO: It should be possible to delete
+// objc_autoreleasePoolPush and objc_autoreleasePoolPop
+// pairs if nothing is actually autoreleased between them. Also, autorelease
+// calls followed by objc_autoreleasePoolPop calls (perhaps in ObjC++ code
+// after inlining) can be turned into plain release calls.
+
+// TODO: Critical-edge splitting. If the optimial insertion point is
+// a critical edge, the current algorithm has to fail, because it doesn't
+// know how to split edges. It should be possible to make the optimizer
+// think in terms of edges, rather than blocks, and then split critical
+// edges on demand.
+
+// TODO: OptimizeSequences could generalized to be Interprocedural.
+
+// TODO: Recognize that a bunch of other objc runtime calls have
+// non-escaping arguments and non-releasing arguments, and may be
+// non-autoreleasing.
+
+// TODO: Sink autorelease calls as far as possible. Unfortunately we
+// usually can't sink them past other calls, which would be the main
+// case where it would be useful.
+
+#include "llvm/GlobalAlias.h"
+#include "llvm/Module.h"
+#include "llvm/Constants.h"
+#include "llvm/LLVMContext.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/ADT/PostOrderIterator.h"
+#include "llvm/ADT/Statistic.h"
+
+STATISTIC(NumNoops,       "Number of no-op objc calls eliminated");
+STATISTIC(NumPartialNoops, "Number of partially no-op objc calls eliminated");
+STATISTIC(NumAutoreleases,"Number of autoreleases converted to releases");
+STATISTIC(NumRets,        "Number of return value forwarding "
+                          "retain+autoreleaes eliminated");
+STATISTIC(NumRRs,         "Number of retain+release paths eliminated");
+STATISTIC(NumPeeps,       "Number of calls peephole-optimized");
+
+namespace {
+  /// ProvenanceAnalysis - This is similar to BasicAliasAnalysis, and it
+  /// uses many of the same techniques, except it uses special ObjC-specific
+  /// reasoning about pointer relationships.
+  class ProvenanceAnalysis {
+    AliasAnalysis *AA;
+
+    typedef std::pair<const Value *, const Value *> ValuePairTy;
+    typedef DenseMap<ValuePairTy, bool> CachedResultsTy;
+    CachedResultsTy CachedResults;
+
+    bool relatedCheck(const Value *A, const Value *B);
+    bool relatedSelect(const SelectInst *A, const Value *B);
+    bool relatedPHI(const PHINode *A, const Value *B);
+
+    // Do not implement.
+    void operator=(const ProvenanceAnalysis &);
+    ProvenanceAnalysis(const ProvenanceAnalysis &);
+
+  public:
+    ProvenanceAnalysis() {}
+
+    void setAA(AliasAnalysis *aa) { AA = aa; }
+
+    AliasAnalysis *getAA() const { return AA; }
+
+    bool related(const Value *A, const Value *B);
+
+    void clear() {
+      CachedResults.clear();
+    }
+  };
+}
+
+bool ProvenanceAnalysis::relatedSelect(const SelectInst *A, const Value *B) {
+  // If the values are Selects with the same condition, we can do a more precise
+  // check: just check for relations between the values on corresponding arms.
+  if (const SelectInst *SB = dyn_cast<SelectInst>(B))
+    if (A->getCondition() == SB->getCondition()) {
+      if (related(A->getTrueValue(), SB->getTrueValue()))
+        return true;
+      if (related(A->getFalseValue(), SB->getFalseValue()))
+        return true;
+      return false;
+    }
+
+  // Check both arms of the Select node individually.
+  if (related(A->getTrueValue(), B))
+    return true;
+  if (related(A->getFalseValue(), B))
+    return true;
+
+  // The arms both checked out.
+  return false;
+}
+
+bool ProvenanceAnalysis::relatedPHI(const PHINode *A, const Value *B) {
+  // If the values are PHIs in the same block, we can do a more precise as well
+  // as efficient check: just check for relations between the values on
+  // corresponding edges.
+  if (const PHINode *PNB = dyn_cast<PHINode>(B))
+    if (PNB->getParent() == A->getParent()) {
+      for (unsigned i = 0, e = A->getNumIncomingValues(); i != e; ++i)
+        if (related(A->getIncomingValue(i),
+                    PNB->getIncomingValueForBlock(A->getIncomingBlock(i))))
+          return true;
+      return false;
+    }
+
+  // Check each unique source of the PHI node against B.
+  SmallPtrSet<const Value *, 4> UniqueSrc;
+  for (unsigned i = 0, e = A->getNumIncomingValues(); i != e; ++i) {
+    const Value *PV1 = A->getIncomingValue(i);
+    if (UniqueSrc.insert(PV1) && related(PV1, B))
+      return true;
+  }
+
+  // All of the arms checked out.
+  return false;
+}
+
+/// isStoredObjCPointer - Test if the value of P, or any value covered by its
+/// provenance, is ever stored within the function (not counting callees).
+static bool isStoredObjCPointer(const Value *P) {
+  SmallPtrSet<const Value *, 8> Visited;
+  SmallVector<const Value *, 8> Worklist;
+  Worklist.push_back(P);
+  Visited.insert(P);
+  do {
+    P = Worklist.pop_back_val();
+    for (Value::const_use_iterator UI = P->use_begin(), UE = P->use_end();
+         UI != UE; ++UI) {
+      const User *Ur = *UI;
+      if (isa<StoreInst>(Ur)) {
+        if (UI.getOperandNo() == 0)
+          // The pointer is stored.
+          return true;
+        // The pointed is stored through.
+        continue;
+      }
+      if (isa<CallInst>(Ur))
+        // The pointer is passed as an argument, ignore this.
+        continue;
+      if (isa<PtrToIntInst>(P))
+        // Assume the worst.
+        return true;
+      if (Visited.insert(Ur))
+        Worklist.push_back(Ur);
+    }
+  } while (!Worklist.empty());
+
+  // Everything checked out.
+  return false;
+}
+
+bool ProvenanceAnalysis::relatedCheck(const Value *A, const Value *B) {
+  // Skip past provenance pass-throughs.
+  A = GetUnderlyingObjCPtr(A);
+  B = GetUnderlyingObjCPtr(B);
+
+  // Quick check.
+  if (A == B)
+    return true;
+
+  // Ask regular AliasAnalysis, for a first approximation.
+  switch (AA->alias(A, B)) {
+  case AliasAnalysis::NoAlias:
+    return false;
+  case AliasAnalysis::MustAlias:
+  case AliasAnalysis::PartialAlias:
+    return true;
+  case AliasAnalysis::MayAlias:
+    break;
+  }
+
+  bool AIsIdentified = IsObjCIdentifiedObject(A);
+  bool BIsIdentified = IsObjCIdentifiedObject(B);
+
+  // An ObjC-Identified object can't alias a load if it is never locally stored.
+  if (AIsIdentified) {
+    if (BIsIdentified) {
+      // If both pointers have provenance, they can be directly compared.
+      if (A != B)
+        return false;
+    } else {
+      if (isa<LoadInst>(B))
+        return isStoredObjCPointer(A);
+    }
+  } else {
+    if (BIsIdentified && isa<LoadInst>(A))
+      return isStoredObjCPointer(B);
+  }
+
+   // Special handling for PHI and Select.
+  if (const PHINode *PN = dyn_cast<PHINode>(A))
+    return relatedPHI(PN, B);
+  if (const PHINode *PN = dyn_cast<PHINode>(B))
+    return relatedPHI(PN, A);
+  if (const SelectInst *S = dyn_cast<SelectInst>(A))
+    return relatedSelect(S, B);
+  if (const SelectInst *S = dyn_cast<SelectInst>(B))
+    return relatedSelect(S, A);
+
+  // Conservative.
+  return true;
+}
+
+bool ProvenanceAnalysis::related(const Value *A, const Value *B) {
+  // Begin by inserting a conservative value into the map. If the insertion
+  // fails, we have the answer already. If it succeeds, leave it there until we
+  // compute the real answer to guard against recursive queries.
+  if (A > B) std::swap(A, B);
+  std::pair<CachedResultsTy::iterator, bool> Pair =
+    CachedResults.insert(std::make_pair(ValuePairTy(A, B), true));
+  if (!Pair.second)
+    return Pair.first->second;
+
+  bool Result = relatedCheck(A, B);
+  CachedResults[ValuePairTy(A, B)] = Result;
+  return Result;
+}
+
+namespace {
+  // Sequence - A sequence of states that a pointer may go through in which an
+  // objc_retain and objc_release are actually needed.
+  enum Sequence {
+    S_None,
+    S_Retain,         ///< objc_retain(x)
+    S_CanRelease,     ///< foo(x) -- x could possibly see a ref count decrement
+    S_Use,            ///< any use of x
+    S_Stop,           ///< like S_Release, but code motion is stopped
+    S_Release,        ///< objc_release(x)
+    S_MovableRelease  ///< objc_release(x), !clang.imprecise_release
+  };
+}
+
+static Sequence MergeSeqs(Sequence A, Sequence B, bool TopDown) {
+  // The easy cases.
+  if (A == B)
+    return A;
+  if (A == S_None || B == S_None)
+    return S_None;
+
+  // Note that we can't merge S_CanRelease and S_Use.
+  if (A > B) std::swap(A, B);
+  if (TopDown) {
+    // Choose the side which is further along in the sequence.
+    if (A == S_Retain && (B == S_CanRelease || B == S_Use))
+      return B;
+  } else {
+    // Choose the side which is further along in the sequence.
+    if ((A == S_Use || A == S_CanRelease) &&
+        (B == S_Release || B == S_Stop || B == S_MovableRelease))
+      return A;
+    // If both sides are releases, choose the more conservative one.
+    if (A == S_Stop && (B == S_Release || B == S_MovableRelease))
+      return A;
+    if (A == S_Release && B == S_MovableRelease)
+      return A;
+  }
+
+  return S_None;
+}
+
+namespace {
+  /// RRInfo - Unidirectional information about either a
+  /// retain-decrement-use-release sequence or release-use-decrement-retain
+  /// reverese sequence.
+  struct RRInfo {
+    /// KnownIncremented - After an objc_retain, the reference count of the
+    /// referenced object is known to be positive. Similarly, before an
+    /// objc_release, the reference count of the referenced object is known to
+    /// be positive. If there are retain-release pairs in code regions where the
+    /// retain count is known to be positive, they can be eliminated, regardless
+    /// of any side effects between them.
+    bool KnownIncremented;
+
+    /// IsRetainBlock - True if the Calls are objc_retainBlock calls (as
+    /// opposed to objc_retain calls).
+    bool IsRetainBlock;
+
+    /// IsTailCallRelease - True of the objc_release calls are all marked
+    /// with the "tail" keyword.
+    bool IsTailCallRelease;
+
+    /// ReleaseMetadata - If the Calls are objc_release calls and they all have
+    /// a clang.imprecise_release tag, this is the metadata tag.
+    MDNode *ReleaseMetadata;
+
+    /// Calls - For a top-down sequence, the set of objc_retains or
+    /// objc_retainBlocks. For bottom-up, the set of objc_releases.
+    SmallPtrSet<Instruction *, 2> Calls;
+
+    /// ReverseInsertPts - The set of optimal insert positions for
+    /// moving calls in the opposite sequence.
+    SmallPtrSet<Instruction *, 2> ReverseInsertPts;
+
+    RRInfo() :
+      KnownIncremented(false), IsRetainBlock(false), IsTailCallRelease(false),
+      ReleaseMetadata(0) {}
+
+    void clear();
+  };
+}
+
+void RRInfo::clear() {
+  KnownIncremented = false;
+  IsRetainBlock = false;
+  IsTailCallRelease = false;
+  ReleaseMetadata = 0;
+  Calls.clear();
+  ReverseInsertPts.clear();
+}
+
+namespace {
+  /// PtrState - This class summarizes several per-pointer runtime properties
+  /// which are propogated through the flow graph.
+  class PtrState {
+    /// RefCount - The known minimum number of reference count increments.
+    unsigned RefCount;
+
+    /// Seq - The current position in the sequence.
+    Sequence Seq;
+
+  public:
+    /// RRI - Unidirectional information about the current sequence.
+    /// TODO: Encapsulate this better.
+    RRInfo RRI;
+
+    PtrState() : RefCount(0), Seq(S_None) {}
+
+    void IncrementRefCount() {
+      if (RefCount != UINT_MAX) ++RefCount;
+    }
+
+    void DecrementRefCount() {
+      if (RefCount != 0) --RefCount;
+    }
+
+    void ClearRefCount() {
+      RefCount = 0;
+    }
+
+    bool IsKnownIncremented() const {
+      return RefCount > 0;
+    }
+
+    void SetSeq(Sequence NewSeq) {
+      Seq = NewSeq;
+    }
+
+    void SetSeqToRelease(MDNode *M) {
+      if (Seq == S_None || Seq == S_Use) {
+        Seq = M ? S_MovableRelease : S_Release;
+        RRI.ReleaseMetadata = M;
+      } else if (Seq != S_MovableRelease || RRI.ReleaseMetadata != M) {
+        Seq = S_Release;
+        RRI.ReleaseMetadata = 0;
+      }
+    }
+
+    Sequence GetSeq() const {
+      return Seq;
+    }
+
+    void ClearSequenceProgress() {
+      Seq = S_None;
+      RRI.clear();
+    }
+
+    void Merge(const PtrState &Other, bool TopDown);
+  };
+}
+
+void
+PtrState::Merge(const PtrState &Other, bool TopDown) {
+  Seq = MergeSeqs(Seq, Other.Seq, TopDown);
+  RefCount = std::min(RefCount, Other.RefCount);
+
+  // We can't merge a plain objc_retain with an objc_retainBlock.
+  if (RRI.IsRetainBlock != Other.RRI.IsRetainBlock)
+    Seq = S_None;
+
+  if (Seq == S_None) {
+    RRI.clear();
+  } else {
+    // Conservatively merge the ReleaseMetadata information.
+    if (RRI.ReleaseMetadata != Other.RRI.ReleaseMetadata)
+      RRI.ReleaseMetadata = 0;
+
+    RRI.KnownIncremented = RRI.KnownIncremented && Other.RRI.KnownIncremented;
+    RRI.IsTailCallRelease = RRI.IsTailCallRelease && Other.RRI.IsTailCallRelease;
+    RRI.Calls.insert(Other.RRI.Calls.begin(), Other.RRI.Calls.end());
+    RRI.ReverseInsertPts.insert(Other.RRI.ReverseInsertPts.begin(),
+                                Other.RRI.ReverseInsertPts.end());
+  }
+}
+
+namespace {
+  /// BBState - Per-BasicBlock state.
+  class BBState {
+    /// TopDownPathCount - The number of unique control paths from the entry
+    /// which can reach this block.
+    unsigned TopDownPathCount;
+
+    /// BottomUpPathCount - The number of unique control paths to exits
+    /// from this block.
+    unsigned BottomUpPathCount;
+
+    /// MapTy - A type for PerPtrTopDown and PerPtrBottomUp.
+    typedef MapVector<const Value *, PtrState> MapTy;
+
+    /// PerPtrTopDown - The top-down traversal uses this to record information
+    /// known about a pointer at the bottom of each block.
+    MapTy PerPtrTopDown;
+
+    /// PerPtrBottomUp - The bottom-up traversal uses this to record information
+    /// known about a pointer at the top of each block.
+    MapTy PerPtrBottomUp;
+
+  public:
+    BBState() : TopDownPathCount(0), BottomUpPathCount(0) {}
+
+    typedef MapTy::iterator ptr_iterator;
+    typedef MapTy::const_iterator ptr_const_iterator;
+
+    ptr_iterator top_down_ptr_begin() { return PerPtrTopDown.begin(); }
+    ptr_iterator top_down_ptr_end() { return PerPtrTopDown.end(); }
+    ptr_const_iterator top_down_ptr_begin() const {
+      return PerPtrTopDown.begin();
+    }
+    ptr_const_iterator top_down_ptr_end() const {
+      return PerPtrTopDown.end();
+    }
+
+    ptr_iterator bottom_up_ptr_begin() { return PerPtrBottomUp.begin(); }
+    ptr_iterator bottom_up_ptr_end() { return PerPtrBottomUp.end(); }
+    ptr_const_iterator bottom_up_ptr_begin() const {
+      return PerPtrBottomUp.begin();
+    }
+    ptr_const_iterator bottom_up_ptr_end() const {
+      return PerPtrBottomUp.end();
+    }
+
+    /// SetAsEntry - Mark this block as being an entry block, which has one
+    /// path from the entry by definition.
+    void SetAsEntry() { TopDownPathCount = 1; }
+
+    /// SetAsExit - Mark this block as being an exit block, which has one
+    /// path to an exit by definition.
+    void SetAsExit()  { BottomUpPathCount = 1; }
+
+    PtrState &getPtrTopDownState(const Value *Arg) {
+      return PerPtrTopDown[Arg];
+    }
+
+    PtrState &getPtrBottomUpState(const Value *Arg) {
+      return PerPtrBottomUp[Arg];
+    }
+
+    void clearBottomUpPointers() {
+      PerPtrTopDown.clear();
+    }
+
+    void clearTopDownPointers() {
+      PerPtrTopDown.clear();
+    }
+
+    void InitFromPred(const BBState &Other);
+    void InitFromSucc(const BBState &Other);
+    void MergePred(const BBState &Other);
+    void MergeSucc(const BBState &Other);
+
+    /// GetAllPathCount - Return the number of possible unique paths from an
+    /// entry to an exit which pass through this block. This is only valid
+    /// after both the top-down and bottom-up traversals are complete.
+    unsigned GetAllPathCount() const {
+      return TopDownPathCount * BottomUpPathCount;
+    }
+  };
+}
+
+void BBState::InitFromPred(const BBState &Other) {
+  PerPtrTopDown = Other.PerPtrTopDown;
+  TopDownPathCount = Other.TopDownPathCount;
+}
+
+void BBState::InitFromSucc(const BBState &Other) {
+  PerPtrBottomUp = Other.PerPtrBottomUp;
+  BottomUpPathCount = Other.BottomUpPathCount;
+}
+
+/// MergePred - The top-down traversal uses this to merge information about
+/// predecessors to form the initial state for a new block.
+void BBState::MergePred(const BBState &Other) {
+  // Other.TopDownPathCount can be 0, in which case it is either dead or a
+  // loop backedge. Loop backedges are special.
+  TopDownPathCount += Other.TopDownPathCount;
+
+  // For each entry in the other set, if our set has an entry with the same key,
+  // merge the entries. Otherwise, copy the entry and merge it with an empty
+  // entry.
+  for (ptr_const_iterator MI = Other.top_down_ptr_begin(),
+       ME = Other.top_down_ptr_end(); MI != ME; ++MI) {
+    std::pair<ptr_iterator, bool> Pair = PerPtrTopDown.insert(*MI);
+    Pair.first->second.Merge(Pair.second ? PtrState() : MI->second,
+                             /*TopDown=*/true);
+  }
+
+  // For each entry in our set, if the other set doens't have an entry with the
+  // same key, force it to merge with an empty entry.
+  for (ptr_iterator MI = top_down_ptr_begin(),
+       ME = top_down_ptr_end(); MI != ME; ++MI)
+    if (Other.PerPtrTopDown.find(MI->first) == Other.PerPtrTopDown.end())
+      MI->second.Merge(PtrState(), /*TopDown=*/true);
+}
+
+/// MergeSucc - The bottom-up traversal uses this to merge information about
+/// successors to form the initial state for a new block.
+void BBState::MergeSucc(const BBState &Other) {
+  // Other.BottomUpPathCount can be 0, in which case it is either dead or a
+  // loop backedge. Loop backedges are special.
+  BottomUpPathCount += Other.BottomUpPathCount;
+
+  // For each entry in the other set, if our set has an entry with the
+  // same key, merge the entries. Otherwise, copy the entry and merge
+  // it with an empty entry.
+  for (ptr_const_iterator MI = Other.bottom_up_ptr_begin(),
+       ME = Other.bottom_up_ptr_end(); MI != ME; ++MI) {
+    std::pair<ptr_iterator, bool> Pair = PerPtrBottomUp.insert(*MI);
+    Pair.first->second.Merge(Pair.second ? PtrState() : MI->second,
+                             /*TopDown=*/false);
+  }
+
+  // For each entry in our set, if the other set doens't have an entry
+  // with the same key, force it to merge with an empty entry.
+  for (ptr_iterator MI = bottom_up_ptr_begin(),
+       ME = bottom_up_ptr_end(); MI != ME; ++MI)
+    if (Other.PerPtrBottomUp.find(MI->first) == Other.PerPtrBottomUp.end())
+      MI->second.Merge(PtrState(), /*TopDown=*/false);
+}
+
+namespace {
+  /// ObjCARCOpt - The main ARC optimization pass.
+  class ObjCARCOpt : public FunctionPass {
+    bool Changed;
+    ProvenanceAnalysis PA;
+
+    /// RetainFunc, RelaseFunc - Declarations for objc_retain,
+    /// objc_retainBlock, and objc_release.
+    Function *RetainFunc, *RetainBlockFunc, *RetainRVFunc, *ReleaseFunc,
+             *AutoreleaseFunc;
+
+    /// RetainRVCallee, etc. - Declarations for ObjC runtime
+    /// functions, for use in creating calls to them. These are initialized
+    /// lazily to avoid cluttering up the Module with unused declarations.
+    Constant *RetainRVCallee, *AutoreleaseRVCallee, *ReleaseCallee,
+             *RetainCallee, *AutoreleaseCallee;
+
+    /// UsedInThisFunciton - Flags which determine whether each of the
+    /// interesting runtine functions is in fact used in the current function.
+    unsigned UsedInThisFunction;
+
+    /// ImpreciseReleaseMDKind - The Metadata Kind for clang.imprecise_release
+    /// metadata.
+    unsigned ImpreciseReleaseMDKind;
+
+    Constant *getRetainRVCallee(Module *M);
+    Constant *getAutoreleaseRVCallee(Module *M);
+    Constant *getReleaseCallee(Module *M);
+    Constant *getRetainCallee(Module *M);
+    Constant *getAutoreleaseCallee(Module *M);
+
+    void OptimizeRetainCall(Function &F, Instruction *Retain);
+    bool OptimizeRetainRVCall(Function &F, Instruction *RetainRV);
+    void OptimizeAutoreleaseRVCall(Function &F, Instruction *AutoreleaseRV);
+    void OptimizeIndividualCalls(Function &F);
+
+    void CheckForCFGHazards(const BasicBlock *BB,
+                            DenseMap<const BasicBlock *, BBState> &BBStates,
+                            BBState &MyStates) const;
+    bool VisitBottomUp(BasicBlock *BB,
+                       DenseMap<const BasicBlock *, BBState> &BBStates,
+                       MapVector<Value *, RRInfo> &Retains);
+    bool VisitTopDown(BasicBlock *BB,
+                      DenseMap<const BasicBlock *, BBState> &BBStates,
+                      DenseMap<Value *, RRInfo> &Releases);
+    bool Visit(Function &F,
+               DenseMap<const BasicBlock *, BBState> &BBStates,
+               MapVector<Value *, RRInfo> &Retains,
+               DenseMap<Value *, RRInfo> &Releases);
+
+    void MoveCalls(Value *Arg, RRInfo &RetainsToMove, RRInfo &ReleasesToMove,
+                   MapVector<Value *, RRInfo> &Retains,
+                   DenseMap<Value *, RRInfo> &Releases,
+                   SmallVectorImpl<Instruction *> &DeadInsts);
+
+    bool PerformCodePlacement(DenseMap<const BasicBlock *, BBState> &BBStates,
+                              MapVector<Value *, RRInfo> &Retains,
+                              DenseMap<Value *, RRInfo> &Releases);
+
+    void OptimizeWeakCalls(Function &F);
+
+    bool OptimizeSequences(Function &F);
+
+    void OptimizeReturns(Function &F);
+
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+    virtual bool doInitialization(Module &M);
+    virtual bool runOnFunction(Function &F);
+    virtual void releaseMemory();
+
+  public:
+    static char ID;
+    ObjCARCOpt() : FunctionPass(ID) {
+      initializeObjCARCOptPass(*PassRegistry::getPassRegistry());
+    }
+  };
+}
+
+char ObjCARCOpt::ID = 0;
+INITIALIZE_PASS_BEGIN(ObjCARCOpt,
+                      "objc-arc", "ObjC ARC optimization", false, false)
+INITIALIZE_PASS_DEPENDENCY(ObjCARCAliasAnalysis)
+INITIALIZE_PASS_END(ObjCARCOpt,
+                    "objc-arc", "ObjC ARC optimization", false, false)
+
+Pass *llvm::createObjCARCOptPass() {
+  return new ObjCARCOpt();
+}
+
+void ObjCARCOpt::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.addRequired<ObjCARCAliasAnalysis>();
+  AU.addRequired<AliasAnalysis>();
+  // ARC optimization doesn't currently split critical edges.
+  AU.setPreservesCFG();
+}
+
+Constant *ObjCARCOpt::getRetainRVCallee(Module *M) {
+  if (!RetainRVCallee) {
+    LLVMContext &C = M->getContext();
+    const Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
+    std::vector<const Type *> Params;
+    Params.push_back(I8X);
+    const FunctionType *FTy =
+      FunctionType::get(I8X, Params, /*isVarArg=*/false);
+    AttrListPtr Attributes;
+    Attributes.addAttr(~0u, Attribute::NoUnwind);
+    RetainRVCallee =
+      M->getOrInsertFunction("objc_retainAutoreleasedReturnValue", FTy,
+                             Attributes);
+  }
+  return RetainRVCallee;
+}
+
+Constant *ObjCARCOpt::getAutoreleaseRVCallee(Module *M) {
+  if (!AutoreleaseRVCallee) {
+    LLVMContext &C = M->getContext();
+    const Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
+    std::vector<const Type *> Params;
+    Params.push_back(I8X);
+    const FunctionType *FTy =
+      FunctionType::get(I8X, Params, /*isVarArg=*/false);
+    AttrListPtr Attributes;
+    Attributes.addAttr(~0u, Attribute::NoUnwind);
+    AutoreleaseRVCallee =
+      M->getOrInsertFunction("objc_autoreleaseReturnValue", FTy,
+                             Attributes);
+  }
+  return AutoreleaseRVCallee;
+}
+
+Constant *ObjCARCOpt::getReleaseCallee(Module *M) {
+  if (!ReleaseCallee) {
+    LLVMContext &C = M->getContext();
+    std::vector<const Type *> Params;
+    Params.push_back(PointerType::getUnqual(Type::getInt8Ty(C)));
+    AttrListPtr Attributes;
+    Attributes.addAttr(~0u, Attribute::NoUnwind);
+    ReleaseCallee =
+      M->getOrInsertFunction(
+        "objc_release",
+        FunctionType::get(Type::getVoidTy(C), Params, /*isVarArg=*/false),
+        Attributes);
+  }
+  return ReleaseCallee;
+}
+
+Constant *ObjCARCOpt::getRetainCallee(Module *M) {
+  if (!RetainCallee) {
+    LLVMContext &C = M->getContext();
+    std::vector<const Type *> Params;
+    Params.push_back(PointerType::getUnqual(Type::getInt8Ty(C)));
+    AttrListPtr Attributes;
+    Attributes.addAttr(~0u, Attribute::NoUnwind);
+    RetainCallee =
+      M->getOrInsertFunction(
+        "objc_retain",
+        FunctionType::get(Params[0], Params, /*isVarArg=*/false),
+        Attributes);
+  }
+  return RetainCallee;
+}
+
+Constant *ObjCARCOpt::getAutoreleaseCallee(Module *M) {
+  if (!AutoreleaseCallee) {
+    LLVMContext &C = M->getContext();
+    std::vector<const Type *> Params;
+    Params.push_back(PointerType::getUnqual(Type::getInt8Ty(C)));
+    AttrListPtr Attributes;
+    Attributes.addAttr(~0u, Attribute::NoUnwind);
+    AutoreleaseCallee =
+      M->getOrInsertFunction(
+        "objc_autorelease",
+        FunctionType::get(Params[0], Params, /*isVarArg=*/false),
+        Attributes);
+  }
+  return AutoreleaseCallee;
+}
+
+/// CanAlterRefCount - Test whether the given instruction can result in a
+/// reference count modification (positive or negative) for the pointer's
+/// object.
+static bool
+CanAlterRefCount(const Instruction *Inst, const Value *Ptr,
+                 ProvenanceAnalysis &PA, InstructionClass Class) {
+  switch (Class) {
+  case IC_Autorelease:
+  case IC_AutoreleaseRV:
+  case IC_User:
+    // These operations never directly modify a reference count.
+    return false;
+  default: break;
+  }
+
+  ImmutableCallSite CS = static_cast<const Value *>(Inst);
+  assert(CS && "Only calls can alter reference counts!");
+
+  // See if AliasAnalysis can help us with the call.
+  AliasAnalysis::ModRefBehavior MRB = PA.getAA()->getModRefBehavior(CS);
+  if (AliasAnalysis::onlyReadsMemory(MRB))
+    return false;
+  if (AliasAnalysis::onlyAccessesArgPointees(MRB)) {
+    for (ImmutableCallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end();
+         I != E; ++I) {
+      const Value *Op = *I;
+      if (IsPotentialUse(Op) && PA.related(Ptr, Op))
+        return true;
+    }
+    return false;
+  }
+
+  // Assume the worst.
+  return true;
+}
+
+/// CanUse - Test whether the given instruction can "use" the given pointer's
+/// object in a way that requires the reference count to be positive.
+static bool
+CanUse(const Instruction *Inst, const Value *Ptr, ProvenanceAnalysis &PA,
+       InstructionClass Class) {
+  // IC_Call operations (as opposed to IC_CallOrUser) never "use" objc pointers.
+  if (Class == IC_Call)
+    return false;
+
+  // Consider various instructions which may have pointer arguments which are
+  // not "uses".
+  if (const ICmpInst *ICI = dyn_cast<ICmpInst>(Inst)) {
+    // Comparing a pointer with null, or any other constant, isn't really a use,
+    // because we don't care what the pointer points to, or about the values
+    // of any other dynamic reference-counted pointers.
+    if (!IsPotentialUse(ICI->getOperand(1)))
+      return false;
+  } else if (ImmutableCallSite CS = static_cast<const Value *>(Inst)) {
+    // For calls, just check the arguments (and not the callee operand).
+    for (ImmutableCallSite::arg_iterator OI = CS.arg_begin(),
+         OE = CS.arg_end(); OI != OE; ++OI) {
+      const Value *Op = *OI;
+      if (IsPotentialUse(Op) && PA.related(Ptr, Op))
+        return true;
+    }
+    return false;
+  } else if (const StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
+    // Special-case stores, because we don't care about the stored value, just
+    // the store address.
+    const Value *Op = GetUnderlyingObjCPtr(SI->getPointerOperand());
+    // If we can't tell what the underlying object was, assume there is a
+    // dependence.
+    return IsPotentialUse(Op) && PA.related(Op, Ptr);
+  }
+
+  // Check each operand for a match.
+  for (User::const_op_iterator OI = Inst->op_begin(), OE = Inst->op_end();
+       OI != OE; ++OI) {
+    const Value *Op = *OI;
+    if (IsPotentialUse(Op) && PA.related(Ptr, Op))
+      return true;
+  }
+  return false;
+}
+
+/// CanInterruptRV - Test whether the given instruction can autorelease
+/// any pointer or cause an autoreleasepool pop.
+static bool
+CanInterruptRV(InstructionClass Class) {
+  switch (Class) {
+  case IC_AutoreleasepoolPop:
+  case IC_CallOrUser:
+  case IC_Call:
+  case IC_Autorelease:
+  case IC_AutoreleaseRV:
+  case IC_FusedRetainAutorelease:
+  case IC_FusedRetainAutoreleaseRV:
+    return true;
+  default:
+    return false;
+  }
+}
+
+namespace {
+  /// DependenceKind - There are several kinds of dependence-like concepts in
+  /// use here.
+  enum DependenceKind {
+    NeedsPositiveRetainCount,
+    CanChangeRetainCount,
+    RetainAutoreleaseDep,       ///< Blocks objc_retainAutorelease.
+    RetainAutoreleaseRVDep,     ///< Blocks objc_retainAutoreleaseReturnValue.
+    RetainRVDep                 ///< Blocks objc_retainAutoreleasedReturnValue.
+  };
+}
+
+/// Depends - Test if there can be dependencies on Inst through Arg. This
+/// function only tests dependencies relevant for removing pairs of calls.
+static bool
+Depends(DependenceKind Flavor, Instruction *Inst, const Value *Arg,
+        ProvenanceAnalysis &PA) {
+  // If we've reached the definition of Arg, stop.
+  if (Inst == Arg)
+    return true;
+
+  switch (Flavor) {
+  case NeedsPositiveRetainCount: {
+    InstructionClass Class = GetInstructionClass(Inst);
+    switch (Class) {
+    case IC_AutoreleasepoolPop:
+    case IC_AutoreleasepoolPush:
+    case IC_None:
+      return false;
+    default:
+      return CanUse(Inst, Arg, PA, Class);
+    }
+  }
+
+  case CanChangeRetainCount: {
+    InstructionClass Class = GetInstructionClass(Inst);
+    switch (Class) {
+    case IC_AutoreleasepoolPop:
+      // Conservatively assume this can decrement any count.
+      return true;
+    case IC_AutoreleasepoolPush:
+    case IC_None:
+      return false;
+    default:
+      return CanAlterRefCount(Inst, Arg, PA, Class);
+    }
+  }
+
+  case RetainAutoreleaseDep:
+    switch (GetBasicInstructionClass(Inst)) {
+    case IC_AutoreleasepoolPop:
+      // Don't merge an objc_autorelease with an objc_retain inside a different
+      // autoreleasepool scope.
+      return true;
+    case IC_Retain:
+    case IC_RetainRV:
+      // Check for a retain of the same pointer for merging.
+      return GetObjCArg(Inst) == Arg;
+    default:
+      // Nothing else matters for objc_retainAutorelease formation.
+      return false;
+    }
+    break;
+
+  case RetainAutoreleaseRVDep: {
+    InstructionClass Class = GetBasicInstructionClass(Inst);
+    switch (Class) {
+    case IC_Retain:
+    case IC_RetainRV:
+      // Check for a retain of the same pointer for merging.
+      return GetObjCArg(Inst) == Arg;
+    default:
+      // Anything that can autorelease interrupts
+      // retainAutoreleaseReturnValue formation.
+      return CanInterruptRV(Class);
+    }
+    break;
+  }
+
+  case RetainRVDep:
+    return CanInterruptRV(GetBasicInstructionClass(Inst));
+  }
+
+  llvm_unreachable("Invalid dependence flavor");
+  return true;
+}
+
+/// FindDependencies - Walk up the CFG from StartPos (which is in StartBB) and
+/// find local and non-local dependencies on Arg.
+/// TODO: Cache results?
+static void
+FindDependencies(DependenceKind Flavor,
+                 const Value *Arg,
+                 BasicBlock *StartBB, Instruction *StartInst,
+                 SmallPtrSet<Instruction *, 4> &DependingInstructions,
+                 SmallPtrSet<const BasicBlock *, 4> &Visited,
+                 ProvenanceAnalysis &PA) {
+  BasicBlock::iterator StartPos = StartInst;
+
+  SmallVector<std::pair<BasicBlock *, BasicBlock::iterator>, 4> Worklist;
+  Worklist.push_back(std::make_pair(StartBB, StartPos));
+  do {
+    std::pair<BasicBlock *, BasicBlock::iterator> Pair =
+      Worklist.pop_back_val();
+    BasicBlock *LocalStartBB = Pair.first;
+    BasicBlock::iterator LocalStartPos = Pair.second;
+    BasicBlock::iterator StartBBBegin = LocalStartBB->begin();
+    for (;;) {
+      if (LocalStartPos == StartBBBegin) {
+        pred_iterator PI(LocalStartBB), PE(LocalStartBB, false);
+        if (PI == PE)
+          // If we've reached the function entry, produce a null dependence.
+          DependingInstructions.insert(0);
+        else
+          // Add the predecessors to the worklist.
+          do {
+            BasicBlock *PredBB = *PI;
+            if (Visited.insert(PredBB))
+              Worklist.push_back(std::make_pair(PredBB, PredBB->end()));
+          } while (++PI != PE);
+        break;
+      }
+
+      Instruction *Inst = --LocalStartPos;
+      if (Depends(Flavor, Inst, Arg, PA)) {
+        DependingInstructions.insert(Inst);
+        break;
+      }
+    }
+  } while (!Worklist.empty());
+
+  // Determine whether the original StartBB post-dominates all of the blocks we
+  // visited. If not, insert a sentinal indicating that most optimizations are
+  // not safe.
+  for (SmallPtrSet<const BasicBlock *, 4>::const_iterator I = Visited.begin(),
+       E = Visited.end(); I != E; ++I) {
+    const BasicBlock *BB = *I;
+    if (BB == StartBB)
+      continue;
+    const TerminatorInst *TI = cast<TerminatorInst>(&BB->back());
+    for (succ_const_iterator SI(TI), SE(TI, false); SI != SE; ++SI) {
+      const BasicBlock *Succ = *SI;
+      if (Succ != StartBB && !Visited.count(Succ)) {
+        DependingInstructions.insert(reinterpret_cast<Instruction *>(-1));
+        return;
+      }
+    }
+  }
+}
+
+static bool isNullOrUndef(const Value *V) {
+  return isa<ConstantPointerNull>(V) || isa<UndefValue>(V);
+}
+
+static bool isNoopInstruction(const Instruction *I) {
+  return isa<BitCastInst>(I) ||
+         (isa<GetElementPtrInst>(I) &&
+          cast<GetElementPtrInst>(I)->hasAllZeroIndices());
+}
+
+/// OptimizeRetainCall - Turn objc_retain into
+/// objc_retainAutoreleasedReturnValue if the operand is a return value.
+void
+ObjCARCOpt::OptimizeRetainCall(Function &F, Instruction *Retain) {
+  CallSite CS(GetObjCArg(Retain));
+  Instruction *Call = CS.getInstruction();
+  if (!Call) return;
+  if (Call->getParent() != Retain->getParent()) return;
+
+  // Check that the call is next to the retain.
+  BasicBlock::iterator I = Call;
+  ++I;
+  while (isNoopInstruction(I)) ++I;
+  if (&*I != Retain)
+    return;
+
+  // Turn it to an objc_retainAutoreleasedReturnValue..
+  Changed = true;
+  ++NumPeeps;
+  cast<CallInst>(Retain)->setCalledFunction(getRetainRVCallee(F.getParent()));
+}
+
+/// OptimizeRetainRVCall - Turn objc_retainAutoreleasedReturnValue into
+/// objc_retain if the operand is not a return value.  Or, if it can be
+/// paired with an objc_autoreleaseReturnValue, delete the pair and
+/// return true.
+bool
+ObjCARCOpt::OptimizeRetainRVCall(Function &F, Instruction *RetainRV) {
+  // Check for the argument being from an immediately preceding call.
+  Value *Arg = GetObjCArg(RetainRV);
+  CallSite CS(Arg);
+  if (Instruction *Call = CS.getInstruction())
+    if (Call->getParent() == RetainRV->getParent()) {
+      BasicBlock::iterator I = Call;
+      ++I;
+      while (isNoopInstruction(I)) ++I;
+      if (&*I == RetainRV)
+        return false;
+    }
+
+  // Check for being preceded by an objc_autoreleaseReturnValue on the same
+  // pointer. In this case, we can delete the pair.
+  BasicBlock::iterator I = RetainRV, Begin = RetainRV->getParent()->begin();
+  if (I != Begin) {
+    do --I; while (I != Begin && isNoopInstruction(I));
+    if (GetBasicInstructionClass(I) == IC_AutoreleaseRV &&
+        GetObjCArg(I) == Arg) {
+      Changed = true;
+      ++NumPeeps;
+      EraseInstruction(I);
+      EraseInstruction(RetainRV);
+      return true;
+    }
+  }
+
+  // Turn it to a plain objc_retain.
+  Changed = true;
+  ++NumPeeps;
+  cast<CallInst>(RetainRV)->setCalledFunction(getRetainCallee(F.getParent()));
+  return false;
+}
+
+/// OptimizeAutoreleaseRVCall - Turn objc_autoreleaseReturnValue into
+/// objc_autorelease if the result is not used as a return value.
+void
+ObjCARCOpt::OptimizeAutoreleaseRVCall(Function &F, Instruction *AutoreleaseRV) {
+  // Check for a return of the pointer value.
+  const Value *Ptr = GetObjCArg(AutoreleaseRV);
+  for (Value::const_use_iterator UI = Ptr->use_begin(), UE = Ptr->use_end();
+       UI != UE; ++UI) {
+    const User *I = *UI;
+    if (isa<ReturnInst>(I) || GetBasicInstructionClass(I) == IC_RetainRV)
+      return;
+  }
+
+  Changed = true;
+  ++NumPeeps;
+  cast<CallInst>(AutoreleaseRV)->
+    setCalledFunction(getAutoreleaseCallee(F.getParent()));
+}
+
+/// OptimizeIndividualCalls - Visit each call, one at a time, and make
+/// simplifications without doing any additional analysis.
+void ObjCARCOpt::OptimizeIndividualCalls(Function &F) {
+  // Reset all the flags in preparation for recomputing them.
+  UsedInThisFunction = 0;
+
+  // Visit all objc_* calls in F.
+  for (inst_iterator I = inst_begin(&F), E = inst_end(&F); I != E; ) {
+    Instruction *Inst = &*I++;
+    InstructionClass Class = GetBasicInstructionClass(Inst);
+
+    switch (Class) {
+    default: break;
+
+    // Delete no-op casts. These function calls have special semantics, but
+    // the semantics are entirely implemented via lowering in the front-end,
+    // so by the time they reach the optimizer, they are just no-op calls
+    // which return their argument.
+    //
+    // There are gray areas here, as the ability to cast reference-counted
+    // pointers to raw void* and back allows code to break ARC assumptions,
+    // however these are currently considered to be unimportant.
+    case IC_NoopCast:
+      Changed = true;
+      ++NumNoops;
+      EraseInstruction(Inst);
+      continue;
+
+    // If the pointer-to-weak-pointer is null, it's undefined behavior.
+    case IC_StoreWeak:
+    case IC_LoadWeak:
+    case IC_LoadWeakRetained:
+    case IC_InitWeak:
+    case IC_DestroyWeak: {
+      CallInst *CI = cast<CallInst>(Inst);
+      if (isNullOrUndef(CI->getArgOperand(0))) {
+        const Type *Ty = CI->getArgOperand(0)->getType();
+        new StoreInst(UndefValue::get(cast<PointerType>(Ty)->getElementType()),
+                      Constant::getNullValue(Ty),
+                      CI);
+        CI->replaceAllUsesWith(UndefValue::get(CI->getType()));
+        CI->eraseFromParent();
+        continue;
+      }
+      break;
+    }
+    case IC_CopyWeak:
+    case IC_MoveWeak: {
+      CallInst *CI = cast<CallInst>(Inst);
+      if (isNullOrUndef(CI->getArgOperand(0)) ||
+          isNullOrUndef(CI->getArgOperand(1))) {
+        const Type *Ty = CI->getArgOperand(0)->getType();
+        new StoreInst(UndefValue::get(cast<PointerType>(Ty)->getElementType()),
+                      Constant::getNullValue(Ty),
+                      CI);
+        CI->replaceAllUsesWith(UndefValue::get(CI->getType()));
+        CI->eraseFromParent();
+        continue;
+      }
+      break;
+    }
+    case IC_Retain:
+      OptimizeRetainCall(F, Inst);
+      break;
+    case IC_RetainRV:
+      if (OptimizeRetainRVCall(F, Inst))
+        continue;
+      break;
+    case IC_AutoreleaseRV:
+      OptimizeAutoreleaseRVCall(F, Inst);
+      break;
+    }
+
+    // objc_autorelease(x) -> objc_release(x) if x is otherwise unused.
+    if (IsAutorelease(Class) && Inst->use_empty()) {
+      CallInst *Call = cast<CallInst>(Inst);
+      const Value *Arg = Call->getArgOperand(0);
+      Arg = FindSingleUseIdentifiedObject(Arg);
+      if (Arg) {
+        Changed = true;
+        ++NumAutoreleases;
+
+        // Create the declaration lazily.
+        LLVMContext &C = Inst->getContext();
+        CallInst *NewCall =
+          CallInst::Create(getReleaseCallee(F.getParent()),
+                           Call->getArgOperand(0), "", Call);
+        NewCall->setMetadata(ImpreciseReleaseMDKind,
+                             MDNode::get(C, ArrayRef<Value *>()));
+        EraseInstruction(Call);
+        Inst = NewCall;
+        Class = IC_Release;
+      }
+    }
+
+    // For functions which can never be passed stack arguments, add
+    // a tail keyword.
+    if (IsAlwaysTail(Class)) {
+      Changed = true;
+      cast<CallInst>(Inst)->setTailCall();
+    }
+
+    // Set nounwind as needed.
+    if (IsNoThrow(Class)) {
+      Changed = true;
+      cast<CallInst>(Inst)->setDoesNotThrow();
+    }
+
+    if (!IsNoopOnNull(Class)) {
+      UsedInThisFunction |= 1 << Class;
+      continue;
+    }
+
+    const Value *Arg = GetObjCArg(Inst);
+
+    // ARC calls with null are no-ops. Delete them.
+    if (isNullOrUndef(Arg)) {
+      Changed = true;
+      ++NumNoops;
+      EraseInstruction(Inst);
+      continue;
+    }
+
+    // Keep track of which of retain, release, autorelease, and retain_block
+    // are actually present in this function.
+    UsedInThisFunction |= 1 << Class;
+
+    // If Arg is a PHI, and one or more incoming values to the
+    // PHI are null, and the call is control-equivalent to the PHI, and there
+    // are no relevant side effects between the PHI and the call, the call
+    // could be pushed up to just those paths with non-null incoming values.
+    // For now, don't bother splitting critical edges for this.
+    SmallVector<std::pair<Instruction *, const Value *>, 4> Worklist;
+    Worklist.push_back(std::make_pair(Inst, Arg));
+    do {
+      std::pair<Instruction *, const Value *> Pair = Worklist.pop_back_val();
+      Inst = Pair.first;
+      Arg = Pair.second;
+
+      const PHINode *PN = dyn_cast<PHINode>(Arg);
+      if (!PN) continue;
+
+      // Determine if the PHI has any null operands, or any incoming
+      // critical edges.
+      bool HasNull = false;
+      bool HasCriticalEdges = false;
+      for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
+        Value *Incoming =
+          StripPointerCastsAndObjCCalls(PN->getIncomingValue(i));
+        if (isNullOrUndef(Incoming))
+          HasNull = true;
+        else if (cast<TerminatorInst>(PN->getIncomingBlock(i)->back())
+                   .getNumSuccessors() != 1) {
+          HasCriticalEdges = true;
+          break;
+        }
+      }
+      // If we have null operands and no critical edges, optimize.
+      if (!HasCriticalEdges && HasNull) {
+        SmallPtrSet<Instruction *, 4> DependingInstructions;
+        SmallPtrSet<const BasicBlock *, 4> Visited;
+
+        // Check that there is nothing that cares about the reference
+        // count between the call and the phi.
+        FindDependencies(NeedsPositiveRetainCount, Arg,
+                         Inst->getParent(), Inst,
+                         DependingInstructions, Visited, PA);
+        if (DependingInstructions.size() == 1 &&
+            *DependingInstructions.begin() == PN) {
+          Changed = true;
+          ++NumPartialNoops;
+          // Clone the call into each predecessor that has a non-null value.
+          CallInst *CInst = cast<CallInst>(Inst);
+          const Type *ParamTy = CInst->getArgOperand(0)->getType();
+          for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
+            Value *Incoming =
+              StripPointerCastsAndObjCCalls(PN->getIncomingValue(i));
+            if (!isNullOrUndef(Incoming)) {
+              CallInst *Clone = cast<CallInst>(CInst->clone());
+              Value *Op = PN->getIncomingValue(i);
+              Instruction *InsertPos = &PN->getIncomingBlock(i)->back();
+              if (Op->getType() != ParamTy)
+                Op = new BitCastInst(Op, ParamTy, "", InsertPos);
+              Clone->setArgOperand(0, Op);
+              Clone->insertBefore(InsertPos);
+              Worklist.push_back(std::make_pair(Clone, Incoming));
+            }
+          }
+          // Erase the original call.
+          EraseInstruction(CInst);
+          continue;
+        }
+      }
+    } while (!Worklist.empty());
+  }
+}
+
+/// CheckForCFGHazards - Check for critical edges, loop boundaries, irreducible
+/// control flow, or other CFG structures where moving code across the edge
+/// would result in it being executed more.
+void
+ObjCARCOpt::CheckForCFGHazards(const BasicBlock *BB,
+                               DenseMap<const BasicBlock *, BBState> &BBStates,
+                               BBState &MyStates) const {
+  // If any top-down local-use or possible-dec has a succ which is earlier in
+  // the sequence, forget it.
+  for (BBState::ptr_const_iterator I = MyStates.top_down_ptr_begin(),
+       E = MyStates.top_down_ptr_end(); I != E; ++I)
+    switch (I->second.GetSeq()) {
+    default: break;
+    case S_Use: {
+      const Value *Arg = I->first;
+      const TerminatorInst *TI = cast<TerminatorInst>(&BB->back());
+      bool SomeSuccHasSame = false;
+      bool AllSuccsHaveSame = true;
+      for (succ_const_iterator SI(TI), SE(TI, false); SI != SE; ++SI)
+        switch (BBStates[*SI].getPtrBottomUpState(Arg).GetSeq()) {
+        case S_None:
+        case S_CanRelease:
+          MyStates.getPtrTopDownState(Arg).ClearSequenceProgress();
+          SomeSuccHasSame = false;
+          break;
+        case S_Use:
+          SomeSuccHasSame = true;
+          break;
+        case S_Stop:
+        case S_Release:
+        case S_MovableRelease:
+          AllSuccsHaveSame = false;
+          break;
+        case S_Retain:
+          llvm_unreachable("bottom-up pointer in retain state!");
+        }
+      // If the state at the other end of any of the successor edges
+      // matches the current state, require all edges to match. This
+      // guards against loops in the middle of a sequence.
+      if (SomeSuccHasSame && !AllSuccsHaveSame)
+        MyStates.getPtrTopDownState(Arg).ClearSequenceProgress();
+    }
+    case S_CanRelease: {
+      const Value *Arg = I->first;
+      const TerminatorInst *TI = cast<TerminatorInst>(&BB->back());
+      bool SomeSuccHasSame = false;
+      bool AllSuccsHaveSame = true;
+      for (succ_const_iterator SI(TI), SE(TI, false); SI != SE; ++SI)
+        switch (BBStates[*SI].getPtrBottomUpState(Arg).GetSeq()) {
+        case S_None:
+          MyStates.getPtrTopDownState(Arg).ClearSequenceProgress();
+          SomeSuccHasSame = false;
+          break;
+        case S_CanRelease:
+          SomeSuccHasSame = true;
+          break;
+        case S_Stop:
+        case S_Release:
+        case S_MovableRelease:
+        case S_Use:
+          AllSuccsHaveSame = false;
+          break;
+        case S_Retain:
+          llvm_unreachable("bottom-up pointer in retain state!");
+        }
+      // If the state at the other end of any of the successor edges
+      // matches the current state, require all edges to match. This
+      // guards against loops in the middle of a sequence.
+      if (SomeSuccHasSame && !AllSuccsHaveSame)
+        MyStates.getPtrTopDownState(Arg).ClearSequenceProgress();
+    }
+    }
+}
+
+bool
+ObjCARCOpt::VisitBottomUp(BasicBlock *BB,
+                          DenseMap<const BasicBlock *, BBState> &BBStates,
+                          MapVector<Value *, RRInfo> &Retains) {
+  bool NestingDetected = false;
+  BBState &MyStates = BBStates[BB];
+
+  // Merge the states from each successor to compute the initial state
+  // for the current block.
+  const TerminatorInst *TI = cast<TerminatorInst>(&BB->back());
+  succ_const_iterator SI(TI), SE(TI, false);
+  if (SI == SE)
+    MyStates.SetAsExit();
+  else
+    do {
+      const BasicBlock *Succ = *SI++;
+      if (Succ == BB)
+        continue;
+      DenseMap<const BasicBlock *, BBState>::iterator I = BBStates.find(Succ);
+      if (I == BBStates.end())
+        continue;
+      MyStates.InitFromSucc(I->second);
+      while (SI != SE) {
+        Succ = *SI++;
+        if (Succ != BB) {
+          I = BBStates.find(Succ);
+          if (I != BBStates.end())
+            MyStates.MergeSucc(I->second);
+        }
+      }
+      break;
+    } while (SI != SE);
+
+  // Visit all the instructions, bottom-up.
+  for (BasicBlock::iterator I = BB->end(), E = BB->begin(); I != E; --I) {
+    Instruction *Inst = llvm::prior(I);
+    InstructionClass Class = GetInstructionClass(Inst);
+    const Value *Arg = 0;
+
+    switch (Class) {
+    case IC_Release: {
+      Arg = GetObjCArg(Inst);
+
+      PtrState &S = MyStates.getPtrBottomUpState(Arg);
+
+      // If we see two releases in a row on the same pointer. If so, make
+      // a note, and we'll cicle back to revisit it after we've
+      // hopefully eliminated the second release, which may allow us to
+      // eliminate the first release too.
+      // Theoretically we could implement removal of nested retain+release
+      // pairs by making PtrState hold a stack of states, but this is
+      // simple and avoids adding overhead for the non-nested case.
+      if (S.GetSeq() == S_Release || S.GetSeq() == S_MovableRelease)
+        NestingDetected = true;
+
+      S.SetSeqToRelease(Inst->getMetadata(ImpreciseReleaseMDKind));
+      S.RRI.clear();
+      S.RRI.KnownIncremented = S.IsKnownIncremented();
+      S.RRI.IsTailCallRelease = cast<CallInst>(Inst)->isTailCall();
+      S.RRI.Calls.insert(Inst);
+
+      S.IncrementRefCount();
+      break;
+    }
+    case IC_RetainBlock:
+    case IC_Retain:
+    case IC_RetainRV: {
+      Arg = GetObjCArg(Inst);
+
+      PtrState &S = MyStates.getPtrBottomUpState(Arg);
+      S.DecrementRefCount();
+
+      switch (S.GetSeq()) {
+      case S_Stop:
+      case S_Release:
+      case S_MovableRelease:
+      case S_Use:
+        S.RRI.ReverseInsertPts.clear();
+        // FALL THROUGH
+      case S_CanRelease:
+        // Don't do retain+release tracking for IC_RetainRV, because it's
+        // better to let it remain as the first instruction after a call.
+        if (Class != IC_RetainRV) {
+          S.RRI.IsRetainBlock = Class == IC_RetainBlock;
+          Retains[Inst] = S.RRI;
+        }
+        S.ClearSequenceProgress();
+        break;
+      case S_None:
+        break;
+      case S_Retain:
+        llvm_unreachable("bottom-up pointer in retain state!");
+      }
+      break;
+    }
+    case IC_AutoreleasepoolPop:
+      // Conservatively, clear MyStates for all known pointers.
+      MyStates.clearBottomUpPointers();
+      continue;
+    case IC_AutoreleasepoolPush:
+    case IC_None:
+      // These are irrelevant.
+      continue;
+    default:
+      break;
+    }
+
+    // Consider any other possible effects of this instruction on each
+    // pointer being tracked.
+    for (BBState::ptr_iterator MI = MyStates.bottom_up_ptr_begin(),
+         ME = MyStates.bottom_up_ptr_end(); MI != ME; ++MI) {
+      const Value *Ptr = MI->first;
+      if (Ptr == Arg)
+        continue; // Handled above.
+      PtrState &S = MI->second;
+      Sequence Seq = S.GetSeq();
+
+      // Check for possible retains and releases.
+      if (CanAlterRefCount(Inst, Ptr, PA, Class)) {
+        // Check for a retain (we're going bottom-up here).
+        S.DecrementRefCount();
+
+        // Check for a release.
+        if (!IsRetain(Class) && Class != IC_RetainBlock)
+          switch (Seq) {
+          case S_Use:
+            S.SetSeq(S_CanRelease);
+            continue;
+          case S_CanRelease:
+          case S_Release:
+          case S_MovableRelease:
+          case S_Stop:
+          case S_None:
+            break;
+          case S_Retain:
+            llvm_unreachable("bottom-up pointer in retain state!");
+          }
+      }
+
+      // Check for possible direct uses.
+      switch (Seq) {
+      case S_Release:
+      case S_MovableRelease:
+        if (CanUse(Inst, Ptr, PA, Class)) {
+          S.RRI.ReverseInsertPts.clear();
+          S.RRI.ReverseInsertPts.insert(Inst);
+          S.SetSeq(S_Use);
+        } else if (Seq == S_Release &&
+                   (Class == IC_User || Class == IC_CallOrUser)) {
+          // Non-movable releases depend on any possible objc pointer use.
+          S.SetSeq(S_Stop);
+          S.RRI.ReverseInsertPts.clear();
+          S.RRI.ReverseInsertPts.insert(Inst);
+        }
+        break;
+      case S_Stop:
+        if (CanUse(Inst, Ptr, PA, Class))
+          S.SetSeq(S_Use);
+        break;
+      case S_CanRelease:
+      case S_Use:
+      case S_None:
+        break;
+      case S_Retain:
+        llvm_unreachable("bottom-up pointer in retain state!");
+      }
+    }
+  }
+
+  return NestingDetected;
+}
+
+bool
+ObjCARCOpt::VisitTopDown(BasicBlock *BB,
+                         DenseMap<const BasicBlock *, BBState> &BBStates,
+                         DenseMap<Value *, RRInfo> &Releases) {
+  bool NestingDetected = false;
+  BBState &MyStates = BBStates[BB];
+
+  // Merge the states from each predecessor to compute the initial state
+  // for the current block.
+  const_pred_iterator PI(BB), PE(BB, false);
+  if (PI == PE)
+    MyStates.SetAsEntry();
+  else
+    do {
+      const BasicBlock *Pred = *PI++;
+      if (Pred == BB)
+        continue;
+      DenseMap<const BasicBlock *, BBState>::iterator I = BBStates.find(Pred);
+      if (I == BBStates.end())
+        continue;
+      MyStates.InitFromPred(I->second);
+      while (PI != PE) {
+        Pred = *PI++;
+        if (Pred != BB) {
+          I = BBStates.find(Pred);
+          if (I != BBStates.end())
+            MyStates.MergePred(I->second);
+        }
+      }
+      break;
+    } while (PI != PE);
+
+  // Visit all the instructions, top-down.
+  for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) {
+    Instruction *Inst = I;
+    InstructionClass Class = GetInstructionClass(Inst);
+    const Value *Arg = 0;
+
+    switch (Class) {
+    case IC_RetainBlock:
+    case IC_Retain:
+    case IC_RetainRV: {
+      Arg = GetObjCArg(Inst);
+
+      PtrState &S = MyStates.getPtrTopDownState(Arg);
+
+      // Don't do retain+release tracking for IC_RetainRV, because it's
+      // better to let it remain as the first instruction after a call.
+      if (Class != IC_RetainRV) {
+        // If we see two retains in a row on the same pointer. If so, make
+        // a note, and we'll cicle back to revisit it after we've
+        // hopefully eliminated the second retain, which may allow us to
+        // eliminate the first retain too.
+        // Theoretically we could implement removal of nested retain+release
+        // pairs by making PtrState hold a stack of states, but this is
+        // simple and avoids adding overhead for the non-nested case.
+        if (S.GetSeq() == S_Retain)
+          NestingDetected = true;
+
+        S.SetSeq(S_Retain);
+        S.RRI.clear();
+        S.RRI.IsRetainBlock = Class == IC_RetainBlock;
+        S.RRI.KnownIncremented = S.IsKnownIncremented();
+        S.RRI.Calls.insert(Inst);
+      }
+
+      S.IncrementRefCount();
+      break;
+    }
+    case IC_Release: {
+      Arg = GetObjCArg(Inst);
+
+      PtrState &S = MyStates.getPtrTopDownState(Arg);
+      S.DecrementRefCount();
+
+      switch (S.GetSeq()) {
+      case S_Retain:
+      case S_CanRelease:
+        S.RRI.ReverseInsertPts.clear();
+        // FALL THROUGH
+      case S_Use:
+        S.RRI.ReleaseMetadata = Inst->getMetadata(ImpreciseReleaseMDKind);
+        S.RRI.IsTailCallRelease = cast<CallInst>(Inst)->isTailCall();
+        Releases[Inst] = S.RRI;
+        S.ClearSequenceProgress();
+        break;
+      case S_None:
+        break;
+      case S_Stop:
+      case S_Release:
+      case S_MovableRelease:
+        llvm_unreachable("top-down pointer in release state!");
+      }
+      break;
+    }
+    case IC_AutoreleasepoolPop:
+      // Conservatively, clear MyStates for all known pointers.
+      MyStates.clearTopDownPointers();
+      continue;
+    case IC_AutoreleasepoolPush:
+    case IC_None:
+      // These are irrelevant.
+      continue;
+    default:
+      break;
+    }
+
+    // Consider any other possible effects of this instruction on each
+    // pointer being tracked.
+    for (BBState::ptr_iterator MI = MyStates.top_down_ptr_begin(),
+         ME = MyStates.top_down_ptr_end(); MI != ME; ++MI) {
+      const Value *Ptr = MI->first;
+      if (Ptr == Arg)
+        continue; // Handled above.
+      PtrState &S = MI->second;
+      Sequence Seq = S.GetSeq();
+
+      // Check for possible releases.
+      if (!IsRetain(Class) && Class != IC_RetainBlock &&
+          CanAlterRefCount(Inst, Ptr, PA, Class)) {
+        // Check for a release.
+        S.DecrementRefCount();
+
+        // Check for a release.
+        switch (Seq) {
+        case S_Retain:
+          S.SetSeq(S_CanRelease);
+          S.RRI.ReverseInsertPts.clear();
+          S.RRI.ReverseInsertPts.insert(Inst);
+
+          // One call can't cause a transition from S_Retain to S_CanRelease
+          // and S_CanRelease to S_Use. If we've made the first transition,
+          // we're done.
+          continue;
+        case S_Use:
+        case S_CanRelease:
+        case S_None:
+          break;
+        case S_Stop:
+        case S_Release:
+        case S_MovableRelease:
+          llvm_unreachable("top-down pointer in release state!");
+        }
+      }
+
+      // Check for possible direct uses.
+      switch (Seq) {
+      case S_CanRelease:
+        if (CanUse(Inst, Ptr, PA, Class))
+          S.SetSeq(S_Use);
+        break;
+      case S_Use:
+      case S_Retain:
+      case S_None:
+        break;
+      case S_Stop:
+      case S_Release:
+      case S_MovableRelease:
+        llvm_unreachable("top-down pointer in release state!");
+      }
+    }
+  }
+
+  CheckForCFGHazards(BB, BBStates, MyStates);
+  return NestingDetected;
+}
+
+// Visit - Visit the function both top-down and bottom-up.
+bool
+ObjCARCOpt::Visit(Function &F,
+                  DenseMap<const BasicBlock *, BBState> &BBStates,
+                  MapVector<Value *, RRInfo> &Retains,
+                  DenseMap<Value *, RRInfo> &Releases) {
+  // Use postorder for bottom-up, and reverse-postorder for top-down, because we
+  // magically know that loops will be well behaved, i.e. they won't repeatedly
+  // call retain on a single pointer without doing a release.
+  bool BottomUpNestingDetected = false;
+  SmallVector<BasicBlock *, 8> PostOrder;
+  for (po_iterator<Function *> I = po_begin(&F), E = po_end(&F); I != E; ++I) {
+    BasicBlock *BB = *I;
+    PostOrder.push_back(BB);
+
+    BottomUpNestingDetected |= VisitBottomUp(BB, BBStates, Retains);
+  }
+
+  // Iterate through the post-order in reverse order, achieving a
+  // reverse-postorder traversal. We don't use the ReversePostOrderTraversal
+  // class here because it works by computing its own full postorder iteration,
+  // recording the sequence, and playing it back in reverse. Since we're already
+  // doing a full iteration above, we can just record the sequence manually and
+  // avoid the cost of having ReversePostOrderTraversal compute it.
+  bool TopDownNestingDetected = false;
+  for (SmallVectorImpl<BasicBlock *>::const_reverse_iterator
+       RI = PostOrder.rbegin(), RE = PostOrder.rend(); RI != RE; ++RI)
+    TopDownNestingDetected |= VisitTopDown(*RI, BBStates, Releases);
+
+  return TopDownNestingDetected && BottomUpNestingDetected;
+}
+
+/// MoveCalls - Move the calls in RetainsToMove and ReleasesToMove.
+void ObjCARCOpt::MoveCalls(Value *Arg,
+                           RRInfo &RetainsToMove,
+                           RRInfo &ReleasesToMove,
+                           MapVector<Value *, RRInfo> &Retains,
+                           DenseMap<Value *, RRInfo> &Releases,
+                           SmallVectorImpl<Instruction *> &DeadInsts) {
+  const Type *ArgTy = Arg->getType();
+  const Type *ParamTy =
+    (RetainRVFunc ? RetainRVFunc :
+     RetainFunc ? RetainFunc :
+     RetainBlockFunc)->arg_begin()->getType();
+
+  // Insert the new retain and release calls.
+  for (SmallPtrSet<Instruction *, 2>::const_iterator
+       PI = ReleasesToMove.ReverseInsertPts.begin(),
+       PE = ReleasesToMove.ReverseInsertPts.end(); PI != PE; ++PI) {
+    Instruction *InsertPt = *PI;
+    Value *MyArg = ArgTy == ParamTy ? Arg :
+                   new BitCastInst(Arg, ParamTy, "", InsertPt);
+    CallInst *Call =
+      CallInst::Create(RetainsToMove.IsRetainBlock ?
+                         RetainBlockFunc : RetainFunc,
+                       MyArg, "", InsertPt);
+    Call->setDoesNotThrow();
+    if (!RetainsToMove.IsRetainBlock)
+      Call->setTailCall();
+  }
+  for (SmallPtrSet<Instruction *, 2>::const_iterator
+       PI = RetainsToMove.ReverseInsertPts.begin(),
+       PE = RetainsToMove.ReverseInsertPts.end(); PI != PE; ++PI) {
+    Instruction *InsertPt = llvm::next(BasicBlock::iterator(*PI));
+    Value *MyArg = ArgTy == ParamTy ? Arg :
+                   new BitCastInst(Arg, ParamTy, "", InsertPt);
+    CallInst *Call = CallInst::Create(ReleaseFunc, MyArg, "", InsertPt);
+    // Attach a clang.imprecise_release metadata tag, if appropriate.
+    if (MDNode *M = ReleasesToMove.ReleaseMetadata)
+      Call->setMetadata(ImpreciseReleaseMDKind, M);
+    Call->setDoesNotThrow();
+    if (ReleasesToMove.IsTailCallRelease)
+      Call->setTailCall();
+  }
+
+  // Delete the original retain and release calls.
+  for (SmallPtrSet<Instruction *, 2>::const_iterator
+       AI = RetainsToMove.Calls.begin(),
+       AE = RetainsToMove.Calls.end(); AI != AE; ++AI) {
+    Instruction *OrigRetain = *AI;
+    Retains.blot(OrigRetain);
+    DeadInsts.push_back(OrigRetain);
+  }
+  for (SmallPtrSet<Instruction *, 2>::const_iterator
+       AI = ReleasesToMove.Calls.begin(),
+       AE = ReleasesToMove.Calls.end(); AI != AE; ++AI) {
+    Instruction *OrigRelease = *AI;
+    Releases.erase(OrigRelease);
+    DeadInsts.push_back(OrigRelease);
+  }
+}
+
+bool
+ObjCARCOpt::PerformCodePlacement(DenseMap<const BasicBlock *, BBState>
+                                   &BBStates,
+                                 MapVector<Value *, RRInfo> &Retains,
+                                 DenseMap<Value *, RRInfo> &Releases) {
+  bool AnyPairsCompletelyEliminated = false;
+  RRInfo RetainsToMove;
+  RRInfo ReleasesToMove;
+  SmallVector<Instruction *, 4> NewRetains;
+  SmallVector<Instruction *, 4> NewReleases;
+  SmallVector<Instruction *, 8> DeadInsts;
+
+  for (MapVector<Value *, RRInfo>::const_iterator I = Retains.begin(),
+       E = Retains.end(); I != E; ) {
+    Value *V = (I++)->first;
+    if (!V) continue; // blotted
+
+    Instruction *Retain = cast<Instruction>(V);
+    Value *Arg = GetObjCArg(Retain);
+
+    // If the object being released is in static or stack storage, we know it's
+    // not being managed by ObjC reference counting, so we can delete pairs
+    // regardless of what possible decrements or uses lie between them.
+    bool KnownSafe = isa<Constant>(Arg) || isa<AllocaInst>(Arg);
+
+    // If a pair happens in a region where it is known that the reference count
+    // is already incremented, we can similarly ignore possible decrements.
+    bool KnownIncrementedTD = true, KnownIncrementedBU = true;
+
+    // Connect the dots between the top-down-collected RetainsToMove and
+    // bottom-up-collected ReleasesToMove to form sets of related calls.
+    // This is an iterative process so that we connect multiple releases
+    // to multiple retains if needed.
+    unsigned OldDelta = 0;
+    unsigned NewDelta = 0;
+    unsigned OldCount = 0;
+    unsigned NewCount = 0;
+    bool FirstRelease = true;
+    bool FirstRetain = true;
+    NewRetains.push_back(Retain);
+    for (;;) {
+      for (SmallVectorImpl<Instruction *>::const_iterator
+           NI = NewRetains.begin(), NE = NewRetains.end(); NI != NE; ++NI) {
+        Instruction *NewRetain = *NI;
+        MapVector<Value *, RRInfo>::const_iterator It = Retains.find(NewRetain);
+        assert(It != Retains.end());
+        const RRInfo &NewRetainRRI = It->second;
+        KnownIncrementedTD &= NewRetainRRI.KnownIncremented;
+        for (SmallPtrSet<Instruction *, 2>::const_iterator
+             LI = NewRetainRRI.Calls.begin(),
+             LE = NewRetainRRI.Calls.end(); LI != LE; ++LI) {
+          Instruction *NewRetainRelease = *LI;
+          DenseMap<Value *, RRInfo>::const_iterator Jt =
+            Releases.find(NewRetainRelease);
+          if (Jt == Releases.end())
+            goto next_retain;
+          const RRInfo &NewRetainReleaseRRI = Jt->second;
+          assert(NewRetainReleaseRRI.Calls.count(NewRetain));
+          if (ReleasesToMove.Calls.insert(NewRetainRelease)) {
+            OldDelta -=
+              BBStates[NewRetainRelease->getParent()].GetAllPathCount();
+
+            // Merge the ReleaseMetadata and IsTailCallRelease values.
+            if (FirstRelease) {
+              ReleasesToMove.ReleaseMetadata =
+                NewRetainReleaseRRI.ReleaseMetadata;
+              ReleasesToMove.IsTailCallRelease =
+                NewRetainReleaseRRI.IsTailCallRelease;
+              FirstRelease = false;
+            } else {
+              if (ReleasesToMove.ReleaseMetadata !=
+                    NewRetainReleaseRRI.ReleaseMetadata)
+                ReleasesToMove.ReleaseMetadata = 0;
+              if (ReleasesToMove.IsTailCallRelease !=
+                    NewRetainReleaseRRI.IsTailCallRelease)
+                ReleasesToMove.IsTailCallRelease = false;
+            }
+
+            // Collect the optimal insertion points.
+            if (!KnownSafe)
+              for (SmallPtrSet<Instruction *, 2>::const_iterator
+                   RI = NewRetainReleaseRRI.ReverseInsertPts.begin(),
+                   RE = NewRetainReleaseRRI.ReverseInsertPts.end();
+                   RI != RE; ++RI) {
+                Instruction *RIP = *RI;
+                if (ReleasesToMove.ReverseInsertPts.insert(RIP))
+                  NewDelta -= BBStates[RIP->getParent()].GetAllPathCount();
+              }
+            NewReleases.push_back(NewRetainRelease);
+          }
+        }
+      }
+      NewRetains.clear();
+      if (NewReleases.empty()) break;
+
+      // Back the other way.
+      for (SmallVectorImpl<Instruction *>::const_iterator
+           NI = NewReleases.begin(), NE = NewReleases.end(); NI != NE; ++NI) {
+        Instruction *NewRelease = *NI;
+        DenseMap<Value *, RRInfo>::const_iterator It =
+          Releases.find(NewRelease);
+        assert(It != Releases.end());
+        const RRInfo &NewReleaseRRI = It->second;
+        KnownIncrementedBU &= NewReleaseRRI.KnownIncremented;
+        for (SmallPtrSet<Instruction *, 2>::const_iterator
+             LI = NewReleaseRRI.Calls.begin(),
+             LE = NewReleaseRRI.Calls.end(); LI != LE; ++LI) {
+          Instruction *NewReleaseRetain = *LI;
+          MapVector<Value *, RRInfo>::const_iterator Jt =
+            Retains.find(NewReleaseRetain);
+          if (Jt == Retains.end())
+            goto next_retain;
+          const RRInfo &NewReleaseRetainRRI = Jt->second;
+          assert(NewReleaseRetainRRI.Calls.count(NewRelease));
+          if (RetainsToMove.Calls.insert(NewReleaseRetain)) {
+            unsigned PathCount =
+              BBStates[NewReleaseRetain->getParent()].GetAllPathCount();
+            OldDelta += PathCount;
+            OldCount += PathCount;
+
+            // Merge the IsRetainBlock values.
+            if (FirstRetain) {
+              RetainsToMove.IsRetainBlock = NewReleaseRetainRRI.IsRetainBlock;
+              FirstRetain = false;
+            } else if (ReleasesToMove.IsRetainBlock !=
+                       NewReleaseRetainRRI.IsRetainBlock)
+              // It's not possible to merge the sequences if one uses
+              // objc_retain and the other uses objc_retainBlock.
+              goto next_retain;
+
+            // Collect the optimal insertion points.
+            if (!KnownSafe)
+              for (SmallPtrSet<Instruction *, 2>::const_iterator
+                   RI = NewReleaseRetainRRI.ReverseInsertPts.begin(),
+                   RE = NewReleaseRetainRRI.ReverseInsertPts.end();
+                   RI != RE; ++RI) {
+                Instruction *RIP = *RI;
+                if (RetainsToMove.ReverseInsertPts.insert(RIP)) {
+                  PathCount = BBStates[RIP->getParent()].GetAllPathCount();
+                  NewDelta += PathCount;
+                  NewCount += PathCount;
+                }
+              }
+            NewRetains.push_back(NewReleaseRetain);
+          }
+        }
+      }
+      NewReleases.clear();
+      if (NewRetains.empty()) break;
+    }
+
+    // If the pointer is known incremented, we can safely delete the pair
+    // regardless of what's between them.
+    if (KnownIncrementedTD || KnownIncrementedBU) {
+      RetainsToMove.ReverseInsertPts.clear();
+      ReleasesToMove.ReverseInsertPts.clear();
+      NewCount = 0;
+    }
+
+    // Determine whether the original call points are balanced in the retain and
+    // release calls through the program. If not, conservatively don't touch
+    // them.
+    // TODO: It's theoretically possible to do code motion in this case, as
+    // long as the existing imbalances are maintained.
+    if (OldDelta != 0)
+      goto next_retain;
+
+    // Determine whether the new insertion points we computed preserve the
+    // balance of retain and release calls through the program.
+    // TODO: If the fully aggressive solution isn't valid, try to find a
+    // less aggressive solution which is.
+    if (NewDelta != 0)
+      goto next_retain;
+
+    // Ok, everything checks out and we're all set. Let's move some code!
+    Changed = true;
+    AnyPairsCompletelyEliminated = NewCount == 0;
+    NumRRs += OldCount - NewCount;
+    MoveCalls(Arg, RetainsToMove, ReleasesToMove, Retains, Releases, DeadInsts);
+
+  next_retain:
+    NewReleases.clear();
+    NewRetains.clear();
+    RetainsToMove.clear();
+    ReleasesToMove.clear();
+  }
+
+  // Now that we're done moving everything, we can delete the newly dead
+  // instructions, as we no longer need them as insert points.
+  while (!DeadInsts.empty())
+    EraseInstruction(DeadInsts.pop_back_val());
+
+  return AnyPairsCompletelyEliminated;
+}
+
+/// OptimizeWeakCalls - Weak pointer optimizations.
+void ObjCARCOpt::OptimizeWeakCalls(Function &F) {
+  // First, do memdep-style RLE and S2L optimizations. We can't use memdep
+  // itself because it uses AliasAnalysis and we need to do provenance
+  // queries instead.
+  for (inst_iterator I = inst_begin(&F), E = inst_end(&F); I != E; ) {
+    Instruction *Inst = &*I++;
+    InstructionClass Class = GetBasicInstructionClass(Inst);
+    if (Class != IC_LoadWeak && Class != IC_LoadWeakRetained)
+      continue;
+
+    // Delete objc_loadWeak calls with no users.
+    if (Class == IC_LoadWeak && Inst->use_empty()) {
+      Inst->eraseFromParent();
+      continue;
+    }
+
+    // TODO: For now, just look for an earlier available version of this value
+    // within the same block. Theoretically, we could do memdep-style non-local
+    // analysis too, but that would want caching. A better approach would be to
+    // use the technique that EarlyCSE uses.
+    inst_iterator Current = llvm::prior(I);
+    BasicBlock *CurrentBB = Current.getBasicBlockIterator();
+    for (BasicBlock::iterator B = CurrentBB->begin(),
+                              J = Current.getInstructionIterator();
+         J != B; --J) {
+      Instruction *EarlierInst = &*llvm::prior(J);
+      InstructionClass EarlierClass = GetInstructionClass(EarlierInst);
+      switch (EarlierClass) {
+      case IC_LoadWeak:
+      case IC_LoadWeakRetained: {
+        // If this is loading from the same pointer, replace this load's value
+        // with that one.
+        CallInst *Call = cast<CallInst>(Inst);
+        CallInst *EarlierCall = cast<CallInst>(EarlierInst);
+        Value *Arg = Call->getArgOperand(0);
+        Value *EarlierArg = EarlierCall->getArgOperand(0);
+        switch (PA.getAA()->alias(Arg, EarlierArg)) {
+        case AliasAnalysis::MustAlias:
+          Changed = true;
+          // If the load has a builtin retain, insert a plain retain for it.
+          if (Class == IC_LoadWeakRetained) {
+            CallInst *CI =
+              CallInst::Create(getRetainCallee(F.getParent()), EarlierCall,
+                               "", Call);
+            CI->setTailCall();
+          }
+          // Zap the fully redundant load.
+          Call->replaceAllUsesWith(EarlierCall);
+          Call->eraseFromParent();
+          goto clobbered;
+        case AliasAnalysis::MayAlias:
+        case AliasAnalysis::PartialAlias:
+          goto clobbered;
+        case AliasAnalysis::NoAlias:
+          break;
+        }
+        break;
+      }
+      case IC_StoreWeak:
+      case IC_InitWeak: {
+        // If this is storing to the same pointer and has the same size etc.
+        // replace this load's value with the stored value.
+        CallInst *Call = cast<CallInst>(Inst);
+        CallInst *EarlierCall = cast<CallInst>(EarlierInst);
+        Value *Arg = Call->getArgOperand(0);
+        Value *EarlierArg = EarlierCall->getArgOperand(0);
+        switch (PA.getAA()->alias(Arg, EarlierArg)) {
+        case AliasAnalysis::MustAlias:
+          Changed = true;
+          // If the load has a builtin retain, insert a plain retain for it.
+          if (Class == IC_LoadWeakRetained) {
+            CallInst *CI =
+              CallInst::Create(getRetainCallee(F.getParent()), EarlierCall,
+                               "", Call);
+            CI->setTailCall();
+          }
+          // Zap the fully redundant load.
+          Call->replaceAllUsesWith(EarlierCall->getArgOperand(1));
+          Call->eraseFromParent();
+          goto clobbered;
+        case AliasAnalysis::MayAlias:
+        case AliasAnalysis::PartialAlias:
+          goto clobbered;
+        case AliasAnalysis::NoAlias:
+          break;
+        }
+        break;
+      }
+      case IC_MoveWeak:
+      case IC_CopyWeak:
+        // TOOD: Grab the copied value.
+        goto clobbered;
+      case IC_AutoreleasepoolPush:
+      case IC_None:
+      case IC_User:
+        // Weak pointers are only modified through the weak entry points
+        // (and arbitrary calls, which could call the weak entry points).
+        break;
+      default:
+        // Anything else could modify the weak pointer.
+        goto clobbered;
+      }
+    }
+  clobbered:;
+  }
+
+  // Then, for each destroyWeak with an alloca operand, check to see if
+  // the alloca and all its users can be zapped.
+  for (inst_iterator I = inst_begin(&F), E = inst_end(&F); I != E; ) {
+    Instruction *Inst = &*I++;
+    InstructionClass Class = GetBasicInstructionClass(Inst);
+    if (Class != IC_DestroyWeak)
+      continue;
+
+    CallInst *Call = cast<CallInst>(Inst);
+    Value *Arg = Call->getArgOperand(0);
+    if (AllocaInst *Alloca = dyn_cast<AllocaInst>(Arg)) {
+      for (Value::use_iterator UI = Alloca->use_begin(),
+           UE = Alloca->use_end(); UI != UE; ++UI) {
+        Instruction *UserInst = cast<Instruction>(*UI);
+        switch (GetBasicInstructionClass(UserInst)) {
+        case IC_InitWeak:
+        case IC_StoreWeak:
+        case IC_DestroyWeak:
+          continue;
+        default:
+          goto done;
+        }
+      }
+      Changed = true;
+      for (Value::use_iterator UI = Alloca->use_begin(),
+           UE = Alloca->use_end(); UI != UE; ) {
+        CallInst *UserInst = cast<CallInst>(*UI++);
+        if (!UserInst->use_empty())
+          UserInst->replaceAllUsesWith(UserInst->getOperand(1));
+        UserInst->eraseFromParent();
+      }
+      Alloca->eraseFromParent();
+    done:;
+    }
+  }
+}
+
+/// OptimizeSequences - Identify program paths which execute sequences of
+/// retains and releases which can be eliminated.
+bool ObjCARCOpt::OptimizeSequences(Function &F) {
+  /// Releases, Retains - These are used to store the results of the main flow
+  /// analysis. These use Value* as the key instead of Instruction* so that the
+  /// map stays valid when we get around to rewriting code and calls get
+  /// replaced by arguments.
+  DenseMap<Value *, RRInfo> Releases;
+  MapVector<Value *, RRInfo> Retains;
+
+  /// BBStates, This is used during the traversal of the function to track the
+  /// states for each identified object at each block.
+  DenseMap<const BasicBlock *, BBState> BBStates;
+
+  // Analyze the CFG of the function, and all instructions.
+  bool NestingDetected = Visit(F, BBStates, Retains, Releases);
+
+  // Transform.
+  return PerformCodePlacement(BBStates, Retains, Releases) && NestingDetected;
+}
+
+/// OptimizeReturns - Look for this pattern:
+///
+///    %call = call i8* @something(...)
+///    %2 = call i8* @objc_retain(i8* %call)
+///    %3 = call i8* @objc_autorelease(i8* %2)
+///    ret i8* %3
+///
+/// And delete the retain and autorelease.
+///
+/// Otherwise if it's just this:
+///
+///    %3 = call i8* @objc_autorelease(i8* %2)
+///    ret i8* %3
+///
+/// convert the autorelease to autoreleaseRV.
+void ObjCARCOpt::OptimizeReturns(Function &F) {
+  if (!F.getReturnType()->isPointerTy())
+    return;
+
+  SmallPtrSet<Instruction *, 4> DependingInstructions;
+  SmallPtrSet<const BasicBlock *, 4> Visited;
+  for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) {
+    BasicBlock *BB = FI;
+    ReturnInst *Ret = dyn_cast<ReturnInst>(&BB->back());
+    if (!Ret) continue;
+
+    const Value *Arg = StripPointerCastsAndObjCCalls(Ret->getOperand(0));
+    FindDependencies(NeedsPositiveRetainCount, Arg,
+                     BB, Ret, DependingInstructions, Visited, PA);
+    if (DependingInstructions.size() != 1)
+      goto next_block;
+
+    {
+      CallInst *Autorelease =
+        dyn_cast_or_null<CallInst>(*DependingInstructions.begin());
+      if (!Autorelease)
+        goto next_block;
+      InstructionClass AutoreleaseClass =
+        GetBasicInstructionClass(Autorelease);
+      if (!IsAutorelease(AutoreleaseClass))
+        goto next_block;
+      if (GetObjCArg(Autorelease) != Arg)
+        goto next_block;
+
+      DependingInstructions.clear();
+      Visited.clear();
+
+      // Check that there is nothing that can affect the reference
+      // count between the autorelease and the retain.
+      FindDependencies(CanChangeRetainCount, Arg,
+                       BB, Autorelease, DependingInstructions, Visited, PA);
+      if (DependingInstructions.size() != 1)
+        goto next_block;
+
+      {
+        CallInst *Retain =
+          dyn_cast_or_null<CallInst>(*DependingInstructions.begin());
+
+        // Check that we found a retain with the same argument.
+        if (!Retain ||
+            !IsRetain(GetBasicInstructionClass(Retain)) ||
+            GetObjCArg(Retain) != Arg)
+          goto next_block;
+
+        DependingInstructions.clear();
+        Visited.clear();
+
+        // Convert the autorelease to an autoreleaseRV, since it's
+        // returning the value.
+        if (AutoreleaseClass == IC_Autorelease) {
+          Autorelease->setCalledFunction(getAutoreleaseRVCallee(F.getParent()));
+          AutoreleaseClass = IC_AutoreleaseRV;
+        }
+
+        // Check that there is nothing that can affect the reference
+        // count between the retain and the call.
+        FindDependencies(CanChangeRetainCount, Arg, BB, Retain,
+                         DependingInstructions, Visited, PA);
+        if (DependingInstructions.size() != 1)
+          goto next_block;
+
+        {
+          CallInst *Call =
+            dyn_cast_or_null<CallInst>(*DependingInstructions.begin());
+
+          // Check that the pointer is the return value of the call.
+          if (!Call || Arg != Call)
+            goto next_block;
+
+          // Check that the call is a regular call.
+          InstructionClass Class = GetBasicInstructionClass(Call);
+          if (Class != IC_CallOrUser && Class != IC_Call)
+            goto next_block;
+
+          // If so, we can zap the retain and autorelease.
+          Changed = true;
+          ++NumRets;
+          EraseInstruction(Retain);
+          EraseInstruction(Autorelease);
+        }
+      }
+    }
+
+  next_block:
+    DependingInstructions.clear();
+    Visited.clear();
+  }
+}
+
+bool ObjCARCOpt::doInitialization(Module &M) {
+  if (!EnableARCOpts)
+    return false;
+
+  // Identify the imprecise release metadata kind.
+  ImpreciseReleaseMDKind =
+    M.getContext().getMDKindID("clang.imprecise_release");
+
+  // Identify the declarations for objc_retain and friends.
+  RetainFunc = M.getFunction("objc_retain");
+  RetainBlockFunc = M.getFunction("objc_retainBlock");
+  RetainRVFunc = M.getFunction("objc_retainAutoreleasedReturnValue");
+  ReleaseFunc = M.getFunction("objc_release");
+  AutoreleaseFunc = M.getFunction("objc_autorelease");
+
+  // Intuitively, objc_retain and others are nocapture, however in practice
+  // they are not, because they return their argument value. And objc_release
+  // calls finalizers.
+
+  // These are initialized lazily.
+  RetainRVCallee = 0;
+  AutoreleaseRVCallee = 0;
+  ReleaseCallee = 0;
+  RetainCallee = 0;
+  AutoreleaseCallee = 0;
+
+  return false;
+}
+
+bool ObjCARCOpt::runOnFunction(Function &F) {
+  if (!EnableARCOpts)
+    return false;
+
+  Changed = false;
+
+  PA.setAA(&getAnalysis<AliasAnalysis>());
+
+  // This pass performs several distinct transformations. As a compile-time aid
+  // when compiling code that isn't ObjC, skip these if the relevant ObjC
+  // library functions aren't declared.
+
+  // Preliminary optimizations. This also computs UsedInThisFunction.
+  OptimizeIndividualCalls(F);
+
+  // Optimizations for weak pointers.
+  if (UsedInThisFunction & ((1 << IC_LoadWeak) |
+                            (1 << IC_LoadWeakRetained) |
+                            (1 << IC_StoreWeak) |
+                            (1 << IC_InitWeak) |
+                            (1 << IC_CopyWeak) |
+                            (1 << IC_MoveWeak) |
+                            (1 << IC_DestroyWeak)))
+    OptimizeWeakCalls(F);
+
+  // Optimizations for retain+release pairs.
+  if (UsedInThisFunction & ((1 << IC_Retain) |
+                            (1 << IC_RetainRV) |
+                            (1 << IC_RetainBlock)))
+    if (UsedInThisFunction & (1 << IC_Release))
+      // Run OptimizeSequences until it either stops making changes or
+      // no retain+release pair nesting is detected.
+      while (OptimizeSequences(F)) {}
+
+  // Optimizations if objc_autorelease is used.
+  if (UsedInThisFunction &
+      ((1 << IC_Autorelease) | (1 << IC_AutoreleaseRV)))
+    OptimizeReturns(F);
+
+  return Changed;
+}
+
+void ObjCARCOpt::releaseMemory() {
+  PA.clear();
+}
+
+//===----------------------------------------------------------------------===//
+// ARC contraction.
+//===----------------------------------------------------------------------===//
+
+// TODO: ObjCARCContract could insert PHI nodes when uses aren't
+// dominated by single calls.
+
+#include "llvm/Operator.h"
+#include "llvm/InlineAsm.h"
+#include "llvm/Analysis/Dominators.h"
+
+STATISTIC(NumStoreStrongs, "Number objc_storeStrong calls formed");
+
+namespace {
+  /// ObjCARCContract - Late ARC optimizations.  These change the IR in a way
+  /// that makes it difficult to be analyzed by ObjCARCOpt, so it's run late.
+  class ObjCARCContract : public FunctionPass {
+    bool Changed;
+    AliasAnalysis *AA;
+    DominatorTree *DT;
+    ProvenanceAnalysis PA;
+
+    /// StoreStrongCallee, etc. - Declarations for ObjC runtime
+    /// functions, for use in creating calls to them. These are initialized
+    /// lazily to avoid cluttering up the Module with unused declarations.
+    Constant *StoreStrongCallee,
+             *RetainAutoreleaseCallee, *RetainAutoreleaseRVCallee;
+
+    /// RetainRVMarker - The inline asm string to insert between calls and
+    /// RetainRV calls to make the optimization work on targets which need it.
+    const MDString *RetainRVMarker;
+
+    Constant *getStoreStrongCallee(Module *M);
+    Constant *getRetainAutoreleaseCallee(Module *M);
+    Constant *getRetainAutoreleaseRVCallee(Module *M);
+
+    bool ContractAutorelease(Function &F, Instruction *Autorelease,
+                             InstructionClass Class,
+                             SmallPtrSet<Instruction *, 4>
+                               &DependingInstructions,
+                             SmallPtrSet<const BasicBlock *, 4>
+                               &Visited);
+
+    void ContractRelease(Instruction *Release,
+                         inst_iterator &Iter);
+
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const;
+    virtual bool doInitialization(Module &M);
+    virtual bool runOnFunction(Function &F);
+
+  public:
+    static char ID;
+    ObjCARCContract() : FunctionPass(ID) {
+      initializeObjCARCContractPass(*PassRegistry::getPassRegistry());
+    }
+  };
+}
+
+char ObjCARCContract::ID = 0;
+INITIALIZE_PASS_BEGIN(ObjCARCContract,
+                      "objc-arc-contract", "ObjC ARC contraction", false, false)
+INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_DEPENDENCY(DominatorTree)
+INITIALIZE_PASS_END(ObjCARCContract,
+                    "objc-arc-contract", "ObjC ARC contraction", false, false)
+
+Pass *llvm::createObjCARCContractPass() {
+  return new ObjCARCContract();
+}
+
+void ObjCARCContract::getAnalysisUsage(AnalysisUsage &AU) const {
+  AU.addRequired<AliasAnalysis>();
+  AU.addRequired<DominatorTree>();
+  AU.setPreservesCFG();
+}
+
+Constant *ObjCARCContract::getStoreStrongCallee(Module *M) {
+  if (!StoreStrongCallee) {
+    LLVMContext &C = M->getContext();
+    const Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
+    const Type *I8XX = PointerType::getUnqual(I8X);
+    std::vector<const Type *> Params;
+    Params.push_back(I8XX);
+    Params.push_back(I8X);
+
+    AttrListPtr Attributes;
+    Attributes.addAttr(~0u, Attribute::NoUnwind);
+    Attributes.addAttr(1, Attribute::NoCapture);
+
+    StoreStrongCallee =
+      M->getOrInsertFunction(
+        "objc_storeStrong",
+        FunctionType::get(Type::getVoidTy(C), Params, /*isVarArg=*/false),
+        Attributes);
+  }
+  return StoreStrongCallee;
+}
+
+Constant *ObjCARCContract::getRetainAutoreleaseCallee(Module *M) {
+  if (!RetainAutoreleaseCallee) {
+    LLVMContext &C = M->getContext();
+    const Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
+    std::vector<const Type *> Params;
+    Params.push_back(I8X);
+    const FunctionType *FTy =
+      FunctionType::get(I8X, Params, /*isVarArg=*/false);
+    AttrListPtr Attributes;
+    Attributes.addAttr(~0u, Attribute::NoUnwind);
+    RetainAutoreleaseCallee =
+      M->getOrInsertFunction("objc_retainAutorelease", FTy, Attributes);
+  }
+  return RetainAutoreleaseCallee;
+}
+
+Constant *ObjCARCContract::getRetainAutoreleaseRVCallee(Module *M) {
+  if (!RetainAutoreleaseRVCallee) {
+    LLVMContext &C = M->getContext();
+    const Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
+    std::vector<const Type *> Params;
+    Params.push_back(I8X);
+    const FunctionType *FTy =
+      FunctionType::get(I8X, Params, /*isVarArg=*/false);
+    AttrListPtr Attributes;
+    Attributes.addAttr(~0u, Attribute::NoUnwind);
+    RetainAutoreleaseRVCallee =
+      M->getOrInsertFunction("objc_retainAutoreleaseReturnValue", FTy,
+                             Attributes);
+  }
+  return RetainAutoreleaseRVCallee;
+}
+
+/// ContractAutorelease - Merge an autorelease with a retain into a fused
+/// call.
+bool
+ObjCARCContract::ContractAutorelease(Function &F, Instruction *Autorelease,
+                                     InstructionClass Class,
+                                     SmallPtrSet<Instruction *, 4>
+                                       &DependingInstructions,
+                                     SmallPtrSet<const BasicBlock *, 4>
+                                       &Visited) {
+  const Value *Arg = GetObjCArg(Autorelease);
+
+  // Check that there are no instructions between the retain and the autorelease
+  // (such as an autorelease_pop) which may change the count.
+  CallInst *Retain = 0;
+  if (Class == IC_AutoreleaseRV)
+    FindDependencies(RetainAutoreleaseRVDep, Arg,
+                     Autorelease->getParent(), Autorelease,
+                     DependingInstructions, Visited, PA);
+  else
+    FindDependencies(RetainAutoreleaseDep, Arg,
+                     Autorelease->getParent(), Autorelease,
+                     DependingInstructions, Visited, PA);
+
+  Visited.clear();
+  if (DependingInstructions.size() != 1) {
+    DependingInstructions.clear();
+    return false;
+  }
+
+  Retain = dyn_cast_or_null<CallInst>(*DependingInstructions.begin());
+  DependingInstructions.clear();
+
+  if (!Retain ||
+      GetBasicInstructionClass(Retain) != IC_Retain ||
+      GetObjCArg(Retain) != Arg)
+    return false;
+
+  Changed = true;
+  ++NumPeeps;
+
+  if (Class == IC_AutoreleaseRV)
+    Retain->setCalledFunction(getRetainAutoreleaseRVCallee(F.getParent()));
+  else
+    Retain->setCalledFunction(getRetainAutoreleaseCallee(F.getParent()));
+
+  EraseInstruction(Autorelease);
+  return true;
+}
+
+/// ContractRelease - Attempt to merge an objc_release with a store, load, and
+/// objc_retain to form an objc_storeStrong. This can be a little tricky because
+/// the instructions don't always appear in order, and there may be unrelated
+/// intervening instructions.
+void ObjCARCContract::ContractRelease(Instruction *Release,
+                                      inst_iterator &Iter) {
+  LoadInst *Load = dyn_cast<LoadInst>(GetObjCArg(Release));
+  if (!Load || Load->isVolatile()) return;
+
+  // For now, require everything to be in one basic block.
+  BasicBlock *BB = Release->getParent();
+  if (Load->getParent() != BB) return;
+
+  // Walk down to find the store.
+  BasicBlock::iterator I = Load, End = BB->end();
+  ++I;
+  AliasAnalysis::Location Loc = AA->getLocation(Load);
+  while (I != End &&
+         (&*I == Release ||
+          IsRetain(GetBasicInstructionClass(I)) ||
+          !(AA->getModRefInfo(I, Loc) & AliasAnalysis::Mod)))
+    ++I;
+  StoreInst *Store = dyn_cast<StoreInst>(I);
+  if (!Store || Store->isVolatile()) return;
+  if (Store->getPointerOperand() != Loc.Ptr) return;
+
+  Value *New = StripPointerCastsAndObjCCalls(Store->getValueOperand());
+
+  // Walk up to find the retain.
+  I = Store;
+  BasicBlock::iterator Begin = BB->begin();
+  while (I != Begin && GetBasicInstructionClass(I) != IC_Retain)
+    --I;
+  Instruction *Retain = I;
+  if (GetBasicInstructionClass(Retain) != IC_Retain) return;
+  if (GetObjCArg(Retain) != New) return;
+
+  Changed = true;
+  ++NumStoreStrongs;
+
+  LLVMContext &C = Release->getContext();
+  const Type *I8X = PointerType::getUnqual(Type::getInt8Ty(C));
+  const Type *I8XX = PointerType::getUnqual(I8X);
+
+  Value *Args[] = { Load->getPointerOperand(), New };
+  if (Args[0]->getType() != I8XX)
+    Args[0] = new BitCastInst(Args[0], I8XX, "", Store);
+  if (Args[1]->getType() != I8X)
+    Args[1] = new BitCastInst(Args[1], I8X, "", Store);
+  CallInst *StoreStrong =
+    CallInst::Create(getStoreStrongCallee(BB->getParent()->getParent()),
+                     Args, array_endof(Args), "", Store);
+  StoreStrong->setDoesNotThrow();
+  StoreStrong->setDebugLoc(Store->getDebugLoc());
+
+  if (&*Iter == Store) ++Iter;
+  Store->eraseFromParent();
+  Release->eraseFromParent();
+  EraseInstruction(Retain);
+  if (Load->use_empty())
+    Load->eraseFromParent();
+}
+
+bool ObjCARCContract::doInitialization(Module &M) {
+  // These are initialized lazily.
+  StoreStrongCallee = 0;
+  RetainAutoreleaseCallee = 0;
+  RetainAutoreleaseRVCallee = 0;
+
+  // Initialize RetainRVMarker.
+  RetainRVMarker = 0;
+  if (NamedMDNode *NMD =
+        M.getNamedMetadata("clang.arc.retainAutoreleasedReturnValueMarker"))
+    if (NMD->getNumOperands() == 1) {
+      const MDNode *N = NMD->getOperand(0);
+      if (N->getNumOperands() == 1)
+        if (const MDString *S = dyn_cast<MDString>(N->getOperand(0)))
+          RetainRVMarker = S;
+    }
+
+  return false;
+}
+
+bool ObjCARCContract::runOnFunction(Function &F) {
+  if (!EnableARCOpts)
+    return false;
+
+  Changed = false;
+  AA = &getAnalysis<AliasAnalysis>();
+  DT = &getAnalysis<DominatorTree>();
+
+  PA.setAA(&getAnalysis<AliasAnalysis>());
+
+  // For ObjC library calls which return their argument, replace uses of the
+  // argument with uses of the call return value, if it dominates the use. This
+  // reduces register pressure.
+  SmallPtrSet<Instruction *, 4> DependingInstructions;
+  SmallPtrSet<const BasicBlock *, 4> Visited;
+  for (inst_iterator I = inst_begin(&F), E = inst_end(&F); I != E; ) {
+    Instruction *Inst = &*I++;
+
+    // Only these library routines return their argument. In particular,
+    // objc_retainBlock does not necessarily return its argument.
+    InstructionClass Class = GetBasicInstructionClass(Inst);
+    switch (Class) {
+    case IC_Retain:
+    case IC_FusedRetainAutorelease:
+    case IC_FusedRetainAutoreleaseRV:
+      break;
+    case IC_Autorelease:
+    case IC_AutoreleaseRV:
+      if (ContractAutorelease(F, Inst, Class, DependingInstructions, Visited))
+        continue;
+      break;
+    case IC_RetainRV: {
+      // If we're compiling for a target which needs a special inline-asm
+      // marker to do the retainAutoreleasedReturnValue optimization,
+      // insert it now.
+      if (!RetainRVMarker)
+        break;
+      BasicBlock::iterator BBI = Inst;
+      --BBI;
+      while (isNoopInstruction(BBI)) --BBI;
+      if (&*BBI == GetObjCArg(Inst)) {
+        InlineAsm *IA =
+          InlineAsm::get(FunctionType::get(Type::getVoidTy(Inst->getContext()),
+                                           /*isVarArg=*/false),
+                         RetainRVMarker->getString(),
+                         /*Constraints=*/"", /*hasSideEffects=*/true);
+        CallInst::Create(IA, "", Inst);
+      }
+      break;
+    }
+    case IC_InitWeak: {
+      // objc_initWeak(p, null) => *p = null
+      CallInst *CI = cast<CallInst>(Inst);
+      if (isNullOrUndef(CI->getArgOperand(1))) {
+        Value *Null =
+          ConstantPointerNull::get(cast<PointerType>(CI->getType()));
+        Changed = true;
+        new StoreInst(Null, CI->getArgOperand(0), CI);
+        CI->replaceAllUsesWith(Null);
+        CI->eraseFromParent();
+      }
+      continue;
+    }
+    case IC_Release:
+      ContractRelease(Inst, I);
+      continue;
+    default:
+      continue;
+    }
+
+    // Don't use GetObjCArg because we don't want to look through bitcasts
+    // and such; to do the replacement, the argument must have type i8*.
+    const Value *Arg = cast<CallInst>(Inst)->getArgOperand(0);
+    for (;;) {
+      // If we're compiling bugpointed code, don't get in trouble.
+      if (!isa<Instruction>(Arg) && !isa<Argument>(Arg))
+        break;
+      // Look through the uses of the pointer.
+      for (Value::const_use_iterator UI = Arg->use_begin(), UE = Arg->use_end();
+           UI != UE; ) {
+        Use &U = UI.getUse();
+        unsigned OperandNo = UI.getOperandNo();
+        ++UI; // Increment UI now, because we may unlink its element.
+        if (Instruction *UserInst = dyn_cast<Instruction>(U.getUser()))
+          if (Inst != UserInst && DT->dominates(Inst, UserInst)) {
+            Changed = true;
+            Instruction *Replacement = Inst;
+            const Type *UseTy = U.get()->getType();
+            if (PHINode *PHI = dyn_cast<PHINode>(UserInst)) {
+              // For PHI nodes, insert the bitcast in the predecessor block.
+              unsigned ValNo =
+                PHINode::getIncomingValueNumForOperand(OperandNo);
+              BasicBlock *BB =
+                PHI->getIncomingBlock(ValNo);
+              if (Replacement->getType() != UseTy)
+                Replacement = new BitCastInst(Replacement, UseTy, "",
+                                              &BB->back());
+              for (unsigned i = 0, e = PHI->getNumIncomingValues();
+                   i != e; ++i)
+                if (PHI->getIncomingBlock(i) == BB) {
+                  // Keep the UI iterator valid.
+                  if (&PHI->getOperandUse(
+                        PHINode::getOperandNumForIncomingValue(i)) ==
+                        &UI.getUse())
+                    ++UI;
+                  PHI->setIncomingValue(i, Replacement);
+                }
+            } else {
+              if (Replacement->getType() != UseTy)
+                Replacement = new BitCastInst(Replacement, UseTy, "", UserInst);
+              U.set(Replacement);
+            }
+          }
+      }
+
+      // If Arg is a no-op casted pointer, strip one level of casts and
+      // iterate.
+      if (const BitCastInst *BI = dyn_cast<BitCastInst>(Arg))
+        Arg = BI->getOperand(0);
+      else if (isa<GEPOperator>(Arg) &&
+               cast<GEPOperator>(Arg)->hasAllZeroIndices())
+        Arg = cast<GEPOperator>(Arg)->getPointerOperand();
+      else if (isa<GlobalAlias>(Arg) &&
+               !cast<GlobalAlias>(Arg)->mayBeOverridden())
+        Arg = cast<GlobalAlias>(Arg)->getAliasee();
+      else
+        break;
+    }
+  }
+
+  return Changed;
+}
diff --git a/lib/Transforms/Scalar/Scalar.cpp b/lib/Transforms/Scalar/Scalar.cpp
index 32a0506174..158d65352f 100644
--- a/lib/Transforms/Scalar/Scalar.cpp
+++ b/lib/Transforms/Scalar/Scalar.cpp
@@ -49,6 +49,10 @@ void llvm::initializeScalarOpts(PassRegistry &Registry) {
   initializeLoopIdiomRecognizePass(Registry);
   initializeLowerAtomicPass(Registry);
   initializeMemCpyOptPass(Registry);
+  initializeObjCARCAliasAnalysisPass(Registry);
+  initializeObjCARCExpandPass(Registry);
+  initializeObjCARCContractPass(Registry);
+  initializeObjCARCOptPass(Registry);
   initializeReassociatePass(Registry);
   initializeRegToMemPass(Registry);
   initializeSCCPPass(Registry);
diff --git a/test/Transforms/ObjCARC/basic.ll b/test/Transforms/ObjCARC/basic.ll
new file mode 100644
index 0000000000..a6bbf8674b
--- /dev/null
+++ b/test/Transforms/ObjCARC/basic.ll
@@ -0,0 +1,1898 @@
+; RUN: opt -objc-arc -S < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64"
+
+declare i8* @objc_retain(i8*)
+declare void @objc_release(i8*)
+declare i8* @objc_autorelease(i8*)
+declare void @objc_autoreleasePoolPop(i8*)
+declare void @objc_autoreleasePoolPush()
+declare i8* @objc_retainBlock(i8*)
+
+declare i8* @objc_retainedObject(i8*)
+declare i8* @objc_unretainedObject(i8*)
+declare i8* @objc_unretainedPointer(i8*)
+
+declare void @use_pointer(i8*)
+declare void @callee()
+declare void @callee_fnptr(void ()*)
+declare void @invokee()
+declare i8* @returner()
+
+declare void @llvm.dbg.value(metadata, i64, metadata)
+
+declare i8* @objc_msgSend(i8*, i8*, ...)
+
+; Simple retain+release pair deletion, with some intervening control
+; flow and harmless instructions.
+
+; CHECK: define void @test0(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test0(i32* %x, i1 %p) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  br i1 %p, label %t, label %f
+
+t:
+  store i8 3, i8* %a
+  %b = bitcast i32* %x to float*
+  store float 2.0, float* %b
+  br label %return
+
+f:
+  store i32 7, i32* %x
+  br label %return
+
+return:
+  %c = bitcast i32* %x to i8*
+  call void @objc_release(i8* %c) nounwind
+  ret void
+}
+
+; Like test0 but the release isn't always executed when the retain is,
+; so the optimization is not safe.
+
+; TODO: Make the objc_release's argument be %0.
+
+; CHECK: define void @test1(
+; CHECK: @objc_retain(i8* %a)
+; CHECK: @objc_release
+; CHECK: }
+define void @test1(i32* %x, i1 %p, i1 %q) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  br i1 %p, label %t, label %f
+
+t:
+  store i8 3, i8* %a
+  %b = bitcast i32* %x to float*
+  store float 2.0, float* %b
+  br label %return
+
+f:
+  store i32 7, i32* %x
+  call void @callee()
+  br i1 %q, label %return, label %alt_return
+
+return:
+  %c = bitcast i32* %x to i8*
+  call void @objc_release(i8* %c) nounwind
+  ret void
+
+alt_return:
+  ret void
+}
+
+; Like test0 but the pointer is passed to an intervening call,
+; so the optimization is not safe.
+
+; CHECK: define void @test2(
+; CHECK: @objc_retain(i8* %a)
+; CHECK: @objc_release
+; CHECK: }
+define void @test2(i32* %x, i1 %p) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  br i1 %p, label %t, label %f
+
+t:
+  store i8 3, i8* %a
+  %b = bitcast i32* %x to float*
+  store float 2.0, float* %b
+  br label %return
+
+f:
+  store i32 7, i32* %x
+  call void @use_pointer(i8* %0)
+  %d = bitcast i32* %x to float*
+  store float 3.0, float* %d
+  br label %return
+
+return:
+  %c = bitcast i32* %x to i8*
+  call void @objc_release(i8* %c) nounwind
+  ret void
+}
+
+; Like test0 but the release is in a loop,
+; so the optimization is not safe.
+
+; TODO: For now, assume this can't happen.
+
+; CHECK: define void @test3(
+; TODO: @objc_retain(i8* %a)
+; TODO: @objc_release
+; CHECK: }
+define void @test3(i32* %x, i1* %q) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  br label %loop
+
+loop:
+  %c = bitcast i32* %x to i8*
+  call void @objc_release(i8* %c) nounwind
+  %j = volatile load i1* %q
+  br i1 %j, label %loop, label %return
+
+return:
+  ret void
+}
+
+; TODO: For now, assume this can't happen.
+
+; Like test0 but the retain is in a loop,
+; so the optimization is not safe.
+
+; CHECK: define void @test4(
+; TODO: @objc_retain(i8* %a)
+; TODO: @objc_release
+; CHECK: }
+define void @test4(i32* %x, i1* %q) nounwind {
+entry:
+  br label %loop
+
+loop:
+  %a = bitcast i32* %x to i8*
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  %j = volatile load i1* %q
+  br i1 %j, label %loop, label %return
+
+return:
+  %c = bitcast i32* %x to i8*
+  call void @objc_release(i8* %c) nounwind
+  ret void
+}
+
+; Like test0 but the pointer is conditionally passed to an intervening call,
+; so the optimization is not safe.
+
+; CHECK: define void @test5(
+; CHECK: @objc_retain(i8*
+; CHECK: @objc_release
+; CHECK: }
+define void @test5(i32* %x, i1 %q, i8* %y) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  %s = select i1 %q, i8* %y, i8* %0
+  call void @use_pointer(i8* %s)
+  store i32 7, i32* %x
+  %c = bitcast i32* %x to i8*
+  call void @objc_release(i8* %c) nounwind
+  ret void
+}
+
+; retain+release pair deletion, where the release happens on two different
+; flow paths.
+
+; CHECK: define void @test6(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test6(i32* %x, i1 %p) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  br i1 %p, label %t, label %f
+
+t:
+  store i8 3, i8* %a
+  %b = bitcast i32* %x to float*
+  store float 2.0, float* %b
+  %ct = bitcast i32* %x to i8*
+  call void @objc_release(i8* %ct) nounwind
+  br label %return
+
+f:
+  store i32 7, i32* %x
+  call void @callee()
+  %cf = bitcast i32* %x to i8*
+  call void @objc_release(i8* %cf) nounwind
+  br label %return
+
+return:
+  ret void
+}
+
+; retain+release pair deletion, where the retain happens on two different
+; flow paths.
+
+; CHECK: define void @test7(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test7(i32* %x, i1 %p) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  br i1 %p, label %t, label %f
+
+t:
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  store i8 3, i8* %a
+  %b = bitcast i32* %x to float*
+  store float 2.0, float* %b
+  br label %return
+
+f:
+  %1 = call i8* @objc_retain(i8* %a) nounwind
+  store i32 7, i32* %x
+  call void @callee()
+  br label %return
+
+return:
+  %c = bitcast i32* %x to i8*
+  call void @objc_release(i8* %c) nounwind
+  ret void
+}
+
+; Like test7, but there's a retain/retainBlock mismatch. Don't delete!
+
+; CHECK: define void @test7b
+; CHECK: t:
+; CHECK: call i8* @objc_retainBlock
+; CHECK: f:
+; CHECK: call i8* @objc_retain
+; CHECK: return:
+; CHECK: call void @objc_release
+; CHECK: }
+define void @test7b(i32* %x, i1 %p) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  br i1 %p, label %t, label %f
+
+t:
+  %0 = call i8* @objc_retainBlock(i8* %a) nounwind
+  store i8 3, i8* %a
+  %b = bitcast i32* %x to float*
+  store float 2.0, float* %b
+  br label %return
+
+f:
+  %1 = call i8* @objc_retain(i8* %a) nounwind
+  store i32 7, i32* %x
+  call void @callee()
+  br label %return
+
+return:
+  %c = bitcast i32* %x to i8*
+  call void @objc_release(i8* %c) nounwind
+  ret void
+}
+
+; retain+release pair deletion, where the retain and release both happen on
+; different flow paths. Wild!
+
+; CHECK: define void @test8(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test8(i32* %x, i1 %p, i1 %q) nounwind {
+entry:
+  %a = bitcast i32* %x to i8*
+  br i1 %p, label %t, label %f
+
+t:
+  %0 = call i8* @objc_retain(i8* %a) nounwind
+  store i8 3, i8* %a
+  %b = bitcast i32* %x to float*
+  store float 2.0, float* %b
+  br label %mid
+
+f:
+  %1 = call i8* @objc_retain(i8* %a) nounwind
+  store i32 7, i32* %x
+  br label %mid
+
+mid:
+  br i1 %q, label %u, label %g
+
+u:
+  call void @callee()
+  %cu = bitcast i32* %x to i8*
+  call void @objc_release(i8* %cu) nounwind
+  br label %return
+
+g:
+  %cg = bitcast i32* %x to i8*
+  call void @objc_release(i8* %cg) nounwind
+  br label %return
+
+return:
+  ret void
+}
+
+; Trivial retain+release pair deletion.
+
+; CHECK: define void @test9(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test9(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  call void @objc_release(i8* %0) nounwind
+  ret void
+}
+
+; Retain+release pair, but on an unknown pointer relationship. Don't delete!
+
+; CHECK: define void @test9b
+; CHECK: @objc_retain(i8* %x)
+; CHECK: @objc_release(i8* %s)
+; CHECK: }
+define void @test9b(i8* %x, i1 %j, i8* %p) nounwind {
+entry:
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  %s = select i1 %j, i8* %x, i8* %p
+  call void @objc_release(i8* %s) nounwind
+  ret void
+}
+
+; Trivial retain+release pair with intervening calls - don't delete!
+
+; CHECK: define void @test10(
+; CHECK: @objc_retain(i8* %x)
+; CHECK: @use_pointer
+; CHECK: @objc_release
+; CHECK: }
+define void @test10(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  call void @use_pointer(i8* %x)
+  call void @objc_release(i8* %0) nounwind
+  ret void
+}
+
+; Trivial retain+autoreleaserelease pair. Don't delete!
+; Also, add a tail keyword, since objc_retain can never be passed
+; a stack argument.
+
+; CHECK: define void @test11(
+; CHECK: tail call i8* @objc_retain(i8* %x) nounwind
+; CHECK: tail call i8* @objc_autorelease(i8* %0) nounwind
+; CHECK: }
+define void @test11(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  call i8* @objc_autorelease(i8* %0) nounwind
+  call void @use_pointer(i8* %x)
+  ret void
+}
+
+; Same as test11 but with no use_pointer call. Delete the pair!
+
+; CHECK: define void @test11a(
+; CHECK: entry:
+; CHECK-NEXT: ret void
+; CHECK: }
+define void @test11a(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  call i8* @objc_autorelease(i8* %0) nounwind
+  ret void
+}
+
+; Same as test11 but the value is returned. Do an RV optimization.
+
+; CHECK: define i8* @test11b(
+; CHECK: tail call i8* @objc_retain(i8* %x) nounwind
+; CHECK: tail call i8* @objc_autoreleaseReturnValue(i8* %0) nounwind
+; CHECK: }
+define i8* @test11b(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  call i8* @objc_autorelease(i8* %0) nounwind
+  ret i8* %x
+}
+
+; Trivial retain,release pair with intervening call, but it's dominated
+; by another retain - delete!
+
+; CHECK: define void @test12(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: @objc_retain(i8* %x)
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test12(i8* %x, i64 %n) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  call void @use_pointer(i8* %x)
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Trivial retain,autorelease pair. Don't delete!
+
+; CHECK: define void @test13(
+; CHECK: tail call i8* @objc_retain(i8* %x) nounwind
+; CHECK: tail call i8* @objc_retain(i8* %x) nounwind
+; CHECK: @use_pointer(i8* %x)
+; CHECK: tail call i8* @objc_autorelease(i8* %x) nounwind
+; CHECK: }
+define void @test13(i8* %x, i64 %n) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  call i8* @objc_autorelease(i8* %x) nounwind
+  ret void
+}
+
+; Delete the retain+release pair.
+
+; CHECK: define void @test13b
+; CHECK-NEXT: entry:
+; CHECK-NEXT: @objc_retain(i8* %x)
+; CHECK-NEXT: @use_pointer
+; CHECK-NEXT: @use_pointer
+; CHECK-NEXT: ret void
+define void @test13b(i8* %x, i64 %n) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  call void @use_pointer(i8* %x)
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Don't delete the retain+release pair because there's an
+; autoreleasePoolPop in the way.
+
+; CHECK: define void @test13c
+; CHECK: @objc_retain(i8* %x)
+; CHECK: @objc_autoreleasePoolPop
+; CHECK: @objc_retain(i8* %x)
+; CHECK: @use_pointer
+; CHECK: @objc_release
+; CHECK: }
+define void @test13c(i8* %x, i64 %n) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @objc_autoreleasePoolPop(i8* undef)
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  call void @use_pointer(i8* %x)
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Like test13c, but there's an autoreleasePoolPush in the way, but that
+; doesn't matter.
+
+; CHECK: define void @test13d
+; CHECK-NEXT: entry:
+; CHECK-NEXT: @objc_retain(i8* %x)
+; CHECK-NEXT: @objc_autoreleasePoolPush
+; CHECK-NEXT: @use_pointer
+; CHECK-NEXT: @use_pointer
+; CHECK-NEXT: ret void
+define void @test13d(i8* %x, i64 %n) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @objc_autoreleasePoolPush()
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  call void @use_pointer(i8* %x)
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Trivial retain,release pair with intervening call, but it's post-dominated
+; by another release - delete!
+
+; CHECK: define void @test14(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: @use_pointer
+; CHECK-NEXT: @use_pointer
+; CHECK-NEXT: @objc_release
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test14(i8* %x, i64 %n) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  call void @use_pointer(i8* %x)
+  call void @objc_release(i8* %x) nounwind
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Trivial retain,autorelease pair with intervening call, but it's post-dominated
+; by another release. Don't delete anything.
+
+; CHECK: define void @test15(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: @objc_retain(i8* %x)
+; CHECK-NEXT: @use_pointer
+; CHECK-NEXT: @objc_autorelease(i8* %x)
+; CHECK-NEXT: @objc_release
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test15(i8* %x, i64 %n) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  call i8* @objc_autorelease(i8* %x) nounwind
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Trivial retain,autorelease pair, post-dominated
+; by another release. Delete the retain and release.
+
+; CHECK: define void @test15b
+; CHECK-NEXT: entry:
+; CHECK-NEXT: @objc_autorelease
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test15b(i8* %x, i64 %n) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  call i8* @objc_autorelease(i8* %x) nounwind
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Retain+release pairs in diamonds, all dominated by a retain.
+
+; CHECK: define void @test16(
+; CHECK: @objc_retain(i8* %x)
+; CHECK-NOT: @objc
+; CHECK: }
+define void @test16(i1 %a, i1 %b, i8* %x) {
+entry:
+  call i8* @objc_retain(i8* %x) nounwind
+  br i1 %a, label %red, label %orange
+
+red:
+  call i8* @objc_retain(i8* %x) nounwind
+  br label %yellow
+
+orange:
+  call i8* @objc_retain(i8* %x) nounwind
+  br label %yellow
+
+yellow:
+  call void @use_pointer(i8* %x)
+  call void @use_pointer(i8* %x)
+  br i1 %b, label %green, label %blue
+
+green:
+  call void @objc_release(i8* %x) nounwind
+  br label %purple
+
+blue:
+  call void @objc_release(i8* %x) nounwind
+  br label %purple
+
+purple:
+  ret void
+}
+
+; Retain+release pairs in diamonds, all post-dominated by a release.
+
+; CHECK: define void @test17(
+; CHECK-NOT: @objc_
+; CHECK: purple:
+; CHECK: @objc_release
+; CHECK: }
+define void @test17(i1 %a, i1 %b, i8* %x) {
+entry:
+  br i1 %a, label %red, label %orange
+
+red:
+  call i8* @objc_retain(i8* %x) nounwind
+  br label %yellow
+
+orange:
+  call i8* @objc_retain(i8* %x) nounwind
+  br label %yellow
+
+yellow:
+  call void @use_pointer(i8* %x)
+  call void @use_pointer(i8* %x)
+  br i1 %b, label %green, label %blue
+
+green:
+  call void @objc_release(i8* %x) nounwind
+  br label %purple
+
+blue:
+  call void @objc_release(i8* %x) nounwind
+  br label %purple
+
+purple:
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Delete no-ops.
+
+; CHECK: define void @test18(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test18() {
+  call i8* @objc_retain(i8* null)
+  call void @objc_release(i8* null)
+  call i8* @objc_autorelease(i8* null)
+  ret void
+}
+
+; Delete no-ops where undef can be assumed to be null.
+
+; CHECK: define void @test18b
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test18b() {
+  call i8* @objc_retain(i8* undef)
+  call void @objc_release(i8* undef)
+  call i8* @objc_autorelease(i8* undef)
+  ret void
+}
+
+; Replace uses of arguments with uses of return values, to reduce
+; register pressure.
+
+; CHECK: define void @test19(i32* %y) {
+; CHECK:   %z = bitcast i32* %y to i8*
+; CHECK:   %0 = bitcast i32* %y to i8*
+; CHECK:   %1 = tail call i8* @objc_retain(i8* %0)
+; CHECK:   call void @use_pointer(i8* %z)
+; CHECK:   call void @use_pointer(i8* %z)
+; CHECK:   %2 = bitcast i32* %y to i8*
+; CHECK:   call void @objc_release(i8* %2)
+; CHECK:   ret void
+; CHECK: }
+define void @test19(i32* %y) {
+entry:
+  %x = bitcast i32* %y to i8*
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  %z = bitcast i32* %y to i8*
+  call void @use_pointer(i8* %z)
+  call void @use_pointer(i8* %z)
+  call void @objc_release(i8* %x)
+  ret void
+}
+
+; Bitcast insertion
+
+; CHECK: define void @test20(
+; CHECK: %tmp1 = tail call i8* @objc_retain(i8* %tmp) nounwind
+; CHECK-NEXT: invoke
+define void @test20(double* %self) {
+if.then12:
+  %tmp = bitcast double* %self to i8*
+  %tmp1 = call i8* @objc_retain(i8* %tmp) nounwind
+  invoke void @invokee()
+          to label %invoke.cont23 unwind label %lpad20
+
+invoke.cont23:                                    ; preds = %if.then12
+  invoke void @invokee()
+          to label %if.end unwind label %lpad20
+
+lpad20:                                           ; preds = %invoke.cont23, %if.then12
+  %tmp502 = phi double* [ undef, %invoke.cont23 ], [ %self, %if.then12 ]
+  unreachable
+
+if.end:                                           ; preds = %invoke.cont23
+  ret void
+}
+
+; Delete a redundant retain,autorelease when forwaring a call result
+; directly to a return value.
+
+; CHECK: define i8* @test21(
+; CHECK: call i8* @returner()
+; CHECK-NEXT: ret i8* %call
+define i8* @test21() {
+entry:
+  %call = call i8* @returner()
+  %0 = call i8* @objc_retain(i8* %call) nounwind
+  %1 = call i8* @objc_autorelease(i8* %0) nounwind
+  ret i8* %1
+}
+
+; Move an objc call up through a phi that has null operands.
+
+; CHECK: define void @test22(
+; CHECK: B:
+; CHECK:   %1 = bitcast double* %p to i8*
+; CHECK:   call void @objc_release(i8* %1)
+; CHECK:   br label %C
+; CHECK: C:                                                ; preds = %B, %A
+; CHECK-NOT: @objc_release
+; CHECK: }
+define void @test22(double* %p, i1 %a) {
+  br i1 %a, label %A, label %B
+A:
+  br label %C
+B:
+  br label %C
+C:
+  %h = phi double* [ null, %A ], [ %p, %B ]
+  %c = bitcast double* %h to i8*
+  call void @objc_release(i8* %c)
+  ret void
+}
+
+; Optimize objc_retainBlock.
+
+; CHECK: define void @test23(
+; CHECK-NOT: @objc_
+; CHECK: }
+%block0 = type { i64, i64, i8*, i8* }
+%block1 = type { i8**, i32, i32, i32 (%struct.__block_literal_1*)*, %block0* }
+%struct.__block_descriptor = type { i64, i64 }
+%struct.__block_literal_1 = type { i8**, i32, i32, i8**, %struct.__block_descriptor* }
+@__block_holder_tmp_1 = external constant %block1
+define void @test23() {
+entry:
+  %0 = call i8* @objc_retainBlock(i8* bitcast (%block1* @__block_holder_tmp_1 to i8*)) nounwind
+  call void @bar(i32 ()* bitcast (%block1* @__block_holder_tmp_1 to i32 ()*))
+  call void @bar(i32 ()* bitcast (%block1* @__block_holder_tmp_1 to i32 ()*))
+  call void @objc_release(i8* bitcast (%block1* @__block_holder_tmp_1 to i8*)) nounwind
+  ret void
+}
+
+; Don't optimize objc_retainBlock.
+
+; CHECK: define void @test23b
+; CHECK: @objc_retainBlock
+; CHECK: @objc_release
+; CHECK: }
+define void @test23b(i8* %p) {
+entry:
+  %0 = call i8* @objc_retainBlock(i8* %p) nounwind
+  call void @use_pointer(i8* %p)
+  call void @use_pointer(i8* %p)
+  call void @objc_release(i8* %p) nounwind
+  ret void
+}
+
+; Any call can decrement a retain count.
+
+; CHECK: define void @test24(
+; CHECK: @objc_retain(i8* %a)
+; CHECK: @objc_release
+; CHECK: }
+define void @test24(i8* %r, i8* %a) {
+  call i8* @objc_retain(i8* %a)
+  call void @use_pointer(i8* %r)
+  %q = load i8* %a
+  call void @objc_release(i8* %a)
+  ret void
+}
+
+; Don't move a retain/release pair if the release can be moved
+; but the retain can't be moved to balance it.
+
+; CHECK: define void @test25(
+; CHECK: entry:
+; CHECK:   call i8* @objc_retain(i8* %p)
+; CHECK: true:
+; CHECK: done:
+; CHECK:   call void @objc_release(i8* %p)
+; CHECK: }
+define void @test25(i8* %p, i1 %x) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  call void @callee()
+  br i1 %x, label %true, label %done
+
+true:
+  store i8 0, i8* %p
+  br label %done
+
+done:
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Don't move a retain/release pair if the retain can be moved
+; but the release can't be moved to balance it.
+
+; CHECK: define void @test26(
+; CHECK: entry:
+; CHECK:   call i8* @objc_retain(i8* %p)
+; CHECK: true:
+; CHECK: done:
+; CHECK:   call void @objc_release(i8* %p)
+; CHECK: }
+define void @test26(i8* %p, i1 %x) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  call void @callee()
+  br label %done
+
+done:
+  store i8 0, i8* %p
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Don't sink the retain,release into the loop.
+
+; CHECK: define void @test27(
+; CHECK: entry:
+; CHECK: call i8* @objc_retain(i8* %p)
+; CHECK: loop:
+; CHECK-NOT: @objc_
+; CHECK: done:
+; CHECK: call void @objc_release
+; CHECK: }
+define void @test27(i8* %p, i1 %x, i1 %y) {
+entry: 
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %loop, label %done
+
+loop:
+  call void @callee()
+  store i8 0, i8* %p
+  br i1 %y, label %done, label %loop
+  
+done: 
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Trivial code motion case: Triangle.
+
+; CHECK: define void @test28(
+; CHECK-NOT: @objc_
+; CHECK: true:
+; CHECK: call i8* @objc_retain(
+; CHECK: call void @callee()
+; CHECK: store
+; CHECK: call void @objc_release
+; CHECK: done:
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test28(i8* %p, i1 %x) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  call void @callee()
+  store i8 0, i8* %p
+  br label %done
+
+done:
+  call void @objc_release(i8* %p), !clang.imprecise_release !0
+  ret void
+}
+
+; Trivial code motion case: Triangle, but no metadata. Don't move past
+; unrelated memory references!
+
+; CHECK: define void @test28b
+; CHECK: call i8* @objc_retain(
+; CHECK: true:
+; CHECK-NOT: @objc_
+; CHECK: call void @callee()
+; CHECK-NOT: @objc_
+; CHECK: store
+; CHECK-NOT: @objc_
+; CHECK: done:
+; CHECK: @objc_release
+; CHECK: }
+define void @test28b(i8* %p, i1 %x, i8* noalias %t) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  call void @callee()
+  store i8 0, i8* %p
+  br label %done
+
+done:
+  store i8 0, i8* %t
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Trivial code motion case: Triangle, with metadata. Do move past
+; unrelated memory references! And preserve the metadata.
+
+; CHECK: define void @test28c
+; CHECK-NOT: @objc_
+; CHECK: true:
+; CHECK: call i8* @objc_retain(
+; CHECK: call void @callee()
+; CHECK: store
+; CHECK: call void @objc_release(i8* %p) nounwind, !clang.imprecise_release
+; CHECK: done:
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test28c(i8* %p, i1 %x, i8* noalias %t) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  call void @callee()
+  store i8 0, i8* %p
+  br label %done
+
+done:
+  store i8 0, i8* %t
+  call void @objc_release(i8* %p), !clang.imprecise_release !0
+  ret void
+}
+
+; Like test28. but with two releases.
+
+; CHECK: define void @test29(
+; CHECK-NOT: @objc_
+; CHECK: true:
+; CHECK: call i8* @objc_retain(
+; CHECK: call void @callee()
+; CHECK: store
+; CHECK: call void @objc_release
+; CHECK-NOT: @objc_release
+; CHECK: done:
+; CHECK-NOT: @objc_
+; CHECK: ohno:
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test29(i8* %p, i1 %x, i1 %y) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  call void @callee()
+  store i8 0, i8* %p
+  br i1 %y, label %done, label %ohno
+
+done:
+  call void @objc_release(i8* %p)
+  ret void
+
+ohno:
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Basic case with the use and call in a diamond
+; with an extra release.
+
+; CHECK: define void @test30(
+; CHECK-NOT: @objc_
+; CHECK: true:
+; CHECK: call i8* @objc_retain(
+; CHECK: call void @callee()
+; CHECK: store
+; CHECK: call void @objc_release
+; CHECK-NOT: @objc_release
+; CHECK: false:
+; CHECK-NOT: @objc_
+; CHECK: done:
+; CHECK-NOT: @objc_
+; CHECK: ohno:
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test30(i8* %p, i1 %x, i1 %y, i1 %z) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %false
+
+true:
+  call void @callee()
+  store i8 0, i8* %p
+  br i1 %y, label %done, label %ohno
+
+false:
+  br i1 %z, label %done, label %ohno
+
+done:
+  call void @objc_release(i8* %p)
+  ret void
+
+ohno:
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Basic case with a mergeable release.
+
+; CHECK: define void @test31(
+; CHECK: call i8* @objc_retain(i8* %p)
+; CHECK: call void @callee()
+; CHECK: store
+; CHECK: call void @objc_release
+; CHECK-NOT: @objc_release
+; CHECK: true:
+; CHECK-NOT: @objc_release
+; CHECK: false:
+; CHECK-NOT: @objc_release
+; CHECK: ret void
+; CHECK-NOT: @objc_release
+; CHECK: }
+define void @test31(i8* %p, i1 %x) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  call void @callee()
+  store i8 0, i8* %p
+  br i1 %x, label %true, label %false
+true:
+  call void @objc_release(i8* %p)
+  ret void
+false:
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Don't consider bitcasts or getelementptrs direct uses.
+
+; CHECK: define void @test32(
+; CHECK-NOT: @objc_
+; CHECK: true:
+; CHECK: call i8* @objc_retain(
+; CHECK: call void @callee()
+; CHECK: store
+; CHECK: call void @objc_release
+; CHECK: done:
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test32(i8* %p, i1 %x) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  call void @callee()
+  store i8 0, i8* %p
+  br label %done
+
+done:
+  %g = bitcast i8* %p to i8*
+  %h = getelementptr i8* %g, i64 0
+  call void @objc_release(i8* %g)
+  ret void
+}
+
+; Do consider icmps to be direct uses.
+
+; CHECK: define void @test33(
+; CHECK-NOT: @objc_
+; CHECK: true:
+; CHECK: call i8* @objc_retain(
+; CHECK: call void @callee()
+; CHECK: icmp
+; CHECK: call void @objc_release
+; CHECK: done:
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test33(i8* %p, i1 %x, i8* %y) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  call void @callee()
+  %v = icmp eq i8* %p, %y
+  br label %done
+
+done:
+  %g = bitcast i8* %p to i8*
+  %h = getelementptr i8* %g, i64 0
+  call void @objc_release(i8* %g)
+  ret void
+}
+
+; Delete retain,release if there's just a possible dec.
+
+; CHECK: define void @test34(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test34(i8* %p, i1 %x, i8* %y) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  call void @callee()
+  br label %done
+
+done:
+  %g = bitcast i8* %p to i8*
+  %h = getelementptr i8* %g, i64 0
+  call void @objc_release(i8* %g)
+  ret void
+}
+
+; Delete retain,release if there's just a use.
+
+; CHECK: define void @test35(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test35(i8* %p, i1 %x, i8* %y) {
+entry:
+  %f0 = call i8* @objc_retain(i8* %p)
+  br i1 %x, label %true, label %done
+
+true:
+  %v = icmp eq i8* %p, %y
+  br label %done
+
+done:
+  %g = bitcast i8* %p to i8*
+  %h = getelementptr i8* %g, i64 0
+  call void @objc_release(i8* %g)
+  ret void
+}
+
+; Delete a retain,release if there's no actual use.
+
+; CHECK: define void @test36(
+; CHECK-NOT: @objc_
+; CHECK: call void @callee()
+; CHECK-NOT: @objc_
+; CHECK: call void @callee()
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test36(i8* %p) {
+entry:
+  call i8* @objc_retain(i8* %p)
+  call void @callee()
+  call void @callee()
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Like test36, but with metadata.
+
+; CHECK: define void @test37(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test37(i8* %p) {
+entry:
+  call i8* @objc_retain(i8* %p)
+  call void @callee()
+  call void @callee()
+  call void @objc_release(i8* %p), !clang.imprecise_release !0
+  ret void
+}
+
+; Be aggressive about analyzing phis to eliminate possible uses.
+
+; CHECK: define void @test38(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test38(i8* %p, i1 %u, i1 %m, i8* %z, i8* %y, i8* %x, i8* %w) {
+entry:
+  call i8* @objc_retain(i8* %p)
+  br i1 %u, label %true, label %false
+true:
+  br i1 %m, label %a, label %b
+false:
+  br i1 %m, label %c, label %d
+a:
+  br label %e
+b:
+  br label %e
+c:
+  br label %f
+d:
+  br label %f
+e:
+  %j = phi i8* [ %z, %a ], [ %y, %b ]
+  br label %g
+f:
+  %k = phi i8* [ %w, %c ], [ %x, %d ]
+  br label %g
+g:
+  %h = phi i8* [ %j, %e ], [ %k, %f ]
+  call void @use_pointer(i8* %h)
+  call void @objc_release(i8* %p), !clang.imprecise_release !0
+  ret void
+}
+
+; Delete retain,release pairs around loops.
+
+; CHECK: define void @test39(
+; CHECK_NOT: @objc_
+; CHECK: }
+define void @test39(i8* %p) {
+entry:
+  %0 = call i8* @objc_retain(i8* %p)
+  br label %loop
+
+loop:                                             ; preds = %loop, %entry
+  br i1 undef, label %loop, label %exit
+
+exit:                                             ; preds = %loop
+  call void @objc_release(i8* %0), !clang.imprecise_release !0
+  ret void
+}
+
+; Delete retain,release pairs around loops containing uses.
+
+; CHECK: define void @test39b(
+; CHECK_NOT: @objc_
+; CHECK: }
+define void @test39b(i8* %p) {
+entry:
+  %0 = call i8* @objc_retain(i8* %p)
+  br label %loop
+
+loop:                                             ; preds = %loop, %entry
+  store i8 0, i8* %0
+  br i1 undef, label %loop, label %exit
+
+exit:                                             ; preds = %loop
+  call void @objc_release(i8* %0), !clang.imprecise_release !0
+  ret void
+}
+
+; Delete retain,release pairs around loops containing potential decrements.
+
+; CHECK: define void @test39c(
+; CHECK_NOT: @objc_
+; CHECK: }
+define void @test39c(i8* %p) {
+entry:
+  %0 = call i8* @objc_retain(i8* %p)
+  br label %loop
+
+loop:                                             ; preds = %loop, %entry
+  call void @use_pointer(i8* %0)
+  br i1 undef, label %loop, label %exit
+
+exit:                                             ; preds = %loop
+  call void @objc_release(i8* %0), !clang.imprecise_release !0
+  ret void
+}
+
+; Delete retain,release pairs around loops even if
+; the successors are in a different order.
+
+; CHECK: define void @test40(
+; CHECK_NOT: @objc_
+; CHECK: }
+define void @test40(i8* %p) {
+entry:
+  %0 = call i8* @objc_retain(i8* %p)
+  br label %loop
+
+loop:                                             ; preds = %loop, %entry
+  call void @use_pointer(i8* %0)
+  br i1 undef, label %exit, label %loop
+
+exit:                                             ; preds = %loop
+  call void @objc_release(i8* %0), !clang.imprecise_release !0
+  ret void
+}
+
+; Do the known-incremented retain+release elimination even if the pointer
+; is also autoreleased.
+
+; CHECK: define void @test42(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call i8* @objc_retain(i8* %p)
+; CHECK-NEXT: call i8* @objc_autorelease(i8* %p)
+; CHECK-NEXT: call void @use_pointer(i8* %p)
+; CHECK-NEXT: call void @use_pointer(i8* %p)
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test42(i8* %p) {
+entry:
+  call i8* @objc_retain(i8* %p)
+  call i8* @objc_autorelease(i8* %p)
+  call i8* @objc_retain(i8* %p)
+  call void @use_pointer(i8* %p)
+  call void @use_pointer(i8* %p)
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Don't the known-incremented retain+release elimination if the pointer is
+; autoreleased and there's an autoreleasePoolPop.
+
+; CHECK: define void @test43(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call i8* @objc_retain(i8* %p)
+; CHECK-NEXT: call i8* @objc_autorelease(i8* %p)
+; CHECK-NEXT: call i8* @objc_retain
+; CHECK-NEXT: call void @use_pointer(i8* %p)
+; CHECK-NEXT: call void @use_pointer(i8* %p)
+; CHECK-NEXT: call void @objc_autoreleasePoolPop(i8* undef)
+; CHECK-NEXT: call void @objc_release
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test43(i8* %p) {
+entry:
+  call i8* @objc_retain(i8* %p)
+  call i8* @objc_autorelease(i8* %p)
+  call i8* @objc_retain(i8* %p)
+  call void @use_pointer(i8* %p)
+  call void @use_pointer(i8* %p)
+  call void @objc_autoreleasePoolPop(i8* undef)
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Do the known-incremented retain+release elimination if the pointer is
+; autoreleased and there's an autoreleasePoolPush.
+
+; CHECK: define void @test43b
+; CHECK-NEXT: entry:
+; CHECK-NEXT: call i8* @objc_retain(i8* %p)
+; CHECK-NEXT: call i8* @objc_autorelease(i8* %p)
+; CHECK-NEXT: call void @use_pointer(i8* %p)
+; CHECK-NEXT: call void @use_pointer(i8* %p)
+; CHECK-NEXT: call void @objc_autoreleasePoolPush()
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test43b(i8* %p) {
+entry:
+  call i8* @objc_retain(i8* %p)
+  call i8* @objc_autorelease(i8* %p)
+  call i8* @objc_retain(i8* %p)
+  call void @use_pointer(i8* %p)
+  call void @use_pointer(i8* %p)
+  call void @objc_autoreleasePoolPush()
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Do retain+release elimination for non-provenance pointers.
+
+; CHECK: define void @test44(
+; CHECK-NOT: objc_
+; CHECK: }
+define void @test44(i8** %pp) {
+  %p = load i8** %pp
+  %q = call i8* @objc_retain(i8* %p)
+  call void @objc_release(i8* %q)
+  ret void
+}
+
+; Don't delete retain+release with an unknown-provenance
+; may-alias objc_release between them.
+
+; CHECK: define void @test45(
+; CHECK: call i8* @objc_retain(i8* %p)
+; CHECK: call void @objc_release(i8* %q)
+; CHECK: call void @use_pointer(i8* %p)
+; CHECK: call void @objc_release(i8* %p)
+define void @test45(i8** %pp, i8** %qq) {
+  %p = load i8** %pp
+  %q = load i8** %qq
+  call i8* @objc_retain(i8* %p)
+  call void @objc_release(i8* %q)
+  call void @use_pointer(i8* %p)
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Don't delete retain and autorelease here.
+
+; CHECK: define void @test46(
+; CHECK: tail call i8* @objc_retain(i8* %p) nounwind
+; CHECK: true:
+; CHECK: tail call i8* @objc_autorelease(i8* %p) nounwind
+define void @test46(i8* %p, i1 %a) {
+entry:
+  call i8* @objc_retain(i8* %p)
+  br i1 %a, label %true, label %false
+
+true:
+  call i8* @objc_autorelease(i8* %p)
+  call void @use_pointer(i8* %p)
+  ret void
+
+false:
+  ret void
+}
+
+; Delete no-op cast calls.
+
+; CHECK: define i8* @test47(
+; CHECK-NOT: call
+; CHECK: ret i8* %p
+define i8* @test47(i8* %p) nounwind {
+  %x = call i8* @objc_retainedObject(i8* %p)
+  ret i8* %x
+}
+
+; Delete no-op cast calls.
+
+; CHECK: define i8* @test48(
+; CHECK-NOT: call
+; CHECK: ret i8* %p
+define i8* @test48(i8* %p) nounwind {
+  %x = call i8* @objc_unretainedObject(i8* %p)
+  ret i8* %x
+}
+
+; Delete no-op cast calls.
+
+; CHECK: define i8* @test49(
+; CHECK-NOT: call
+; CHECK: ret i8* %p
+define i8* @test49(i8* %p) nounwind {
+  %x = call i8* @objc_unretainedPointer(i8* %p)
+  ret i8* %x
+}
+
+; Do delete retain+release with intervening stores of the
+; address value;
+
+; CHECK: define void @test50(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test50(i8* %p, i8** %pp) {
+  call i8* @objc_retain(i8* %p)
+  call void @callee()
+  store i8* %p, i8** %pp
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Don't delete retain+release with intervening stores through the
+; address value.
+
+; CHECK: define void @test51(
+; CHECK: call i8* @objc_retain(i8* %p)
+; CHECK: call void @objc_release(i8* %p)
+define void @test51(i8* %p) {
+  call i8* @objc_retain(i8* %p)
+  call void @callee()
+  store i8 0, i8* %p
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Don't delete retain+release with intervening use of a pointer of
+; unknown provenance.
+
+; CHECK: define void @test52(
+; CHECK: call i8* @objc_retain
+; CHECK: call void @callee()
+; CHECK: call void @use_pointer(i8* %z)
+; CHECK: call void @objc_release
+define void @test52(i8** %zz, i8** %pp) {
+  %p = load i8** %pp
+  %1 = call i8* @objc_retain(i8* %p)
+  call void @callee()
+  %z = load i8** %zz
+  call void @use_pointer(i8* %z)
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Like test52, but the pointer has function type, so it's assumed to
+; be not reference counted.
+
+; CHECK: define void @test53(
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test53(void ()** %zz, i8** %pp) {
+  %p = load i8** %pp
+  %1 = call i8* @objc_retain(i8* %p)
+  call void @callee()
+  %z = load void ()** %zz
+  call void @callee_fnptr(void ()* %z)
+  call void @objc_release(i8* %p)
+  ret void
+}
+
+; Convert autorelease to release if the value is unused.
+
+; CHECK: define void @test54(
+; CHECK: call i8* @returner()
+; CHECK-NEXT: call void @objc_release(i8* %t) nounwind, !clang.imprecise_release !0
+; CHECK-NEXT: ret void
+define void @test54() {
+  %t = call i8* @returner()
+  call i8* @objc_autorelease(i8* %t)
+  ret void
+}
+
+; Nested retain+release pairs. Delete them both.
+
+; CHECK: define void @test55(
+; CHECK-NOT: @objc
+; CHECK: }
+define void @test55(i8* %x) { 
+entry: 
+  %0 = call i8* @objc_retain(i8* %x) nounwind 
+  %1 = call i8* @objc_retain(i8* %x) nounwind 
+  call void @objc_release(i8* %x) nounwind 
+  call void @objc_release(i8* %x) nounwind 
+  ret void 
+}
+
+; Nested retain+release pairs where the inner pair depends
+; on the outer pair to be removed, and then the outer pair
+; can be partially eliminated. Plus an extra outer pair to
+; eliminate, for fun.
+
+; CHECK: define void @test56(
+; CHECK-NOT: @objc
+; CHECK: if.then:
+; CHECK-NEXT: %0 = tail call i8* @objc_retain(i8* %x) nounwind
+; CHECK-NEXT: tail call void @use_pointer(i8* %x)
+; CHECK-NEXT: tail call void @use_pointer(i8* %x)
+; CHECK-NEXT: tail call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
+; CHECK-NEXT: br label %if.end
+; CHECK-NOT: @objc
+; CHECK: }
+define void @test56(i8* %x, i32 %n) {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %x) nounwind
+  %1 = tail call i8* @objc_retain(i8* %0) nounwind
+  %tobool = icmp eq i32 %n, 0
+  br i1 %tobool, label %if.end, label %if.then
+
+if.then:                                          ; preds = %entry
+  %2 = tail call i8* @objc_retain(i8* %1) nounwind
+  tail call void @use_pointer(i8* %2)
+  tail call void @use_pointer(i8* %2)
+  tail call void @objc_release(i8* %2) nounwind, !clang.imprecise_release !0
+  br label %if.end
+
+if.end:                                           ; preds = %entry, %if.then
+  tail call void @objc_release(i8* %1) nounwind, !clang.imprecise_release !0
+  tail call void @objc_release(i8* %0) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+declare void @bar(i32 ()*)
+
+; A few real-world testcases.
+
+@.str4 = private unnamed_addr constant [33 x i8] c"-[A z] = { %f, %f, { %f, %f } }\0A\00"
+@"OBJC_IVAR_$_A.myZ" = global i64 20, section "__DATA, __objc_const", align 8
+declare i32 @printf(i8* nocapture, ...) nounwind
+declare i32 @puts(i8* nocapture) nounwind
+@str = internal constant [16 x i8] c"-[ Top0 _getX ]\00"
+
+; CHECK: @"\01-[A z]"
+; CHECK-NOT: @objc_
+; CHECK: }
+
+define {<2 x float>, <2 x float>} @"\01-[A z]"({}* %self, i8* nocapture %_cmd) nounwind {
+invoke.cont:
+  %0 = bitcast {}* %self to i8*
+  %1 = tail call i8* @objc_retain(i8* %0) nounwind
+  tail call void @llvm.dbg.value(metadata !{{}* %self}, i64 0, metadata !0)
+  tail call void @llvm.dbg.value(metadata !{{}* %self}, i64 0, metadata !0)
+  %ivar = load i64* @"OBJC_IVAR_$_A.myZ", align 8
+  %add.ptr = getelementptr i8* %0, i64 %ivar
+  %tmp1 = bitcast i8* %add.ptr to float*
+  %tmp2 = load float* %tmp1, align 4
+  %conv = fpext float %tmp2 to double
+  %add.ptr.sum = add i64 %ivar, 4
+  %tmp6 = getelementptr inbounds i8* %0, i64 %add.ptr.sum
+  %2 = bitcast i8* %tmp6 to float*
+  %tmp7 = load float* %2, align 4
+  %conv8 = fpext float %tmp7 to double
+  %add.ptr.sum36 = add i64 %ivar, 8
+  %tmp12 = getelementptr inbounds i8* %0, i64 %add.ptr.sum36
+  %arrayidx = bitcast i8* %tmp12 to float*
+  %tmp13 = load float* %arrayidx, align 4
+  %conv14 = fpext float %tmp13 to double
+  %tmp12.sum = add i64 %ivar, 12
+  %arrayidx19 = getelementptr inbounds i8* %0, i64 %tmp12.sum
+  %3 = bitcast i8* %arrayidx19 to float*
+  %tmp20 = load float* %3, align 4
+  %conv21 = fpext float %tmp20 to double
+  %call = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([33 x i8]* @.str4, i64 0, i64 0), double %conv, double %conv8, double %conv14, double %conv21)
+  %ivar23 = load i64* @"OBJC_IVAR_$_A.myZ", align 8
+  %add.ptr24 = getelementptr i8* %0, i64 %ivar23
+  %4 = bitcast i8* %add.ptr24 to i128*
+  %srcval = load i128* %4, align 4
+  tail call void @objc_release(i8* %0) nounwind
+  %tmp29 = trunc i128 %srcval to i64
+  %tmp30 = bitcast i64 %tmp29 to <2 x float>
+  %tmp31 = insertvalue {<2 x float>, <2 x float>} undef, <2 x float> %tmp30, 0
+  %tmp32 = lshr i128 %srcval, 64
+  %tmp33 = trunc i128 %tmp32 to i64
+  %tmp34 = bitcast i64 %tmp33 to <2 x float>
+  %tmp35 = insertvalue {<2 x float>, <2 x float>} %tmp31, <2 x float> %tmp34, 1
+  ret {<2 x float>, <2 x float>} %tmp35
+}
+
+; CHECK: @"\01-[Top0 _getX]"
+; CHECK-NOT: @objc_
+; CHECK: }
+
+define i32 @"\01-[Top0 _getX]"({}* %self, i8* nocapture %_cmd) nounwind {
+invoke.cont:
+  %0 = bitcast {}* %self to i8*
+  %1 = tail call i8* @objc_retain(i8* %0) nounwind
+  %puts = tail call i32 @puts(i8* getelementptr inbounds ([16 x i8]* @str, i64 0, i64 0))
+  tail call void @objc_release(i8* %0) nounwind
+  ret i32 0
+}
+
+@"\01L_OBJC_METH_VAR_NAME_" = internal global [5 x i8] c"frob\00", section "__TEXT,__cstring,cstring_literals", align 1@"\01L_OBJC_SELECTOR_REFERENCES_" = internal global i8* getelementptr inbounds ([5 x i8]* @"\01L_OBJC_METH_VAR_NAME_", i64 0, i64 0), section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_IMAGE_INFO" = internal constant [2 x i32] [i32 0, i32 16], section "__DATA, __objc_imageinfo, regular, no_dead_strip"
+@llvm.used = appending global [3 x i8*] [i8* getelementptr inbounds ([5 x i8]* @"\01L_OBJC_METH_VAR_NAME_", i32 0, i32 0), i8* bitcast (i8** @"\01L_OBJC_SELECTOR_REFERENCES_" to i8*), i8* bitcast ([2 x i32]* @"\01L_OBJC_IMAGE_INFO" to i8*)], section "llvm.metadata"
+
+; A simple loop. Eliminate the retain and release inside of it!
+
+; CHECK: define void @loop
+; CHECK: for.body:
+; CHECK-NOT: @objc_
+; CHECK: @objc_msgSend
+; CHECK-NOT: @objc_
+; CHECK: for.end:
+define void @loop(i8* %x, i64 %n) {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %x) nounwind
+  %cmp9 = icmp sgt i64 %n, 0
+  br i1 %cmp9, label %for.body, label %for.end
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.010 = phi i64 [ %inc, %for.body ], [ 0, %entry ]
+  %1 = tail call i8* @objc_retain(i8* %x) nounwind
+  %tmp5 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_", align 8
+  %call = tail call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %1, i8* %tmp5)
+  tail call void @objc_release(i8* %1) nounwind, !clang.imprecise_release !0
+  %inc = add nsw i64 %i.010, 1
+  %exitcond = icmp eq i64 %inc, %n
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  tail call void @objc_release(i8* %x) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; ObjCARCOpt can delete the retain,release on self.
+
+; CHECK: define void @TextEditTest
+; CHECK-NOT: call i8* @objc_retain(i8* %tmp7)
+; CHECK: }
+
+%0 = type { i8* (i8*, %struct._message_ref_t*, ...)*, i8* }
+%1 = type opaque
+%2 = type opaque
+%3 = type opaque
+%4 = type opaque
+%5 = type opaque
+%struct.NSConstantString = type { i32*, i32, i8*, i64 }
+%struct._NSRange = type { i64, i64 }
+%struct.__CFString = type opaque
+%struct.__method_list_t = type { i32, i32, [0 x %struct._objc_method] }
+%struct._class_ro_t = type { i32, i32, i32, i8*, i8*, %struct.__method_list_t*, %struct._objc_protocol_list*, %struct._ivar_list_t*, i8*, %struct._prop_list_t* }
+%struct._class_t = type { %struct._class_t*, %struct._class_t*, %struct._objc_cache*, i8* (i8*, i8*)**, %struct._class_ro_t* }
+%struct._ivar_list_t = type { i32, i32, [0 x %struct._ivar_t] }
+%struct._ivar_t = type { i64*, i8*, i8*, i32, i32 }
+%struct._message_ref_t = type { i8*, i8* }
+%struct._objc_cache = type opaque
+%struct._objc_method = type { i8*, i8*, i8* }
+%struct._objc_protocol_list = type { i64, [0 x %struct._protocol_t*] }
+%struct._prop_list_t = type { i32, i32, [0 x %struct._message_ref_t] }
+%struct._protocol_t = type { i8*, i8*, %struct._objc_protocol_list*, %struct.__method_list_t*, %struct.__method_list_t*, %struct.__method_list_t*, %struct.__method_list_t*, %struct._prop_list_t*, i32, i32 }
+
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_17" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@kUTTypePlainText = external constant %struct.__CFString*
+@"\01L_OBJC_SELECTOR_REFERENCES_19" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_21" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_23" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_25" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_26" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@"\01L_OBJC_SELECTOR_REFERENCES_28" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_29" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@"\01L_OBJC_SELECTOR_REFERENCES_31" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_33" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_35" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_37" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_38" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@"\01L_OBJC_SELECTOR_REFERENCES_40" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_42" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@_unnamed_cfstring_44 = external hidden constant %struct.NSConstantString, section "__DATA,__cfstring"
+@"\01L_OBJC_SELECTOR_REFERENCES_46" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_48" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01l_objc_msgSend_fixup_isEqual_" = external hidden global %0, section "__DATA, __objc_msgrefs, coalesced", align 16
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_50" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@NSCocoaErrorDomain = external constant %1*
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_51" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@NSFilePathErrorKey = external constant %1*
+@"\01L_OBJC_SELECTOR_REFERENCES_53" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_55" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_56" = external hidden global %struct._class_t*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@"\01L_OBJC_SELECTOR_REFERENCES_58" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_60" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+
+declare %1* @truncatedString(%1*, i64)
+define void @TextEditTest(%2* %self, %3* %pboard) {
+entry:
+  %err = alloca %4*, align 8
+  %tmp7 = bitcast %2* %self to i8*
+  %tmp8 = call i8* @objc_retain(i8* %tmp7) nounwind
+  store %4* null, %4** %err, align 8
+  %tmp1 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_17", align 8
+  %tmp2 = load %struct.__CFString** @kUTTypePlainText, align 8
+  %tmp3 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_19", align 8
+  %tmp4 = bitcast %struct._class_t* %tmp1 to i8*
+  %call5 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp4, i8* %tmp3, %struct.__CFString* %tmp2)
+  %tmp5 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_21", align 8
+  %tmp6 = bitcast %3* %pboard to i8*
+  %call76 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp6, i8* %tmp5, i8* %call5)
+  %tmp9 = call i8* @objc_retain(i8* %call76) nounwind
+  %tobool = icmp eq i8* %tmp9, null
+  br i1 %tobool, label %end, label %land.lhs.true
+
+land.lhs.true:                                    ; preds = %entry
+  %tmp11 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_23", align 8
+  %call137 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp6, i8* %tmp11, i8* %tmp9)
+  %tmp = bitcast i8* %call137 to %1*
+  %tmp10 = call i8* @objc_retain(i8* %call137) nounwind
+  call void @objc_release(i8* null) nounwind
+  %tmp12 = call i8* @objc_retain(i8* %call137) nounwind
+  call void @objc_release(i8* null) nounwind
+  %tobool16 = icmp eq i8* %call137, null
+  br i1 %tobool16, label %end, label %if.then
+
+if.then:                                          ; preds = %land.lhs.true
+  %tmp19 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_25", align 8
+  %call21 = call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*)*)(i8* %call137, i8* %tmp19)
+  %tobool22 = icmp eq i8 %call21, 0
+  br i1 %tobool22, label %if.then44, label %land.lhs.true23
+
+land.lhs.true23:                                  ; preds = %if.then
+  %tmp24 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_26", align 8
+  %tmp26 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_28", align 8
+  %tmp27 = bitcast %struct._class_t* %tmp24 to i8*
+  %call2822 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp27, i8* %tmp26, i8* %call137)
+  %tmp13 = bitcast i8* %call2822 to %5*
+  %tmp14 = call i8* @objc_retain(i8* %call2822) nounwind
+  call void @objc_release(i8* null) nounwind
+  %tobool30 = icmp eq i8* %call2822, null
+  br i1 %tobool30, label %if.then44, label %if.end
+
+if.end:                                           ; preds = %land.lhs.true23
+  %tmp32 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_29", align 8
+  %tmp33 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_31", align 8
+  %tmp34 = bitcast %struct._class_t* %tmp32 to i8*
+  %call35 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp34, i8* %tmp33)
+  %tmp37 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_33", align 8
+  %call3923 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %call35, i8* %tmp37, i8* %call2822, i32 signext 1, %4** %err)
+  %cmp = icmp eq i8* %call3923, null
+  br i1 %cmp, label %if.then44, label %end
+
+if.then44:                                        ; preds = %if.end, %land.lhs.true23, %if.then
+  %url.025 = phi %5* [ %tmp13, %if.end ], [ %tmp13, %land.lhs.true23 ], [ null, %if.then ]
+  %tmp49 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_35", align 8
+  %call51 = call %struct._NSRange bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %struct._NSRange (i8*, i8*, i64, i64)*)(i8* %call137, i8* %tmp49, i64 0, i64 0)
+  %call513 = extractvalue %struct._NSRange %call51, 0
+  %call514 = extractvalue %struct._NSRange %call51, 1
+  %tmp52 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_37", align 8
+  %call548 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %call137, i8* %tmp52, i64 %call513, i64 %call514)
+  %tmp55 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_38", align 8
+  %tmp56 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_40", align 8
+  %tmp57 = bitcast %struct._class_t* %tmp55 to i8*
+  %call58 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp57, i8* %tmp56)
+  %tmp59 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_42", align 8
+  %call6110 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %call548, i8* %tmp59, i8* %call58)
+  %tmp15 = call i8* @objc_retain(i8* %call6110) nounwind
+  call void @objc_release(i8* %call137) nounwind
+  %tmp64 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_46", align 8
+  %call66 = call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*, %1*)*)(i8* %call6110, i8* %tmp64, %1* bitcast (%struct.NSConstantString* @_unnamed_cfstring_44 to %1*))
+  %tobool67 = icmp eq i8 %call66, 0
+  br i1 %tobool67, label %if.end74, label %if.then68
+
+if.then68:                                        ; preds = %if.then44
+  %tmp70 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_48", align 8
+  %call7220 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %call6110, i8* %tmp70)
+  %tmp16 = call i8* @objc_retain(i8* %call7220) nounwind
+  call void @objc_release(i8* %call6110) nounwind
+  br label %if.end74
+
+if.end74:                                         ; preds = %if.then68, %if.then44
+  %filename.0.in = phi i8* [ %call7220, %if.then68 ], [ %call6110, %if.then44 ]
+  %filename.0 = bitcast i8* %filename.0.in to %1*
+  %tmp17 = load i8** bitcast (%0* @"\01l_objc_msgSend_fixup_isEqual_" to i8**), align 16
+  %tmp18 = bitcast i8* %tmp17 to i8 (i8*, %struct._message_ref_t*, i8*, ...)*
+  %call78 = call signext i8 (i8*, %struct._message_ref_t*, i8*, ...)* %tmp18(i8* %call137, %struct._message_ref_t* bitcast (%0* @"\01l_objc_msgSend_fixup_isEqual_" to %struct._message_ref_t*), i8* %filename.0.in)
+  %tobool79 = icmp eq i8 %call78, 0
+  br i1 %tobool79, label %land.lhs.true80, label %if.then109
+
+land.lhs.true80:                                  ; preds = %if.end74
+  %tmp82 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_25", align 8
+  %call84 = call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*)*)(i8* %filename.0.in, i8* %tmp82)
+  %tobool86 = icmp eq i8 %call84, 0
+  br i1 %tobool86, label %if.then109, label %if.end106
+
+if.end106:                                        ; preds = %land.lhs.true80
+  %tmp88 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_26", align 8
+  %tmp90 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_28", align 8
+  %tmp91 = bitcast %struct._class_t* %tmp88 to i8*
+  %call9218 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp91, i8* %tmp90, i8* %filename.0.in)
+  %tmp20 = bitcast i8* %call9218 to %5*
+  %tmp21 = call i8* @objc_retain(i8* %call9218) nounwind
+  %tmp22 = bitcast %5* %url.025 to i8*
+  call void @objc_release(i8* %tmp22) nounwind
+  %tmp94 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_29", align 8
+  %tmp95 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_31", align 8
+  %tmp96 = bitcast %struct._class_t* %tmp94 to i8*
+  %call97 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp96, i8* %tmp95)
+  %tmp99 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_33", align 8
+  %call10119 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %call97, i8* %tmp99, i8* %call9218, i32 signext 1, %4** %err)
+  %phitmp = icmp eq i8* %call10119, null
+  br i1 %phitmp, label %if.then109, label %end
+
+if.then109:                                       ; preds = %if.end106, %land.lhs.true80, %if.end74
+  %url.129 = phi %5* [ %tmp20, %if.end106 ], [ %url.025, %if.end74 ], [ %url.025, %land.lhs.true80 ]
+  %tmp110 = load %4** %err, align 8
+  %tobool111 = icmp eq %4* %tmp110, null
+  br i1 %tobool111, label %if.then112, label %if.end125
+
+if.then112:                                       ; preds = %if.then109
+  %tmp113 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_50", align 8
+  %tmp114 = load %1** @NSCocoaErrorDomain, align 8
+  %tmp115 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_51", align 8
+  %call117 = call %1* @truncatedString(%1* %filename.0, i64 1034)
+  %tmp118 = load %1** @NSFilePathErrorKey, align 8
+  %tmp119 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_53", align 8
+  %tmp120 = bitcast %struct._class_t* %tmp115 to i8*
+  %call12113 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp120, i8* %tmp119, %1* %call117, %1* %tmp118, i8* null)
+  %tmp122 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_55", align 8
+  %tmp123 = bitcast %struct._class_t* %tmp113 to i8*
+  %call12414 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp123, i8* %tmp122, %1* %tmp114, i64 258, i8* %call12113)
+  %tmp23 = call i8* @objc_retain(i8* %call12414) nounwind
+  %tmp25 = call i8* @objc_autorelease(i8* %tmp23) nounwind
+  %tmp28 = bitcast i8* %tmp25 to %4*
+  store %4* %tmp28, %4** %err, align 8
+  br label %if.end125
+
+if.end125:                                        ; preds = %if.then112, %if.then109
+  %tmp127 = phi %4* [ %tmp110, %if.then109 ], [ %tmp28, %if.then112 ]
+  %tmp126 = load %struct._class_t** @"\01L_OBJC_CLASSLIST_REFERENCES_$_56", align 8
+  %tmp128 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_58", align 8
+  %tmp129 = bitcast %struct._class_t* %tmp126 to i8*
+  %call13015 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %tmp129, i8* %tmp128, %4* %tmp127)
+  %tmp131 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_60", align 8
+  %call13317 = call i8* (i8*, i8*, ...)* @objc_msgSend(i8* %call13015, i8* %tmp131)
+  br label %end
+
+end:                                              ; preds = %if.end125, %if.end106, %if.end, %land.lhs.true, %entry
+  %filename.2 = phi %1* [ %filename.0, %if.end106 ], [ %filename.0, %if.end125 ], [ %tmp, %land.lhs.true ], [ null, %entry ], [ %tmp, %if.end ]
+  %origFilename.0 = phi %1* [ %tmp, %if.end106 ], [ %tmp, %if.end125 ], [ %tmp, %land.lhs.true ], [ null, %entry ], [ %tmp, %if.end ]
+  %url.2 = phi %5* [ %tmp20, %if.end106 ], [ %url.129, %if.end125 ], [ null, %land.lhs.true ], [ null, %entry ], [ %tmp13, %if.end ]
+  call void @objc_release(i8* %tmp9) nounwind, !clang.imprecise_release !0
+  %tmp29 = bitcast %5* %url.2 to i8*
+  call void @objc_release(i8* %tmp29) nounwind, !clang.imprecise_release !0
+  %tmp30 = bitcast %1* %origFilename.0 to i8*
+  call void @objc_release(i8* %tmp30) nounwind, !clang.imprecise_release !0
+  %tmp31 = bitcast %1* %filename.2 to i8*
+  call void @objc_release(i8* %tmp31) nounwind, !clang.imprecise_release !0
+  call void @objc_release(i8* %tmp7) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+!0 = metadata !{}
diff --git a/test/Transforms/ObjCARC/cfg-hazards.ll b/test/Transforms/ObjCARC/cfg-hazards.ll
new file mode 100644
index 0000000000..e3624df57b
--- /dev/null
+++ b/test/Transforms/ObjCARC/cfg-hazards.ll
@@ -0,0 +1,86 @@
+; RUN: opt -S -objc-arc < %s | FileCheck %s
+; rdar://9503416
+
+; Detect loop boundaries and don't move retains and releases
+; across them.
+
+declare void @use_pointer(i8*)
+declare i8* @objc_retain(i8*)
+declare void @objc_release(i8*)
+
+; CHECK: define void @test0(
+; CHECK:   call i8* @objc_retain(
+; CHECK: for.body:
+; CHECK-NOT: @objc
+; CHECK: for.end:
+; CHECK:   call void @objc_release(
+; CHECK: }
+define void @test0(i8* %digits) {
+entry:
+  %tmp1 = call i8* @objc_retain(i8* %digits) nounwind
+  call void @use_pointer(i8* %tmp1)
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %upcDigitIndex.01 = phi i64 [ 2, %entry ], [ %inc, %for.body ]
+  call void @use_pointer(i8* %tmp1)
+  %inc = add i64 %upcDigitIndex.01, 1
+  %cmp = icmp ult i64 %inc, 12
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body
+  call void @objc_release(i8* %tmp1) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; CHECK: define void @test1(
+; CHECK:   call i8* @objc_retain(
+; CHECK: for.body:
+; CHECK-NOT: @objc
+; CHECK: for.end:
+; CHECK:   void @objc_release(
+; CHECK: }
+define void @test1(i8* %digits) {
+entry:
+  %tmp1 = call i8* @objc_retain(i8* %digits) nounwind
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %upcDigitIndex.01 = phi i64 [ 2, %entry ], [ %inc, %for.body ]
+  call void @use_pointer(i8* %tmp1)
+  call void @use_pointer(i8* %tmp1)
+  %inc = add i64 %upcDigitIndex.01, 1
+  %cmp = icmp ult i64 %inc, 12
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body
+  call void @objc_release(i8* %tmp1) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+; CHECK: define void @test2(
+; CHECK:   call i8* @objc_retain(
+; CHECK: for.body:
+; CHECK-NOT: @objc
+; CHECK: for.end:
+; CHECK:   void @objc_release(
+; CHECK: }
+define void @test2(i8* %digits) {
+entry:
+  %tmp1 = call i8* @objc_retain(i8* %digits) nounwind
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %upcDigitIndex.01 = phi i64 [ 2, %entry ], [ %inc, %for.body ]
+  call void @use_pointer(i8* %tmp1)
+  %inc = add i64 %upcDigitIndex.01, 1
+  %cmp = icmp ult i64 %inc, 12
+  br i1 %cmp, label %for.body, label %for.end
+
+for.end:                                          ; preds = %for.body
+  call void @use_pointer(i8* %tmp1)
+  call void @objc_release(i8* %tmp1) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+!0 = metadata !{}
diff --git a/test/Transforms/ObjCARC/contract-marker.ll b/test/Transforms/ObjCARC/contract-marker.ll
new file mode 100644
index 0000000000..01d978a0e2
--- /dev/null
+++ b/test/Transforms/ObjCARC/contract-marker.ll
@@ -0,0 +1,23 @@
+; RUN: opt -S -objc-arc-contract < %s | FileCheck %s
+
+; CHECK:      %call = tail call i32* @qux()
+; CHECK-NEXT: %tcall = bitcast i32* %call to i8*
+; CHECK-NEXT: call void asm sideeffect "mov\09r7, r7\09\09@ marker for objc_retainAutoreleaseReturnValue", ""()
+; CHECK-NEXT: %0 = tail call i8* @objc_retainAutoreleasedReturnValue(i8* %tcall) nounwind
+
+define void @foo() {
+entry:
+  %call = tail call i32* @qux()
+  %tcall = bitcast i32* %call to i8*
+  %0 = tail call i8* @objc_retainAutoreleasedReturnValue(i8* %tcall) nounwind
+  tail call void @bar(i8* %0)
+  ret void
+}
+
+declare i32* @qux()
+declare i8* @objc_retainAutoreleasedReturnValue(i8*)
+declare void @bar(i8*)
+
+!clang.arc.retainAutoreleasedReturnValueMarker = !{!0}
+
+!0 = metadata !{metadata !"mov\09r7, r7\09\09@ marker for objc_retainAutoreleaseReturnValue"}
diff --git a/test/Transforms/ObjCARC/contract-storestrong-ivar.ll b/test/Transforms/ObjCARC/contract-storestrong-ivar.ll
new file mode 100644
index 0000000000..4ad78e7534
--- /dev/null
+++ b/test/Transforms/ObjCARC/contract-storestrong-ivar.ll
@@ -0,0 +1,31 @@
+; RUN: opt -objc-arc-contract -S < %s | FileCheck %s
+
+; CHECK: call void @objc_storeStrong(i8**
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+target triple = "x86_64-apple-darwin11.0.0"
+
+%0 = type opaque
+%1 = type opaque
+
+@"OBJC_IVAR_$_Controller.preferencesController" = external global i64, section "__DATA, __objc_const", align 8
+
+declare i8* @objc_retain(i8*)
+
+declare void @objc_release(i8*)
+
+define hidden void @y(%0* nocapture %self, %1* %preferencesController) nounwind {
+entry:
+  %ivar = load i64* @"OBJC_IVAR_$_Controller.preferencesController", align 8
+  %tmp = bitcast %0* %self to i8*
+  %add.ptr = getelementptr inbounds i8* %tmp, i64 %ivar
+  %tmp1 = bitcast i8* %add.ptr to %1**
+  %tmp2 = load %1** %tmp1, align 8
+  %tmp3 = bitcast %1* %preferencesController to i8*
+  %tmp4 = tail call i8* @objc_retain(i8* %tmp3) nounwind
+  %tmp5 = bitcast %1* %tmp2 to i8*
+  tail call void @objc_release(i8* %tmp5) nounwind
+  %tmp6 = bitcast i8* %tmp4 to %1*
+  store %1* %tmp6, %1** %tmp1, align 8
+  ret void
+}
diff --git a/test/Transforms/ObjCARC/contract-storestrong.ll b/test/Transforms/ObjCARC/contract-storestrong.ll
new file mode 100644
index 0000000000..50ed260eb0
--- /dev/null
+++ b/test/Transforms/ObjCARC/contract-storestrong.ll
@@ -0,0 +1,59 @@
+; RUN: opt -objc-arc-contract -S < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64"
+
+declare i8* @objc_retain(i8*)
+declare void @objc_release(i8*)
+
+@x = external global i8*
+
+; CHECK: define void @test0(
+; CHECK: entry:
+; CHECK-NEXT: call void @objc_storeStrong(i8** @x, i8* %p) nounwind
+; CHECK-NEXT: ret void
+define void @test0(i8* %p) {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %p) nounwind
+  %tmp = load i8** @x, align 8
+  store i8* %0, i8** @x, align 8
+  tail call void @objc_release(i8* %tmp) nounwind
+  ret void
+}
+
+; Don't do this if the load is volatile.
+
+;      CHECK: define void @test1(i8* %p) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %0 = tail call i8* @objc_retain(i8* %p) nounwind
+; CHECK-NEXT:   %tmp = volatile load i8** @x, align 8
+; CHECK-NEXT:   store i8* %0, i8** @x, align 8
+; CHECK-NEXT:   tail call void @objc_release(i8* %tmp) nounwind
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test1(i8* %p) {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %p) nounwind
+  %tmp = volatile load i8** @x, align 8
+  store i8* %0, i8** @x, align 8
+  tail call void @objc_release(i8* %tmp) nounwind
+  ret void
+}
+
+; Don't do this if the store is volatile.
+
+;      CHECK: define void @test2(i8* %p) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %0 = tail call i8* @objc_retain(i8* %p) nounwind
+; CHECK-NEXT:   %tmp = load i8** @x, align 8
+; CHECK-NEXT:   volatile store i8* %0, i8** @x, align 8
+; CHECK-NEXT:   tail call void @objc_release(i8* %tmp) nounwind
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test2(i8* %p) {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %p) nounwind
+  %tmp = load i8** @x, align 8
+  volatile store i8* %0, i8** @x, align 8
+  tail call void @objc_release(i8* %tmp) nounwind
+  ret void
+}
diff --git a/test/Transforms/ObjCARC/contract-testcases.ll b/test/Transforms/ObjCARC/contract-testcases.ll
new file mode 100644
index 0000000000..69fa837675
--- /dev/null
+++ b/test/Transforms/ObjCARC/contract-testcases.ll
@@ -0,0 +1,63 @@
+; RUN: opt -objc-arc-contract -S < %s | FileCheck %s
+; rdar://9511608
+
+%0 = type opaque
+%1 = type opaque
+%2 = type { i64, i64 }
+%3 = type { i8*, i8* }
+%4 = type opaque
+
+declare %0* @"\01-[NSAttributedString(Terminal) pathAtIndex:effectiveRange:]"(%1*, i8* nocapture, i64, %2*) optsize
+declare i8* @objc_retainAutoreleasedReturnValue(i8*)
+declare i8* @objc_msgSend_fixup(i8*, %3*, ...)
+declare void @objc_release(i8*)
+declare %2 @NSUnionRange(i64, i64, i64, i64) optsize
+declare i8* @objc_autoreleaseReturnValue(i8*)
+declare i8* @objc_autorelease(i8*)
+declare i8* @objc_msgSend() nonlazybind
+
+; Don't get in trouble on bugpointed code.
+
+; CHECK: define void @test0(
+define void @test0() {
+bb:
+  %tmp = bitcast %4* undef to i8*
+  %tmp1 = tail call i8* @objc_retainAutoreleasedReturnValue(i8* %tmp) nounwind
+  br label %bb3
+
+bb3:                                              ; preds = %bb2
+  br i1 undef, label %bb6, label %bb4
+
+bb4:                                              ; preds = %bb3
+  switch i64 undef, label %bb5 [
+    i64 9223372036854775807, label %bb6
+    i64 0, label %bb6
+  ]
+
+bb5:                                              ; preds = %bb4
+  br label %bb6
+
+bb6:                                              ; preds = %bb5, %bb4, %bb4, %bb3
+  %tmp7 = phi %4* [ undef, %bb5 ], [ undef, %bb4 ], [ undef, %bb3 ], [ undef, %bb4 ]
+  unreachable
+}
+
+; When rewriting operands for a phi which has multiple operands
+; for the same block, use the exactly same value in each block.
+
+; CHECK: define void @test1(
+; CHECK: %0 = bitcast i8* %tmp3 to %0* 
+; CHECK: br i1 undef, label %bb7, label %bb7
+; CHECK: bb7:
+; CHECK: %tmp8 = phi %0* [ %0, %bb ], [ %0, %bb ]
+define void @test1() {
+bb:
+  %tmp = tail call %0* bitcast (i8* ()* @objc_msgSend to %0* ()*)()
+  %tmp2 = bitcast %0* %tmp to i8*
+  %tmp3 = tail call i8* @objc_retainAutoreleasedReturnValue(i8* %tmp2) nounwind
+  br i1 undef, label %bb7, label %bb7
+
+bb7:                                              ; preds = %bb6, %bb6, %bb5
+  %tmp8 = phi %0* [ %tmp, %bb ], [ %tmp, %bb ]
+  unreachable
+}
diff --git a/test/Transforms/ObjCARC/contract.ll b/test/Transforms/ObjCARC/contract.ll
new file mode 100644
index 0000000000..04ae3ca505
--- /dev/null
+++ b/test/Transforms/ObjCARC/contract.ll
@@ -0,0 +1,145 @@
+; RUN: opt -objc-arc-contract -S < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64"
+
+declare i8* @objc_retain(i8*)
+declare void @objc_release(i8*)
+declare i8* @objc_autorelease(i8*)
+declare i8* @objc_autoreleaseReturnValue(i8*)
+declare i8* @objc_retainAutoreleasedReturnValue(i8*)
+
+declare void @use_pointer(i8*)
+declare i8* @returner()
+
+; CHECK: define void @test0
+; CHECK: call void @use_pointer(i8* %0)
+; CHECK: }
+define void @test0(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  ret void
+}
+
+; CHECK: define void @test1
+; CHECK: call void @use_pointer(i8* %0)
+; CHECK: }
+define void @test1(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_autorelease(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  ret void
+}
+
+; Merge objc_retain and objc_autorelease into objc_retainAutorelease.
+
+; CHECK: define void @test2(
+; CHECK: tail call i8* @objc_retainAutorelease(i8* %x) nounwind
+; CHECK: }
+define void @test2(i8* %x) nounwind {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %x) nounwind
+  tail call i8* @objc_autorelease(i8* %0) nounwind
+  call void @use_pointer(i8* %x)
+  ret void
+}
+
+; Same as test2 but the value is returned. Do an RV optimization.
+
+; CHECK: define i8* @test2b(
+; CHECK: tail call i8* @objc_retainAutoreleaseReturnValue(i8* %x) nounwind
+; CHECK: }
+define i8* @test2b(i8* %x) nounwind {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %x) nounwind
+  tail call i8* @objc_autoreleaseReturnValue(i8* %0) nounwind
+  ret i8* %x
+}
+
+; Merge a retain,autorelease pair around a call.
+
+; CHECK: define void @test3(
+; CHECK: tail call i8* @objc_retainAutorelease(i8* %x) nounwind
+; CHECK: @use_pointer(i8* %0)
+; CHECK: }
+define void @test3(i8* %x, i64 %n) {
+entry:
+  tail call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  tail call i8* @objc_autorelease(i8* %x) nounwind
+  ret void
+}
+
+; Trivial retain,autorelease pair with intervening call, but it's post-dominated
+; by another release. The retain and autorelease can be merged.
+
+; CHECK: define void @test4(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: @objc_retainAutorelease(i8* %x) nounwind
+; CHECK-NEXT: @use_pointer
+; CHECK-NEXT: @objc_release
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test4(i8* %x, i64 %n) {
+entry:
+  tail call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  tail call i8* @objc_autorelease(i8* %x) nounwind
+  tail call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Don't merge retain and autorelease if they're not control-equivalent.
+
+; CHECK: define void @test5(
+; CHECK: tail call i8* @objc_retain(i8* %p) nounwind
+; CHECK: true:
+; CHECK: tail call i8* @objc_autorelease(i8* %0) nounwind
+; CHECK: }
+define void @test5(i8* %p, i1 %a) {
+entry:
+  tail call i8* @objc_retain(i8* %p) nounwind
+  br i1 %a, label %true, label %false
+
+true:
+  tail call i8* @objc_autorelease(i8* %p) nounwind
+  call void @use_pointer(i8* %p)
+  ret void
+
+false:
+  ret void
+}
+
+; Don't eliminate objc_retainAutoreleasedReturnValue by merging it into
+; an objc_autorelease.
+; TODO? Merge objc_retainAutoreleasedReturnValue and objc_autorelease into
+; objc_retainAutoreleasedReturnValueAutorelease and merge
+; objc_retainAutoreleasedReturnValue and objc_autoreleaseReturnValue
+; into objc_retainAutoreleasedReturnValueAutoreleaseReturnValue?
+; Those entrypoints don't exist yet though.
+
+; CHECK: define i8* @test6(
+; CHECK: call i8* @objc_retainAutoreleasedReturnValue(i8* %p) nounwind
+; CHECK: %t = tail call i8* @objc_autoreleaseReturnValue(i8* %1) nounwind
+; CHECK: }
+define i8* @test6() {
+  %p = call i8* @returner()
+  tail call i8* @objc_retainAutoreleasedReturnValue(i8* %p) nounwind
+  %t = tail call i8* @objc_autoreleaseReturnValue(i8* %p) nounwind
+  call void @use_pointer(i8* %t)
+  ret i8* %t
+}
+
+; Don't spoil the RV optimization.
+
+; CHECK: define i8* @test7(i8* %p)
+; CHECK: tail call i8* @objc_retain(i8* %p)
+; CHECK: call void @use_pointer(i8* %1)
+; CHECK: tail call i8* @objc_autoreleaseReturnValue(i8* %1)
+; CHECK: ret i8* %2
+define i8* @test7(i8* %p) {
+  %1 = tail call i8* @objc_retain(i8* %p)
+  call void @use_pointer(i8* %p)
+  %2 = tail call i8* @objc_autoreleaseReturnValue(i8* %p)
+  ret i8* %p
+}
diff --git a/test/Transforms/ObjCARC/dg.exp b/test/Transforms/ObjCARC/dg.exp
new file mode 100644
index 0000000000..f2005891a5
--- /dev/null
+++ b/test/Transforms/ObjCARC/dg.exp
@@ -0,0 +1,3 @@
+load_lib llvm.exp
+
+RunLLVMTests [lsort [glob -nocomplain $srcdir/$subdir/*.{ll,c,cpp}]]
diff --git a/test/Transforms/ObjCARC/expand.ll b/test/Transforms/ObjCARC/expand.ll
new file mode 100644
index 0000000000..5388673f2b
--- /dev/null
+++ b/test/Transforms/ObjCARC/expand.ll
@@ -0,0 +1,28 @@
+; RUN: opt -objc-arc-expand -S < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64"
+
+declare i8* @objc_retain(i8*)
+declare i8* @objc_autorelease(i8*)
+
+declare void @use_pointer(i8*)
+
+; CHECK: define void @test0
+; CHECK: call void @use_pointer(i8* %x)
+; CHECK: }
+define void @test0(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_retain(i8* %x) nounwind
+  call void @use_pointer(i8* %0)
+  ret void
+}
+
+; CHECK: define void @test1
+; CHECK: call void @use_pointer(i8* %x)
+; CHECK: }
+define void @test1(i8* %x) nounwind {
+entry:
+  %0 = call i8* @objc_autorelease(i8* %x) nounwind
+  call void @use_pointer(i8* %x)
+  ret void
+}
diff --git a/test/Transforms/ObjCARC/gvn.ll b/test/Transforms/ObjCARC/gvn.ll
new file mode 100644
index 0000000000..6917b02e03
--- /dev/null
+++ b/test/Transforms/ObjCARC/gvn.ll
@@ -0,0 +1,21 @@
+; RUN: opt -S -basicaa -objc-arc -gvn < %s | FileCheck %s
+
+@x = common global i8* null, align 8
+
+declare i8* @objc_retain(i8*)
+
+; GVN should be able to eliminate this redundant load, with ARC-specific
+; alias analysis.
+
+; CHECK: @foo
+; CHECK-NEXT: entry:
+; CHECK-NEXT: %s = load i8** @x
+; CHECK-NOT: load
+; CHECK: ret i8* %s
+define i8* @foo(i32 %n) nounwind {
+entry:
+  %s = load i8** @x
+  %0 = tail call i8* @objc_retain(i8* %s) nounwind
+  %t = load i8** @x
+  ret i8* %s
+}
diff --git a/test/Transforms/ObjCARC/move-and-form-retain-autorelease.ll b/test/Transforms/ObjCARC/move-and-form-retain-autorelease.ll
new file mode 100644
index 0000000000..170d0a99c9
--- /dev/null
+++ b/test/Transforms/ObjCARC/move-and-form-retain-autorelease.ll
@@ -0,0 +1,221 @@
+; RUN: opt -S -objc-arc-contract < %s | FileCheck %s
+
+; The optimizer should be able to move the autorelease past a control triangle
+; and various scary looking things and fold it into an objc_retainAutorelease.
+
+; CHECK: bb57:
+; CHECK: tail call i8* @objc_retainAutorelease(i8* %tmp71x) nounwind
+; CHECK: bb99:
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+target triple = "x86_64-apple-darwin11.0.0"
+
+%0 = type { i8* (i8*, %1*, ...)*, i8* }
+%1 = type { i8*, i8* }
+%2 = type { %2*, %2*, %3*, i8* (i8*, i8*)**, %4* }
+%3 = type opaque
+%4 = type { i32, i32, i32, i8*, i8*, %5*, %7*, %10*, i8*, %9* }
+%5 = type { i32, i32, [0 x %6] }
+%6 = type { i8*, i8*, i8* }
+%7 = type { i64, [0 x %8*] }
+%8 = type { i8*, i8*, %7*, %5*, %5*, %5*, %5*, %9*, i32, i32 }
+%9 = type { i32, i32, [0 x %1] }
+%10 = type { i32, i32, [0 x %11] }
+%11 = type { i64*, i8*, i8*, i32, i32 }
+%12 = type { i32*, i32, i8*, i64 }
+%13 = type opaque
+%14 = type opaque
+%15 = type opaque
+%16 = type opaque
+%17 = type opaque
+%18 = type opaque
+%19 = type opaque
+%20 = type opaque
+%21 = type opaque
+%22 = type opaque
+%23 = type opaque
+%24 = type opaque
+%25 = type opaque
+
+@"\01l_objc_msgSend_fixup_alloc" = external hidden global %0, section "__DATA, __objc_msgrefs, coalesced", align 16
+@"\01L_OBJC_SELECTOR_REFERENCES_8" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_3725" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_40" = external hidden global %2*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@"\01L_OBJC_SELECTOR_REFERENCES_4227" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_4631" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_70" = external hidden global %2*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@"\01L_OBJC_SELECTOR_REFERENCES_148" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_159" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_188" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_328" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01l_objc_msgSend_fixup_objectAtIndex_" = external hidden global %0, section "__DATA, __objc_msgrefs, coalesced", align 16
+@_unnamed_cfstring_386 = external hidden constant %12, section "__DATA,__cfstring"
+@"\01l_objc_msgSend_fixup_count" = external hidden global %0, section "__DATA, __objc_msgrefs, coalesced", align 16
+@"\01L_OBJC_SELECTOR_REFERENCES_389" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_391" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_393" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@NSPrintHeaderAndFooter = external constant %13*
+@"\01L_OBJC_SELECTOR_REFERENCES_395" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_396" = external hidden global %2*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@"\01L_OBJC_SELECTOR_REFERENCES_398" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_400" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_402" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_404" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_406" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_408" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_CLASSLIST_REFERENCES_$_409" = external hidden global %2*, section "__DATA, __objc_classrefs, regular, no_dead_strip", align 8
+@"\01L_OBJC_SELECTOR_REFERENCES_411" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_413" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_415" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+
+declare i8* @objc_msgSend(i8*, i8*, ...)
+
+declare i8* @objc_retain(i8*)
+
+declare void @objc_release(i8*)
+
+declare i8* @objc_autorelease(i8*)
+
+declare i8* @objc_explicit_autorelease(i8*)
+
+define hidden %14* @foo(%15* %arg, %16* %arg2) {
+bb:
+  %tmp = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_3725", align 8
+  %tmp4 = bitcast %15* %arg to i8*
+  %tmp5 = tail call %18* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %18* (i8*, i8*)*)(i8* %tmp4, i8* %tmp)
+  %tmp6 = bitcast %18* %tmp5 to i8*
+  %tmp7 = tail call i8* @objc_retain(i8* %tmp6) nounwind
+  %tmp8 = load %2** @"\01L_OBJC_CLASSLIST_REFERENCES_$_40", align 8
+  %tmp9 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_4227", align 8
+  %tmp10 = bitcast %2* %tmp8 to i8*
+  %tmp11 = tail call %19* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %19* (i8*, i8*)*)(i8* %tmp10, i8* %tmp9)
+  %tmp12 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_4631", align 8
+  %tmp13 = bitcast %19* %tmp11 to i8*
+  %tmp14 = tail call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*, %13*)*)(i8* %tmp13, i8* %tmp12, %13* bitcast (%12* @_unnamed_cfstring_386 to %13*))
+  %tmp15 = bitcast %16* %arg2 to i8*
+  %tmp16 = load i8** bitcast (%0* @"\01l_objc_msgSend_fixup_count" to i8**), align 16
+  %tmp17 = bitcast i8* %tmp16 to i64 (i8*, %1*)*
+  %tmp18 = tail call i64 %tmp17(i8* %tmp15, %1* bitcast (%0* @"\01l_objc_msgSend_fixup_count" to %1*))
+  %tmp19 = icmp eq i64 %tmp18, 0
+  br i1 %tmp19, label %bb22, label %bb20
+
+bb20:                                             ; preds = %bb
+  %tmp21 = icmp eq i8 %tmp14, 0
+  br label %bb25
+
+bb22:                                             ; preds = %bb
+  %tmp23 = bitcast i8* %tmp7 to %18*
+  %tmp24 = icmp eq i8 %tmp14, 0
+  br i1 %tmp24, label %bb46, label %bb25
+
+bb25:                                             ; preds = %bb22, %bb20
+  %tmp26 = phi i1 [ %tmp21, %bb20 ], [ false, %bb22 ]
+  %tmp27 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_188", align 8
+  %tmp28 = tail call i8* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8* (i8*, i8*)*)(i8* %tmp7, i8* %tmp27)
+  %tmp29 = tail call i8* @objc_explicit_autorelease(i8* %tmp28) nounwind
+  %tmp30 = bitcast i8* %tmp29 to %18*
+  tail call void @objc_release(i8* %tmp7) nounwind
+  %tmp31 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_389", align 8
+  %tmp32 = tail call %20* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %20* (i8*, i8*)*)(i8* %tmp29, i8* %tmp31)
+  %tmp33 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_391", align 8
+  %tmp34 = bitcast %20* %tmp32 to i8*
+  tail call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*, %16*)*)(i8* %tmp34, i8* %tmp33, %16* %arg2)
+  br i1 %tmp26, label %bb46, label %bb35
+
+bb35:                                             ; preds = %bb25
+  %tmp36 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_389", align 8
+  %tmp37 = tail call %20* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %20* (i8*, i8*)*)(i8* %tmp29, i8* %tmp36)
+  %tmp38 = load %2** @"\01L_OBJC_CLASSLIST_REFERENCES_$_70", align 8
+  %tmp39 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_393", align 8
+  %tmp40 = bitcast %2* %tmp38 to i8*
+  %tmp41 = tail call %21* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %21* (i8*, i8*, i8)*)(i8* %tmp40, i8* %tmp39, i8 signext 1)
+  %tmp42 = bitcast %21* %tmp41 to i8*
+  %tmp43 = load %13** @NSPrintHeaderAndFooter, align 8
+  %tmp44 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_159", align 8
+  %tmp45 = bitcast %20* %tmp37 to i8*
+  tail call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*, i8*, %13*)*)(i8* %tmp45, i8* %tmp44, i8* %tmp42, %13* %tmp43)
+  br label %bb46
+
+bb46:                                             ; preds = %bb35, %bb25, %bb22
+  %tmp47 = phi %18* [ %tmp30, %bb35 ], [ %tmp30, %bb25 ], [ %tmp23, %bb22 ]
+  %tmp48 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_328", align 8
+  %tmp49 = tail call %22* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %22* (i8*, i8*)*)(i8* %tmp4, i8* %tmp48)
+  %tmp50 = bitcast %22* %tmp49 to i8*
+  %tmp51 = load i8** bitcast (%0* @"\01l_objc_msgSend_fixup_count" to i8**), align 16
+  %tmp52 = bitcast i8* %tmp51 to i64 (i8*, %1*)*
+  %tmp53 = tail call i64 %tmp52(i8* %tmp50, %1* bitcast (%0* @"\01l_objc_msgSend_fixup_count" to %1*))
+  %tmp54 = icmp eq i64 %tmp53, 0
+  br i1 %tmp54, label %bb55, label %bb57
+
+bb55:                                             ; preds = %bb46
+  %tmp56 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_395", align 8
+  tail call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*)*)(i8* %tmp4, i8* %tmp56)
+  br label %bb57
+
+bb57:                                             ; preds = %bb55, %bb46
+  %tmp58 = load %2** @"\01L_OBJC_CLASSLIST_REFERENCES_$_396", align 8
+  %tmp59 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_328", align 8
+  %tmp60 = tail call %22* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %22* (i8*, i8*)*)(i8* %tmp4, i8* %tmp59)
+  %tmp61 = bitcast %22* %tmp60 to i8*
+  %tmp62 = load i8** bitcast (%0* @"\01l_objc_msgSend_fixup_objectAtIndex_" to i8**), align 16
+  %tmp63 = bitcast i8* %tmp62 to i8* (i8*, %1*, i64)*
+  %tmp64 = tail call i8* %tmp63(i8* %tmp61, %1* bitcast (%0* @"\01l_objc_msgSend_fixup_objectAtIndex_" to %1*), i64 0)
+  %tmp65 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_398", align 8
+  %tmp66 = tail call i8* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8* (i8*, i8*)*)(i8* %tmp64, i8* %tmp65)
+  %tmp67 = bitcast i8* %tmp66 to %23*
+  %tmp68 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_400", align 8
+  %tmp69 = bitcast %2* %tmp58 to i8*
+  %tmp70 = tail call %14* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %14* (i8*, i8*, %23*, %18*)*)(i8* %tmp69, i8* %tmp68, %23* %tmp67, %18* %tmp47)
+  %tmp71 = bitcast %14* %tmp70 to i8*
+  ; hack to prevent the optimize from using objc_retainAutoreleasedReturnValue.
+  %tmp71x = getelementptr i8* %tmp71, i64 1
+  %tmp72 = tail call i8* @objc_retain(i8* %tmp71x) nounwind
+  %tmp73 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_402", align 8
+  tail call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*, i8)*)(i8* %tmp72, i8* %tmp73, i8 signext 1)
+  %tmp74 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_404", align 8
+  tail call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*, i8)*)(i8* %tmp72, i8* %tmp74, i8 signext 1)
+  %tmp75 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_328", align 8
+  %tmp76 = tail call %22* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %22* (i8*, i8*)*)(i8* %tmp4, i8* %tmp75)
+  %tmp77 = bitcast %22* %tmp76 to i8*
+  %tmp78 = load i8** bitcast (%0* @"\01l_objc_msgSend_fixup_objectAtIndex_" to i8**), align 16
+  %tmp79 = bitcast i8* %tmp78 to i8* (i8*, %1*, i64)*
+  %tmp80 = tail call i8* %tmp79(i8* %tmp77, %1* bitcast (%0* @"\01l_objc_msgSend_fixup_objectAtIndex_" to %1*), i64 0)
+  %tmp81 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_406", align 8
+  tail call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*, i64)*)(i8* %tmp80, i8* %tmp81, i64 9223372036854775807)
+  %tmp82 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_408", align 8
+  %tmp83 = tail call %24* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %24* (i8*, i8*)*)(i8* %tmp72, i8* %tmp82)
+  %tmp84 = bitcast %24* %tmp83 to i8*
+  %tmp85 = tail call i8* @objc_retain(i8* %tmp84) nounwind
+  %tmp86 = load %2** @"\01L_OBJC_CLASSLIST_REFERENCES_$_409", align 8
+  %tmp87 = bitcast %2* %tmp86 to i8*
+  %tmp88 = load i8** bitcast (%0* @"\01l_objc_msgSend_fixup_alloc" to i8**), align 16
+  %tmp89 = bitcast i8* %tmp88 to i8* (i8*, %1*)*
+  %tmp90 = tail call i8* %tmp89(i8* %tmp87, %1* bitcast (%0* @"\01l_objc_msgSend_fixup_alloc" to %1*))
+  %tmp91 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_8", align 8
+  %tmp92 = tail call i8* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8* (i8*, i8*)*)(i8* %tmp90, i8* %tmp91)
+  %tmp93 = tail call i8* @objc_explicit_autorelease(i8* %tmp92) nounwind
+  %tmp94 = bitcast i8* %tmp93 to %25*
+  %tmp95 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_411", align 8
+  tail call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*, %25*)*)(i8* %tmp85, i8* %tmp95, %25* %tmp94)
+  tail call void @objc_release(i8* %tmp93) nounwind
+  %tmp96 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_148", align 8
+  %tmp97 = tail call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*)*)(i8* %tmp4, i8* %tmp96)
+  %tmp98 = icmp eq i8 %tmp97, 0
+  br i1 %tmp98, label %bb99, label %bb104
+
+bb99:                                             ; preds = %bb57
+  %tmp100 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_413", align 8
+  %tmp101 = tail call i64 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i64 (i8*, i8*)*)(i8* %tmp85, i8* %tmp100)
+  %tmp102 = or i64 %tmp101, 12
+  %tmp103 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_415", align 8
+  tail call void bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to void (i8*, i8*, i64)*)(i8* %tmp85, i8* %tmp103, i64 %tmp102)
+  br label %bb104
+
+bb104:                                            ; preds = %bb99, %bb57
+  %tmp105 = tail call i8* @objc_autorelease(i8* %tmp72) nounwind
+  %tmp106 = bitcast i8* %tmp105 to %14*
+  tail call void @objc_release(i8* %tmp85) nounwind
+  %tmp107 = bitcast %18* %tmp47 to i8*
+  tail call void @objc_release(i8* %tmp107) nounwind
+  ret %14* %tmp106
+}
diff --git a/test/Transforms/ObjCARC/move-and-merge-autorelease.ll b/test/Transforms/ObjCARC/move-and-merge-autorelease.ll
new file mode 100644
index 0000000000..8462c70a48
--- /dev/null
+++ b/test/Transforms/ObjCARC/move-and-merge-autorelease.ll
@@ -0,0 +1,108 @@
+; RUN: opt -S -objc-arc < %s | FileCheck %s
+
+; The optimizer should be able to move the autorelease past two phi nodes
+; and fold it with the release in bb65.
+
+; CHECK: bb65:
+; CHECK: call i8* @objc_retainAutorelease
+; CHECK: br label %bb76
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+target triple = "x86_64-apple-darwin11.0.0"
+
+%0 = type opaque
+%1 = type opaque
+%2 = type opaque
+%3 = type opaque
+%4 = type opaque
+%5 = type opaque
+
+@"\01L_OBJC_SELECTOR_REFERENCES_11" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_421455" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_598" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_620" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_622" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_624" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+@"\01L_OBJC_SELECTOR_REFERENCES_626" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
+
+declare i8* @objc_msgSend(i8*, i8*, ...)
+
+declare i8* @objc_retain(i8*)
+
+declare void @objc_release(i8*)
+
+declare i8* @objc_autorelease(i8*)
+
+define hidden %0* @foo(%1* %arg, %3* %arg3) {
+bb:
+  %tmp16 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_620", align 8
+  %tmp17 = bitcast %3* %arg3 to i8*
+  %tmp18 = call %4* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %4* (i8*, i8*)*)(i8* %tmp17, i8* %tmp16)
+  %tmp19 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_622", align 8
+  %tmp20 = bitcast %4* %tmp18 to i8*
+  %tmp21 = call %5* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %5* (i8*, i8*)*)(i8* %tmp20, i8* %tmp19)
+  %tmp22 = bitcast %5* %tmp21 to i8*
+  %tmp23 = call i8* @objc_retain(i8* %tmp22) nounwind
+  %tmp24 = bitcast i8* %tmp23 to %5*
+  %tmp26 = icmp eq i8* %tmp23, null
+  br i1 %tmp26, label %bb81, label %bb27
+
+bb27:                                             ; preds = %bb
+  %tmp29 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_11", align 8
+  %tmp30 = bitcast %1* %arg to i8*
+  %tmp31 = call i8* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8* (i8*, i8*)*)(i8* %tmp30, i8* %tmp29)
+  %tmp34 = call i8* @objc_retain(i8* %tmp31) nounwind
+  %tmp37 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_421455", align 8
+  %tmp39 = call %0* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %0* (i8*, i8*)*)(i8* %tmp34, i8* %tmp37)
+  %tmp40 = bitcast %0* %tmp39 to i8*
+  %tmp41 = call i8* @objc_retain(i8* %tmp40) nounwind
+  %tmp42 = bitcast i8* %tmp41 to %0*
+  %tmp44 = icmp eq i8* %tmp41, null
+  br i1 %tmp44, label %bb45, label %bb55
+
+bb45:                                             ; preds = %bb27
+  %tmp47 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_624", align 8
+  %tmp49 = call %0* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %0* (i8*, i8*)*)(i8* %tmp34, i8* %tmp47)
+  %tmp51 = bitcast %0* %tmp49 to i8*
+  %tmp52 = call i8* @objc_retain(i8* %tmp51) nounwind
+  call void @objc_release(i8* %tmp41) nounwind
+  br label %bb55
+
+bb55:                                             ; preds = %bb27, %bb45
+  %tmp13.0 = phi %0* [ %tmp42, %bb27 ], [ %tmp49, %bb45 ]
+  %tmp57 = icmp eq %0* %tmp13.0, null
+  br i1 %tmp57, label %bb76, label %bb58
+
+bb58:                                             ; preds = %bb55
+  %tmp60 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_598", align 8
+  %tmp61 = bitcast %0* %tmp13.0 to i8*
+  %tmp62 = call signext i8 bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to i8 (i8*, i8*)*)(i8* %tmp61, i8* %tmp60)
+  %tmp64 = icmp eq i8 %tmp62, 0
+  br i1 %tmp64, label %bb76, label %bb65
+
+bb65:                                             ; preds = %bb58
+  %tmp68 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_626", align 8
+  %tmp69 = bitcast %0* %tmp13.0 to i8*
+  %tmp70 = call %0* bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to %0* (i8*, i8*, %5*)*)(i8* %tmp69, i8* %tmp68, %5* %tmp24)
+  %tmp72 = bitcast %0* %tmp70 to i8*
+  %tmp73 = call i8* @objc_retain(i8* %tmp72) nounwind
+  br label %bb76
+
+bb76:                                             ; preds = %bb58, %bb55, %bb65
+  %tmp10.0 = phi %0* [ %tmp70, %bb65 ], [ null, %bb58 ], [ null, %bb55 ]
+  %tmp78 = bitcast %0* %tmp13.0 to i8*
+  call void @objc_release(i8* %tmp78) nounwind
+  call void @objc_release(i8* %tmp34) nounwind
+  br label %bb81
+
+bb81:                                             ; preds = %bb, %bb76
+  %tmp10.1 = phi %0* [ %tmp10.0, %bb76 ], [ null, %bb ]
+  %tmp83 = bitcast %0* %tmp10.1 to i8*
+  %tmp84 = call i8* @objc_retain(i8* %tmp83) nounwind
+  %tmp88 = bitcast i8* %tmp87 to %0*
+  call void @objc_release(i8* %tmp23) nounwind
+  %tmp87 = call i8* @objc_autorelease(i8* %tmp84) nounwind
+  %tmp92 = bitcast %0* %tmp10.1 to i8*
+  call void @objc_release(i8* %tmp92) nounwind
+  ret %0* %tmp88
+}
diff --git a/test/Transforms/ObjCARC/post-inlining.ll b/test/Transforms/ObjCARC/post-inlining.ll
new file mode 100644
index 0000000000..ad69ccdd79
--- /dev/null
+++ b/test/Transforms/ObjCARC/post-inlining.ll
@@ -0,0 +1,48 @@
+; RUN: opt -S -objc-arc < %s | FileCheck %s
+
+declare void @use_pointer(i8*)
+declare i8* @returner()
+declare i8* @objc_retain(i8*)
+declare i8* @objc_autoreleaseReturnValue(i8*)
+declare i8* @objc_retainAutoreleasedReturnValue(i8*)
+
+; Clean up residue left behind after inlining.
+
+; CHECK: define void @test0(
+; CHECK: entry:
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test0(i8* %call.i) {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %call.i) nounwind
+  %1 = tail call i8* @objc_autoreleaseReturnValue(i8* %0) nounwind
+  ret void
+}
+
+; Same as test0, but with slightly different use arrangements.
+
+; CHECK: define void @test1(
+; CHECK: entry:
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test1(i8* %call.i) {
+entry:
+  %0 = tail call i8* @objc_retain(i8* %call.i) nounwind
+  %1 = tail call i8* @objc_autoreleaseReturnValue(i8* %call.i) nounwind
+  ret void
+}
+
+; Delete a retainRV+autoreleaseRV even if the pointer is used.
+
+; CHECK: define void @test24(
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   call void @use_pointer(i8* %p)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test24(i8* %p) {
+entry:
+  call i8* @objc_autoreleaseReturnValue(i8* %p) nounwind
+  call i8* @objc_retainAutoreleasedReturnValue(i8* %p) nounwind
+  call void @use_pointer(i8* %p)
+  ret void
+}
diff --git a/test/Transforms/ObjCARC/retain-not-declared.ll b/test/Transforms/ObjCARC/retain-not-declared.ll
new file mode 100644
index 0000000000..e1fe117063
--- /dev/null
+++ b/test/Transforms/ObjCARC/retain-not-declared.ll
@@ -0,0 +1,25 @@
+; RUN: opt -S -objc-arc -objc-arc-contract < %s | FileCheck %s
+
+; Test that the optimizer can create an objc_retainAutoreleaseReturnValue
+; declaration even if no objc_retain declaration exists.
+; rdar://9401303
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+declare i8* @objc_unretainedObject(i8*)
+declare i8* @objc_retainAutoreleasedReturnValue(i8*)
+declare i8* @objc_autoreleaseReturnValue(i8*)
+
+; CHECK:      define i8* @foo(i8* %p) {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %0 = tail call i8* @objc_retainAutoreleaseReturnValue(i8* %p) nounwind
+; CHECK-NEXT:   ret i8* %0
+; CHECK-NEXT: }
+
+define i8* @foo(i8* %p) {
+entry:
+  %call = tail call i8* @objc_unretainedObject(i8* %p)
+  %0 = tail call i8* @objc_retainAutoreleasedReturnValue(i8* %call) nounwind
+  %1 = tail call i8* @objc_autoreleaseReturnValue(i8* %0) nounwind
+  ret i8* %1
+}
+
diff --git a/test/Transforms/ObjCARC/rle-s2l.ll b/test/Transforms/ObjCARC/rle-s2l.ll
new file mode 100644
index 0000000000..8f8d5c0d38
--- /dev/null
+++ b/test/Transforms/ObjCARC/rle-s2l.ll
@@ -0,0 +1,135 @@
+; RUN: opt -S -basicaa -objc-arc < %s | FileCheck %s
+
+declare i8* @objc_loadWeak(i8**)
+declare i8* @objc_loadWeakRetained(i8**)
+declare i8* @objc_storeWeak(i8**, i8*)
+declare i8* @objc_initWeak(i8**, i8*)
+declare void @use_pointer(i8*)
+declare void @callee()
+
+; Basic redundant @objc_loadWeak elimination.
+
+; CHECK:      define void @test0(i8** %p) {
+; CHECK-NEXT:   %y = call i8* @objc_loadWeak(i8** %p)
+; CHECK-NEXT:   call void @use_pointer(i8* %y)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test0(i8** %p) {
+  %x = call i8* @objc_loadWeak(i8** %p)
+  %y = call i8* @objc_loadWeak(i8** %p)
+  call void @use_pointer(i8* %y)
+  ret void
+}
+
+; DCE the @objc_loadWeak.
+
+; CHECK:      define void @test1(i8** %p) {
+; CHECK-NEXT:   %y = call i8* @objc_loadWeakRetained(i8** %p)
+; CHECK-NEXT:   call void @use_pointer(i8* %y)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test1(i8** %p) {
+  %x = call i8* @objc_loadWeak(i8** %p)
+  %y = call i8* @objc_loadWeakRetained(i8** %p)
+  call void @use_pointer(i8* %y)
+  ret void
+}
+
+; Basic redundant @objc_loadWeakRetained elimination.
+
+; CHECK:      define void @test2(i8** %p) {
+; CHECK-NEXT:   %x = call i8* @objc_loadWeak(i8** %p)
+; CHECK-NEXT:   store i8 3, i8* %x
+; CHECK-NEXT:   %1 = tail call i8* @objc_retain(i8* %x)
+; CHECK-NEXT:   call void @use_pointer(i8* %x)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test2(i8** %p) {
+  %x = call i8* @objc_loadWeak(i8** %p)
+  store i8 3, i8* %x
+  %y = call i8* @objc_loadWeakRetained(i8** %p)
+  call void @use_pointer(i8* %y)
+  ret void
+}
+
+; Basic redundant @objc_loadWeakRetained elimination, this time
+; with a readonly call instead of a store.
+
+; CHECK:      define void @test3(i8** %p) {
+; CHECK-NEXT:   %x = call i8* @objc_loadWeak(i8** %p)
+; CHECK-NEXT:   call void @use_pointer(i8* %x) readonly
+; CHECK-NEXT:   %1 = tail call i8* @objc_retain(i8* %x)
+; CHECK-NEXT:   call void @use_pointer(i8* %x)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test3(i8** %p) {
+  %x = call i8* @objc_loadWeak(i8** %p)
+  call void @use_pointer(i8* %x) readonly
+  %y = call i8* @objc_loadWeakRetained(i8** %p)
+  call void @use_pointer(i8* %y)
+  ret void
+}
+
+; A regular call blocks redundant weak load elimination.
+
+; CHECK:      define void @test4(i8** %p) {
+; CHECK-NEXT:   %x = call i8* @objc_loadWeak(i8** %p)
+; CHECK-NEXT:   call void @use_pointer(i8* %x) readonly
+; CHECK-NEXT:   call void @callee()
+; CHECK-NEXT:   %y = call i8* @objc_loadWeak(i8** %p)
+; CHECK-NEXT:   call void @use_pointer(i8* %y)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test4(i8** %p) {
+  %x = call i8* @objc_loadWeak(i8** %p)
+  call void @use_pointer(i8* %x) readonly
+  call void @callee()
+  %y = call i8* @objc_loadWeak(i8** %p)
+  call void @use_pointer(i8* %y)
+  ret void
+}
+
+; Store to load forwarding.
+
+; CHECK:      define void @test5(i8** %p, i8* %n) {
+; CHECK-NEXT:   %1 = call i8* @objc_storeWeak(i8** %p, i8* %n)
+; CHECK-NEXT:   call void @use_pointer(i8* %n)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test5(i8** %p, i8* %n) {
+  call i8* @objc_storeWeak(i8** %p, i8* %n)
+  %y = call i8* @objc_loadWeak(i8** %p)
+  call void @use_pointer(i8* %y)
+  ret void
+}
+
+; Store to load forwarding with objc_initWeak.
+
+; CHECK:      define void @test6(i8** %p, i8* %n) {
+; CHECK-NEXT:   %1 = call i8* @objc_initWeak(i8** %p, i8* %n)
+; CHECK-NEXT:   call void @use_pointer(i8* %n)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test6(i8** %p, i8* %n) {
+  call i8* @objc_initWeak(i8** %p, i8* %n)
+  %y = call i8* @objc_loadWeak(i8** %p)
+  call void @use_pointer(i8* %y)
+  ret void
+}
+
+; Don't forward if there's a may-alias store in the way.
+
+; CHECK:      define void @test7(i8** %p, i8* %n, i8** %q, i8* %m) {
+; CHECK-NEXT:   call i8* @objc_initWeak(i8** %p, i8* %n)
+; CHECK-NEXT:   call i8* @objc_storeWeak(i8** %q, i8* %m)
+; CHECK-NEXT:   %y = call i8* @objc_loadWeak(i8** %p)
+; CHECK-NEXT:   call void @use_pointer(i8* %y)
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @test7(i8** %p, i8* %n, i8** %q, i8* %m) {
+  call i8* @objc_initWeak(i8** %p, i8* %n)
+  call i8* @objc_storeWeak(i8** %q, i8* %m)
+  %y = call i8* @objc_loadWeak(i8** %p)
+  call void @use_pointer(i8* %y)
+  ret void
+}
diff --git a/test/Transforms/ObjCARC/rv.ll b/test/Transforms/ObjCARC/rv.ll
new file mode 100644
index 0000000000..da53a86b10
--- /dev/null
+++ b/test/Transforms/ObjCARC/rv.ll
@@ -0,0 +1,331 @@
+; RUN: opt -objc-arc -S < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64"
+
+declare i8* @objc_retain(i8*)
+declare i8* @objc_retainAutoreleasedReturnValue(i8*)
+declare void @objc_release(i8*)
+declare i8* @objc_autorelease(i8*)
+declare i8* @objc_autoreleaseReturnValue(i8*)
+declare i8* @objc_retainAutoreleaseReturnValue(i8*)
+declare void @objc_autoreleasePoolPop(i8*)
+declare void @objc_autoreleasePoolPush()
+declare i8* @objc_retainBlock(i8*)
+
+declare i8* @objc_retainedObject(i8*)
+declare i8* @objc_unretainedObject(i8*)
+declare i8* @objc_unretainedPointer(i8*)
+
+declare void @use_pointer(i8*)
+declare void @callee()
+declare void @callee_fnptr(void ()*)
+declare void @invokee()
+declare i8* @returner()
+
+; Test that retain+release elimination is suppressed when the
+; retain is an objc_retainAutoreleasedReturnValue, since it's
+; better to do the RV optimization.
+
+; CHECK:      define void @test0(
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %x = call i8* @returner
+; CHECK-NEXT:   %0 = tail call i8* @objc_retainAutoreleasedReturnValue(i8* %x) nounwind
+; CHECK: t:
+; CHECK-NOT: @objc_
+; CHECK: return:
+; CHECK-NEXT: call void @objc_release(i8* %x)
+; CHECK-NEXT: ret void
+; CHECK-NEXT: }
+define void @test0(i1 %p) nounwind {
+entry:
+  %x = call i8* @returner()
+  %0 = call i8* @objc_retainAutoreleasedReturnValue(i8* %x)
+  br i1 %p, label %t, label %return
+
+t:
+  call void @use_pointer(i8* %x)
+  store i8 0, i8* %x
+  br label %return
+
+return:
+  call void @objc_release(i8* %x) nounwind
+  ret void
+}
+
+; Delete no-ops.
+
+; CHECK: define void @test2
+; CHECK-NOT: @objc_
+; CHECK: }
+define void @test2() {
+  call i8* @objc_retainAutoreleasedReturnValue(i8* null)
+  call i8* @objc_autoreleaseReturnValue(i8* null)
+  ; call i8* @objc_retainAutoreleaseReturnValue(i8* null) ; TODO
+  ret void
+}
+
+; Delete a redundant retainRV,autoreleaseRV when forwaring a call result
+; directly to a return value.
+
+; CHECK: define i8* @test3
+; CHECK: call i8* @returner()
+; CHECK-NEXT: ret i8* %call
+define i8* @test3() {
+entry:
+  %call = call i8* @returner()
+  %0 = call i8* @objc_retainAutoreleasedReturnValue(i8* %call) nounwind
+  %1 = call i8* @objc_autoreleaseReturnValue(i8* %0) nounwind
+  ret i8* %1
+}
+
+; Delete a redundant retain,autoreleaseRV when forwaring a call result
+; directly to a return value.
+
+; CHECK: define i8* @test4
+; CHECK: call i8* @returner()
+; CHECK-NEXT: ret i8* %call
+define i8* @test4() {
+entry:
+  %call = call i8* @returner()
+  %0 = call i8* @objc_retain(i8* %call) nounwind
+  %1 = call i8* @objc_autoreleaseReturnValue(i8* %0) nounwind
+  ret i8* %1
+}
+
+; Delete a redundant fused retain+autoreleaseRV when forwaring a call result
+; directly to a return value.
+
+; TODO
+; HECK: define i8* @test5
+; HECK: call i8* @returner()
+; HECK-NEXT: ret i8* %call
+;define i8* @test5() {
+;entry:
+;  %call = call i8* @returner()
+;  %0 = call i8* @objc_retainAutoreleaseReturnValue(i8* %call) nounwind
+;  ret i8* %0
+;}
+
+; Don't eliminate objc_retainAutoreleasedReturnValue by merging it into
+; an objc_autorelease.
+; TODO? Merge objc_retainAutoreleasedReturnValue and objc_autorelease into
+; objc_retainAutoreleasedReturnValueAutorelease and merge
+; objc_retainAutoreleasedReturnValue and objc_autoreleaseReturnValue
+; into objc_retainAutoreleasedReturnValueAutoreleaseReturnValue?
+; Those entrypoints don't exist yet though.
+
+; CHECK: define i8* @test7(
+; CHECK: call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+; CHECK: %t = tail call i8* @objc_autoreleaseReturnValue(i8* %p)
+define i8* @test7() {
+  %p = call i8* @returner()
+  call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+  %t = call i8* @objc_autoreleaseReturnValue(i8* %p)
+  call void @use_pointer(i8* %t)
+  ret i8* %t
+}
+
+; CHECK: define i8* @test7b(
+; CHECK: call i8* @objc_retain(i8* %p)
+; CHECK: %t = tail call i8* @objc_autoreleaseReturnValue(i8* %p)
+define i8* @test7b() {
+  %p = call i8* @returner()
+  call void @use_pointer(i8* %p)
+  call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+  %t = call i8* @objc_autoreleaseReturnValue(i8* %p)
+  ret i8* %t
+}
+
+; Turn objc_retain into objc_retainAutoreleasedReturnValue if its operand
+; is a return value.
+
+; CHECK: define void @test8()
+; CHECK: tail call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+define void @test8() {
+  %p = call i8* @returner()
+  call i8* @objc_retain(i8* %p)
+  ret void
+}
+
+; Don't apply the RV optimization to autorelease if there's no retain.
+
+; CHECK: define i8* @test9(i8* %p)
+; CHECK: tail call i8* @objc_autorelease(i8* %p)
+define i8* @test9(i8* %p) {
+  call i8* @objc_autorelease(i8* %p)
+  ret i8* %p
+}
+
+; Apply the RV optimization.
+
+; CHECK: define i8* @test10(i8* %p)
+; CHECK: tail call i8* @objc_retain(i8* %p) nounwind
+; CHECK: tail call i8* @objc_autoreleaseReturnValue(i8* %p) nounwind
+; CHECK-NEXT: ret i8* %p
+define i8* @test10(i8* %p) {
+  %1 = call i8* @objc_retain(i8* %p)
+  %2 = call i8* @objc_autorelease(i8* %p)
+  ret i8* %p
+}
+
+; Don't do the autoreleaseRV optimization because @use_pointer
+; could undo the retain.
+
+; CHECK: define i8* @test11(i8* %p)
+; CHECK: tail call i8* @objc_retain(i8* %p)
+; CHECK-NEXT: call void @use_pointer(i8* %p)
+; CHECK: tail call i8* @objc_autorelease(i8* %p)
+; CHECK-NEXT: ret i8* %p
+define i8* @test11(i8* %p) {
+  %1 = call i8* @objc_retain(i8* %p)
+  call void @use_pointer(i8* %p)
+  %2 = call i8* @objc_autorelease(i8* %p)
+  ret i8* %p
+}
+
+; Don't spoil the RV optimization.
+
+; CHECK: define i8* @test12(i8* %p)
+; CHECK: tail call i8* @objc_retain(i8* %p)
+; CHECK: call void @use_pointer(i8* %p)
+; CHECK: tail call i8* @objc_autoreleaseReturnValue(i8* %p)
+; CHECK: ret i8* %p
+define i8* @test12(i8* %p) {
+  %1 = call i8* @objc_retain(i8* %p)
+  call void @use_pointer(i8* %p)
+  %2 = call i8* @objc_autoreleaseReturnValue(i8* %p)
+  ret i8* %p
+}
+
+; Don't zap the objc_retainAutoreleasedReturnValue.
+
+; CHECK: define i8* @test13(
+; CHECK: tail call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+; CHECK: tail call i8* @objc_autorelease(i8* %p)
+; CHECK: ret i8* %p
+define i8* @test13() {
+  %p = call i8* @returner()
+  %1 = call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+  call void @callee()
+  %2 = call i8* @objc_autorelease(i8* %p)
+  ret i8* %p
+}
+
+; Convert objc_retainAutoreleasedReturnValue to objc_retain if its
+; argument is not a return value.
+
+; CHECK: define void @test14(
+; CHECK-NEXT: tail call i8* @objc_retain(i8* %p) nounwind
+; CHECK-NEXT: ret void
+define void @test14(i8* %p) {
+  call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+  ret void
+}
+
+; Don't convert objc_retainAutoreleasedReturnValue to objc_retain if its
+; argument is a return value.
+
+; CHECK: define void @test15(
+; CHECK-NEXT: %y = call i8* @returner()
+; CHECK-NEXT: tail call i8* @objc_retainAutoreleasedReturnValue(i8* %y) nounwind
+; CHECK-NEXT: ret void
+define void @test15() {
+  %y = call i8* @returner()
+  call i8* @objc_retainAutoreleasedReturnValue(i8* %y)
+  ret void
+}
+
+; Convert objc_retain to objc_retainAutoreleasedReturnValue if its
+; argument is a return value.
+
+; CHECK: define void @test16(
+; CHECK-NEXT: %y = call i8* @returner()
+; CHECK-NEXT: tail call i8* @objc_retainAutoreleasedReturnValue(i8* %y) nounwind
+; CHECK-NEXT: ret void
+define void @test16() {
+  %y = call i8* @returner()
+  call i8* @objc_retain(i8* %y)
+  ret void
+}
+
+; Don't convert objc_retain to objc_retainAutoreleasedReturnValue if its
+; argument is not a return value.
+
+; CHECK: define void @test17(
+; CHECK-NEXT: tail call i8* @objc_retain(i8* %y) nounwind
+; CHECK-NEXT: ret void
+define void @test17(i8* %y) {
+  call i8* @objc_retain(i8* %y)
+  ret void
+}
+
+; Don't Convert objc_retain to objc_retainAutoreleasedReturnValue if it
+; isn't next to the call providing its return value.
+
+; CHECK: define void @test18(
+; CHECK-NEXT: %y = call i8* @returner()
+; CHECK-NEXT: call void @callee()
+; CHECK-NEXT: tail call i8* @objc_retain(i8* %y) nounwind
+; CHECK-NEXT: ret void
+define void @test18() {
+  %y = call i8* @returner()
+  call void @callee()
+  call i8* @objc_retain(i8* %y)
+  ret void
+}
+
+; Delete autoreleaseRV+retainRV pairs.
+
+; CHECK: define i8* @test19(i8* %p) {
+; CHECK-NEXT: ret i8* %p
+define i8* @test19(i8* %p) {
+  call i8* @objc_autoreleaseReturnValue(i8* %p)
+  call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+  ret i8* %p
+}
+
+; Like test19 but with plain autorelease.
+
+; CHECK: define i8* @test20(i8* %p) {
+; CHECK-NEXT: call i8* @objc_autorelease(i8* %p)
+; CHECK-NEXT: call i8* @objc_retain(i8* %p)
+; CHECK-NEXT: ret i8* %p
+define i8* @test20(i8* %p) {
+  call i8* @objc_autorelease(i8* %p)
+  call i8* @objc_retainAutoreleasedReturnValue(i8* %p)
+  ret i8* %p
+}
+
+; Like test19 but with plain retain.
+
+; CHECK: define i8* @test21(i8* %p) {
+; CHECK-NEXT: call i8* @objc_autoreleaseReturnValue(i8* %p)
+; CHECK-NEXT: call i8* @objc_retain(i8* %p)
+; CHECK-NEXT: ret i8* %p
+define i8* @test21(i8* %p) {
+  call i8* @objc_autoreleaseReturnValue(i8* %p)
+  call i8* @objc_retain(i8* %p)
+  ret i8* %p
+}
+
+; Like test19 but with plain retain and autorelease.
+
+; CHECK: define i8* @test22(i8* %p) {
+; CHECK-NEXT: call i8* @objc_autorelease(i8* %p)
+; CHECK-NEXT: call i8* @objc_retain(i8* %p)
+; CHECK-NEXT: ret i8* %p
+define i8* @test22(i8* %p) {
+  call i8* @objc_autorelease(i8* %p)
+  call i8* @objc_retain(i8* %p)
+  ret i8* %p
+}
+
+; Convert autoreleaseRV to autorelease.
+
+; CHECK: define void @test23(
+; CHECK: tail call i8* @objc_autorelease(i8* %p) nounwind
+define void @test23(i8* %p) {
+  store i8 0, i8* %p
+  call i8* @objc_autoreleaseReturnValue(i8* %p)
+  ret void
+}
diff --git a/test/Transforms/ObjCARC/weak-contract.ll b/test/Transforms/ObjCARC/weak-contract.ll
new file mode 100644
index 0000000000..ca69c7087d
--- /dev/null
+++ b/test/Transforms/ObjCARC/weak-contract.ll
@@ -0,0 +1,14 @@
+; RUN: opt -objc-arc-contract -S < %s | FileCheck %s
+
+declare i8* @objc_initWeak(i8**, i8*)
+
+; Convert objc_initWeak(p, null) to *p = null.
+
+; CHECK:      define i8* @test0(i8** %p) {
+; CHECK-NEXT:   store i8* null, i8** %p
+; CHECK-NEXT:   ret i8* null
+; CHECK-NEXT: }
+define i8* @test0(i8** %p) {
+  %t = call i8* @objc_initWeak(i8** %p, i8* null)
+  ret i8* %t
+}
diff --git a/test/Transforms/ObjCARC/weak-copies.ll b/test/Transforms/ObjCARC/weak-copies.ll
new file mode 100644
index 0000000000..e1a94bb474
--- /dev/null
+++ b/test/Transforms/ObjCARC/weak-copies.ll
@@ -0,0 +1,87 @@
+; RUN: opt -S -basicaa -objc-arc < %s | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+target triple = "x86_64-apple-darwin11.0.0"
+
+%0 = type { i64, i64, i8*, i8*, i8*, i8* }
+%1 = type <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i8* }>
+%struct.__block_descriptor = type { i64, i64 }
+
+@_NSConcreteStackBlock = external global i8*
+@.str = private unnamed_addr constant [6 x i8] c"v8@?0\00"
+@"\01L_OBJC_CLASS_NAME_" = internal global [3 x i8] c"\01@\00", section "__TEXT,__objc_classname,cstring_literals", align 1
+@__block_descriptor_tmp = internal constant %0 { i64 0, i64 40, i8* bitcast (void (i8*, i8*)* @__copy_helper_block_ to i8*), i8* bitcast (void (i8*)* @__destroy_helper_block_ to i8*), i8* getelementptr inbounds ([6 x i8]* @.str, i32 0, i32 0), i8* getelementptr inbounds ([3 x i8]* @"\01L_OBJC_CLASS_NAME_", i32 0, i32 0) }
+@"\01L_OBJC_IMAGE_INFO" = internal constant [2 x i32] [i32 0, i32 16], section "__DATA, __objc_imageinfo, regular, no_dead_strip"
+@llvm.used = appending global [2 x i8*] [i8* getelementptr inbounds ([3 x i8]* @"\01L_OBJC_CLASS_NAME_", i32 0, i32 0), i8* bitcast ([2 x i32]* @"\01L_OBJC_IMAGE_INFO" to i8*)], section "llvm.metadata"
+
+; Eliminate unnecessary weak pointer copies.
+
+; CHECK:      define void @foo() {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %call = call i8* @bar()
+; CHECK-NEXT:   call void @use(i8* %call) nounwind
+; CHECK-NEXT:   ret void
+; CHECK-NEXT: }
+define void @foo() {
+entry:
+  %w = alloca i8*, align 8
+  %x = alloca i8*, align 8
+  %call = call i8* @bar()
+  %0 = call i8* @objc_initWeak(i8** %w, i8* %call) nounwind
+  %1 = call i8* @objc_loadWeak(i8** %w) nounwind
+  %2 = call i8* @objc_initWeak(i8** %x, i8* %1) nounwind
+  %3 = call i8* @objc_loadWeak(i8** %x) nounwind
+  call void @use(i8* %3) nounwind
+  call void @objc_destroyWeak(i8** %x) nounwind
+  call void @objc_destroyWeak(i8** %w) nounwind
+  ret void
+}
+
+; Eliminate unnecessary weak pointer copies in a block initialization.
+
+; CHECK:      define void @qux(i8* %me) nounwind {
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   %block = alloca %1, align 8
+; CHECK-NOT:    alloca
+; CHECK:      }
+define void @qux(i8* %me) nounwind {
+entry:
+  %w = alloca i8*, align 8
+  %block = alloca %1, align 8
+  %0 = call i8* @objc_retain(i8* %me) nounwind
+  %1 = call i8* @objc_initWeak(i8** %w, i8* %0) nounwind
+  %block.isa = getelementptr inbounds %1* %block, i64 0, i32 0
+  store i8* bitcast (i8** @_NSConcreteStackBlock to i8*), i8** %block.isa, align 8
+  %block.flags = getelementptr inbounds %1* %block, i64 0, i32 1
+  store i32 1107296256, i32* %block.flags, align 8
+  %block.reserved = getelementptr inbounds %1* %block, i64 0, i32 2
+  store i32 0, i32* %block.reserved, align 4
+  %block.invoke = getelementptr inbounds %1* %block, i64 0, i32 3
+  store i8* bitcast (void (i8*)* @__qux_block_invoke_0 to i8*), i8** %block.invoke, align 8
+  %block.descriptor = getelementptr inbounds %1* %block, i64 0, i32 4
+  store %struct.__block_descriptor* bitcast (%0* @__block_descriptor_tmp to %struct.__block_descriptor*), %struct.__block_descriptor** %block.descriptor, align 8
+  %block.captured = getelementptr inbounds %1* %block, i64 0, i32 5
+  %2 = call i8* @objc_loadWeak(i8** %w) nounwind
+  %3 = call i8* @objc_initWeak(i8** %block.captured, i8* %2) nounwind
+  %4 = bitcast %1* %block to void ()*
+  call void @use_block(void ()* %4) nounwind
+  call void @objc_destroyWeak(i8** %block.captured) nounwind
+  call void @objc_destroyWeak(i8** %w) nounwind
+  call void @objc_release(i8* %0) nounwind, !clang.imprecise_release !0
+  ret void
+}
+
+declare i8* @objc_retain(i8*)
+declare void @use_block(void ()*) nounwind
+declare void @__qux_block_invoke_0(i8* %.block_descriptor) nounwind
+declare void @__copy_helper_block_(i8*, i8*) nounwind
+declare void @objc_copyWeak(i8**, i8**)
+declare void @__destroy_helper_block_(i8*) nounwind
+declare void @objc_release(i8*)
+declare i8* @bar()
+declare i8* @objc_initWeak(i8**, i8*)
+declare i8* @objc_loadWeak(i8**)
+declare void @use(i8*) nounwind
+declare void @objc_destroyWeak(i8**)
+
+!0 = metadata !{}
diff --git a/test/Transforms/ObjCARC/weak.ll b/test/Transforms/ObjCARC/weak.ll
new file mode 100644
index 0000000000..85a290c098
--- /dev/null
+++ b/test/Transforms/ObjCARC/weak.ll
@@ -0,0 +1,57 @@
+; RUN: opt -objc-arc -S < %s | FileCheck %s
+
+declare i8* @objc_initWeak(i8**, i8*)
+declare i8* @objc_storeWeak(i8**, i8*)
+declare i8* @objc_loadWeak(i8**)
+declare void @objc_destroyWeak(i8**)
+declare i8* @objc_loadWeakRetained(i8**)
+declare void @objc_moveWeak(i8**, i8**)
+declare void @objc_copyWeak(i8**, i8**)
+
+; If the pointer-to-weak-pointer is null, it's undefined behavior.
+
+; CHECK: define void @test0(
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: store i8* undef, i8** null
+; CHECK: ret void
+define void @test0(i8* %p, i8** %q) {
+entry:
+  call i8* @objc_storeWeak(i8** null, i8* %p)
+  call i8* @objc_storeWeak(i8** undef, i8* %p)
+  call i8* @objc_loadWeakRetained(i8** null)
+  call i8* @objc_loadWeakRetained(i8** undef)
+  call i8* @objc_loadWeak(i8** null)
+  call i8* @objc_loadWeak(i8** undef)
+  call i8* @objc_initWeak(i8** null, i8* %p)
+  call i8* @objc_initWeak(i8** undef, i8* %p)
+  call void @objc_destroyWeak(i8** null)
+  call void @objc_destroyWeak(i8** undef)
+
+  call void @objc_copyWeak(i8** null, i8** %q)
+  call void @objc_copyWeak(i8** undef, i8** %q)
+  call void @objc_copyWeak(i8** %q, i8** null)
+  call void @objc_copyWeak(i8** %q, i8** undef)
+
+  call void @objc_moveWeak(i8** null, i8** %q)
+  call void @objc_moveWeak(i8** undef, i8** %q)
+  call void @objc_moveWeak(i8** %q, i8** null)
+  call void @objc_moveWeak(i8** %q, i8** undef)
+
+  ret void
+}