vendor/llvm-project/llvmorg-11-init-20887-g2e10b7a39b9vendor/llvm-project/master

55 files changed, 4967 insertions, 1502 deletions

diff --git a/llvm/lib/ExecutionEngine/ExecutionEngine.cpp b/llvm/lib/ExecutionEngine/ExecutionEngine.cpp
index ee7a7cb60bc9..d8bd671c6661 100644
--- a/llvm/lib/ExecutionEngine/ExecutionEngine.cpp
+++ b/llvm/lib/ExecutionEngine/ExecutionEngine.cpp
@@ -9,6 +9,8 @@
 // This file defines the common interface used by the various execution engine
 // subclasses.
 //
+// FIXME: This file needs to be updated to support scalable vectors
+//
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ExecutionEngine/ExecutionEngine.h"
@@ -108,7 +110,7 @@ public:
     Type *ElTy = GV->getValueType();
     size_t GVSize = (size_t)TD.getTypeAllocSize(ElTy);
     void *RawMemory = ::operator new(
-        alignTo(sizeof(GVMemoryBlock), TD.getPreferredAlignment(GV)) + GVSize);
+        alignTo(sizeof(GVMemoryBlock), TD.getPreferredAlign(GV)) + GVSize);
     new(RawMemory) GVMemoryBlock(GV);
     return static_cast<char*>(RawMemory) + sizeof(GVMemoryBlock);
   }
@@ -200,7 +202,7 @@ std::string ExecutionEngine::getMangledName(const GlobalValue *GV) {
       : GV->getParent()->getDataLayout();
 
   Mangler::getNameWithPrefix(FullName, GV->getName(), DL);
-  return FullName.str();
+  return std::string(FullName.str());
 }
 
 void ExecutionEngine::addGlobalMapping(const GlobalValue *GV, void *Addr) {
@@ -223,7 +225,7 @@ void ExecutionEngine::addGlobalMapping(StringRef Name, uint64_t Addr) {
     std::string &V = EEState.getGlobalAddressReverseMap()[CurVal];
     assert((!V.empty() || !Name.empty()) &&
            "GlobalMapping already established!");
-    V = Name;
+    V = std::string(Name);
   }
 }
 
@@ -269,7 +271,7 @@ uint64_t ExecutionEngine::updateGlobalMapping(StringRef Name, uint64_t Addr) {
     std::string &V = EEState.getGlobalAddressReverseMap()[CurVal];
     assert((!V.empty() || !Name.empty()) &&
            "GlobalMapping already established!");
-    V = Name;
+    V = std::string(Name);
   }
   return OldVal;
 }
@@ -307,8 +309,8 @@ const GlobalValue *ExecutionEngine::getGlobalValueAtAddress(void *Addr) {
            E = EEState.getGlobalAddressMap().end(); I != E; ++I) {
       StringRef Name = I->first();
       uint64_t Addr = I->second;
-      EEState.getGlobalAddressReverseMap().insert(std::make_pair(
-                                                          Addr, Name));
+      EEState.getGlobalAddressReverseMap().insert(
+          std::make_pair(Addr, std::string(Name)));
     }
   }
 
@@ -582,7 +584,7 @@ void *ExecutionEngine::getPointerToGlobal(const GlobalValue *GV) {
   // Global variable might have been added since interpreter started.
   if (GlobalVariable *GVar =
           const_cast<GlobalVariable *>(dyn_cast<GlobalVariable>(GV)))
-    EmitGlobalVariable(GVar);
+    emitGlobalVariable(GVar);
   else
     llvm_unreachable("Global hasn't had an address allocated yet!");
 
@@ -624,17 +626,20 @@ GenericValue ExecutionEngine::getConstantValue(const Constant *C) {
         }
       }
       break;
-    case Type::VectorTyID:
-      // if the whole vector is 'undef' just reserve memory for the value.
-      auto* VTy = cast<VectorType>(C->getType());
-      Type *ElemTy = VTy->getElementType();
-      unsigned int elemNum = VTy->getNumElements();
-      Result.AggregateVal.resize(elemNum);
-      if (ElemTy->isIntegerTy())
-        for (unsigned int i = 0; i < elemNum; ++i)
-          Result.AggregateVal[i].IntVal =
-            APInt(ElemTy->getPrimitiveSizeInBits(), 0);
-      break;
+      case Type::ScalableVectorTyID:
+        report_fatal_error(
+            "Scalable vector support not yet implemented in ExecutionEngine");
+      case Type::FixedVectorTyID:
+        // if the whole vector is 'undef' just reserve memory for the value.
+        auto *VTy = cast<FixedVectorType>(C->getType());
+        Type *ElemTy = VTy->getElementType();
+        unsigned int elemNum = VTy->getNumElements();
+        Result.AggregateVal.resize(elemNum);
+        if (ElemTy->isIntegerTy())
+          for (unsigned int i = 0; i < elemNum; ++i)
+            Result.AggregateVal[i].IntVal =
+                APInt(ElemTy->getPrimitiveSizeInBits(), 0);
+        break;
     }
     return Result;
   }
@@ -914,7 +919,10 @@ GenericValue ExecutionEngine::getConstantValue(const Constant *C) {
     else
       llvm_unreachable("Unknown constant pointer type!");
     break;
-  case Type::VectorTyID: {
+  case Type::ScalableVectorTyID:
+    report_fatal_error(
+        "Scalable vector support not yet implemented in ExecutionEngine");
+  case Type::FixedVectorTyID: {
     unsigned elemNum;
     Type* ElemTy;
     const ConstantDataVector *CDV = dyn_cast<ConstantDataVector>(C);
@@ -925,9 +933,9 @@ GenericValue ExecutionEngine::getConstantValue(const Constant *C) {
         elemNum = CDV->getNumElements();
         ElemTy = CDV->getElementType();
     } else if (CV || CAZ) {
-        auto* VTy = cast<VectorType>(C->getType());
-        elemNum = VTy->getNumElements();
-        ElemTy = VTy->getElementType();
+      auto *VTy = cast<FixedVectorType>(C->getType());
+      elemNum = VTy->getNumElements();
+      ElemTy = VTy->getElementType();
     } else {
         llvm_unreachable("Unknown constant vector type!");
     }
@@ -1006,8 +1014,7 @@ GenericValue ExecutionEngine::getConstantValue(const Constant *C) {
       break;
     }
     llvm_unreachable("Unknown constant pointer type!");
-  }
-  break;
+  } break;
 
   default:
     SmallString<256> Msg;
@@ -1046,7 +1053,8 @@ void ExecutionEngine::StoreValueToMemory(const GenericValue &Val,
 
     *((PointerTy*)Ptr) = Val.PointerVal;
     break;
-  case Type::VectorTyID:
+  case Type::FixedVectorTyID:
+  case Type::ScalableVectorTyID:
     for (unsigned i = 0; i < Val.AggregateVal.size(); ++i) {
       if (cast<VectorType>(Ty)->getElementType()->isDoubleTy())
         *(((double*)Ptr)+i) = Val.AggregateVal[i].DoubleVal;
@@ -1096,8 +1104,11 @@ void ExecutionEngine::LoadValueFromMemory(GenericValue &Result,
     Result.IntVal = APInt(80, y);
     break;
   }
-  case Type::VectorTyID: {
-    auto *VT = cast<VectorType>(Ty);
+  case Type::ScalableVectorTyID:
+    report_fatal_error(
+        "Scalable vector support not yet implemented in ExecutionEngine");
+  case Type::FixedVectorTyID: {
+    auto *VT = cast<FixedVectorType>(Ty);
     Type *ElemT = VT->getElementType();
     const unsigned numElems = VT->getNumElements();
     if (ElemT->isFloatTy()) {
@@ -1200,8 +1211,8 @@ void ExecutionEngine::emitGlobals() {
             GV.hasAppendingLinkage() || !GV.hasName())
           continue;// Ignore external globals and globals with internal linkage.
 
-        const GlobalValue *&GVEntry =
-          LinkedGlobalsMap[std::make_pair(GV.getName(), GV.getType())];
+        const GlobalValue *&GVEntry = LinkedGlobalsMap[std::make_pair(
+            std::string(GV.getName()), GV.getType())];
 
         // If this is the first time we've seen this global, it is the canonical
         // version.
@@ -1228,8 +1239,8 @@ void ExecutionEngine::emitGlobals() {
     for (const auto &GV : M.globals()) {
       // In the multi-module case, see what this global maps to.
       if (!LinkedGlobalsMap.empty()) {
-        if (const GlobalValue *GVEntry =
-              LinkedGlobalsMap[std::make_pair(GV.getName(), GV.getType())]) {
+        if (const GlobalValue *GVEntry = LinkedGlobalsMap[std::make_pair(
+                std::string(GV.getName()), GV.getType())]) {
           // If something else is the canonical global, ignore this one.
           if (GVEntry != &GV) {
             NonCanonicalGlobals.push_back(&GV);
@@ -1243,8 +1254,8 @@ void ExecutionEngine::emitGlobals() {
       } else {
         // External variable reference. Try to use the dynamic loader to
         // get a pointer to it.
-        if (void *SymAddr =
-            sys::DynamicLibrary::SearchForAddressOfSymbol(GV.getName()))
+        if (void *SymAddr = sys::DynamicLibrary::SearchForAddressOfSymbol(
+                std::string(GV.getName())))
           addGlobalMapping(&GV, SymAddr);
         else {
           report_fatal_error("Could not resolve external global address: "
@@ -1258,8 +1269,8 @@ void ExecutionEngine::emitGlobals() {
     if (!NonCanonicalGlobals.empty()) {
       for (unsigned i = 0, e = NonCanonicalGlobals.size(); i != e; ++i) {
         const GlobalValue *GV = NonCanonicalGlobals[i];
-        const GlobalValue *CGV =
-          LinkedGlobalsMap[std::make_pair(GV->getName(), GV->getType())];
+        const GlobalValue *CGV = LinkedGlobalsMap[std::make_pair(
+            std::string(GV->getName()), GV->getType())];
         void *Ptr = getPointerToGlobalIfAvailable(CGV);
         assert(Ptr && "Canonical global wasn't codegen'd!");
         addGlobalMapping(GV, Ptr);
@@ -1271,12 +1282,12 @@ void ExecutionEngine::emitGlobals() {
     for (const auto &GV : M.globals()) {
       if (!GV.isDeclaration()) {
         if (!LinkedGlobalsMap.empty()) {
-          if (const GlobalValue *GVEntry =
-                LinkedGlobalsMap[std::make_pair(GV.getName(), GV.getType())])
+          if (const GlobalValue *GVEntry = LinkedGlobalsMap[std::make_pair(
+                  std::string(GV.getName()), GV.getType())])
             if (GVEntry != &GV)  // Not the canonical variable.
               continue;
         }
-        EmitGlobalVariable(&GV);
+        emitGlobalVariable(&GV);
       }
     }
   }
@@ -1285,7 +1296,7 @@ void ExecutionEngine::emitGlobals() {
 // EmitGlobalVariable - This method emits the specified global variable to the
 // address specified in GlobalAddresses, or allocates new memory if it's not
 // already in the map.
-void ExecutionEngine::EmitGlobalVariable(const GlobalVariable *GV) {
+void ExecutionEngine::emitGlobalVariable(const GlobalVariable *GV) {
   void *GA = getPointerToGlobalIfAvailable(GV);
 
   if (!GA) {
diff --git a/llvm/lib/ExecutionEngine/ExecutionEngineBindings.cpp b/llvm/lib/ExecutionEngine/ExecutionEngineBindings.cpp
index ff1e8050c7e7..addec6871fa1 100644
--- a/llvm/lib/ExecutionEngine/ExecutionEngineBindings.cpp
+++ b/llvm/lib/ExecutionEngine/ExecutionEngineBindings.cpp
@@ -308,6 +308,18 @@ uint64_t LLVMGetFunctionAddress(LLVMExecutionEngineRef EE, const char *Name) {
   return unwrap(EE)->getFunctionAddress(Name);
 }
 
+LLVMBool LLVMExecutionEngineGetErrMsg(LLVMExecutionEngineRef EE,
+                                      char **OutError) {
+  assert(OutError && "OutError must be non-null");
+  auto *ExecEngine = unwrap(EE);
+  if (ExecEngine->hasError()) {
+    *OutError = strdup(ExecEngine->getErrorMessage().c_str());
+    ExecEngine->clearErrorMessage();
+    return true;
+  }
+  return false;
+}
+
 /*===-- Operations on memory managers -------------------------------------===*/
 
 namespace {
diff --git a/llvm/lib/ExecutionEngine/Interpreter/Execution.cpp b/llvm/lib/ExecutionEngine/Interpreter/Execution.cpp
index 51f31d3d5d8f..62e1ea6e0f0a 100644
--- a/llvm/lib/ExecutionEngine/Interpreter/Execution.cpp
+++ b/llvm/lib/ExecutionEngine/Interpreter/Execution.cpp
@@ -169,13 +169,14 @@ static void executeFRemInst(GenericValue &Dest, GenericValue Src1,
       Dest.IntVal = APInt(1,Src1.IntVal.OP(Src2.IntVal)); \
       break;
 
-#define IMPLEMENT_VECTOR_INTEGER_ICMP(OP, TY)                        \
-  case Type::VectorTyID: {                                           \
-    assert(Src1.AggregateVal.size() == Src2.AggregateVal.size());    \
-    Dest.AggregateVal.resize( Src1.AggregateVal.size() );            \
-    for( uint32_t _i=0;_i<Src1.AggregateVal.size();_i++)             \
-      Dest.AggregateVal[_i].IntVal = APInt(1,                        \
-      Src1.AggregateVal[_i].IntVal.OP(Src2.AggregateVal[_i].IntVal));\
+#define IMPLEMENT_VECTOR_INTEGER_ICMP(OP, TY)                                  \
+  case Type::FixedVectorTyID:                                                  \
+  case Type::ScalableVectorTyID: {                                             \
+    assert(Src1.AggregateVal.size() == Src2.AggregateVal.size());              \
+    Dest.AggregateVal.resize(Src1.AggregateVal.size());                        \
+    for (uint32_t _i = 0; _i < Src1.AggregateVal.size(); _i++)                 \
+      Dest.AggregateVal[_i].IntVal = APInt(                                    \
+          1, Src1.AggregateVal[_i].IntVal.OP(Src2.AggregateVal[_i].IntVal));   \
   } break;
 
 // Handle pointers specially because they must be compared with only as much
@@ -367,12 +368,13 @@ void Interpreter::visitICmpInst(ICmpInst &I) {
     Src1.AggregateVal[_i].TY##Val OP Src2.AggregateVal[_i].TY##Val);\
   break;
 
-#define IMPLEMENT_VECTOR_FCMP(OP)                                   \
-  case Type::VectorTyID:                                            \
-    if (cast<VectorType>(Ty)->getElementType()->isFloatTy()) {      \
-      IMPLEMENT_VECTOR_FCMP_T(OP, Float);                           \
-    } else {                                                        \
-        IMPLEMENT_VECTOR_FCMP_T(OP, Double);                        \
+#define IMPLEMENT_VECTOR_FCMP(OP)                                              \
+  case Type::FixedVectorTyID:                                                  \
+  case Type::ScalableVectorTyID:                                               \
+    if (cast<VectorType>(Ty)->getElementType()->isFloatTy()) {                 \
+      IMPLEMENT_VECTOR_FCMP_T(OP, Float);                                      \
+    } else {                                                                   \
+      IMPLEMENT_VECTOR_FCMP_T(OP, Double);                                     \
     }
 
 static GenericValue executeFCMP_OEQ(GenericValue Src1, GenericValue Src2,
@@ -902,13 +904,13 @@ void Interpreter::popStackAndReturnValueToCaller(Type *RetTy,
     // If we have a previous stack frame, and we have a previous call,
     // fill in the return value...
     ExecutionContext &CallingSF = ECStack.back();
-    if (Instruction *I = CallingSF.Caller.getInstruction()) {
+    if (CallingSF.Caller) {
       // Save result...
-      if (!CallingSF.Caller.getType()->isVoidTy())
-        SetValue(I, Result, CallingSF);
-      if (InvokeInst *II = dyn_cast<InvokeInst> (I))
+      if (!CallingSF.Caller->getType()->isVoidTy())
+        SetValue(CallingSF.Caller, Result, CallingSF);
+      if (InvokeInst *II = dyn_cast<InvokeInst>(CallingSF.Caller))
         SwitchToNewBasicBlock (II->getNormalDest (), CallingSF);
-      CallingSF.Caller = CallSite();          // We returned from the call...
+      CallingSF.Caller = nullptr;             // We returned from the call...
     }
   }
 }
@@ -1113,64 +1115,59 @@ void Interpreter::visitStoreInst(StoreInst &I) {
 //                 Miscellaneous Instruction Implementations
 //===----------------------------------------------------------------------===//
 
-void Interpreter::visitCallSite(CallSite CS) {
+void Interpreter::visitVAStartInst(VAStartInst &I) {
   ExecutionContext &SF = ECStack.back();
+  GenericValue ArgIndex;
+  ArgIndex.UIntPairVal.first = ECStack.size() - 1;
+  ArgIndex.UIntPairVal.second = 0;
+  SetValue(&I, ArgIndex, SF);
+}
 
-  // Check to see if this is an intrinsic function call...
-  Function *F = CS.getCalledFunction();
-  if (F && F->isDeclaration())
-    switch (F->getIntrinsicID()) {
-    case Intrinsic::not_intrinsic:
-      break;
-    case Intrinsic::vastart: { // va_start
-      GenericValue ArgIndex;
-      ArgIndex.UIntPairVal.first = ECStack.size() - 1;
-      ArgIndex.UIntPairVal.second = 0;
-      SetValue(CS.getInstruction(), ArgIndex, SF);
-      return;
-    }
-    case Intrinsic::vaend:    // va_end is a noop for the interpreter
-      return;
-    case Intrinsic::vacopy:   // va_copy: dest = src
-      SetValue(CS.getInstruction(), getOperandValue(*CS.arg_begin(), SF), SF);
-      return;
-    default:
-      // If it is an unknown intrinsic function, use the intrinsic lowering
-      // class to transform it into hopefully tasty LLVM code.
-      //
-      BasicBlock::iterator me(CS.getInstruction());
-      BasicBlock *Parent = CS.getInstruction()->getParent();
-      bool atBegin(Parent->begin() == me);
-      if (!atBegin)
-        --me;
-      IL->LowerIntrinsicCall(cast<CallInst>(CS.getInstruction()));
-
-      // Restore the CurInst pointer to the first instruction newly inserted, if
-      // any.
-      if (atBegin) {
-        SF.CurInst = Parent->begin();
-      } else {
-        SF.CurInst = me;
-        ++SF.CurInst;
-      }
-      return;
-    }
+void Interpreter::visitVAEndInst(VAEndInst &I) {
+  // va_end is a noop for the interpreter
+}
+
+void Interpreter::visitVACopyInst(VACopyInst &I) {
+  ExecutionContext &SF = ECStack.back();
+  SetValue(&I, getOperandValue(*I.arg_begin(), SF), SF);
+}
+
+void Interpreter::visitIntrinsicInst(IntrinsicInst &I) {
+  ExecutionContext &SF = ECStack.back();
+
+  // If it is an unknown intrinsic function, use the intrinsic lowering
+  // class to transform it into hopefully tasty LLVM code.
+  //
+  BasicBlock::iterator Me(&I);
+  BasicBlock *Parent = I.getParent();
+  bool atBegin(Parent->begin() == Me);
+  if (!atBegin)
+    --Me;
+  IL->LowerIntrinsicCall(&I);
+
+  // Restore the CurInst pointer to the first instruction newly inserted, if
+  // any.
+  if (atBegin) {
+    SF.CurInst = Parent->begin();
+  } else {
+    SF.CurInst = Me;
+    ++SF.CurInst;
+  }
+}
 
+void Interpreter::visitCallBase(CallBase &I) {
+  ExecutionContext &SF = ECStack.back();
 
-  SF.Caller = CS;
+  SF.Caller = &I;
   std::vector<GenericValue> ArgVals;
-  const unsigned NumArgs = SF.Caller.arg_size();
+  const unsigned NumArgs = SF.Caller->arg_size();
   ArgVals.reserve(NumArgs);
-  uint16_t pNum = 1;
-  for (CallSite::arg_iterator i = SF.Caller.arg_begin(),
-         e = SF.Caller.arg_end(); i != e; ++i, ++pNum) {
-    Value *V = *i;
+  for (Value *V : SF.Caller->args())
     ArgVals.push_back(getOperandValue(V, SF));
-  }
 
   // To handle indirect calls, we must get the pointer value from the argument
   // and treat it as a function pointer.
-  GenericValue SRC = getOperandValue(SF.Caller.getCalledValue(), SF);
+  GenericValue SRC = getOperandValue(SF.Caller->getCalledOperand(), SF);
   callFunction((Function*)GVTOP(SRC), ArgVals);
 }
 
@@ -1332,7 +1329,7 @@ GenericValue Interpreter::executeFPTruncInst(Value *SrcVal, Type *DstTy,
                                              ExecutionContext &SF) {
   GenericValue Dest, Src = getOperandValue(SrcVal, SF);
 
-  if (SrcVal->getType()->getTypeID() == Type::VectorTyID) {
+  if (isa<VectorType>(SrcVal->getType())) {
     assert(SrcVal->getType()->getScalarType()->isDoubleTy() &&
            DstTy->getScalarType()->isFloatTy() &&
            "Invalid FPTrunc instruction");
@@ -1355,7 +1352,7 @@ GenericValue Interpreter::executeFPExtInst(Value *SrcVal, Type *DstTy,
                                            ExecutionContext &SF) {
   GenericValue Dest, Src = getOperandValue(SrcVal, SF);
 
-  if (SrcVal->getType()->getTypeID() == Type::VectorTyID) {
+  if (isa<VectorType>(SrcVal->getType())) {
     assert(SrcVal->getType()->getScalarType()->isFloatTy() &&
            DstTy->getScalarType()->isDoubleTy() && "Invalid FPExt instruction");
 
@@ -1378,7 +1375,7 @@ GenericValue Interpreter::executeFPToUIInst(Value *SrcVal, Type *DstTy,
   Type *SrcTy = SrcVal->getType();
   GenericValue Dest, Src = getOperandValue(SrcVal, SF);
 
-  if (SrcTy->getTypeID() == Type::VectorTyID) {
+  if (isa<VectorType>(SrcTy)) {
     Type *DstVecTy = DstTy->getScalarType();
     Type *SrcVecTy = SrcTy->getScalarType();
     uint32_t DBitWidth = cast<IntegerType>(DstVecTy)->getBitWidth();
@@ -1416,7 +1413,7 @@ GenericValue Interpreter::executeFPToSIInst(Value *SrcVal, Type *DstTy,
   Type *SrcTy = SrcVal->getType();
   GenericValue Dest, Src = getOperandValue(SrcVal, SF);
 
-  if (SrcTy->getTypeID() == Type::VectorTyID) {
+  if (isa<VectorType>(SrcTy)) {
     Type *DstVecTy = DstTy->getScalarType();
     Type *SrcVecTy = SrcTy->getScalarType();
     uint32_t DBitWidth = cast<IntegerType>(DstVecTy)->getBitWidth();
@@ -1452,7 +1449,7 @@ GenericValue Interpreter::executeUIToFPInst(Value *SrcVal, Type *DstTy,
                                             ExecutionContext &SF) {
   GenericValue Dest, Src = getOperandValue(SrcVal, SF);
 
-  if (SrcVal->getType()->getTypeID() == Type::VectorTyID) {
+  if (isa<VectorType>(SrcVal->getType())) {
     Type *DstVecTy = DstTy->getScalarType();
     unsigned size = Src.AggregateVal.size();
     // the sizes of src and dst vectors must be equal
@@ -1484,7 +1481,7 @@ GenericValue Interpreter::executeSIToFPInst(Value *SrcVal, Type *DstTy,
                                             ExecutionContext &SF) {
   GenericValue Dest, Src = getOperandValue(SrcVal, SF);
 
-  if (SrcVal->getType()->getTypeID() == Type::VectorTyID) {
+  if (isa<VectorType>(SrcVal->getType())) {
     Type *DstVecTy = DstTy->getScalarType();
     unsigned size = Src.AggregateVal.size();
     // the sizes of src and dst vectors must be equal
@@ -1545,8 +1542,7 @@ GenericValue Interpreter::executeBitCastInst(Value *SrcVal, Type *DstTy,
   Type *SrcTy = SrcVal->getType();
   GenericValue Dest, Src = getOperandValue(SrcVal, SF);
 
-  if ((SrcTy->getTypeID() == Type::VectorTyID) ||
-      (DstTy->getTypeID() == Type::VectorTyID)) {
+  if (isa<VectorType>(SrcTy) || isa<VectorType>(DstTy)) {
     // vector src bitcast to vector dst or vector src bitcast to scalar dst or
     // scalar src bitcast to vector dst
     bool isLittleEndian = getDataLayout().isLittleEndian();
@@ -1558,7 +1554,7 @@ GenericValue Interpreter::executeBitCastInst(Value *SrcVal, Type *DstTy,
     unsigned SrcNum;
     unsigned DstNum;
 
-    if (SrcTy->getTypeID() == Type::VectorTyID) {
+    if (isa<VectorType>(SrcTy)) {
       SrcElemTy = SrcTy->getScalarType();
       SrcBitSize = SrcTy->getScalarSizeInBits();
       SrcNum = Src.AggregateVal.size();
@@ -1571,7 +1567,7 @@ GenericValue Interpreter::executeBitCastInst(Value *SrcVal, Type *DstTy,
       SrcVec.AggregateVal.push_back(Src);
     }
 
-    if (DstTy->getTypeID() == Type::VectorTyID) {
+    if (isa<VectorType>(DstTy)) {
       DstElemTy = DstTy->getScalarType();
       DstBitSize = DstTy->getScalarSizeInBits();
       DstNum = (SrcNum * SrcBitSize) / DstBitSize;
@@ -1644,7 +1640,7 @@ GenericValue Interpreter::executeBitCastInst(Value *SrcVal, Type *DstTy,
     }
 
     // convert result from integer to specified type
-    if (DstTy->getTypeID() == Type::VectorTyID) {
+    if (isa<VectorType>(DstTy)) {
       if (DstElemTy->isDoubleTy()) {
         Dest.AggregateVal.resize(DstNum);
         for (unsigned i = 0; i < DstNum; i++)
@@ -1667,8 +1663,7 @@ GenericValue Interpreter::executeBitCastInst(Value *SrcVal, Type *DstTy,
         Dest.IntVal = TempDst.AggregateVal[0].IntVal;
       }
     }
-  } else { //  if ((SrcTy->getTypeID() == Type::VectorTyID) ||
-           //     (DstTy->getTypeID() == Type::VectorTyID))
+  } else { //  if (isa<VectorType>(SrcTy)) || isa<VectorType>(DstTy))
 
     // scalar src bitcast to scalar dst
     if (DstTy->isPointerTy()) {
@@ -1868,7 +1863,6 @@ void Interpreter::visitShuffleVectorInst(ShuffleVectorInst &I){
 
   GenericValue Src1 = getOperandValue(I.getOperand(0), SF);
   GenericValue Src2 = getOperandValue(I.getOperand(1), SF);
-  GenericValue Src3 = getOperandValue(I.getOperand(2), SF);
   GenericValue Dest;
 
   // There is no need to check types of src1 and src2, because the compiled
@@ -1878,7 +1872,7 @@ void Interpreter::visitShuffleVectorInst(ShuffleVectorInst &I){
   Type *TyContained = Ty->getElementType();
   unsigned src1Size = (unsigned)Src1.AggregateVal.size();
   unsigned src2Size = (unsigned)Src2.AggregateVal.size();
-  unsigned src3Size = (unsigned)Src3.AggregateVal.size();
+  unsigned src3Size = I.getShuffleMask().size();
 
   Dest.AggregateVal.resize(src3Size);
 
@@ -1888,7 +1882,7 @@ void Interpreter::visitShuffleVectorInst(ShuffleVectorInst &I){
       break;
     case Type::IntegerTyID:
       for( unsigned i=0; i<src3Size; i++) {
-        unsigned j = Src3.AggregateVal[i].IntVal.getZExtValue();
+        unsigned j = std::max(0, I.getMaskValue(i));
         if(j < src1Size)
           Dest.AggregateVal[i].IntVal = Src1.AggregateVal[j].IntVal;
         else if(j < src1Size + src2Size)
@@ -1904,7 +1898,7 @@ void Interpreter::visitShuffleVectorInst(ShuffleVectorInst &I){
       break;
     case Type::FloatTyID:
       for( unsigned i=0; i<src3Size; i++) {
-        unsigned j = Src3.AggregateVal[i].IntVal.getZExtValue();
+        unsigned j = std::max(0, I.getMaskValue(i));
         if(j < src1Size)
           Dest.AggregateVal[i].FloatVal = Src1.AggregateVal[j].FloatVal;
         else if(j < src1Size + src2Size)
@@ -1915,7 +1909,7 @@ void Interpreter::visitShuffleVectorInst(ShuffleVectorInst &I){
       break;
     case Type::DoubleTyID:
       for( unsigned i=0; i<src3Size; i++) {
-        unsigned j = Src3.AggregateVal[i].IntVal.getZExtValue();
+        unsigned j = std::max(0, I.getMaskValue(i));
         if(j < src1Size)
           Dest.AggregateVal[i].DoubleVal = Src1.AggregateVal[j].DoubleVal;
         else if(j < src1Size + src2Size)
@@ -1960,7 +1954,8 @@ void Interpreter::visitExtractValueInst(ExtractValueInst &I) {
     break;
     case Type::ArrayTyID:
     case Type::StructTyID:
-    case Type::VectorTyID:
+    case Type::FixedVectorTyID:
+    case Type::ScalableVectorTyID:
       Dest.AggregateVal = pSrc->AggregateVal;
     break;
     case Type::PointerTyID:
@@ -2007,7 +2002,8 @@ void Interpreter::visitInsertValueInst(InsertValueInst &I) {
     break;
     case Type::ArrayTyID:
     case Type::StructTyID:
-    case Type::VectorTyID:
+    case Type::FixedVectorTyID:
+    case Type::ScalableVectorTyID:
       pDest->AggregateVal = Src2.AggregateVal;
     break;
     case Type::PointerTyID:
@@ -2121,8 +2117,8 @@ GenericValue Interpreter::getOperandValue(Value *V, ExecutionContext &SF) {
 // callFunction - Execute the specified function...
 //
 void Interpreter::callFunction(Function *F, ArrayRef<GenericValue> ArgVals) {
-  assert((ECStack.empty() || !ECStack.back().Caller.getInstruction() ||
-          ECStack.back().Caller.arg_size() == ArgVals.size()) &&
+  assert((ECStack.empty() || !ECStack.back().Caller ||
+          ECStack.back().Caller->arg_size() == ArgVals.size()) &&
          "Incorrect number of arguments passed into function call!");
   // Make a new stack frame... and fill it in.
   ECStack.emplace_back();
diff --git a/llvm/lib/ExecutionEngine/Interpreter/ExternalFunctions.cpp b/llvm/lib/ExecutionEngine/Interpreter/ExternalFunctions.cpp
index 71b7f893d712..cb1b35d62388 100644
--- a/llvm/lib/ExecutionEngine/Interpreter/ExternalFunctions.cpp
+++ b/llvm/lib/ExecutionEngine/Interpreter/ExternalFunctions.cpp
@@ -274,7 +274,7 @@ GenericValue Interpreter::callExternalFunction(Function *F,
   RawFunc RawFn;
   if (RF == RawFunctions->end()) {
     RawFn = (RawFunc)(intptr_t)
-      sys::DynamicLibrary::SearchForAddressOfSymbol(F->getName());
+      sys::DynamicLibrary::SearchForAddressOfSymbol(std::string(F->getName()));
     if (!RawFn)
       RawFn = (RawFunc)(intptr_t)getPointerToGlobalIfAvailable(F);
     if (RawFn != 0)
diff --git a/llvm/lib/ExecutionEngine/Interpreter/Interpreter.h b/llvm/lib/ExecutionEngine/Interpreter/Interpreter.h
index e72d778317d6..fd7fa21df196 100644
--- a/llvm/lib/ExecutionEngine/Interpreter/Interpreter.h
+++ b/llvm/lib/ExecutionEngine/Interpreter/Interpreter.h
@@ -15,7 +15,6 @@
 
 #include "llvm/ExecutionEngine/ExecutionEngine.h"
 #include "llvm/ExecutionEngine/GenericValue.h"
-#include "llvm/IR/CallSite.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/InstVisitor.h"
@@ -61,8 +60,8 @@ struct ExecutionContext {
   Function             *CurFunction;// The currently executing function
   BasicBlock           *CurBB;      // The currently executing BB
   BasicBlock::iterator  CurInst;    // The next instruction to execute
-  CallSite             Caller;     // Holds the call that called subframes.
-                                   // NULL if main func or debugger invoked fn
+  CallBase             *Caller;     // Holds the call that called subframes.
+                                    // NULL if main func or debugger invoked fn
   std::map<Value *, GenericValue> Values; // LLVM values used in this invocation
   std::vector<GenericValue>  VarArgs; // Values passed through an ellipsis
   AllocaHolder Allocas;            // Track memory allocated by alloca
@@ -149,10 +148,11 @@ public:
   void visitBitCastInst(BitCastInst &I);
   void visitSelectInst(SelectInst &I);
 
-
-  void visitCallSite(CallSite CS);
-  void visitCallInst(CallInst &I) { visitCallSite (CallSite (&I)); }
-  void visitInvokeInst(InvokeInst &I) { visitCallSite (CallSite (&I)); }
+  void visitVAStartInst(VAStartInst &I);
+  void visitVAEndInst(VAEndInst &I);
+  void visitVACopyInst(VACopyInst &I);
+  void visitIntrinsicInst(IntrinsicInst &I);
+  void visitCallBase(CallBase &I);
   void visitUnreachableInst(UnreachableInst &I);
 
   void visitShl(BinaryOperator &I);
diff --git a/llvm/lib/ExecutionEngine/JITLink/BasicGOTAndStubsBuilder.h b/llvm/lib/ExecutionEngine/JITLink/BasicGOTAndStubsBuilder.h
index b47a798c7603..82258a35a675 100644
--- a/llvm/lib/ExecutionEngine/JITLink/BasicGOTAndStubsBuilder.h
+++ b/llvm/lib/ExecutionEngine/JITLink/BasicGOTAndStubsBuilder.h
@@ -15,6 +15,8 @@
 
 #include "llvm/ExecutionEngine/JITLink/JITLink.h"
 
+#define DEBUG_TYPE "jitlink"
+
 namespace llvm {
 namespace jitlink {
 
@@ -27,12 +29,25 @@ public:
     // the newly added ones, so just copy the existing blocks out.
     std::vector<Block *> Blocks(G.blocks().begin(), G.blocks().end());
 
+    LLVM_DEBUG(dbgs() << "Creating GOT entries and stubs:\n");
+
     for (auto *B : Blocks)
       for (auto &E : B->edges())
-        if (impl().isGOTEdge(E))
+        if (impl().isGOTEdge(E)) {
+          LLVM_DEBUG({
+            dbgs() << "  Updating GOT edge ";
+            printEdge(dbgs(), *B, E, "<target GOT>");
+            dbgs() << "\n";
+          });
           impl().fixGOTEdge(E, getGOTEntrySymbol(E.getTarget()));
-        else if (impl().isExternalBranchEdge(E))
+        } else if (impl().isExternalBranchEdge(E)) {
+          LLVM_DEBUG({
+            dbgs() << "  Updating external branch edge ";
+            printEdge(dbgs(), *B, E, "<target PC-rel>");
+            dbgs() << "\n";
+          });
           impl().fixExternalBranchEdge(E, getStubSymbol(E.getTarget()));
+        }
   }
 
 protected:
@@ -44,11 +59,17 @@ protected:
     // Build the entry if it doesn't exist.
     if (GOTEntryI == GOTEntries.end()) {
       auto &GOTEntry = impl().createGOTEntry(Target);
+      LLVM_DEBUG({
+        dbgs() << "    Created GOT entry for " << Target.getName() << ": "
+               << GOTEntry << "\n";
+      });
       GOTEntryI =
           GOTEntries.insert(std::make_pair(Target.getName(), &GOTEntry)).first;
     }
 
     assert(GOTEntryI != GOTEntries.end() && "Could not get GOT entry symbol");
+    LLVM_DEBUG(
+        { dbgs() << "    Using GOT entry " << *GOTEntryI->second << "\n"; });
     return *GOTEntryI->second;
   }
 
@@ -59,10 +80,15 @@ protected:
 
     if (StubI == Stubs.end()) {
       auto &StubSymbol = impl().createStub(Target);
+      LLVM_DEBUG({
+        dbgs() << "    Created stub for " << Target.getName() << ": "
+               << StubSymbol << "\n";
+      });
       StubI = Stubs.insert(std::make_pair(Target.getName(), &StubSymbol)).first;
     }
 
     assert(StubI != Stubs.end() && "Count not get stub symbol");
+    LLVM_DEBUG({ dbgs() << "    Using stub " << *StubI->second << "\n"; });
     return *StubI->second;
   }
 
diff --git a/llvm/lib/ExecutionEngine/JITLink/ELF.cpp b/llvm/lib/ExecutionEngine/JITLink/ELF.cpp
new file mode 100644
index 000000000000..6160583b13fe
--- /dev/null
+++ b/llvm/lib/ExecutionEngine/JITLink/ELF.cpp
@@ -0,0 +1,51 @@
+//===-------------- ELF.cpp - JIT linker function for ELF -------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// ELF jit-link function.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ExecutionEngine/JITLink/ELF.h"
+
+#include "llvm/BinaryFormat/ELF.h"
+#include "llvm/ExecutionEngine/JITLink/ELF_x86_64.h"
+#include "llvm/Support/Endian.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/MemoryBuffer.h"
+#include <cstring>
+
+using namespace llvm;
+
+#define DEBUG_TYPE "jitlink"
+
+namespace llvm {
+namespace jitlink {
+
+void jitLink_ELF(std::unique_ptr<JITLinkContext> Ctx) {
+
+  // We don't want to do full ELF validation here. We just verify it is elf'ish.
+  // Probably should parse into an elf header when we support more than x86 :)
+
+  StringRef Data = Ctx->getObjectBuffer().getBuffer();
+  if (Data.size() < llvm::ELF::EI_MAG3 + 1) {
+    Ctx->notifyFailed(make_error<JITLinkError>("Truncated ELF buffer"));
+    return;
+  }
+
+  if (!memcmp(Data.data(), llvm::ELF::ElfMagic, strlen(llvm::ELF::ElfMagic))) {
+    if (Data.data()[llvm::ELF::EI_CLASS] == ELF::ELFCLASS64) {
+      return jitLink_ELF_x86_64(std::move(Ctx));
+    }
+  }
+
+  Ctx->notifyFailed(make_error<JITLinkError>("ELF magic not valid"));
+}
+
+} // end namespace jitlink
+} // end namespace llvm
diff --git a/llvm/lib/ExecutionEngine/JITLink/ELF_x86_64.cpp b/llvm/lib/ExecutionEngine/JITLink/ELF_x86_64.cpp
new file mode 100644
index 000000000000..505f03590b6b
--- /dev/null
+++ b/llvm/lib/ExecutionEngine/JITLink/ELF_x86_64.cpp
@@ -0,0 +1,463 @@
+//===---- ELF_x86_64.cpp -JIT linker implementation for ELF/x86-64 ----===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// ELF/x86-64 jit-link implementation.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ExecutionEngine/JITLink/ELF_x86_64.h"
+#include "JITLinkGeneric.h"
+#include "llvm/ExecutionEngine/JITLink/JITLink.h"
+#include "llvm/Object/ELFObjectFile.h"
+
+#define DEBUG_TYPE "jitlink"
+
+using namespace llvm;
+using namespace llvm::jitlink;
+
+static const char *CommonSectionName = "__common";
+
+namespace llvm {
+namespace jitlink {
+
+// This should become a template as the ELFFile is so a lot of this could become
+// generic
+class ELFLinkGraphBuilder_x86_64 {
+
+private:
+  Section *CommonSection = nullptr;
+  // TODO hack to get this working
+  // Find a better way
+  using SymbolTable = object::ELFFile<object::ELF64LE>::Elf_Shdr;
+  // For now we just assume
+  std::map<int32_t, Symbol *> JITSymbolTable;
+
+  Section &getCommonSection() {
+    if (!CommonSection) {
+      auto Prot = static_cast<sys::Memory::ProtectionFlags>(
+          sys::Memory::MF_READ | sys::Memory::MF_WRITE);
+      CommonSection = &G->createSection(CommonSectionName, Prot);
+    }
+    return *CommonSection;
+  }
+
+  static Expected<ELF_x86_64_Edges::ELFX86RelocationKind>
+  getRelocationKind(const uint32_t Type) {
+    switch (Type) {
+    case ELF::R_X86_64_PC32:
+      return ELF_x86_64_Edges::ELFX86RelocationKind::PCRel32;
+    }
+    return make_error<JITLinkError>("Unsupported x86-64 relocation:" +
+                                    formatv("{0:d}", Type));
+  }
+
+  std::unique_ptr<LinkGraph> G;
+  // This could be a template
+  const object::ELFFile<object::ELF64LE> &Obj;
+  object::ELFFile<object::ELF64LE>::Elf_Shdr_Range sections;
+  SymbolTable SymTab;
+
+  bool isRelocatable() { return Obj.getHeader()->e_type == llvm::ELF::ET_REL; }
+
+  support::endianness
+  getEndianness(const object::ELFFile<object::ELF64LE> &Obj) {
+    return Obj.isLE() ? support::little : support::big;
+  }
+
+  // This could also just become part of a template
+  unsigned getPointerSize(const object::ELFFile<object::ELF64LE> &Obj) {
+    return Obj.getHeader()->getFileClass() == ELF::ELFCLASS64 ? 8 : 4;
+  }
+
+  // We don't technically need this right now
+  // But for now going to keep it as it helps me to debug things
+
+  Error createNormalizedSymbols() {
+    LLVM_DEBUG(dbgs() << "Creating normalized symbols...\n");
+
+    for (auto SecRef : sections) {
+      if (SecRef.sh_type != ELF::SHT_SYMTAB &&
+          SecRef.sh_type != ELF::SHT_DYNSYM)
+        continue;
+
+      auto Symbols = Obj.symbols(&SecRef);
+      // TODO: Currently I use this function to test things 
+      // I also want to leave it to see if its common between MACH and elf
+      // so for now I just want to continue even if there is an error
+      if (errorToBool(Symbols.takeError()))
+        continue;
+
+      auto StrTabSec = Obj.getSection(SecRef.sh_link);
+      if (!StrTabSec)
+        return StrTabSec.takeError();
+      auto StringTable = Obj.getStringTable(*StrTabSec);
+      if (!StringTable)
+        return StringTable.takeError();
+
+      for (auto SymRef : *Symbols) {
+        Optional<StringRef> Name;
+        uint64_t Size = 0;
+
+        // FIXME: Read size.
+        (void)Size;
+
+        if (auto NameOrErr = SymRef.getName(*StringTable))
+          Name = *NameOrErr;
+        else
+          return NameOrErr.takeError();
+
+        LLVM_DEBUG({
+          dbgs() << "  ";
+          if (!Name)
+            dbgs() << "<anonymous symbol>";
+          else
+            dbgs() << *Name;
+          dbgs() << ": value = " << formatv("{0:x16}", SymRef.getValue())
+                 << ", type = " << formatv("{0:x2}", SymRef.getType())
+                 << ", binding = " << SymRef.getBinding()
+                 << ", size =" << Size;
+          dbgs() << "\n";
+        });
+      }
+    }
+    return Error::success();
+  }
+
+  Error createNormalizedSections() {
+    LLVM_DEBUG(dbgs() << "Creating normalized sections...\n");
+    for (auto &SecRef : sections) {
+      auto Name = Obj.getSectionName(&SecRef);
+      if (!Name)
+        return Name.takeError();
+      sys::Memory::ProtectionFlags Prot;
+      if (SecRef.sh_flags & ELF::SHF_EXECINSTR) {
+        Prot = static_cast<sys::Memory::ProtectionFlags>(sys::Memory::MF_READ |
+                                                         sys::Memory::MF_EXEC);
+      } else {
+        Prot = static_cast<sys::Memory::ProtectionFlags>(sys::Memory::MF_READ |
+                                                         sys::Memory::MF_WRITE);
+      }
+      uint64_t Address = SecRef.sh_addr;
+      uint64_t Size = SecRef.sh_size;
+      uint64_t Flags = SecRef.sh_flags;
+      uint64_t Alignment = SecRef.sh_addralign;
+      const char *Data = nullptr;
+      // TODO: figure out what it is that has 0 size no name and address
+      // 0000-0000
+      if (Size == 0)
+        continue;
+
+      // FIXME: Use flags.
+      (void)Flags;
+
+      LLVM_DEBUG({
+        dbgs() << "  " << *Name << ": " << formatv("{0:x16}", Address) << " -- "
+               << formatv("{0:x16}", Address + Size) << ", align: " << Alignment
+               << " Flags:" << Flags << "\n";
+      });
+
+      if (SecRef.sh_type != ELF::SHT_NOBITS) {
+        // .sections() already checks that the data is not beyond the end of
+        // file
+        auto contents = Obj.getSectionContentsAsArray<char>(&SecRef);
+        if (!contents)
+          return contents.takeError();
+
+        Data = contents->data();
+        // TODO protection flags.
+        // for now everything is
+        auto &section = G->createSection(*Name, Prot);
+        // Do this here because we have it, but move it into graphify later
+        G->createContentBlock(section, StringRef(Data, Size), Address,
+                              Alignment, 0);
+        if (SecRef.sh_type == ELF::SHT_SYMTAB)
+          // TODO: Dynamic?
+          SymTab = SecRef;
+      }
+    }
+
+    return Error::success();
+  }
+
+  Error addRelocations() {
+    LLVM_DEBUG(dbgs() << "Adding relocations\n");
+    // TODO a partern is forming of iterate some sections but only give me
+    // ones I am interested, i should abstract that concept some where
+    for (auto &SecRef : sections) {
+      if (SecRef.sh_type != ELF::SHT_RELA && SecRef.sh_type != ELF::SHT_REL)
+        continue;
+      // TODO can the elf obj file do this for me?
+      if (SecRef.sh_type == ELF::SHT_REL)
+        return make_error<llvm::StringError>("Shouldn't have REL in x64",
+                                             llvm::inconvertibleErrorCode());
+
+      auto RelSectName = Obj.getSectionName(&SecRef);
+      if (!RelSectName)
+        return RelSectName.takeError();
+      // Deal with .eh_frame later
+      if (*RelSectName == StringRef(".rela.eh_frame"))
+        continue;
+
+      auto UpdateSection = Obj.getSection(SecRef.sh_info);
+      if (!UpdateSection)
+        return UpdateSection.takeError();
+
+      auto UpdateSectionName = Obj.getSectionName(*UpdateSection);
+      if (!UpdateSectionName)
+        return UpdateSectionName.takeError();
+
+      auto JITSection = G->findSectionByName(*UpdateSectionName);
+      if (!JITSection)
+        return make_error<llvm::StringError>(
+            "Refencing a a section that wasn't added to graph" +
+                *UpdateSectionName,
+            llvm::inconvertibleErrorCode());
+
+      auto Relocations = Obj.relas(&SecRef);
+      if (!Relocations)
+        return Relocations.takeError();
+
+      for (const auto &Rela : *Relocations) {
+        auto Type = Rela.getType(false);
+
+        LLVM_DEBUG({
+          dbgs() << "Relocation Type: " << Type << "\n"
+                 << "Name: " << Obj.getRelocationTypeName(Type) << "\n";
+        });
+
+        auto Symbol = Obj.getRelocationSymbol(&Rela, &SymTab);
+        if (!Symbol)
+          return Symbol.takeError();
+
+        auto BlockToFix = *(JITSection->blocks().begin());
+        auto TargetSymbol = JITSymbolTable[(*Symbol)->st_shndx];
+        uint64_t Addend = Rela.r_addend;
+        JITTargetAddress FixupAddress =
+            (*UpdateSection)->sh_addr + Rela.r_offset;
+
+        LLVM_DEBUG({
+          dbgs() << "Processing relocation at "
+                 << format("0x%016" PRIx64, FixupAddress) << "\n";
+        });
+        auto Kind = getRelocationKind(Type);
+        if (!Kind)
+          return Kind.takeError();
+
+        LLVM_DEBUG({
+          Edge GE(*Kind, FixupAddress - BlockToFix->getAddress(), *TargetSymbol,
+                  Addend);
+          // TODO a mapping of KIND => type then call getRelocationTypeName4
+          printEdge(dbgs(), *BlockToFix, GE, StringRef(""));
+          dbgs() << "\n";
+        });
+        BlockToFix->addEdge(*Kind, FixupAddress - BlockToFix->getAddress(),
+                            *TargetSymbol, Addend);
+      }
+    }
+    return Error::success();
+  }
+
+  Error graphifyRegularSymbols() {
+
+    // TODO: ELF supports beyond SHN_LORESERVE,
+    // need to perf test how a vector vs map handles those cases
+
+    std::vector<std::vector<object::ELFFile<object::ELF64LE>::Elf_Shdr_Range *>>
+        SecIndexToSymbols;
+
+    LLVM_DEBUG(dbgs() << "Creating graph symbols...\n");
+
+    for (auto SecRef : sections) {
+
+      if (SecRef.sh_type != ELF::SHT_SYMTAB &&
+          SecRef.sh_type != ELF::SHT_DYNSYM)
+        continue;
+      auto Symbols = Obj.symbols(&SecRef);
+      if (!Symbols)
+        return Symbols.takeError();
+
+      auto StrTabSec = Obj.getSection(SecRef.sh_link);
+      if (!StrTabSec)
+        return StrTabSec.takeError();
+      auto StringTable = Obj.getStringTable(*StrTabSec);
+      if (!StringTable)
+        return StringTable.takeError();
+      auto Name = Obj.getSectionName(&SecRef);
+      if (!Name)
+        return Name.takeError();
+      auto Section = G->findSectionByName(*Name);
+      if (!Section)
+        return make_error<llvm::StringError>("Could not find a section",
+                                             llvm::inconvertibleErrorCode());
+      // we only have one for now
+      auto blocks = Section->blocks();
+      if (blocks.empty())
+        return make_error<llvm::StringError>("Section has no block",
+                                             llvm::inconvertibleErrorCode());
+
+      for (auto SymRef : *Symbols) {
+        auto Type = SymRef.getType();
+        if (Type == ELF::STT_NOTYPE || Type == ELF::STT_FILE)
+          continue;
+        // these should do it for now
+        // if(Type != ELF::STT_NOTYPE &&
+        //   Type != ELF::STT_OBJECT &&
+        //   Type != ELF::STT_FUNC    &&
+        //   Type != ELF::STT_SECTION &&
+        //   Type != ELF::STT_COMMON) {
+        //     continue;
+        //   }
+        std::pair<Linkage, Scope> bindings;
+        auto Name = SymRef.getName(*StringTable);
+        // I am not sure on If this is going to hold as an invariant. Revisit.
+        if (!Name)
+          return Name.takeError();
+        // TODO: weak and hidden
+        if (SymRef.isExternal())
+          bindings = {Linkage::Strong, Scope::Default};
+        else
+          bindings = {Linkage::Strong, Scope::Local};
+
+        if (SymRef.isDefined() &&
+            (Type == ELF::STT_FUNC || Type == ELF::STT_OBJECT)) {
+
+          auto DefinedSection = Obj.getSection(SymRef.st_shndx);
+          if (!DefinedSection)
+            return DefinedSection.takeError();
+          auto sectName = Obj.getSectionName(*DefinedSection);
+          if (!sectName)
+            return Name.takeError();
+
+          auto JitSection = G->findSectionByName(*sectName);
+          if (!JitSection)
+            return make_error<llvm::StringError>(
+                "Could not find a section", llvm::inconvertibleErrorCode());
+          auto bs = JitSection->blocks();
+          if (bs.empty())
+            return make_error<llvm::StringError>(
+                "Section has no block", llvm::inconvertibleErrorCode());
+
+          auto B = *bs.begin();
+          LLVM_DEBUG({ dbgs() << "  " << *Name << ": "; });
+
+          auto &S = G->addDefinedSymbol(
+              *B, SymRef.getValue(), *Name, SymRef.st_size, bindings.first,
+              bindings.second, SymRef.getType() == ELF::STT_FUNC, false);
+          JITSymbolTable[SymRef.st_shndx] = &S;
+        }
+        //TODO: The following has to be implmented.
+        // leaving commented out to save time for future patchs
+        /*
+          G->addAbsoluteSymbol(*Name, SymRef.getValue(), SymRef.st_size,
+          Linkage::Strong, Scope::Default, false);
+
+          if(SymRef.isCommon()) {
+            G->addCommonSymbol(*Name, Scope::Default, getCommonSection(), 0, 0,
+          SymRef.getValue(), false);
+          }
+
+
+          //G->addExternalSymbol(*Name, SymRef.st_size, Linkage::Strong);
+  */
+      }
+    }
+    return Error::success();
+  }
+
+public:
+  ELFLinkGraphBuilder_x86_64(std::string filename,
+                             const object::ELFFile<object::ELF64LE> &Obj)
+      : G(std::make_unique<LinkGraph>(filename, getPointerSize(Obj),
+                                      getEndianness(Obj))),
+        Obj(Obj) {}
+
+  Expected<std::unique_ptr<LinkGraph>> buildGraph() {
+    // Sanity check: we only operate on relocatable objects.
+    if (!isRelocatable())
+      return make_error<JITLinkError>("Object is not a relocatable ELF");
+
+    auto Secs = Obj.sections();
+
+    if (!Secs) {
+      return Secs.takeError();
+    }
+    sections = *Secs;
+
+    if (auto Err = createNormalizedSections())
+      return std::move(Err);
+
+    if (auto Err = createNormalizedSymbols())
+      return std::move(Err);
+
+    if (auto Err = graphifyRegularSymbols())
+      return std::move(Err);
+
+    if (auto Err = addRelocations())
+      return std::move(Err);
+
+    return std::move(G);
+  }
+};
+
+class ELFJITLinker_x86_64 : public JITLinker<ELFJITLinker_x86_64> {
+  friend class JITLinker<ELFJITLinker_x86_64>;
+
+public:
+  ELFJITLinker_x86_64(std::unique_ptr<JITLinkContext> Ctx,
+                      PassConfiguration PassConfig)
+      : JITLinker(std::move(Ctx), std::move(PassConfig)) {}
+
+private:
+  StringRef getEdgeKindName(Edge::Kind R) const override { return StringRef(); }
+
+  Expected<std::unique_ptr<LinkGraph>>
+  buildGraph(MemoryBufferRef ObjBuffer) override {
+    auto ELFObj = object::ObjectFile::createELFObjectFile(ObjBuffer);
+    if (!ELFObj)
+      return ELFObj.takeError();
+
+    auto &ELFObjFile = cast<object::ELFObjectFile<object::ELF64LE>>(**ELFObj);
+    std::string fileName(ELFObj->get()->getFileName());
+    return ELFLinkGraphBuilder_x86_64(std::move(fileName),
+                                      *ELFObjFile.getELFFile())
+        .buildGraph();
+  }
+
+  Error applyFixup(Block &B, const Edge &E, char *BlockWorkingMem) const {
+    using namespace ELF_x86_64_Edges;
+    char *FixupPtr = BlockWorkingMem + E.getOffset();
+    JITTargetAddress FixupAddress = B.getAddress() + E.getOffset();
+    switch (E.getKind()) {
+
+    case ELFX86RelocationKind::PCRel32:
+      int64_t Value = E.getTarget().getAddress() + E.getAddend() - FixupAddress;
+      // verify
+      *(support::little32_t *)FixupPtr = Value;
+      break;
+    }
+    return Error::success();
+  }
+};
+
+void jitLink_ELF_x86_64(std::unique_ptr<JITLinkContext> Ctx) {
+  PassConfiguration Config;
+  Triple TT("x86_64-linux");
+  // Construct a JITLinker and run the link function.
+  // Add a mark-live pass.
+  if (auto MarkLive = Ctx->getMarkLivePass(TT))
+    Config.PrePrunePasses.push_back(std::move(MarkLive));
+  else
+    Config.PrePrunePasses.push_back(markAllSymbolsLive);
+
+  if (auto Err = Ctx->modifyPassConfig(TT, Config))
+    return Ctx->notifyFailed(std::move(Err));
+
+  ELFJITLinker_x86_64::link(std::move(Ctx), std::move(Config));
+}
+} // end namespace jitlink
+} // end namespace llvm
diff --git a/llvm/lib/ExecutionEngine/JITLink/JITLink.cpp b/llvm/lib/ExecutionEngine/JITLink/JITLink.cpp
index 6c924f889577..5105ec495148 100644
--- a/llvm/lib/ExecutionEngine/JITLink/JITLink.cpp
+++ b/llvm/lib/ExecutionEngine/JITLink/JITLink.cpp
@@ -10,6 +10,7 @@
 #include "llvm/ExecutionEngine/JITLink/JITLink.h"
 
 #include "llvm/BinaryFormat/Magic.h"
+#include "llvm/ExecutionEngine/JITLink/ELF.h"
 #include "llvm/ExecutionEngine/JITLink/MachO.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Support/ManagedStatic.h"
@@ -180,18 +181,14 @@ Block &LinkGraph::splitBlock(Block &B, size_t SplitIndex,
     // Copy edges to NewBlock (recording their iterators so that we can remove
     // them from B), and update of Edges remaining on B.
     std::vector<Block::edge_iterator> EdgesToRemove;
-    for (auto I = B.edges().begin(), E = B.edges().end(); I != E; ++I) {
+    for (auto I = B.edges().begin(); I != B.edges().end();) {
       if (I->getOffset() < SplitIndex) {
         NewBlock.addEdge(*I);
-        EdgesToRemove.push_back(I);
-      } else
+        I = B.removeEdge(I);
+      } else {
         I->setOffset(I->getOffset() - SplitIndex);
-    }
-
-    // Remove edges that were transfered to NewBlock from B.
-    while (!EdgesToRemove.empty()) {
-      B.removeEdge(EdgesToRemove.back());
-      EdgesToRemove.pop_back();
+        ++I;
+      }
     }
   }
 
@@ -304,6 +301,8 @@ void jitLink(std::unique_ptr<JITLinkContext> Ctx) {
   switch (Magic) {
   case file_magic::macho_object:
     return jitLink_MachO(std::move(Ctx));
+  case file_magic::elf_relocatable:
+    return jitLink_ELF(std::move(Ctx));
   default:
     Ctx->notifyFailed(make_error<JITLinkError>("Unsupported file format"));
   };
diff --git a/llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.cpp b/llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.cpp
index 7b594fd2c0ea..1d76a49939dc 100644
--- a/llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.cpp
+++ b/llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.cpp
@@ -24,6 +24,8 @@ JITLinkerBase::~JITLinkerBase() {}
 
 void JITLinkerBase::linkPhase1(std::unique_ptr<JITLinkerBase> Self) {
 
+  LLVM_DEBUG({ dbgs() << "Building jitlink graph for new input...\n"; });
+
   // Build the link graph.
   if (auto GraphOrErr = buildGraph(Ctx->getObjectBuffer()))
     G = std::move(*GraphOrErr);
@@ -31,6 +33,10 @@ void JITLinkerBase::linkPhase1(std::unique_ptr<JITLinkerBase> Self) {
     return Ctx->notifyFailed(GraphOrErr.takeError());
   assert(G && "Graph should have been created by buildGraph above");
 
+  LLVM_DEBUG({
+    dbgs() << "Starting link phase 1 for graph " << G->getName() << "\n";
+  });
+
   // Prune and optimize the graph.
   if (auto Err = runPasses(Passes.PrePrunePasses))
     return Ctx->notifyFailed(std::move(Err));
@@ -59,10 +65,17 @@ void JITLinkerBase::linkPhase1(std::unique_ptr<JITLinkerBase> Self) {
     return Ctx->notifyFailed(std::move(Err));
 
   // Notify client that the defined symbols have been assigned addresses.
+  LLVM_DEBUG(
+      { dbgs() << "Resolving symbols defined in " << G->getName() << "\n"; });
   Ctx->notifyResolved(*G);
 
   auto ExternalSymbols = getExternalSymbolNames();
 
+  LLVM_DEBUG({
+    dbgs() << "Issuing lookup for external symbols for " << G->getName()
+           << " (may trigger materialization/linking of other graphs)...\n";
+  });
+
   // We're about to hand off ownership of ourself to the continuation. Grab a
   // pointer to the context so that we can call it to initiate the lookup.
   //
@@ -87,6 +100,11 @@ void JITLinkerBase::linkPhase1(std::unique_ptr<JITLinkerBase> Self) {
 void JITLinkerBase::linkPhase2(std::unique_ptr<JITLinkerBase> Self,
                                Expected<AsyncLookupResult> LR,
                                SegmentLayoutMap Layout) {
+
+  LLVM_DEBUG({
+    dbgs() << "Starting link phase 2 for graph " << G->getName() << "\n";
+  });
+
   // If the lookup failed, bail out.
   if (!LR)
     return deallocateAndBailOut(LR.takeError());
@@ -94,13 +112,25 @@ void JITLinkerBase::linkPhase2(std::unique_ptr<JITLinkerBase> Self,
   // Assign addresses to external addressables.
   applyLookupResult(*LR);
 
+  // Copy block content to working memory.
+  copyBlockContentToWorkingMemory(Layout, *Alloc);
+
+  LLVM_DEBUG({
+    dbgs() << "Link graph \"" << G->getName()
+           << "\" before post-allocation passes:\n";
+    dumpGraph(dbgs());
+  });
+
+  if (auto Err = runPasses(Passes.PostAllocationPasses))
+    return deallocateAndBailOut(std::move(Err));
+
   LLVM_DEBUG({
     dbgs() << "Link graph \"" << G->getName() << "\" before copy-and-fixup:\n";
     dumpGraph(dbgs());
   });
 
-  // Copy block content to working memory and fix up.
-  if (auto Err = copyAndFixUpBlocks(Layout, *Alloc))
+  // Fix up block content.
+  if (auto Err = fixUpBlocks(*G))
     return deallocateAndBailOut(std::move(Err));
 
   LLVM_DEBUG({
@@ -122,9 +152,16 @@ void JITLinkerBase::linkPhase2(std::unique_ptr<JITLinkerBase> Self,
 }
 
 void JITLinkerBase::linkPhase3(std::unique_ptr<JITLinkerBase> Self, Error Err) {
+
+  LLVM_DEBUG({
+    dbgs() << "Starting link phase 3 for graph " << G->getName() << "\n";
+  });
+
   if (Err)
     return deallocateAndBailOut(std::move(Err));
   Ctx->notifyFinalized(std::move(Alloc));
+
+  LLVM_DEBUG({ dbgs() << "Link of graph " << G->getName() << " complete\n"; });
 }
 
 Error JITLinkerBase::runPasses(LinkGraphPassList &Passes) {
@@ -165,7 +202,7 @@ JITLinkerBase::SegmentLayoutMap JITLinkerBase::layOutBlocks() {
   }
 
   LLVM_DEBUG({
-    dbgs() << "Segment ordering:\n";
+    dbgs() << "Computed segment ordering:\n";
     for (auto &KV : Layout) {
       dbgs() << "  Segment "
              << static_cast<sys::Memory::ProtectionFlags>(KV.first) << ":\n";
@@ -230,11 +267,12 @@ Error JITLinkerBase::allocateSegments(const SegmentLayoutMap &Layout) {
     return AllocOrErr.takeError();
 
   LLVM_DEBUG({
-    dbgs() << "JIT linker got working memory:\n";
+    dbgs() << "JIT linker got memory (working -> target):\n";
     for (auto &KV : Layout) {
       auto Prot = static_cast<sys::Memory::ProtectionFlags>(KV.first);
       dbgs() << "  " << Prot << ": "
-             << (const void *)Alloc->getWorkingMemory(Prot).data() << "\n";
+             << (const void *)Alloc->getWorkingMemory(Prot).data() << " -> "
+             << formatv("{0:x16}", Alloc->getTargetMemory(Prot)) << "\n";
     }
   });
 
@@ -302,6 +340,77 @@ void JITLinkerBase::applyLookupResult(AsyncLookupResult Result) {
          "All strong external symbols should have been resolved by now");
 }
 
+void JITLinkerBase::copyBlockContentToWorkingMemory(
+    const SegmentLayoutMap &Layout, JITLinkMemoryManager::Allocation &Alloc) {
+
+  LLVM_DEBUG(dbgs() << "Copying block content:\n");
+  for (auto &KV : Layout) {
+    auto &Prot = KV.first;
+    auto &SegLayout = KV.second;
+
+    auto SegMem =
+        Alloc.getWorkingMemory(static_cast<sys::Memory::ProtectionFlags>(Prot));
+    char *LastBlockEnd = SegMem.data();
+    char *BlockDataPtr = LastBlockEnd;
+
+    LLVM_DEBUG({
+      dbgs() << "  Processing segment "
+             << static_cast<sys::Memory::ProtectionFlags>(Prot) << " [ "
+             << (const void *)SegMem.data() << " .. "
+             << (const void *)((char *)SegMem.data() + SegMem.size())
+             << " ]\n    Processing content sections:\n";
+    });
+
+    for (auto *B : SegLayout.ContentBlocks) {
+      LLVM_DEBUG(dbgs() << "    " << *B << ":\n");
+
+      // Pad to alignment/alignment-offset.
+      BlockDataPtr = alignToBlock(BlockDataPtr, *B);
+
+      LLVM_DEBUG({
+        dbgs() << "      Bumped block pointer to " << (const void *)BlockDataPtr
+               << " to meet block alignment " << B->getAlignment()
+               << " and alignment offset " << B->getAlignmentOffset() << "\n";
+      });
+
+      // Zero pad up to alignment.
+      LLVM_DEBUG({
+        if (LastBlockEnd != BlockDataPtr)
+          dbgs() << "      Zero padding from " << (const void *)LastBlockEnd
+                 << " to " << (const void *)BlockDataPtr << "\n";
+      });
+
+      while (LastBlockEnd != BlockDataPtr)
+        *LastBlockEnd++ = 0;
+
+      // Copy initial block content.
+      LLVM_DEBUG({
+        dbgs() << "      Copying block " << *B << " content, "
+               << B->getContent().size() << " bytes, from "
+               << (const void *)B->getContent().data() << " to "
+               << (const void *)BlockDataPtr << "\n";
+      });
+      memcpy(BlockDataPtr, B->getContent().data(), B->getContent().size());
+
+      // Point the block's content to the fixed up buffer.
+      B->setContent(StringRef(BlockDataPtr, B->getContent().size()));
+
+      // Update block end pointer.
+      LastBlockEnd = BlockDataPtr + B->getContent().size();
+      BlockDataPtr = LastBlockEnd;
+    }
+
+    // Zero pad the rest of the segment.
+    LLVM_DEBUG({
+      dbgs() << "    Zero padding end of segment from "
+             << (const void *)LastBlockEnd << " to "
+             << (const void *)((char *)SegMem.data() + SegMem.size()) << "\n";
+    });
+    while (LastBlockEnd != SegMem.data() + SegMem.size())
+      *LastBlockEnd++ = 0;
+  }
+}
+
 void JITLinkerBase::deallocateAndBailOut(Error Err) {
   assert(Err && "Should not be bailing out on success value");
   assert(Alloc && "can not call deallocateAndBailOut before allocation");
diff --git a/llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.h b/llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.h
index d5687b7afc96..87e5e8bbc98d 100644
--- a/llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.h
+++ b/llvm/lib/ExecutionEngine/JITLink/JITLinkGeneric.h
@@ -80,13 +80,13 @@ protected:
   // For debug dumping of the link graph.
   virtual StringRef getEdgeKindName(Edge::Kind K) const = 0;
 
-  // Alight a JITTargetAddress to conform with block alignment requirements.
+  // Align a JITTargetAddress to conform with block alignment requirements.
   static JITTargetAddress alignToBlock(JITTargetAddress Addr, Block &B) {
     uint64_t Delta = (B.getAlignmentOffset() - Addr) % B.getAlignment();
     return Addr + Delta;
   }
 
-  // Alight a pointer to conform with block alignment requirements.
+  // Align a pointer to conform with block alignment requirements.
   static char *alignToBlock(char *P, Block &B) {
     uint64_t PAddr = static_cast<uint64_t>(reinterpret_cast<uintptr_t>(P));
     uint64_t Delta = (B.getAlignmentOffset() - PAddr) % B.getAlignment();
@@ -100,14 +100,14 @@ private:
 
   // Copy block contents and apply relocations.
   // Implemented in JITLinker.
-  virtual Error
-  copyAndFixUpBlocks(const SegmentLayoutMap &Layout,
-                     JITLinkMemoryManager::Allocation &Alloc) const = 0;
+  virtual Error fixUpBlocks(LinkGraph &G) const = 0;
 
   SegmentLayoutMap layOutBlocks();
   Error allocateSegments(const SegmentLayoutMap &Layout);
   JITLinkContext::LookupMap getExternalSymbolNames() const;
   void applyLookupResult(AsyncLookupResult LR);
+  void copyBlockContentToWorkingMemory(const SegmentLayoutMap &Layout,
+                                       JITLinkMemoryManager::Allocation &Alloc);
   void deallocateAndBailOut(Error Err);
 
   void dumpGraph(raw_ostream &OS);
@@ -144,88 +144,25 @@ private:
     return static_cast<const LinkerImpl &>(*this);
   }
 
-  Error
-  copyAndFixUpBlocks(const SegmentLayoutMap &Layout,
-                     JITLinkMemoryManager::Allocation &Alloc) const override {
-    LLVM_DEBUG(dbgs() << "Copying and fixing up blocks:\n");
-    for (auto &KV : Layout) {
-      auto &Prot = KV.first;
-      auto &SegLayout = KV.second;
-
-      auto SegMem = Alloc.getWorkingMemory(
-          static_cast<sys::Memory::ProtectionFlags>(Prot));
-      char *LastBlockEnd = SegMem.data();
-      char *BlockDataPtr = LastBlockEnd;
-
-      LLVM_DEBUG({
-        dbgs() << "  Processing segment "
-               << static_cast<sys::Memory::ProtectionFlags>(Prot) << " [ "
-               << (const void *)SegMem.data() << " .. "
-               << (const void *)((char *)SegMem.data() + SegMem.size())
-               << " ]\n    Processing content sections:\n";
-      });
-
-      for (auto *B : SegLayout.ContentBlocks) {
-        LLVM_DEBUG(dbgs() << "    " << *B << ":\n");
-
-        // Pad to alignment/alignment-offset.
-        BlockDataPtr = alignToBlock(BlockDataPtr, *B);
-
-        LLVM_DEBUG({
-          dbgs() << "      Bumped block pointer to "
-                 << (const void *)BlockDataPtr << " to meet block alignment "
-                 << B->getAlignment() << " and alignment offset "
-                 << B->getAlignmentOffset() << "\n";
-        });
-
-        // Zero pad up to alignment.
-        LLVM_DEBUG({
-          if (LastBlockEnd != BlockDataPtr)
-            dbgs() << "      Zero padding from " << (const void *)LastBlockEnd
-                   << " to " << (const void *)BlockDataPtr << "\n";
-        });
-
-        while (LastBlockEnd != BlockDataPtr)
-          *LastBlockEnd++ = 0;
-
-        // Copy initial block content.
-        LLVM_DEBUG({
-          dbgs() << "      Copying block " << *B << " content, "
-                 << B->getContent().size() << " bytes, from "
-                 << (const void *)B->getContent().data() << " to "
-                 << (const void *)BlockDataPtr << "\n";
-        });
-        memcpy(BlockDataPtr, B->getContent().data(), B->getContent().size());
-
-        // Copy Block data and apply fixups.
-        LLVM_DEBUG(dbgs() << "      Applying fixups.\n");
-        for (auto &E : B->edges()) {
-
-          // Skip non-relocation edges.
-          if (!E.isRelocation())
-            continue;
-
-          // Dispatch to LinkerImpl for fixup.
-          if (auto Err = impl().applyFixup(*B, E, BlockDataPtr))
-            return Err;
-        }
-
-        // Point the block's content to the fixed up buffer.
-        B->setContent(StringRef(BlockDataPtr, B->getContent().size()));
-
-        // Update block end pointer.
-        LastBlockEnd = BlockDataPtr + B->getContent().size();
-        BlockDataPtr = LastBlockEnd;
-      }
+  Error fixUpBlocks(LinkGraph &G) const override {
+    LLVM_DEBUG(dbgs() << "Fixing up blocks:\n");
+
+    for (auto *B : G.blocks()) {
+      LLVM_DEBUG(dbgs() << "  " << *B << ":\n");
+
+      // Copy Block data and apply fixups.
+      LLVM_DEBUG(dbgs() << "    Applying fixups.\n");
+      for (auto &E : B->edges()) {
 
-      // Zero pad the rest of the segment.
-      LLVM_DEBUG({
-        dbgs() << "    Zero padding end of segment from "
-               << (const void *)LastBlockEnd << " to "
-               << (const void *)((char *)SegMem.data() + SegMem.size()) << "\n";
-      });
-      while (LastBlockEnd != SegMem.data() + SegMem.size())
-        *LastBlockEnd++ = 0;
+        // Skip non-relocation edges.
+        if (!E.isRelocation())
+          continue;
+
+        // Dispatch to LinkerImpl for fixup.
+        auto *BlockData = const_cast<char *>(B->getContent().data());
+        if (auto Err = impl().applyFixup(*B, E, BlockData))
+          return Err;
+      }
     }
 
     return Error::success();
diff --git a/llvm/lib/ExecutionEngine/JITLink/JITLinkMemoryManager.cpp b/llvm/lib/ExecutionEngine/JITLink/JITLinkMemoryManager.cpp
index 9e0d207e8bdb..68ec9d79af9b 100644
--- a/llvm/lib/ExecutionEngine/JITLink/JITLinkMemoryManager.cpp
+++ b/llvm/lib/ExecutionEngine/JITLink/JITLinkMemoryManager.cpp
@@ -32,7 +32,7 @@ InProcessMemoryManager::allocate(const SegmentsRequestMap &Request) {
     }
     JITTargetAddress getTargetMemory(ProtectionFlags Seg) override {
       assert(SegBlocks.count(Seg) && "No allocation for segment");
-      return reinterpret_cast<JITTargetAddress>(SegBlocks[Seg].base());
+      return pointerToJITTargetAddress(SegBlocks[Seg].base());
     }
     void finalizeAsync(FinalizeContinuation OnFinalize) override {
       OnFinalize(applyProtections());
diff --git a/llvm/lib/ExecutionEngine/JITLink/MachO.cpp b/llvm/lib/ExecutionEngine/JITLink/MachO.cpp
index 58bc0f56e155..b3e45868ab22 100644
--- a/llvm/lib/ExecutionEngine/JITLink/MachO.cpp
+++ b/llvm/lib/ExecutionEngine/JITLink/MachO.cpp
@@ -16,9 +16,9 @@
 #include "llvm/BinaryFormat/MachO.h"
 #include "llvm/ExecutionEngine/JITLink/MachO_arm64.h"
 #include "llvm/ExecutionEngine/JITLink/MachO_x86_64.h"
-#include "llvm/Support/Endian.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/SwapByteOrder.h"
 
 using namespace llvm;
 
@@ -34,7 +34,9 @@ void jitLink_MachO(std::unique_ptr<JITLinkContext> Ctx) {
 
   StringRef Data = Ctx->getObjectBuffer().getBuffer();
   if (Data.size() < 4) {
-    Ctx->notifyFailed(make_error<JITLinkError>("Truncated MachO buffer"));
+    StringRef BufferName = Ctx->getObjectBuffer().getBufferIdentifier();
+    Ctx->notifyFailed(make_error<JITLinkError>("Truncated MachO buffer \"" +
+                                               BufferName + "\""));
     return;
   }
 
@@ -51,20 +53,26 @@ void jitLink_MachO(std::unique_ptr<JITLinkContext> Ctx) {
         make_error<JITLinkError>("MachO 32-bit platforms not supported"));
     return;
   } else if (Magic == MachO::MH_MAGIC_64 || Magic == MachO::MH_CIGAM_64) {
-    MachO::mach_header_64 Header;
 
-    memcpy(&Header, Data.data(), sizeof(MachO::mach_header_64));
+    if (Data.size() < sizeof(MachO::mach_header_64)) {
+      StringRef BufferName = Ctx->getObjectBuffer().getBufferIdentifier();
+      Ctx->notifyFailed(make_error<JITLinkError>("Truncated MachO buffer \"" +
+                                                 BufferName + "\""));
+      return;
+    }
+
+    // Read the CPU type from the header.
+    uint32_t CPUType;
+    memcpy(&CPUType, Data.data() + 4, sizeof(uint32_t));
     if (Magic == MachO::MH_CIGAM_64)
-      swapStruct(Header);
+      CPUType = ByteSwap_32(CPUType);
 
     LLVM_DEBUG({
-      dbgs() << "jitLink_MachO: cputype = "
-             << format("0x%08" PRIx32, Header.cputype)
-             << ", cpusubtype = " << format("0x%08" PRIx32, Header.cpusubtype)
+      dbgs() << "jitLink_MachO: cputype = " << format("0x%08" PRIx32, CPUType)
              << "\n";
     });
 
-    switch (Header.cputype) {
+    switch (CPUType) {
     case MachO::CPU_TYPE_ARM64:
       return jitLink_MachO_arm64(std::move(Ctx));
     case MachO::CPU_TYPE_X86_64:
diff --git a/llvm/lib/ExecutionEngine/JITLink/MachOLinkGraphBuilder.cpp b/llvm/lib/ExecutionEngine/JITLink/MachOLinkGraphBuilder.cpp
index 701f108a9a21..fa3f403b717c 100644
--- a/llvm/lib/ExecutionEngine/JITLink/MachOLinkGraphBuilder.cpp
+++ b/llvm/lib/ExecutionEngine/JITLink/MachOLinkGraphBuilder.cpp
@@ -47,8 +47,8 @@ Expected<std::unique_ptr<LinkGraph>> MachOLinkGraphBuilder::buildGraph() {
 
 MachOLinkGraphBuilder::MachOLinkGraphBuilder(const object::MachOObjectFile &Obj)
     : Obj(Obj),
-      G(std::make_unique<LinkGraph>(Obj.getFileName(), getPointerSize(Obj),
-                                    getEndianness(Obj))) {}
+      G(std::make_unique<LinkGraph>(std::string(Obj.getFileName()),
+                                    getPointerSize(Obj), getEndianness(Obj))) {}
 
 void MachOLinkGraphBuilder::addCustomSectionParser(
     StringRef SectionName, SectionParserFunction Parser) {
@@ -64,12 +64,14 @@ Linkage MachOLinkGraphBuilder::getLinkage(uint16_t Desc) {
 }
 
 Scope MachOLinkGraphBuilder::getScope(StringRef Name, uint8_t Type) {
-  if (Name.startswith("l"))
-    return Scope::Local;
   if (Type & MachO::N_PEXT)
     return Scope::Hidden;
-  if (Type & MachO::N_EXT)
-    return Scope::Default;
+  if (Type & MachO::N_EXT) {
+    if (Name.startswith("l"))
+      return Scope::Hidden;
+    else
+      return Scope::Default;
+  }
   return Scope::Local;
 }
 
@@ -77,6 +79,11 @@ bool MachOLinkGraphBuilder::isAltEntry(const NormalizedSymbol &NSym) {
   return NSym.Desc & MachO::N_ALT_ENTRY;
 }
 
+bool MachOLinkGraphBuilder::isDebugSection(const NormalizedSection &NSec) {
+  return (NSec.Flags & MachO::S_ATTR_DEBUG &&
+          strcmp(NSec.SegName, "__DWARF") == 0);
+}
+
 unsigned
 MachOLinkGraphBuilder::getPointerSize(const object::MachOObjectFile &Obj) {
   return Obj.is64Bit() ? 8 : 4;
@@ -116,6 +123,11 @@ Error MachOLinkGraphBuilder::createNormalizedSections() {
       const MachO::section_64 &Sec64 =
           Obj.getSection64(SecRef.getRawDataRefImpl());
 
+      memcpy(&NSec.SectName, &Sec64.sectname, 16);
+      NSec.SectName[16] = '\0';
+      memcpy(&NSec.SegName, Sec64.segname, 16);
+      NSec.SegName[16] = '\0';
+
       NSec.Address = Sec64.addr;
       NSec.Size = Sec64.size;
       NSec.Alignment = 1ULL << Sec64.align;
@@ -123,6 +135,12 @@ Error MachOLinkGraphBuilder::createNormalizedSections() {
       DataOffset = Sec64.offset;
     } else {
       const MachO::section &Sec32 = Obj.getSection(SecRef.getRawDataRefImpl());
+
+      memcpy(&NSec.SectName, &Sec32.sectname, 16);
+      NSec.SectName[16] = '\0';
+      memcpy(&NSec.SegName, Sec32.segname, 16);
+      NSec.SegName[16] = '\0';
+
       NSec.Address = Sec32.addr;
       NSec.Size = Sec32.size;
       NSec.Alignment = 1ULL << Sec32.align;
@@ -162,7 +180,14 @@ Error MachOLinkGraphBuilder::createNormalizedSections() {
       Prot = static_cast<sys::Memory::ProtectionFlags>(sys::Memory::MF_READ |
                                                        sys::Memory::MF_WRITE);
 
-    NSec.GraphSection = &G->createSection(*Name, Prot);
+    if (!isDebugSection(NSec))
+      NSec.GraphSection = &G->createSection(*Name, Prot);
+    else
+      LLVM_DEBUG({
+        dbgs() << "    " << *Name
+               << " is a debug section: No graph section will be created.\n";
+      });
+
     IndexToSection.insert(std::make_pair(SecIndex, std::move(NSec)));
   }
 
@@ -189,12 +214,12 @@ Error MachOLinkGraphBuilder::createNormalizedSections() {
     auto &Next = *Sections[I + 1];
     if (Next.Address < Cur.Address + Cur.Size)
       return make_error<JITLinkError>(
-          "Address range for section " + Cur.GraphSection->getName() +
-          formatv(" [ {0:x16} -- {1:x16} ] ", Cur.Address,
-                  Cur.Address + Cur.Size) +
-          "overlaps " +
-          formatv(" [ {0:x16} -- {1:x16} ] ", Next.Address,
-                  Next.Address + Next.Size));
+          "Address range for section " +
+          formatv("\"{0}/{1}\" [ {2:x16} -- {3:x16} ] ", Cur.SegName,
+                  Cur.SectName, Cur.Address, Cur.Address + Cur.Size) +
+          "overlaps section \"" + Next.SegName + "/" + Next.SectName + "\"" +
+          formatv("\"{0}/{1}\" [ {2:x16} -- {3:x16} ] ", Next.SegName,
+                  Next.SectName, Next.Address, Next.Address + Next.Size));
   }
 
   return Error::success();
@@ -260,21 +285,28 @@ Error MachOLinkGraphBuilder::createNormalizedSymbols() {
     });
 
     // If this symbol has a section, sanity check that the addresses line up.
-    NormalizedSection *NSec = nullptr;
     if (Sect != 0) {
-      if (auto NSecOrErr = findSectionByIndex(Sect - 1))
-        NSec = &*NSecOrErr;
-      else
-        return NSecOrErr.takeError();
+      auto NSec = findSectionByIndex(Sect - 1);
+      if (!NSec)
+        return NSec.takeError();
 
       if (Value < NSec->Address || Value > NSec->Address + NSec->Size)
         return make_error<JITLinkError>("Symbol address does not fall within "
                                         "section");
+
+      if (!NSec->GraphSection) {
+        LLVM_DEBUG({
+          dbgs() << "  Skipping: Symbol is in section " << NSec->SegName << "/"
+                 << NSec->SectName
+                 << " which has no associated graph section.\n";
+        });
+        continue;
+      }
     }
 
     IndexToSymbol[SymbolIndex] =
         &createNormalizedSymbol(*Name, Value, Type, Sect, Desc,
-                                getLinkage(Type), getScope(*Name, Type));
+                                getLinkage(Desc), getScope(*Name, Type));
   }
 
   return Error::success();
@@ -362,6 +394,14 @@ Error MachOLinkGraphBuilder::graphifyRegularSymbols() {
     auto SecIndex = KV.first;
     auto &NSec = KV.second;
 
+    if (!NSec.GraphSection) {
+      LLVM_DEBUG({
+        dbgs() << "  " << NSec.SegName << "/" << NSec.SectName
+               << " has no graph section. Skipping.\n";
+      });
+      continue;
+    }
+
     // Skip sections with custom parsers.
     if (CustomSectionParserFunctions.count(NSec.GraphSection->getName())) {
       LLVM_DEBUG({
@@ -524,6 +564,10 @@ Error MachOLinkGraphBuilder::graphifySectionsWithCustomParsers() {
   for (auto &KV : IndexToSection) {
     auto &NSec = KV.second;
 
+    // Skip non-graph sections.
+    if (!NSec.GraphSection)
+      continue;
+
     auto HI = CustomSectionParserFunctions.find(NSec.GraphSection->getName());
     if (HI != CustomSectionParserFunctions.end()) {
       auto &Parse = HI->second;
diff --git a/llvm/lib/ExecutionEngine/JITLink/MachOLinkGraphBuilder.h b/llvm/lib/ExecutionEngine/JITLink/MachOLinkGraphBuilder.h
index 91b1d5a22387..dd3bcf27494c 100644
--- a/llvm/lib/ExecutionEngine/JITLink/MachOLinkGraphBuilder.h
+++ b/llvm/lib/ExecutionEngine/JITLink/MachOLinkGraphBuilder.h
@@ -13,6 +13,8 @@
 #ifndef LIB_EXECUTIONENGINE_JITLINK_MACHOLINKGRAPHBUILDER_H
 #define LIB_EXECUTIONENGINE_JITLINK_MACHOLINKGRAPHBUILDER_H
 
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/StringMap.h"
 #include "llvm/ExecutionEngine/JITLink/JITLink.h"
 
 #include "EHFrameSupportImpl.h"
@@ -58,6 +60,8 @@ protected:
     Symbol *GraphSymbol = nullptr;
   };
 
+  // Normalized section representation. Section and segment names are guaranteed
+  // to be null-terminated, hence the extra bytes on SegName and SectName.
   class NormalizedSection {
     friend class MachOLinkGraphBuilder;
 
@@ -65,12 +69,14 @@ protected:
     NormalizedSection() = default;
 
   public:
-    Section *GraphSection = nullptr;
+    char SectName[17];
+    char SegName[17];
     uint64_t Address = 0;
     uint64_t Size = 0;
     uint64_t Alignment = 0;
     uint32_t Flags = 0;
     const char *Data = nullptr;
+    Section *GraphSection = nullptr;
   };
 
   using SectionParserFunction = std::function<Error(NormalizedSection &S)>;
@@ -110,7 +116,7 @@ protected:
     auto I = IndexToSection.find(Index);
     if (I == IndexToSection.end())
       return make_error<JITLinkError>("No section recorded for index " +
-                                      formatv("{0:u}", Index));
+                                      formatv("{0:d}", Index));
     return I->second;
   }
 
@@ -123,7 +129,7 @@ protected:
     auto *Sym = IndexToSymbol[Index];
     if (!Sym)
       return make_error<JITLinkError>("No symbol at index " +
-                                      formatv("{0:u}", Index));
+                                      formatv("{0:d}", Index));
     return *Sym;
   }
 
@@ -151,6 +157,22 @@ protected:
   static Scope getScope(StringRef Name, uint8_t Type);
   static bool isAltEntry(const NormalizedSymbol &NSym);
 
+  static bool isDebugSection(const NormalizedSection &NSec);
+
+  MachO::relocation_info
+  getRelocationInfo(const object::relocation_iterator RelItr) {
+    MachO::any_relocation_info ARI =
+        getObject().getRelocation(RelItr->getRawDataRefImpl());
+    MachO::relocation_info RI;
+    RI.r_address = ARI.r_word0;
+    RI.r_symbolnum = ARI.r_word1 & 0xffffff;
+    RI.r_pcrel = (ARI.r_word1 >> 24) & 1;
+    RI.r_length = (ARI.r_word1 >> 25) & 3;
+    RI.r_extern = (ARI.r_word1 >> 27) & 1;
+    RI.r_type = (ARI.r_word1 >> 28);
+    return RI;
+  }
+
 private:
   static unsigned getPointerSize(const object::MachOObjectFile &Obj);
   static support::endianness getEndianness(const object::MachOObjectFile &Obj);
diff --git a/llvm/lib/ExecutionEngine/JITLink/MachO_arm64.cpp b/llvm/lib/ExecutionEngine/JITLink/MachO_arm64.cpp
index 944767449ce2..463845a5b8cb 100644
--- a/llvm/lib/ExecutionEngine/JITLink/MachO_arm64.cpp
+++ b/llvm/lib/ExecutionEngine/JITLink/MachO_arm64.cpp
@@ -92,15 +92,6 @@ private:
         ", length=" + formatv("{0:d}", RI.r_length));
   }
 
-  MachO::relocation_info
-  getRelocationInfo(const object::relocation_iterator RelItr) {
-    MachO::any_relocation_info ARI =
-        getObject().getRelocation(RelItr->getRawDataRefImpl());
-    MachO::relocation_info RI;
-    memcpy(&RI, &ARI, sizeof(MachO::relocation_info));
-    return RI;
-  }
-
   using PairRelocInfo =
       std::tuple<MachOARM64RelocationKind, Symbol *, uint64_t>;
 
@@ -194,6 +185,28 @@ private:
 
       JITTargetAddress SectionAddress = S.getAddress();
 
+      // Skip relocations virtual sections.
+      if (S.isVirtual()) {
+        if (S.relocation_begin() != S.relocation_end())
+          return make_error<JITLinkError>("Virtual section contains "
+                                          "relocations");
+        continue;
+      }
+
+      // Skip relocations for debug symbols.
+      {
+        auto &NSec =
+            getSectionByIndex(Obj.getSectionIndex(S.getRawDataRefImpl()));
+        if (!NSec.GraphSection) {
+          LLVM_DEBUG({
+            dbgs() << "Skipping relocations for MachO section " << NSec.SegName
+                   << "/" << NSec.SectName
+                   << " which has no associated graph section\n";
+          });
+          continue;
+        }
+      }
+
       for (auto RelItr = S.relocation_begin(), RelEnd = S.relocation_end();
            RelItr != RelEnd; ++RelItr) {
 
@@ -560,7 +573,8 @@ private:
       *(ulittle32_t *)FixupPtr = Value;
       break;
     }
-    case Pointer64: {
+    case Pointer64:
+    case Pointer64Anon: {
       uint64_t Value = E.getTarget().getAddress() + E.getAddend();
       *(ulittle64_t *)FixupPtr = Value;
       break;
diff --git a/llvm/lib/ExecutionEngine/JITLink/MachO_x86_64.cpp b/llvm/lib/ExecutionEngine/JITLink/MachO_x86_64.cpp
index 69ec72aae292..a91bc3b6033c 100644
--- a/llvm/lib/ExecutionEngine/JITLink/MachO_x86_64.cpp
+++ b/llvm/lib/ExecutionEngine/JITLink/MachO_x86_64.cpp
@@ -95,15 +95,6 @@ private:
         ", length=" + formatv("{0:d}", RI.r_length));
   }
 
-  MachO::relocation_info
-  getRelocationInfo(const object::relocation_iterator RelItr) {
-    MachO::any_relocation_info ARI =
-        getObject().getRelocation(RelItr->getRawDataRefImpl());
-    MachO::relocation_info RI;
-    memcpy(&RI, &ARI, sizeof(MachO::relocation_info));
-    return RI;
-  }
-
   using PairRelocInfo = std::tuple<MachOX86RelocationKind, Symbol *, uint64_t>;
 
   // Parses paired SUBTRACTOR/UNSIGNED relocations and, on success,
@@ -196,6 +187,7 @@ private:
 
       JITTargetAddress SectionAddress = S.getAddress();
 
+      // Skip relocations virtual sections.
       if (S.isVirtual()) {
         if (S.relocation_begin() != S.relocation_end())
           return make_error<JITLinkError>("Virtual section contains "
@@ -203,6 +195,21 @@ private:
         continue;
       }
 
+      // Skip relocations for debug symbols.
+      {
+        auto &NSec =
+            getSectionByIndex(Obj.getSectionIndex(S.getRawDataRefImpl()));
+        if (!NSec.GraphSection) {
+          LLVM_DEBUG({
+            dbgs() << "Skipping relocations for MachO section " << NSec.SegName
+                   << "/" << NSec.SectName
+                   << " which has no associated graph section\n";
+          });
+          continue;
+        }
+      }
+
+      // Add relocations for section.
       for (auto RelItr = S.relocation_begin(), RelEnd = S.relocation_end();
            RelItr != RelEnd; ++RelItr) {
 
@@ -350,6 +357,9 @@ private:
 class MachO_x86_64_GOTAndStubsBuilder
     : public BasicGOTAndStubsBuilder<MachO_x86_64_GOTAndStubsBuilder> {
 public:
+  static const uint8_t NullGOTEntryContent[8];
+  static const uint8_t StubContent[6];
+
   MachO_x86_64_GOTAndStubsBuilder(LinkGraph &G)
       : BasicGOTAndStubsBuilder<MachO_x86_64_GOTAndStubsBuilder>(G) {}
 
@@ -367,7 +377,13 @@ public:
   void fixGOTEdge(Edge &E, Symbol &GOTEntry) {
     assert((E.getKind() == PCRel32GOT || E.getKind() == PCRel32GOTLoad) &&
            "Not a GOT edge?");
-    E.setKind(PCRel32);
+    // If this is a PCRel32GOT then change it to an ordinary PCRel32. If it is
+    // a PCRel32GOTLoad then leave it as-is for now. We will use the kind to
+    // check for GOT optimization opportunities in the
+    // optimizeMachO_x86_64_GOTAndStubs pass below.
+    if (E.getKind() == PCRel32GOT)
+      E.setKind(PCRel32);
+
     E.setTarget(GOTEntry);
     // Leave the edge addend as-is.
   }
@@ -388,6 +404,11 @@ public:
   void fixExternalBranchEdge(Edge &E, Symbol &Stub) {
     assert(E.getKind() == Branch32 && "Not a Branch32 edge?");
     assert(E.getAddend() == 0 && "Branch32 edge has non-zero addend?");
+
+    // Set the edge kind to Branch32ToStub. We will use this to check for stub
+    // optimization opportunities in the optimizeMachO_x86_64_GOTAndStubs pass
+    // below.
+    E.setKind(Branch32ToStub);
     E.setTarget(Stub);
   }
 
@@ -417,8 +438,6 @@ private:
                      sizeof(StubContent));
   }
 
-  static const uint8_t NullGOTEntryContent[8];
-  static const uint8_t StubContent[6];
   Section *GOTSection = nullptr;
   Section *StubsSection = nullptr;
 };
@@ -429,6 +448,89 @@ const uint8_t MachO_x86_64_GOTAndStubsBuilder::StubContent[6] = {
     0xFF, 0x25, 0x00, 0x00, 0x00, 0x00};
 } // namespace
 
+static Error optimizeMachO_x86_64_GOTAndStubs(LinkGraph &G) {
+  LLVM_DEBUG(dbgs() << "Optimizing GOT entries and stubs:\n");
+
+  for (auto *B : G.blocks())
+    for (auto &E : B->edges())
+      if (E.getKind() == PCRel32GOTLoad) {
+        assert(E.getOffset() >= 3 && "GOT edge occurs too early in block");
+
+        // Switch the edge kind to PCRel32: Whether we change the edge target
+        // or not this will be the desired kind.
+        E.setKind(PCRel32);
+
+        // Optimize GOT references.
+        auto &GOTBlock = E.getTarget().getBlock();
+        assert(GOTBlock.getSize() == G.getPointerSize() &&
+               "GOT entry block should be pointer sized");
+        assert(GOTBlock.edges_size() == 1 &&
+               "GOT entry should only have one outgoing edge");
+
+        auto &GOTTarget = GOTBlock.edges().begin()->getTarget();
+        JITTargetAddress EdgeAddr = B->getAddress() + E.getOffset();
+        JITTargetAddress TargetAddr = GOTTarget.getAddress();
+
+        // Check that this is a recognized MOV instruction.
+        // FIXME: Can we assume this?
+        constexpr uint8_t MOVQRIPRel[] = {0x48, 0x8b};
+        if (strncmp(B->getContent().data() + E.getOffset() - 3,
+                    reinterpret_cast<const char *>(MOVQRIPRel), 2) != 0)
+          continue;
+
+        int64_t Displacement = TargetAddr - EdgeAddr + 4;
+        if (Displacement >= std::numeric_limits<int32_t>::min() &&
+            Displacement <= std::numeric_limits<int32_t>::max()) {
+          E.setTarget(GOTTarget);
+          auto *BlockData = reinterpret_cast<uint8_t *>(
+              const_cast<char *>(B->getContent().data()));
+          BlockData[E.getOffset() - 2] = 0x8d;
+          LLVM_DEBUG({
+            dbgs() << "  Replaced GOT load wih LEA:\n    ";
+            printEdge(dbgs(), *B, E,
+                      getMachOX86RelocationKindName(E.getKind()));
+            dbgs() << "\n";
+          });
+        }
+      } else if (E.getKind() == Branch32ToStub) {
+
+        // Switch the edge kind to PCRel32: Whether we change the edge target
+        // or not this will be the desired kind.
+        E.setKind(Branch32);
+
+        auto &StubBlock = E.getTarget().getBlock();
+        assert(StubBlock.getSize() ==
+                   sizeof(MachO_x86_64_GOTAndStubsBuilder::StubContent) &&
+               "Stub block should be stub sized");
+        assert(StubBlock.edges_size() == 1 &&
+               "Stub block should only have one outgoing edge");
+
+        auto &GOTBlock = StubBlock.edges().begin()->getTarget().getBlock();
+        assert(GOTBlock.getSize() == G.getPointerSize() &&
+               "GOT block should be pointer sized");
+        assert(GOTBlock.edges_size() == 1 &&
+               "GOT block should only have one outgoing edge");
+
+        auto &GOTTarget = GOTBlock.edges().begin()->getTarget();
+        JITTargetAddress EdgeAddr = B->getAddress() + E.getOffset();
+        JITTargetAddress TargetAddr = GOTTarget.getAddress();
+
+        int64_t Displacement = TargetAddr - EdgeAddr + 4;
+        if (Displacement >= std::numeric_limits<int32_t>::min() &&
+            Displacement <= std::numeric_limits<int32_t>::max()) {
+          E.setTarget(GOTTarget);
+          LLVM_DEBUG({
+            dbgs() << "  Replaced stub branch with direct branch:\n    ";
+            printEdge(dbgs(), *B, E,
+                      getMachOX86RelocationKindName(E.getKind()));
+            dbgs() << "\n";
+          });
+        }
+      }
+
+  return Error::success();
+}
+
 namespace llvm {
 namespace jitlink {
 
@@ -570,6 +672,9 @@ void jitLink_MachO_x86_64(std::unique_ptr<JITLinkContext> Ctx) {
       MachO_x86_64_GOTAndStubsBuilder(G).run();
       return Error::success();
     });
+
+    // Add GOT/Stubs optimizer pass.
+    Config.PostAllocationPasses.push_back(optimizeMachO_x86_64_GOTAndStubs);
   }
 
   if (auto Err = Ctx->modifyPassConfig(TT, Config))
@@ -583,6 +688,8 @@ StringRef getMachOX86RelocationKindName(Edge::Kind R) {
   switch (R) {
   case Branch32:
     return "Branch32";
+  case Branch32ToStub:
+    return "Branch32ToStub";
   case Pointer32:
     return "Pointer32";
   case Pointer64:
diff --git a/llvm/lib/ExecutionEngine/MCJIT/MCJIT.cpp b/llvm/lib/ExecutionEngine/MCJIT/MCJIT.cpp
index 94741f5f01d5..144329aa8bea 100644
--- a/llvm/lib/ExecutionEngine/MCJIT/MCJIT.cpp
+++ b/llvm/lib/ExecutionEngine/MCJIT/MCJIT.cpp
@@ -11,6 +11,7 @@
 #include "llvm/ExecutionEngine/GenericValue.h"
 #include "llvm/ExecutionEngine/JITEventListener.h"
 #include "llvm/ExecutionEngine/MCJIT.h"
+#include "llvm/ExecutionEngine/ObjectCache.h"
 #include "llvm/ExecutionEngine/SectionMemoryManager.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/DerivedTypes.h"
@@ -23,6 +24,7 @@
 #include "llvm/Support/DynamicLibrary.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/SmallVectorMemoryBuffer.h"
 #include <mutex>
 
 using namespace llvm;
@@ -239,6 +241,10 @@ void MCJIT::finalizeLoadedModules() {
   // Resolve any outstanding relocations.
   Dyld.resolveRelocations();
 
+  // Check for Dyld error.
+  if (Dyld.hasError())
+    ErrMsg = Dyld.getErrorString().str();
+
   OwnedModules.markAllLoadedModulesAsFinalized();
 
   // Register EH frame data for any module we own which has been loaded
@@ -609,7 +615,7 @@ GenericValue MCJIT::runFunction(Function *F, ArrayRef<GenericValue> ArgValues) {
 
 void *MCJIT::getPointerToNamedFunction(StringRef Name, bool AbortOnFailure) {
   if (!isSymbolSearchingDisabled()) {
-    if (auto Sym = Resolver.findSymbol(Name)) {
+    if (auto Sym = Resolver.findSymbol(std::string(Name))) {
       if (auto AddrOrErr = Sym.getAddress())
         return reinterpret_cast<void*>(
                  static_cast<uintptr_t>(*AddrOrErr));
@@ -619,7 +625,7 @@ void *MCJIT::getPointerToNamedFunction(StringRef Name, bool AbortOnFailure) {
 
   /// If a LazyFunctionCreator is installed, use it to get/create the function.
   if (LazyFunctionCreator)
-    if (void *RP = LazyFunctionCreator(Name))
+    if (void *RP = LazyFunctionCreator(std::string(Name)))
       return RP;
 
   if (AbortOnFailure) {
diff --git a/llvm/lib/ExecutionEngine/MCJIT/MCJIT.h b/llvm/lib/ExecutionEngine/MCJIT/MCJIT.h
index 77097fc0d17e..83b64b5171c0 100644
--- a/llvm/lib/ExecutionEngine/MCJIT/MCJIT.h
+++ b/llvm/lib/ExecutionEngine/MCJIT/MCJIT.h
@@ -12,14 +12,13 @@
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ExecutionEngine/ExecutionEngine.h"
-#include "llvm/ExecutionEngine/ObjectCache.h"
 #include "llvm/ExecutionEngine/RTDyldMemoryManager.h"
 #include "llvm/ExecutionEngine/RuntimeDyld.h"
-#include "llvm/IR/Module.h"
-#include "llvm/Support/SmallVectorMemoryBuffer.h"
 
 namespace llvm {
 class MCJIT;
+class Module;
+class ObjectCache;
 
 // This is a helper class that the MCJIT execution engine uses for linking
 // functions across modules that it owns.  It aggregates the memory manager
diff --git a/llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp b/llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp
index f26835ff8a08..9e38dc36faae 100644
--- a/llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/CompileOnDemandLayer.cpp
@@ -7,8 +7,12 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ExecutionEngine/Orc/CompileOnDemandLayer.h"
+
+#include "llvm/ADT/Hashing.h"
+#include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
 #include "llvm/IR/Mangler.h"
 #include "llvm/IR/Module.h"
+#include "llvm/Support/FormatVariadic.h"
 
 using namespace llvm;
 using namespace llvm::orc;
@@ -35,7 +39,7 @@ static ThreadSafeModule extractSubModule(ThreadSafeModule &TSM,
       Constant *Aliasee = A.getAliasee();
       assert(A.hasName() && "Anonymous alias?");
       assert(Aliasee->hasName() && "Anonymous aliasee");
-      std::string AliasName = A.getName();
+      std::string AliasName = std::string(A.getName());
 
       if (isa<Function>(Aliasee)) {
         auto *F = cloneFunctionDecl(*A.getParent(), *cast<Function>(Aliasee));
@@ -67,17 +71,19 @@ namespace orc {
 
 class PartitioningIRMaterializationUnit : public IRMaterializationUnit {
 public:
-  PartitioningIRMaterializationUnit(ExecutionSession &ES, ThreadSafeModule TSM,
-                                    VModuleKey K, CompileOnDemandLayer &Parent)
-      : IRMaterializationUnit(ES, std::move(TSM), std::move(K)),
+  PartitioningIRMaterializationUnit(ExecutionSession &ES,
+                                    const IRSymbolMapper::ManglingOptions &MO,
+                                    ThreadSafeModule TSM, VModuleKey K,
+                                    CompileOnDemandLayer &Parent)
+      : IRMaterializationUnit(ES, MO, std::move(TSM), std::move(K)),
         Parent(Parent) {}
 
   PartitioningIRMaterializationUnit(
-      ThreadSafeModule TSM, SymbolFlagsMap SymbolFlags,
-      SymbolNameToDefinitionMap SymbolToDefinition,
+      ThreadSafeModule TSM, VModuleKey K, SymbolFlagsMap SymbolFlags,
+      SymbolStringPtr InitSymbol, SymbolNameToDefinitionMap SymbolToDefinition,
       CompileOnDemandLayer &Parent)
       : IRMaterializationUnit(std::move(TSM), std::move(K),
-                              std::move(SymbolFlags),
+                              std::move(SymbolFlags), std::move(InitSymbol),
                               std::move(SymbolToDefinition)),
         Parent(Parent) {}
 
@@ -111,7 +117,8 @@ CompileOnDemandLayer::compileWholeModule(GlobalValueSet Requested) {
 CompileOnDemandLayer::CompileOnDemandLayer(
     ExecutionSession &ES, IRLayer &BaseLayer, LazyCallThroughManager &LCTMgr,
     IndirectStubsManagerBuilder BuildIndirectStubsManager)
-    : IRLayer(ES), BaseLayer(BaseLayer), LCTMgr(LCTMgr),
+    : IRLayer(ES, BaseLayer.getManglingOptions()), BaseLayer(BaseLayer),
+      LCTMgr(LCTMgr),
       BuildIndirectStubsManager(std::move(BuildIndirectStubsManager)) {}
 
 void CompileOnDemandLayer::setPartitionFunction(PartitionFunction Partition) {
@@ -136,36 +143,34 @@ void CompileOnDemandLayer::emit(MaterializationResponsibility R,
   TSM.withModuleDo([&](Module &M) {
     // First, do some cleanup on the module:
     cleanUpModule(M);
-
-    MangleAndInterner Mangle(ES, M.getDataLayout());
-    for (auto &GV : M.global_values()) {
-      if (GV.isDeclaration() || GV.hasLocalLinkage() ||
-          GV.hasAppendingLinkage())
-        continue;
-
-      auto Name = Mangle(GV.getName());
-      auto Flags = JITSymbolFlags::fromGlobalValue(GV);
-      if (Flags.isCallable())
-        Callables[Name] = SymbolAliasMapEntry(Name, Flags);
-      else
-        NonCallables[Name] = SymbolAliasMapEntry(Name, Flags);
-    }
   });
 
+  for (auto &KV : R.getSymbols()) {
+    auto &Name = KV.first;
+    auto &Flags = KV.second;
+    if (Flags.isCallable())
+      Callables[Name] = SymbolAliasMapEntry(Name, Flags);
+    else
+      NonCallables[Name] = SymbolAliasMapEntry(Name, Flags);
+  }
+
   // Create a partitioning materialization unit and lodge it with the
   // implementation dylib.
   if (auto Err = PDR.getImplDylib().define(
           std::make_unique<PartitioningIRMaterializationUnit>(
-              ES, std::move(TSM), R.getVModuleKey(), *this))) {
+              ES, *getManglingOptions(), std::move(TSM), R.getVModuleKey(),
+              *this))) {
     ES.reportError(std::move(Err));
     R.failMaterialization();
     return;
   }
 
-  R.replace(reexports(PDR.getImplDylib(), std::move(NonCallables),
-                      JITDylibLookupFlags::MatchAllSymbols));
-  R.replace(lazyReexports(LCTMgr, PDR.getISManager(), PDR.getImplDylib(),
-                          std::move(Callables), AliaseeImpls));
+  if (!NonCallables.empty())
+    R.replace(reexports(PDR.getImplDylib(), std::move(NonCallables),
+                        JITDylibLookupFlags::MatchAllSymbols));
+  if (!Callables.empty())
+    R.replace(lazyReexports(LCTMgr, PDR.getISManager(), PDR.getImplDylib(),
+                            std::move(Callables), AliaseeImpls));
 }
 
 CompileOnDemandLayer::PerDylibResources &
@@ -173,21 +178,22 @@ CompileOnDemandLayer::getPerDylibResources(JITDylib &TargetD) {
   auto I = DylibResources.find(&TargetD);
   if (I == DylibResources.end()) {
     auto &ImplD =
-        getExecutionSession().createJITDylib(TargetD.getName() + ".impl");
-    TargetD.withSearchOrderDo(
-        [&](const JITDylibSearchOrder &TargetSearchOrder) {
-          auto NewSearchOrder = TargetSearchOrder;
-          assert(
-              !NewSearchOrder.empty() &&
-              NewSearchOrder.front().first == &TargetD &&
-              NewSearchOrder.front().second ==
-                  JITDylibLookupFlags::MatchAllSymbols &&
-              "TargetD must be at the front of its own search order and match "
-              "non-exported symbol");
-          NewSearchOrder.insert(std::next(NewSearchOrder.begin()),
-                                {&ImplD, JITDylibLookupFlags::MatchAllSymbols});
-          ImplD.setSearchOrder(std::move(NewSearchOrder), false);
-        });
+        getExecutionSession().createBareJITDylib(TargetD.getName() + ".impl");
+    JITDylibSearchOrder NewLinkOrder;
+    TargetD.withLinkOrderDo([&](const JITDylibSearchOrder &TargetLinkOrder) {
+      NewLinkOrder = TargetLinkOrder;
+    });
+
+    assert(!NewLinkOrder.empty() && NewLinkOrder.front().first == &TargetD &&
+           NewLinkOrder.front().second ==
+               JITDylibLookupFlags::MatchAllSymbols &&
+           "TargetD must be at the front of its own search order and match "
+           "non-exported symbol");
+    NewLinkOrder.insert(std::next(NewLinkOrder.begin()),
+                        {&ImplD, JITDylibLookupFlags::MatchAllSymbols});
+    ImplD.setLinkOrder(NewLinkOrder, false);
+    TargetD.setLinkOrder(std::move(NewLinkOrder), false);
+
     PerDylibResources PDR(ImplD, BuildIndirectStubsManager());
     I = DylibResources.insert(std::make_pair(&TargetD, std::move(PDR))).first;
   }
@@ -252,8 +258,15 @@ void CompileOnDemandLayer::emitPartition(
   auto &ES = getExecutionSession();
   GlobalValueSet RequestedGVs;
   for (auto &Name : R.getRequestedSymbols()) {
-    assert(Defs.count(Name) && "No definition for symbol");
-    RequestedGVs.insert(Defs[Name]);
+    if (Name == R.getInitializerSymbol())
+      TSM.withModuleDo([&](Module &M) {
+        for (auto &GV : getStaticInitGVs(M))
+          RequestedGVs.insert(&GV);
+      });
+    else {
+      assert(Defs.count(Name) && "No definition for symbol");
+      RequestedGVs.insert(Defs[Name]);
+    }
   }
 
   /// Perform partitioning with the context lock held, since the partition
@@ -273,7 +286,8 @@ void CompileOnDemandLayer::emitPartition(
   // If the partition is empty, return the whole module to the symbol table.
   if (GVsToExtract->empty()) {
     R.replace(std::make_unique<PartitioningIRMaterializationUnit>(
-        std::move(TSM), R.getSymbols(), std::move(Defs), *this));
+        std::move(TSM), R.getVModuleKey(), R.getSymbols(),
+        R.getInitializerSymbol(), std::move(Defs), *this));
     return;
   }
 
@@ -284,29 +298,52 @@ void CompileOnDemandLayer::emitPartition(
   //
   // FIXME: We apply this promotion once per partitioning. It's safe, but
   // overkill.
-
   auto ExtractedTSM =
       TSM.withModuleDo([&](Module &M) -> Expected<ThreadSafeModule> {
         auto PromotedGlobals = PromoteSymbols(M);
         if (!PromotedGlobals.empty()) {
+
           MangleAndInterner Mangle(ES, M.getDataLayout());
           SymbolFlagsMap SymbolFlags;
-          for (auto &GV : PromotedGlobals)
-            SymbolFlags[Mangle(GV->getName())] =
-                JITSymbolFlags::fromGlobalValue(*GV);
+          IRSymbolMapper::add(ES, *getManglingOptions(),
+                              PromotedGlobals, SymbolFlags);
+
           if (auto Err = R.defineMaterializing(SymbolFlags))
             return std::move(Err);
         }
 
         expandPartition(*GVsToExtract);
 
+        // Submodule name is given by hashing the names of the globals.
+        std::string SubModuleName;
+        {
+          std::vector<const GlobalValue*> HashGVs;
+          HashGVs.reserve(GVsToExtract->size());
+          for (auto *GV : *GVsToExtract)
+            HashGVs.push_back(GV);
+          llvm::sort(HashGVs, [](const GlobalValue *LHS, const GlobalValue *RHS) {
+              return LHS->getName() < RHS->getName();
+            });
+          hash_code HC(0);
+          for (auto *GV : HashGVs) {
+            assert(GV->hasName() && "All GVs to extract should be named by now");
+            auto GVName = GV->getName();
+            HC = hash_combine(HC, hash_combine_range(GVName.begin(), GVName.end()));
+          }
+          raw_string_ostream(SubModuleName)
+            << ".submodule."
+            << formatv(sizeof(size_t) == 8 ? "{0:x16}" : "{0:x8}",
+                       static_cast<size_t>(HC))
+            << ".ll";
+        }
+
         // Extract the requested partiton (plus any necessary aliases) and
         // put the rest back into the impl dylib.
         auto ShouldExtract = [&](const GlobalValue &GV) -> bool {
           return GVsToExtract->count(&GV);
         };
 
-        return extractSubModule(TSM, ".submodule", ShouldExtract);
+        return extractSubModule(TSM, SubModuleName , ShouldExtract);
       });
 
   if (!ExtractedTSM) {
@@ -316,7 +353,7 @@ void CompileOnDemandLayer::emitPartition(
   }
 
   R.replace(std::make_unique<PartitioningIRMaterializationUnit>(
-      ES, std::move(TSM), R.getVModuleKey(), *this));
+      ES, *getManglingOptions(), std::move(TSM), R.getVModuleKey(), *this));
   BaseLayer.emit(std::move(R), std::move(*ExtractedTSM));
 }
 
diff --git a/llvm/lib/ExecutionEngine/Orc/CompileUtils.cpp b/llvm/lib/ExecutionEngine/Orc/CompileUtils.cpp
index f5671d90420a..f8efed15edea 100644
--- a/llvm/lib/ExecutionEngine/Orc/CompileUtils.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/CompileUtils.cpp
@@ -24,11 +24,20 @@
 namespace llvm {
 namespace orc {
 
+IRSymbolMapper::ManglingOptions
+irManglingOptionsFromTargetOptions(const TargetOptions &Opts) {
+  IRSymbolMapper::ManglingOptions MO;
+
+  MO.EmulatedTLS = Opts.EmulatedTLS;
+
+  return MO;
+}
+
 /// Compile a Module to an ObjectFile.
-SimpleCompiler::CompileResult SimpleCompiler::operator()(Module &M) {
+Expected<SimpleCompiler::CompileResult> SimpleCompiler::operator()(Module &M) {
   CompileResult CachedObject = tryToLoadFromObjectCache(M);
   if (CachedObject)
-    return CachedObject;
+    return std::move(CachedObject);
 
   SmallVector<char, 0> ObjBufferSV;
 
@@ -38,7 +47,8 @@ SimpleCompiler::CompileResult SimpleCompiler::operator()(Module &M) {
     legacy::PassManager PM;
     MCContext *Ctx;
     if (TM.addPassesToEmitMC(PM, Ctx, ObjStream))
-      llvm_unreachable("Target does not support MC emission.");
+      return make_error<StringError>("Target does not support MC emission",
+                                     inconvertibleErrorCode());
     PM.run(M);
   }
 
@@ -47,14 +57,11 @@ SimpleCompiler::CompileResult SimpleCompiler::operator()(Module &M) {
 
   auto Obj = object::ObjectFile::createObjectFile(ObjBuffer->getMemBufferRef());
 
-  if (Obj) {
-    notifyObjectCompiled(M, *ObjBuffer);
-    return std::move(ObjBuffer);
-  }
+  if (!Obj)
+    return Obj.takeError();
 
-  // TODO: Actually report errors helpfully.
-  consumeError(Obj.takeError());
-  return nullptr;
+  notifyObjectCompiled(M, *ObjBuffer);
+  return std::move(ObjBuffer);
 }
 
 SimpleCompiler::CompileResult
@@ -73,9 +80,11 @@ void SimpleCompiler::notifyObjectCompiled(const Module &M,
 
 ConcurrentIRCompiler::ConcurrentIRCompiler(JITTargetMachineBuilder JTMB,
                                            ObjectCache *ObjCache)
-    : JTMB(std::move(JTMB)), ObjCache(ObjCache) {}
+    : IRCompiler(irManglingOptionsFromTargetOptions(JTMB.getOptions())),
+      JTMB(std::move(JTMB)), ObjCache(ObjCache) {}
 
-std::unique_ptr<MemoryBuffer> ConcurrentIRCompiler::operator()(Module &M) {
+Expected<std::unique_ptr<MemoryBuffer>>
+ConcurrentIRCompiler::operator()(Module &M) {
   auto TM = cantFail(JTMB.createTargetMachine());
   SimpleCompiler C(*TM, ObjCache);
   return C(M);
diff --git a/llvm/lib/ExecutionEngine/Orc/Core.cpp b/llvm/lib/ExecutionEngine/Orc/Core.cpp
index 63ef889dae46..bad13cfebbc6 100644
--- a/llvm/lib/ExecutionEngine/Orc/Core.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/Core.cpp
@@ -10,302 +10,30 @@
 
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/Config/llvm-config.h"
+#include "llvm/ExecutionEngine/Orc/DebugUtils.h"
 #include "llvm/ExecutionEngine/Orc/OrcError.h"
-#include "llvm/IR/Mangler.h"
-#include "llvm/Support/CommandLine.h"
-#include "llvm/Support/Debug.h"
-#include "llvm/Support/Format.h"
 
+#include <condition_variable>
 #if LLVM_ENABLE_THREADS
 #include <future>
 #endif
 
 #define DEBUG_TYPE "orc"
 
-using namespace llvm;
-
-namespace {
-
-#ifndef NDEBUG
-
-cl::opt<bool> PrintHidden("debug-orc-print-hidden", cl::init(true),
-                          cl::desc("debug print hidden symbols defined by "
-                                   "materialization units"),
-                          cl::Hidden);
-
-cl::opt<bool> PrintCallable("debug-orc-print-callable", cl::init(true),
-                            cl::desc("debug print callable symbols defined by "
-                                     "materialization units"),
-                            cl::Hidden);
-
-cl::opt<bool> PrintData("debug-orc-print-data", cl::init(true),
-                        cl::desc("debug print data symbols defined by "
-                                 "materialization units"),
-                        cl::Hidden);
-
-#endif // NDEBUG
-
-// SetPrinter predicate that prints every element.
-template <typename T> struct PrintAll {
-  bool operator()(const T &E) { return true; }
-};
-
-bool anyPrintSymbolOptionSet() {
-#ifndef NDEBUG
-  return PrintHidden || PrintCallable || PrintData;
-#else
-  return false;
-#endif // NDEBUG
-}
-
-bool flagsMatchCLOpts(const JITSymbolFlags &Flags) {
-#ifndef NDEBUG
-  // Bail out early if this is a hidden symbol and we're not printing hiddens.
-  if (!PrintHidden && !Flags.isExported())
-    return false;
-
-  // Return true if this is callable and we're printing callables.
-  if (PrintCallable && Flags.isCallable())
-    return true;
-
-  // Return true if this is data and we're printing data.
-  if (PrintData && !Flags.isCallable())
-    return true;
-
-  // otherwise return false.
-  return false;
-#else
-  return false;
-#endif // NDEBUG
-}
-
-// Prints a sequence of items, filtered by an user-supplied predicate.
-template <typename Sequence,
-          typename Pred = PrintAll<typename Sequence::value_type>>
-class SequencePrinter {
-public:
-  SequencePrinter(const Sequence &S, char OpenSeq, char CloseSeq,
-                  Pred ShouldPrint = Pred())
-      : S(S), OpenSeq(OpenSeq), CloseSeq(CloseSeq),
-        ShouldPrint(std::move(ShouldPrint)) {}
-
-  void printTo(llvm::raw_ostream &OS) const {
-    bool PrintComma = false;
-    OS << OpenSeq;
-    for (auto &E : S) {
-      if (ShouldPrint(E)) {
-        if (PrintComma)
-          OS << ',';
-        OS << ' ' << E;
-        PrintComma = true;
-      }
-    }
-    OS << ' ' << CloseSeq;
-  }
-
-private:
-  const Sequence &S;
-  char OpenSeq;
-  char CloseSeq;
-  mutable Pred ShouldPrint;
-};
-
-template <typename Sequence, typename Pred>
-SequencePrinter<Sequence, Pred> printSequence(const Sequence &S, char OpenSeq,
-                                              char CloseSeq, Pred P = Pred()) {
-  return SequencePrinter<Sequence, Pred>(S, OpenSeq, CloseSeq, std::move(P));
-}
-
-// Render a SequencePrinter by delegating to its printTo method.
-template <typename Sequence, typename Pred>
-llvm::raw_ostream &operator<<(llvm::raw_ostream &OS,
-                              const SequencePrinter<Sequence, Pred> &Printer) {
-  Printer.printTo(OS);
-  return OS;
-}
-
-struct PrintSymbolFlagsMapElemsMatchingCLOpts {
-  bool operator()(const orc::SymbolFlagsMap::value_type &KV) {
-    return flagsMatchCLOpts(KV.second);
-  }
-};
-
-struct PrintSymbolMapElemsMatchingCLOpts {
-  bool operator()(const orc::SymbolMap::value_type &KV) {
-    return flagsMatchCLOpts(KV.second.getFlags());
-  }
-};
-
-} // end anonymous namespace
-
 namespace llvm {
 namespace orc {
 
 char FailedToMaterialize::ID = 0;
 char SymbolsNotFound::ID = 0;
 char SymbolsCouldNotBeRemoved::ID = 0;
+char MissingSymbolDefinitions::ID = 0;
+char UnexpectedSymbolDefinitions::ID = 0;
 
 RegisterDependenciesFunction NoDependenciesToRegister =
     RegisterDependenciesFunction();
 
 void MaterializationUnit::anchor() {}
 
-raw_ostream &operator<<(raw_ostream &OS, const SymbolStringPtr &Sym) {
-  return OS << *Sym;
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolNameSet &Symbols) {
-  return OS << printSequence(Symbols, '{', '}', PrintAll<SymbolStringPtr>());
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolNameVector &Symbols) {
-  return OS << printSequence(Symbols, '[', ']', PrintAll<SymbolStringPtr>());
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const JITSymbolFlags &Flags) {
-  if (Flags.hasError())
-    OS << "[*ERROR*]";
-  if (Flags.isCallable())
-    OS << "[Callable]";
-  else
-    OS << "[Data]";
-  if (Flags.isWeak())
-    OS << "[Weak]";
-  else if (Flags.isCommon())
-    OS << "[Common]";
-
-  if (!Flags.isExported())
-    OS << "[Hidden]";
-
-  return OS;
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const JITEvaluatedSymbol &Sym) {
-  return OS << format("0x%016" PRIx64, Sym.getAddress()) << " "
-            << Sym.getFlags();
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolFlagsMap::value_type &KV) {
-  return OS << "(\"" << KV.first << "\", " << KV.second << ")";
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolMap::value_type &KV) {
-  return OS << "(\"" << KV.first << "\": " << KV.second << ")";
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolFlagsMap &SymbolFlags) {
-  return OS << printSequence(SymbolFlags, '{', '}',
-                             PrintSymbolFlagsMapElemsMatchingCLOpts());
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolMap &Symbols) {
-  return OS << printSequence(Symbols, '{', '}',
-                             PrintSymbolMapElemsMatchingCLOpts());
-}
-
-raw_ostream &operator<<(raw_ostream &OS,
-                        const SymbolDependenceMap::value_type &KV) {
-  return OS << "(" << KV.first << ", " << KV.second << ")";
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolDependenceMap &Deps) {
-  return OS << printSequence(Deps, '{', '}',
-                             PrintAll<SymbolDependenceMap::value_type>());
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const MaterializationUnit &MU) {
-  OS << "MU@" << &MU << " (\"" << MU.getName() << "\"";
-  if (anyPrintSymbolOptionSet())
-    OS << ", " << MU.getSymbols();
-  return OS << ")";
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const LookupKind &K) {
-  switch (K) {
-  case LookupKind::Static:
-    return OS << "Static";
-  case LookupKind::DLSym:
-    return OS << "DLSym";
-  }
-  llvm_unreachable("Invalid lookup kind");
-}
-
-raw_ostream &operator<<(raw_ostream &OS,
-                        const JITDylibLookupFlags &JDLookupFlags) {
-  switch (JDLookupFlags) {
-  case JITDylibLookupFlags::MatchExportedSymbolsOnly:
-    return OS << "MatchExportedSymbolsOnly";
-  case JITDylibLookupFlags::MatchAllSymbols:
-    return OS << "MatchAllSymbols";
-  }
-  llvm_unreachable("Invalid JITDylib lookup flags");
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolLookupFlags &LookupFlags) {
-  switch (LookupFlags) {
-  case SymbolLookupFlags::RequiredSymbol:
-    return OS << "RequiredSymbol";
-  case SymbolLookupFlags::WeaklyReferencedSymbol:
-    return OS << "WeaklyReferencedSymbol";
-  }
-  llvm_unreachable("Invalid symbol lookup flags");
-}
-
-raw_ostream &operator<<(raw_ostream &OS,
-                        const SymbolLookupSet::value_type &KV) {
-  return OS << "(" << KV.first << ", " << KV.second << ")";
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolLookupSet &LookupSet) {
-  return OS << printSequence(LookupSet, '{', '}',
-                             PrintAll<SymbolLookupSet::value_type>());
-}
-
-raw_ostream &operator<<(raw_ostream &OS,
-                        const JITDylibSearchOrder &SearchOrder) {
-  OS << "[";
-  if (!SearchOrder.empty()) {
-    assert(SearchOrder.front().first &&
-           "JITDylibList entries must not be null");
-    OS << " (\"" << SearchOrder.front().first->getName() << "\", "
-       << SearchOrder.begin()->second << ")";
-    for (auto &KV :
-         make_range(std::next(SearchOrder.begin(), 1), SearchOrder.end())) {
-      assert(KV.first && "JITDylibList entries must not be null");
-      OS << ", (\"" << KV.first->getName() << "\", " << KV.second << ")";
-    }
-  }
-  OS << " ]";
-  return OS;
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolAliasMap &Aliases) {
-  OS << "{";
-  for (auto &KV : Aliases)
-    OS << " " << *KV.first << ": " << KV.second.Aliasee << " "
-       << KV.second.AliasFlags;
-  OS << " }";
-  return OS;
-}
-
-raw_ostream &operator<<(raw_ostream &OS, const SymbolState &S) {
-  switch (S) {
-  case SymbolState::Invalid:
-    return OS << "Invalid";
-  case SymbolState::NeverSearched:
-    return OS << "Never-Searched";
-  case SymbolState::Materializing:
-    return OS << "Materializing";
-  case SymbolState::Resolved:
-    return OS << "Resolved";
-  case SymbolState::Emitted:
-    return OS << "Emitted";
-  case SymbolState::Ready:
-    return OS << "Ready";
-  }
-  llvm_unreachable("Invalid state");
-}
-
 FailedToMaterialize::FailedToMaterialize(
     std::shared_ptr<SymbolDependenceMap> Symbols)
     : Symbols(std::move(Symbols)) {
@@ -352,6 +80,24 @@ void SymbolsCouldNotBeRemoved::log(raw_ostream &OS) const {
   OS << "Symbols could not be removed: " << Symbols;
 }
 
+std::error_code MissingSymbolDefinitions::convertToErrorCode() const {
+  return orcError(OrcErrorCode::MissingSymbolDefinitions);
+}
+
+void MissingSymbolDefinitions::log(raw_ostream &OS) const {
+  OS << "Missing definitions in module " << ModuleName
+     << ": " << Symbols;
+}
+
+std::error_code UnexpectedSymbolDefinitions::convertToErrorCode() const {
+  return orcError(OrcErrorCode::UnexpectedSymbolDefinitions);
+}
+
+void UnexpectedSymbolDefinitions::log(raw_ostream &OS) const {
+  OS << "Unexpected definitions in module " << ModuleName
+     << ": " << Symbols;
+}
+
 AsynchronousSymbolQuery::AsynchronousSymbolQuery(
     const SymbolLookupSet &Symbols, SymbolState RequiredState,
     SymbolsResolvedCallback NotifyComplete)
@@ -372,7 +118,13 @@ void AsynchronousSymbolQuery::notifySymbolMetRequiredState(
   assert(I != ResolvedSymbols.end() &&
          "Resolving symbol outside the requested set");
   assert(I->second.getAddress() == 0 && "Redundantly resolving symbol Name");
-  I->second = std::move(Sym);
+
+  // If this is a materialization-side-effects-only symbol then drop it,
+  // otherwise update its map entry with its resolved address.
+  if (Sym.getFlags().hasMaterializationSideEffectsOnly())
+    ResolvedSymbols.erase(I);
+  else
+    I->second = std::move(Sym);
   --OutstandingSymbolsCount;
 }
 
@@ -413,6 +165,14 @@ void AsynchronousSymbolQuery::removeQueryDependence(
     QueryRegistrations.erase(QRI);
 }
 
+void AsynchronousSymbolQuery::dropSymbol(const SymbolStringPtr &Name) {
+  auto I = ResolvedSymbols.find(Name);
+  assert(I != ResolvedSymbols.end() &&
+         "Redundant removal of weakly-referenced symbol");
+  ResolvedSymbols.erase(I);
+  --OutstandingSymbolsCount;
+}
+
 void AsynchronousSymbolQuery::detach() {
   ResolvedSymbols.clear();
   OutstandingSymbolsCount = 0;
@@ -421,24 +181,18 @@ void AsynchronousSymbolQuery::detach() {
   QueryRegistrations.clear();
 }
 
-MaterializationResponsibility::MaterializationResponsibility(
-    JITDylib &JD, SymbolFlagsMap SymbolFlags, VModuleKey K)
-    : JD(JD), SymbolFlags(std::move(SymbolFlags)), K(std::move(K)) {
-  assert(!this->SymbolFlags.empty() && "Materializing nothing?");
-}
-
 MaterializationResponsibility::~MaterializationResponsibility() {
   assert(SymbolFlags.empty() &&
          "All symbols should have been explicitly materialized or failed");
 }
 
 SymbolNameSet MaterializationResponsibility::getRequestedSymbols() const {
-  return JD.getRequestedSymbols(SymbolFlags);
+  return JD->getRequestedSymbols(SymbolFlags);
 }
 
 Error MaterializationResponsibility::notifyResolved(const SymbolMap &Symbols) {
   LLVM_DEBUG({
-    dbgs() << "In " << JD.getName() << " resolving " << Symbols << "\n";
+    dbgs() << "In " << JD->getName() << " resolving " << Symbols << "\n";
   });
 #ifndef NDEBUG
   for (auto &KV : Symbols) {
@@ -446,21 +200,23 @@ Error MaterializationResponsibility::notifyResolved(const SymbolMap &Symbols) {
     auto I = SymbolFlags.find(KV.first);
     assert(I != SymbolFlags.end() &&
            "Resolving symbol outside this responsibility set");
+    assert(!I->second.hasMaterializationSideEffectsOnly() &&
+           "Can't resolve materialization-side-effects-only symbol");
     assert((KV.second.getFlags() & ~WeakFlags) == (I->second & ~WeakFlags) &&
            "Resolving symbol with incorrect flags");
   }
 #endif
 
-  return JD.resolve(Symbols);
+  return JD->resolve(Symbols);
 }
 
 Error MaterializationResponsibility::notifyEmitted() {
 
   LLVM_DEBUG({
-    dbgs() << "In " << JD.getName() << " emitting " << SymbolFlags << "\n";
+    dbgs() << "In " << JD->getName() << " emitting " << SymbolFlags << "\n";
   });
 
-  if (auto Err = JD.emit(SymbolFlags))
+  if (auto Err = JD->emit(SymbolFlags))
     return Err;
 
   SymbolFlags.clear();
@@ -468,44 +224,59 @@ Error MaterializationResponsibility::notifyEmitted() {
 }
 
 Error MaterializationResponsibility::defineMaterializing(
-    const SymbolFlagsMap &NewSymbolFlags) {
-  // Add the given symbols to this responsibility object.
-  // It's ok if we hit a duplicate here: In that case the new version will be
-  // discarded, and the JITDylib::defineMaterializing method will return a
-  // duplicate symbol error.
-  for (auto &KV : NewSymbolFlags)
-    SymbolFlags.insert(KV);
+    SymbolFlagsMap NewSymbolFlags) {
 
-  return JD.defineMaterializing(NewSymbolFlags);
+  LLVM_DEBUG({
+    dbgs() << "In " << JD->getName() << " defining materializing symbols "
+           << NewSymbolFlags << "\n";
+  });
+  if (auto AcceptedDefs = JD->defineMaterializing(std::move(NewSymbolFlags))) {
+    // Add all newly accepted symbols to this responsibility object.
+    for (auto &KV : *AcceptedDefs)
+      SymbolFlags.insert(KV);
+    return Error::success();
+  } else
+    return AcceptedDefs.takeError();
 }
 
 void MaterializationResponsibility::failMaterialization() {
 
   LLVM_DEBUG({
-    dbgs() << "In " << JD.getName() << " failing materialization for "
+    dbgs() << "In " << JD->getName() << " failing materialization for "
            << SymbolFlags << "\n";
   });
 
   JITDylib::FailedSymbolsWorklist Worklist;
 
   for (auto &KV : SymbolFlags)
-    Worklist.push_back(std::make_pair(&JD, KV.first));
+    Worklist.push_back(std::make_pair(JD.get(), KV.first));
   SymbolFlags.clear();
 
-  JD.notifyFailed(std::move(Worklist));
+  JD->notifyFailed(std::move(Worklist));
 }
 
 void MaterializationResponsibility::replace(
     std::unique_ptr<MaterializationUnit> MU) {
-  for (auto &KV : MU->getSymbols())
+
+  // If the replacement MU is empty then return.
+  if (MU->getSymbols().empty())
+    return;
+
+  for (auto &KV : MU->getSymbols()) {
+    assert(SymbolFlags.count(KV.first) &&
+           "Replacing definition outside this responsibility set");
     SymbolFlags.erase(KV.first);
+  }
 
-  LLVM_DEBUG(JD.getExecutionSession().runSessionLocked([&]() {
-    dbgs() << "In " << JD.getName() << " replacing symbols with " << *MU
+  if (MU->getInitializerSymbol() == InitSymbol)
+    InitSymbol = nullptr;
+
+  LLVM_DEBUG(JD->getExecutionSession().runSessionLocked([&]() {
+    dbgs() << "In " << JD->getName() << " replacing symbols with " << *MU
            << "\n";
   }););
 
-  JD.replace(std::move(MU));
+  JD->replace(std::move(MU));
 }
 
 MaterializationResponsibility
@@ -515,6 +286,7 @@ MaterializationResponsibility::delegate(const SymbolNameSet &Symbols,
   if (NewKey == VModuleKey())
     NewKey = K;
 
+  SymbolStringPtr DelegatedInitSymbol;
   SymbolFlagsMap DelegatedFlags;
 
   for (auto &Name : Symbols) {
@@ -524,29 +296,41 @@ MaterializationResponsibility::delegate(const SymbolNameSet &Symbols,
            "instance");
 
     DelegatedFlags[Name] = std::move(I->second);
+    if (Name == InitSymbol)
+      std::swap(InitSymbol, DelegatedInitSymbol);
+
     SymbolFlags.erase(I);
   }
 
   return MaterializationResponsibility(JD, std::move(DelegatedFlags),
+                                       std::move(DelegatedInitSymbol),
                                        std::move(NewKey));
 }
 
 void MaterializationResponsibility::addDependencies(
     const SymbolStringPtr &Name, const SymbolDependenceMap &Dependencies) {
+  LLVM_DEBUG({
+    dbgs() << "Adding dependencies for " << Name << ": " << Dependencies
+           << "\n";
+  });
   assert(SymbolFlags.count(Name) &&
          "Symbol not covered by this MaterializationResponsibility instance");
-  JD.addDependencies(Name, Dependencies);
+  JD->addDependencies(Name, Dependencies);
 }
 
 void MaterializationResponsibility::addDependenciesForAll(
     const SymbolDependenceMap &Dependencies) {
+  LLVM_DEBUG({
+    dbgs() << "Adding dependencies for all symbols in " << SymbolFlags << ": "
+           << Dependencies << "\n";
+  });
   for (auto &KV : SymbolFlags)
-    JD.addDependencies(KV.first, Dependencies);
+    JD->addDependencies(KV.first, Dependencies);
 }
 
 AbsoluteSymbolsMaterializationUnit::AbsoluteSymbolsMaterializationUnit(
     SymbolMap Symbols, VModuleKey K)
-    : MaterializationUnit(extractFlags(Symbols), std::move(K)),
+    : MaterializationUnit(extractFlags(Symbols), nullptr, std::move(K)),
       Symbols(std::move(Symbols)) {}
 
 StringRef AbsoluteSymbolsMaterializationUnit::getName() const {
@@ -577,7 +361,7 @@ AbsoluteSymbolsMaterializationUnit::extractFlags(const SymbolMap &Symbols) {
 ReExportsMaterializationUnit::ReExportsMaterializationUnit(
     JITDylib *SourceJD, JITDylibLookupFlags SourceJDLookupFlags,
     SymbolAliasMap Aliases, VModuleKey K)
-    : MaterializationUnit(extractFlags(Aliases), std::move(K)),
+    : MaterializationUnit(extractFlags(Aliases), nullptr, std::move(K)),
       SourceJD(SourceJD), SourceJDLookupFlags(SourceJDLookupFlags),
       Aliases(std::move(Aliases)) {}
 
@@ -630,11 +414,13 @@ void ReExportsMaterializationUnit::materialize(
     SymbolAliasMap Aliases;
   };
 
-  // Build a list of queries to issue. In each round we build the largest set of
-  // aliases that we can resolve without encountering a chain definition of the
-  // form Foo -> Bar, Bar -> Baz. Such a form would deadlock as the query would
-  // be waitin on a symbol that it itself had to resolve. Usually this will just
-  // involve one round and a single query.
+  // Build a list of queries to issue. In each round we build a query for the
+  // largest set of aliases that we can resolve without encountering a chain of
+  // aliases (e.g. Foo -> Bar, Bar -> Baz). Such a chain would deadlock as the
+  // query would be waiting on a symbol that it itself had to resolve. Creating
+  // a new query for each link in such a chain eliminates the possibility of
+  // deadlock. In practice chains are likely to be rare, and this algorithm will
+  // usually result in a single query to issue.
 
   std::vector<std::pair<SymbolLookupSet, std::shared_ptr<OnResolveInfo>>>
       QueryInfos;
@@ -651,7 +437,10 @@ void ReExportsMaterializationUnit::materialize(
         continue;
 
       ResponsibilitySymbols.insert(KV.first);
-      QuerySymbols.add(KV.second.Aliasee);
+      QuerySymbols.add(KV.second.Aliasee,
+                       KV.second.AliasFlags.hasMaterializationSideEffectsOnly()
+                           ? SymbolLookupFlags::WeaklyReferencedSymbol
+                           : SymbolLookupFlags::RequiredSymbol);
       QueryAliases[KV.first] = std::move(KV.second);
     }
 
@@ -700,8 +489,13 @@ void ReExportsMaterializationUnit::materialize(
       if (Result) {
         SymbolMap ResolutionMap;
         for (auto &KV : QueryInfo->Aliases) {
-          assert(Result->count(KV.second.Aliasee) &&
+          assert((KV.second.AliasFlags.hasMaterializationSideEffectsOnly() ||
+                  Result->count(KV.second.Aliasee)) &&
                  "Result map missing entry?");
+          // Don't try to resolve materialization-side-effects-only symbols.
+          if (KV.second.AliasFlags.hasMaterializationSideEffectsOnly())
+            continue;
+
           ResolutionMap[KV.first] = JITEvaluatedSymbol(
               (*Result)[KV.second.Aliasee].getAddress(), KV.second.AliasFlags);
         }
@@ -809,31 +603,52 @@ void JITDylib::removeGenerator(DefinitionGenerator &G) {
   });
 }
 
-Error JITDylib::defineMaterializing(const SymbolFlagsMap &SymbolFlags) {
-  return ES.runSessionLocked([&]() -> Error {
+Expected<SymbolFlagsMap>
+JITDylib::defineMaterializing(SymbolFlagsMap SymbolFlags) {
+
+  return ES.runSessionLocked([&]() -> Expected<SymbolFlagsMap> {
     std::vector<SymbolTable::iterator> AddedSyms;
+    std::vector<SymbolFlagsMap::iterator> RejectedWeakDefs;
 
-    for (auto &KV : SymbolFlags) {
-      SymbolTable::iterator EntryItr;
-      bool Added;
+    for (auto SFItr = SymbolFlags.begin(), SFEnd = SymbolFlags.end();
+         SFItr != SFEnd; ++SFItr) {
 
-      std::tie(EntryItr, Added) =
-          Symbols.insert(std::make_pair(KV.first, SymbolTableEntry(KV.second)));
+      auto &Name = SFItr->first;
+      auto &Flags = SFItr->second;
 
-      if (Added) {
-        AddedSyms.push_back(EntryItr);
-        EntryItr->second.setState(SymbolState::Materializing);
-      } else {
-        // Remove any symbols already added.
-        for (auto &SI : AddedSyms)
-          Symbols.erase(SI);
+      auto EntryItr = Symbols.find(Name);
 
-        // FIXME: Return all duplicates.
-        return make_error<DuplicateDefinition>(*KV.first);
-      }
+      // If the entry already exists...
+      if (EntryItr != Symbols.end()) {
+
+        // If this is a strong definition then error out.
+        if (!Flags.isWeak()) {
+          // Remove any symbols already added.
+          for (auto &SI : AddedSyms)
+            Symbols.erase(SI);
+
+          // FIXME: Return all duplicates.
+          return make_error<DuplicateDefinition>(std::string(*Name));
+        }
+
+        // Otherwise just make a note to discard this symbol after the loop.
+        RejectedWeakDefs.push_back(SFItr);
+        continue;
+      } else
+        EntryItr =
+          Symbols.insert(std::make_pair(Name, SymbolTableEntry(Flags))).first;
+
+      AddedSyms.push_back(EntryItr);
+      EntryItr->second.setState(SymbolState::Materializing);
     }
 
-    return Error::success();
+    // Remove any rejected weak definitions from the SymbolFlags map.
+    while (!RejectedWeakDefs.empty()) {
+      SymbolFlags.erase(RejectedWeakDefs.back());
+      RejectedWeakDefs.pop_back();
+    }
+
+    return SymbolFlags;
   });
 }
 
@@ -847,8 +662,8 @@ void JITDylib::replace(std::unique_ptr<MaterializationUnit> MU) {
         for (auto &KV : MU->getSymbols()) {
           auto SymI = Symbols.find(KV.first);
           assert(SymI != Symbols.end() && "Replacing unknown symbol");
-          assert(SymI->second.isInMaterializationPhase() &&
-                 "Can not call replace on a symbol that is not materializing");
+          assert(SymI->second.getState() == SymbolState::Materializing &&
+                 "Can not replace a symbol that ha is not materializing");
           assert(!SymI->second.hasMaterializerAttached() &&
                  "Symbol should not have materializer attached already");
           assert(UnmaterializedInfos.count(KV.first) == 0 &&
@@ -878,14 +693,21 @@ void JITDylib::replace(std::unique_ptr<MaterializationUnit> MU) {
                  "Unexpected materializer entry in map");
           SymI->second.setAddress(SymI->second.getAddress());
           SymI->second.setMaterializerAttached(true);
-          UnmaterializedInfos[KV.first] = UMI;
+
+          auto &UMIEntry = UnmaterializedInfos[KV.first];
+          assert((!UMIEntry || !UMIEntry->MU) &&
+                 "Replacing symbol with materializer still attached");
+          UMIEntry = UMI;
         }
 
         return nullptr;
       });
 
-  if (MustRunMU)
-    ES.dispatchMaterialization(*this, std::move(MustRunMU));
+  if (MustRunMU) {
+    auto MR =
+        MustRunMU->createMaterializationResponsibility(shared_from_this());
+    ES.dispatchMaterialization(std::move(MustRunMU), std::move(MR));
+  }
 }
 
 SymbolNameSet
@@ -895,7 +717,9 @@ JITDylib::getRequestedSymbols(const SymbolFlagsMap &SymbolFlags) const {
 
     for (auto &KV : SymbolFlags) {
       assert(Symbols.count(KV.first) && "JITDylib does not cover this symbol?");
-      assert(Symbols.find(KV.first)->second.isInMaterializationPhase() &&
+      assert(Symbols.find(KV.first)->second.getState() !=
+                 SymbolState::NeverSearched &&
+             Symbols.find(KV.first)->second.getState() != SymbolState::Ready &&
              "getRequestedSymbols can only be called for symbols that have "
              "started materializing");
       auto I = MaterializingInfos.find(KV.first);
@@ -913,9 +737,14 @@ JITDylib::getRequestedSymbols(const SymbolFlagsMap &SymbolFlags) const {
 void JITDylib::addDependencies(const SymbolStringPtr &Name,
                                const SymbolDependenceMap &Dependencies) {
   assert(Symbols.count(Name) && "Name not in symbol table");
-  assert(Symbols[Name].isInMaterializationPhase() &&
+  assert(Symbols[Name].getState() < SymbolState::Emitted &&
          "Can not add dependencies for a symbol that is not materializing");
 
+  LLVM_DEBUG({
+      dbgs() << "In " << getName() << " adding dependencies for "
+             << *Name << ": " << Dependencies << "\n";
+    });
+
   // If Name is already in an error state then just bail out.
   if (Symbols[Name].getFlags().hasError())
     return;
@@ -938,16 +767,18 @@ void JITDylib::addDependencies(const SymbolStringPtr &Name,
       // Check the sym entry for the dependency.
       auto OtherSymI = OtherJITDylib.Symbols.find(OtherSymbol);
 
-#ifndef NDEBUG
       // Assert that this symbol exists and has not reached the ready state
       // already.
       assert(OtherSymI != OtherJITDylib.Symbols.end() &&
-             (OtherSymI->second.getState() != SymbolState::Ready &&
-              "Dependency on emitted/ready symbol"));
-#endif
+             "Dependency on unknown symbol");
 
       auto &OtherSymEntry = OtherSymI->second;
 
+      // If the other symbol is already in the Ready state then there's no
+      // dependency to add.
+      if (OtherSymEntry.getState() == SymbolState::Ready)
+        continue;
+
       // If the dependency is in an error state then note this and continue,
       // we will move this symbol to the error state below.
       if (OtherSymEntry.getFlags().hasError()) {
@@ -957,8 +788,6 @@ void JITDylib::addDependencies(const SymbolStringPtr &Name,
 
       // If the dependency was not in the error state then add it to
       // our list of dependencies.
-      assert(OtherJITDylib.MaterializingInfos.count(OtherSymbol) &&
-             "No MaterializingInfo for dependency");
       auto &OtherMI = OtherJITDylib.MaterializingInfos[OtherSymbol];
 
       if (OtherSymEntry.getState() == SymbolState::Emitted)
@@ -1039,7 +868,11 @@ Error JITDylib::resolve(const SymbolMap &Resolved) {
       SymI->second.setFlags(ResolvedFlags);
       SymI->second.setState(SymbolState::Resolved);
 
-      auto &MI = MaterializingInfos[Name];
+      auto MII = MaterializingInfos.find(Name);
+      if (MII == MaterializingInfos.end())
+        continue;
+
+      auto &MI = MII->second;
       for (auto &Q : MI.takeQueriesMeeting(SymbolState::Resolved)) {
         Q->notifySymbolMetRequiredState(Name, ResolvedSym);
         Q->removeQueryDependence(*this, Name);
@@ -1071,6 +904,7 @@ Error JITDylib::resolve(const SymbolMap &Resolved) {
 Error JITDylib::emit(const SymbolFlagsMap &Emitted) {
   AsynchronousSymbolQuerySet CompletedQueries;
   SymbolNameSet SymbolsInErrorState;
+  DenseMap<JITDylib *, SymbolNameVector> ReadySymbols;
 
   ES.runSessionLocked([&, this]() {
     std::vector<SymbolTable::iterator> Worklist;
@@ -1101,13 +935,21 @@ Error JITDylib::emit(const SymbolFlagsMap &Emitted) {
       auto &SymEntry = SymI->second;
 
       // Move symbol to the emitted state.
-      assert(SymEntry.getState() == SymbolState::Resolved &&
+      assert(((SymEntry.getFlags().hasMaterializationSideEffectsOnly() &&
+               SymEntry.getState() == SymbolState::Materializing) ||
+              SymEntry.getState() == SymbolState::Resolved) &&
              "Emitting from state other than Resolved");
       SymEntry.setState(SymbolState::Emitted);
 
       auto MII = MaterializingInfos.find(Name);
-      assert(MII != MaterializingInfos.end() &&
-             "Missing MaterializingInfo entry");
+
+      // If this symbol has no MaterializingInfo then it's trivially ready.
+      // Update its state and continue.
+      if (MII == MaterializingInfos.end()) {
+        SymEntry.setState(SymbolState::Ready);
+        continue;
+      }
+
       auto &MI = MII->second;
 
       // For each dependant, transfer this node's emitted dependencies to
@@ -1115,6 +957,7 @@ Error JITDylib::emit(const SymbolFlagsMap &Emitted) {
       // dependencies) then notify any pending queries.
       for (auto &KV : MI.Dependants) {
         auto &DependantJD = *KV.first;
+        auto &DependantJDReadySymbols = ReadySymbols[&DependantJD];
         for (auto &DependantName : KV.second) {
           auto DependantMII =
               DependantJD.MaterializingInfos.find(DependantName);
@@ -1154,6 +997,7 @@ Error JITDylib::emit(const SymbolFlagsMap &Emitted) {
             // Since this dependant is now ready, we erase its MaterializingInfo
             // and update its materializing state.
             DependantSymEntry.setState(SymbolState::Ready);
+            DependantJDReadySymbols.push_back(DependantName);
 
             for (auto &Q : DependantMI.takeQueriesMeeting(SymbolState::Ready)) {
               Q->notifySymbolMetRequiredState(
@@ -1162,22 +1006,21 @@ Error JITDylib::emit(const SymbolFlagsMap &Emitted) {
                 CompletedQueries.insert(Q);
               Q->removeQueryDependence(DependantJD, DependantName);
             }
-
-            DependantJD.MaterializingInfos.erase(DependantMII);
           }
         }
       }
 
+      auto &ThisJDReadySymbols = ReadySymbols[this];
       MI.Dependants.clear();
       if (MI.UnemittedDependencies.empty()) {
         SymI->second.setState(SymbolState::Ready);
+        ThisJDReadySymbols.push_back(Name);
         for (auto &Q : MI.takeQueriesMeeting(SymbolState::Ready)) {
           Q->notifySymbolMetRequiredState(Name, SymI->second.getSymbol());
           if (Q->isComplete())
             CompletedQueries.insert(Q);
           Q->removeQueryDependence(*this, Name);
         }
-        MaterializingInfos.erase(MII);
       }
     }
   });
@@ -1317,30 +1160,29 @@ void JITDylib::notifyFailed(FailedSymbolsWorklist Worklist) {
     Q->handleFailed(make_error<FailedToMaterialize>(FailedSymbolsMap));
 }
 
-void JITDylib::setSearchOrder(JITDylibSearchOrder NewSearchOrder,
-                              bool SearchThisJITDylibFirst) {
+void JITDylib::setLinkOrder(JITDylibSearchOrder NewLinkOrder,
+                            bool LinkAgainstThisJITDylibFirst) {
   ES.runSessionLocked([&]() {
-    if (SearchThisJITDylibFirst) {
-      SearchOrder.clear();
-      if (NewSearchOrder.empty() || NewSearchOrder.front().first != this)
-        SearchOrder.push_back(
+    if (LinkAgainstThisJITDylibFirst) {
+      LinkOrder.clear();
+      if (NewLinkOrder.empty() || NewLinkOrder.front().first != this)
+        LinkOrder.push_back(
             std::make_pair(this, JITDylibLookupFlags::MatchAllSymbols));
-      SearchOrder.insert(SearchOrder.end(), NewSearchOrder.begin(),
-                         NewSearchOrder.end());
+      LinkOrder.insert(LinkOrder.end(), NewLinkOrder.begin(),
+                       NewLinkOrder.end());
     } else
-      SearchOrder = std::move(NewSearchOrder);
+      LinkOrder = std::move(NewLinkOrder);
   });
 }
 
-void JITDylib::addToSearchOrder(JITDylib &JD,
-                                JITDylibLookupFlags JDLookupFlags) {
-  ES.runSessionLocked([&]() { SearchOrder.push_back({&JD, JDLookupFlags}); });
+void JITDylib::addToLinkOrder(JITDylib &JD, JITDylibLookupFlags JDLookupFlags) {
+  ES.runSessionLocked([&]() { LinkOrder.push_back({&JD, JDLookupFlags}); });
 }
 
-void JITDylib::replaceInSearchOrder(JITDylib &OldJD, JITDylib &NewJD,
-                                    JITDylibLookupFlags JDLookupFlags) {
+void JITDylib::replaceInLinkOrder(JITDylib &OldJD, JITDylib &NewJD,
+                                  JITDylibLookupFlags JDLookupFlags) {
   ES.runSessionLocked([&]() {
-    for (auto &KV : SearchOrder)
+    for (auto &KV : LinkOrder)
       if (KV.first == &OldJD) {
         KV = {&NewJD, JDLookupFlags};
         break;
@@ -1348,14 +1190,14 @@ void JITDylib::replaceInSearchOrder(JITDylib &OldJD, JITDylib &NewJD,
   });
 }
 
-void JITDylib::removeFromSearchOrder(JITDylib &JD) {
+void JITDylib::removeFromLinkOrder(JITDylib &JD) {
   ES.runSessionLocked([&]() {
-    auto I = std::find_if(SearchOrder.begin(), SearchOrder.end(),
+    auto I = std::find_if(LinkOrder.begin(), LinkOrder.end(),
                           [&](const JITDylibSearchOrder::value_type &KV) {
                             return KV.first == &JD;
                           });
-    if (I != SearchOrder.end())
-      SearchOrder.erase(I);
+    if (I != LinkOrder.end())
+      LinkOrder.erase(I);
   });
 }
 
@@ -1377,7 +1219,8 @@ Error JITDylib::remove(const SymbolNameSet &Names) {
       }
 
       // Note symbol materializing.
-      if (I->second.isInMaterializationPhase()) {
+      if (I->second.getState() != SymbolState::NeverSearched &&
+          I->second.getState() != SymbolState::Ready) {
         Materializing.insert(Name);
         continue;
       }
@@ -1498,6 +1341,12 @@ Error JITDylib::lodgeQueryImpl(MaterializationUnitList &MUs,
         if (SymI == Symbols.end())
           return false;
 
+        // If we match against a materialization-side-effects only symbol then
+        // make sure it is weakly-referenced. Otherwise bail out with an error.
+        if (SymI->second.getFlags().hasMaterializationSideEffectsOnly() &&
+            SymLookupFlags != SymbolLookupFlags::WeaklyReferencedSymbol)
+          return make_error<SymbolsNotFound>(SymbolNameVector({Name}));
+
         // If this is a non exported symbol and we're matching exported symbols
         // only then skip this symbol without removal.
         if (!SymI->second.getFlags().isExported() &&
@@ -1545,7 +1394,8 @@ Error JITDylib::lodgeQueryImpl(MaterializationUnitList &MUs,
 
         // Add the query to the PendingQueries list and continue, deleting the
         // element.
-        assert(SymI->second.isInMaterializationPhase() &&
+        assert(SymI->second.getState() != SymbolState::NeverSearched &&
+               SymI->second.getState() != SymbolState::Ready &&
                "By this line the symbol should be materializing");
         auto &MI = MaterializingInfos[Name];
         MI.addQuery(Q);
@@ -1601,8 +1451,11 @@ JITDylib::legacyLookup(std::shared_ptr<AsynchronousSymbolQuery> Q,
   // Add MUs to the OutstandingMUs list.
   {
     std::lock_guard<std::recursive_mutex> Lock(ES.OutstandingMUsMutex);
-    for (auto &MU : MUs)
-      ES.OutstandingMUs.push_back(make_pair(this, std::move(MU)));
+    auto ThisJD = shared_from_this();
+    for (auto &MU : MUs) {
+      auto MR = MU->createMaterializationResponsibility(ThisJD);
+      ES.OutstandingMUs.push_back(make_pair(std::move(MU), std::move(MR)));
+    }
   }
   ES.runOutstandingMUs();
 
@@ -1665,7 +1518,8 @@ bool JITDylib::lookupImpl(
         }
 
         // Add the query to the PendingQueries list.
-        assert(SymI->second.isInMaterializationPhase() &&
+        assert(SymI->second.getState() != SymbolState::NeverSearched &&
+               SymI->second.getState() != SymbolState::Ready &&
                "By this line the symbol should be materializing");
         auto &MI = MaterializingInfos[Name];
         MI.addQuery(Q);
@@ -1680,7 +1534,7 @@ void JITDylib::dump(raw_ostream &OS) {
   ES.runSessionLocked([&, this]() {
     OS << "JITDylib \"" << JITDylibName << "\" (ES: "
        << format("0x%016" PRIx64, reinterpret_cast<uintptr_t>(&ES)) << "):\n"
-       << "Search order: " << SearchOrder << "\n"
+       << "Link order: " << LinkOrder << "\n"
        << "Symbol table:\n";
 
     for (auto &KV : Symbols) {
@@ -1691,14 +1545,14 @@ void JITDylib::dump(raw_ostream &OS) {
       else
         OS << "<not resolved> ";
 
-      OS << KV.second.getState();
+      OS << KV.second.getFlags() << " " << KV.second.getState();
 
       if (KV.second.hasMaterializerAttached()) {
         OS << " (Materializer ";
         auto I = UnmaterializedInfos.find(KV.first);
         assert(I != UnmaterializedInfos.end() &&
                "Lazy symbol should have UnmaterializedInfo");
-        OS << I->second->MU.get() << ")\n";
+        OS << I->second->MU.get() << ", " << I->second->MU->getName() << ")\n";
       } else
         OS << "\n";
     }
@@ -1761,10 +1615,13 @@ JITDylib::MaterializingInfo::takeQueriesMeeting(SymbolState RequiredState) {
 
 JITDylib::JITDylib(ExecutionSession &ES, std::string Name)
     : ES(ES), JITDylibName(std::move(Name)) {
-  SearchOrder.push_back({this, JITDylibLookupFlags::MatchAllSymbols});
+  LinkOrder.push_back({this, JITDylibLookupFlags::MatchAllSymbols});
 }
 
 Error JITDylib::defineImpl(MaterializationUnit &MU) {
+
+  LLVM_DEBUG({ dbgs() << "  " << MU.getSymbols() << "\n"; });
+
   SymbolNameSet Duplicates;
   std::vector<SymbolStringPtr> ExistingDefsOverridden;
   std::vector<SymbolStringPtr> MUDefsOverridden;
@@ -1789,14 +1646,26 @@ Error JITDylib::defineImpl(MaterializationUnit &MU) {
   }
 
   // If there were any duplicate definitions then bail out.
-  if (!Duplicates.empty())
-    return make_error<DuplicateDefinition>(**Duplicates.begin());
+  if (!Duplicates.empty()) {
+    LLVM_DEBUG(
+        { dbgs() << "  Error: Duplicate symbols " << Duplicates << "\n"; });
+    return make_error<DuplicateDefinition>(std::string(**Duplicates.begin()));
+  }
 
   // Discard any overridden defs in this MU.
+  LLVM_DEBUG({
+    if (!MUDefsOverridden.empty())
+      dbgs() << "  Defs in this MU overridden: " << MUDefsOverridden << "\n";
+  });
   for (auto &S : MUDefsOverridden)
     MU.doDiscard(*this, S);
 
   // Discard existing overridden defs.
+  LLVM_DEBUG({
+    if (!ExistingDefsOverridden.empty())
+      dbgs() << "  Existing defs overridden by this MU: " << MUDefsOverridden
+             << "\n";
+  });
   for (auto &S : ExistingDefsOverridden) {
 
     auto UMII = UnmaterializedInfos.find(S);
@@ -1852,6 +1721,57 @@ void JITDylib::transferEmittedNodeDependencies(
   }
 }
 
+Platform::~Platform() {}
+
+Expected<DenseMap<JITDylib *, SymbolMap>> Platform::lookupInitSymbols(
+    ExecutionSession &ES,
+    const DenseMap<JITDylib *, SymbolLookupSet> &InitSyms) {
+
+  DenseMap<JITDylib *, SymbolMap> CompoundResult;
+  Error CompoundErr = Error::success();
+  std::mutex LookupMutex;
+  std::condition_variable CV;
+  uint64_t Count = InitSyms.size();
+
+  LLVM_DEBUG({
+    dbgs() << "Issuing init-symbol lookup:\n";
+    for (auto &KV : InitSyms)
+      dbgs() << "  " << KV.first->getName() << ": " << KV.second << "\n";
+  });
+
+  for (auto &KV : InitSyms) {
+    auto *JD = KV.first;
+    auto Names = std::move(KV.second);
+    ES.lookup(
+        LookupKind::Static,
+        JITDylibSearchOrder({{JD, JITDylibLookupFlags::MatchAllSymbols}}),
+        std::move(Names), SymbolState::Ready,
+        [&, JD](Expected<SymbolMap> Result) {
+          {
+            std::lock_guard<std::mutex> Lock(LookupMutex);
+            --Count;
+            if (Result) {
+              assert(!CompoundResult.count(JD) &&
+                     "Duplicate JITDylib in lookup?");
+              CompoundResult[JD] = std::move(*Result);
+            } else
+              CompoundErr =
+                  joinErrors(std::move(CompoundErr), Result.takeError());
+          }
+          CV.notify_one();
+        },
+        NoDependenciesToRegister);
+  }
+
+  std::unique_lock<std::mutex> Lock(LookupMutex);
+  CV.wait(Lock, [&] { return Count == 0 || CompoundErr; });
+
+  if (CompoundErr)
+    return std::move(CompoundErr);
+
+  return std::move(CompoundResult);
+}
+
 ExecutionSession::ExecutionSession(std::shared_ptr<SymbolStringPool> SSP)
     : SSP(SSP ? std::move(SSP) : std::make_shared<SymbolStringPool>()) {
 }
@@ -1865,15 +1785,23 @@ JITDylib *ExecutionSession::getJITDylibByName(StringRef Name) {
   });
 }
 
-JITDylib &ExecutionSession::createJITDylib(std::string Name) {
+JITDylib &ExecutionSession::createBareJITDylib(std::string Name) {
   assert(!getJITDylibByName(Name) && "JITDylib with that name already exists");
   return runSessionLocked([&, this]() -> JITDylib & {
     JDs.push_back(
-        std::unique_ptr<JITDylib>(new JITDylib(*this, std::move(Name))));
+        std::shared_ptr<JITDylib>(new JITDylib(*this, std::move(Name))));
     return *JDs.back();
   });
 }
 
+Expected<JITDylib &> ExecutionSession::createJITDylib(std::string Name) {
+  auto &JD = createBareJITDylib(Name);
+  if (P)
+    if (auto Err = P->setupJITDylib(JD))
+      return std::move(Err);
+  return JD;
+}
+
 void ExecutionSession::legacyFailQuery(AsynchronousSymbolQuery &Q, Error Err) {
   assert(!!Err && "Error should be in failure state");
 
@@ -2050,9 +1978,13 @@ void ExecutionSession::lookup(
   {
     std::lock_guard<std::recursive_mutex> Lock(OutstandingMUsMutex);
 
-    for (auto &KV : CollectedMUsMap)
-      for (auto &MU : KV.second)
-        OutstandingMUs.push_back(std::make_pair(KV.first, std::move(MU)));
+    for (auto &KV : CollectedMUsMap) {
+      auto JD = KV.first->shared_from_this();
+      for (auto &MU : KV.second) {
+        auto MR = MU->createMaterializationResponsibility(JD);
+        OutstandingMUs.push_back(std::make_pair(std::move(MU), std::move(MR)));
+      }
+    }
   }
 
   runOutstandingMUs();
@@ -2114,11 +2046,11 @@ ExecutionSession::lookup(const JITDylibSearchOrder &SearchOrder,
 
 Expected<JITEvaluatedSymbol>
 ExecutionSession::lookup(const JITDylibSearchOrder &SearchOrder,
-                         SymbolStringPtr Name) {
+                         SymbolStringPtr Name, SymbolState RequiredState) {
   SymbolLookupSet Names({Name});
 
   if (auto ResultMap = lookup(SearchOrder, std::move(Names), LookupKind::Static,
-                              SymbolState::Ready, NoDependenciesToRegister)) {
+                              RequiredState, NoDependenciesToRegister)) {
     assert(ResultMap->size() == 1 && "Unexpected number of results");
     assert(ResultMap->count(Name) && "Missing result for symbol");
     return std::move(ResultMap->begin()->second);
@@ -2127,14 +2059,15 @@ ExecutionSession::lookup(const JITDylibSearchOrder &SearchOrder,
 }
 
 Expected<JITEvaluatedSymbol>
-ExecutionSession::lookup(ArrayRef<JITDylib *> SearchOrder,
-                         SymbolStringPtr Name) {
-  return lookup(makeJITDylibSearchOrder(SearchOrder), Name);
+ExecutionSession::lookup(ArrayRef<JITDylib *> SearchOrder, SymbolStringPtr Name,
+                         SymbolState RequiredState) {
+  return lookup(makeJITDylibSearchOrder(SearchOrder), Name, RequiredState);
 }
 
 Expected<JITEvaluatedSymbol>
-ExecutionSession::lookup(ArrayRef<JITDylib *> SearchOrder, StringRef Name) {
-  return lookup(SearchOrder, intern(Name));
+ExecutionSession::lookup(ArrayRef<JITDylib *> SearchOrder, StringRef Name,
+                         SymbolState RequiredState) {
+  return lookup(SearchOrder, intern(Name), RequiredState);
 }
 
 void ExecutionSession::dump(raw_ostream &OS) {
@@ -2146,36 +2079,33 @@ void ExecutionSession::dump(raw_ostream &OS) {
 
 void ExecutionSession::runOutstandingMUs() {
   while (1) {
-    std::pair<JITDylib *, std::unique_ptr<MaterializationUnit>> JITDylibAndMU;
+    Optional<std::pair<std::unique_ptr<MaterializationUnit>,
+                       MaterializationResponsibility>>
+        JMU;
 
     {
       std::lock_guard<std::recursive_mutex> Lock(OutstandingMUsMutex);
       if (!OutstandingMUs.empty()) {
-        JITDylibAndMU = std::move(OutstandingMUs.back());
+        JMU.emplace(std::move(OutstandingMUs.back()));
         OutstandingMUs.pop_back();
       }
     }
 
-    if (JITDylibAndMU.first) {
-      assert(JITDylibAndMU.second && "JITDylib, but no MU?");
-      dispatchMaterialization(*JITDylibAndMU.first,
-                              std::move(JITDylibAndMU.second));
-    } else
+    if (!JMU)
       break;
+
+    assert(JMU->first && "No MU?");
+    dispatchMaterialization(std::move(JMU->first), std::move(JMU->second));
   }
 }
 
-MangleAndInterner::MangleAndInterner(ExecutionSession &ES, const DataLayout &DL)
-    : ES(ES), DL(DL) {}
-
-SymbolStringPtr MangleAndInterner::operator()(StringRef Name) {
-  std::string MangledName;
-  {
-    raw_string_ostream MangledNameStream(MangledName);
-    Mangler::getNameWithPrefix(MangledNameStream, Name, DL);
-  }
-  return ES.intern(MangledName);
+#ifndef NDEBUG
+void ExecutionSession::dumpDispatchInfo(JITDylib &JD, MaterializationUnit &MU) {
+  runSessionLocked([&]() {
+    dbgs() << "Dispatching " << MU << " for " << JD.getName() << "\n";
+  });
 }
+#endif // NDEBUG
 
 } // End namespace orc.
 } // End namespace llvm.
diff --git a/llvm/lib/ExecutionEngine/Orc/DebugUtils.cpp b/llvm/lib/ExecutionEngine/Orc/DebugUtils.cpp
index c9e87ff737fc..6247158919fa 100644
--- a/llvm/lib/ExecutionEngine/Orc/DebugUtils.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/DebugUtils.cpp
@@ -7,16 +7,297 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ExecutionEngine/Orc/DebugUtils.h"
+
+#include "llvm/ExecutionEngine/Orc/Core.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/FileSystem.h"
+#include "llvm/Support/Format.h"
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/raw_ostream.h"
 
 #define DEBUG_TYPE "orc"
 
+using namespace llvm;
+
+namespace {
+
+#ifndef NDEBUG
+
+cl::opt<bool> PrintHidden("debug-orc-print-hidden", cl::init(true),
+                          cl::desc("debug print hidden symbols defined by "
+                                   "materialization units"),
+                          cl::Hidden);
+
+cl::opt<bool> PrintCallable("debug-orc-print-callable", cl::init(true),
+                            cl::desc("debug print callable symbols defined by "
+                                     "materialization units"),
+                            cl::Hidden);
+
+cl::opt<bool> PrintData("debug-orc-print-data", cl::init(true),
+                        cl::desc("debug print data symbols defined by "
+                                 "materialization units"),
+                        cl::Hidden);
+
+#endif // NDEBUG
+
+// SetPrinter predicate that prints every element.
+template <typename T> struct PrintAll {
+  bool operator()(const T &E) { return true; }
+};
+
+bool anyPrintSymbolOptionSet() {
+#ifndef NDEBUG
+  return PrintHidden || PrintCallable || PrintData;
+#else
+  return false;
+#endif // NDEBUG
+}
+
+bool flagsMatchCLOpts(const JITSymbolFlags &Flags) {
+#ifndef NDEBUG
+  // Bail out early if this is a hidden symbol and we're not printing hiddens.
+  if (!PrintHidden && !Flags.isExported())
+    return false;
+
+  // Return true if this is callable and we're printing callables.
+  if (PrintCallable && Flags.isCallable())
+    return true;
+
+  // Return true if this is data and we're printing data.
+  if (PrintData && !Flags.isCallable())
+    return true;
+
+  // otherwise return false.
+  return false;
+#else
+  return false;
+#endif // NDEBUG
+}
+
+// Prints a sequence of items, filtered by an user-supplied predicate.
+template <typename Sequence,
+          typename Pred = PrintAll<typename Sequence::value_type>>
+class SequencePrinter {
+public:
+  SequencePrinter(const Sequence &S, char OpenSeq, char CloseSeq,
+                  Pred ShouldPrint = Pred())
+      : S(S), OpenSeq(OpenSeq), CloseSeq(CloseSeq),
+        ShouldPrint(std::move(ShouldPrint)) {}
+
+  void printTo(llvm::raw_ostream &OS) const {
+    bool PrintComma = false;
+    OS << OpenSeq;
+    for (auto &E : S) {
+      if (ShouldPrint(E)) {
+        if (PrintComma)
+          OS << ',';
+        OS << ' ' << E;
+        PrintComma = true;
+      }
+    }
+    OS << ' ' << CloseSeq;
+  }
+
+private:
+  const Sequence &S;
+  char OpenSeq;
+  char CloseSeq;
+  mutable Pred ShouldPrint;
+};
+
+template <typename Sequence, typename Pred>
+SequencePrinter<Sequence, Pred> printSequence(const Sequence &S, char OpenSeq,
+                                              char CloseSeq, Pred P = Pred()) {
+  return SequencePrinter<Sequence, Pred>(S, OpenSeq, CloseSeq, std::move(P));
+}
+
+// Render a SequencePrinter by delegating to its printTo method.
+template <typename Sequence, typename Pred>
+llvm::raw_ostream &operator<<(llvm::raw_ostream &OS,
+                              const SequencePrinter<Sequence, Pred> &Printer) {
+  Printer.printTo(OS);
+  return OS;
+}
+
+struct PrintSymbolFlagsMapElemsMatchingCLOpts {
+  bool operator()(const orc::SymbolFlagsMap::value_type &KV) {
+    return flagsMatchCLOpts(KV.second);
+  }
+};
+
+struct PrintSymbolMapElemsMatchingCLOpts {
+  bool operator()(const orc::SymbolMap::value_type &KV) {
+    return flagsMatchCLOpts(KV.second.getFlags());
+  }
+};
+
+} // end anonymous namespace
+
 namespace llvm {
 namespace orc {
 
+raw_ostream &operator<<(raw_ostream &OS, const SymbolStringPtr &Sym) {
+  return OS << *Sym;
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolNameSet &Symbols) {
+  return OS << printSequence(Symbols, '{', '}', PrintAll<SymbolStringPtr>());
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolNameVector &Symbols) {
+  return OS << printSequence(Symbols, '[', ']', PrintAll<SymbolStringPtr>());
+}
+
+raw_ostream &operator<<(raw_ostream &OS, ArrayRef<SymbolStringPtr> Symbols) {
+  return OS << printSequence(Symbols, '[', ']', PrintAll<SymbolStringPtr>());
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const JITSymbolFlags &Flags) {
+  if (Flags.hasError())
+    OS << "[*ERROR*]";
+  if (Flags.isCallable())
+    OS << "[Callable]";
+  else
+    OS << "[Data]";
+  if (Flags.isWeak())
+    OS << "[Weak]";
+  else if (Flags.isCommon())
+    OS << "[Common]";
+
+  if (!Flags.isExported())
+    OS << "[Hidden]";
+
+  return OS;
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const JITEvaluatedSymbol &Sym) {
+  return OS << format("0x%016" PRIx64, Sym.getAddress()) << " "
+            << Sym.getFlags();
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolFlagsMap::value_type &KV) {
+  return OS << "(\"" << KV.first << "\", " << KV.second << ")";
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolMap::value_type &KV) {
+  return OS << "(\"" << KV.first << "\": " << KV.second << ")";
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolFlagsMap &SymbolFlags) {
+  return OS << printSequence(SymbolFlags, '{', '}',
+                             PrintSymbolFlagsMapElemsMatchingCLOpts());
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolMap &Symbols) {
+  return OS << printSequence(Symbols, '{', '}',
+                             PrintSymbolMapElemsMatchingCLOpts());
+}
+
+raw_ostream &operator<<(raw_ostream &OS,
+                        const SymbolDependenceMap::value_type &KV) {
+  return OS << "(" << KV.first->getName() << ", " << KV.second << ")";
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolDependenceMap &Deps) {
+  return OS << printSequence(Deps, '{', '}',
+                             PrintAll<SymbolDependenceMap::value_type>());
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const MaterializationUnit &MU) {
+  OS << "MU@" << &MU << " (\"" << MU.getName() << "\"";
+  if (anyPrintSymbolOptionSet())
+    OS << ", " << MU.getSymbols();
+  return OS << ")";
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const LookupKind &K) {
+  switch (K) {
+  case LookupKind::Static:
+    return OS << "Static";
+  case LookupKind::DLSym:
+    return OS << "DLSym";
+  }
+  llvm_unreachable("Invalid lookup kind");
+}
+
+raw_ostream &operator<<(raw_ostream &OS,
+                        const JITDylibLookupFlags &JDLookupFlags) {
+  switch (JDLookupFlags) {
+  case JITDylibLookupFlags::MatchExportedSymbolsOnly:
+    return OS << "MatchExportedSymbolsOnly";
+  case JITDylibLookupFlags::MatchAllSymbols:
+    return OS << "MatchAllSymbols";
+  }
+  llvm_unreachable("Invalid JITDylib lookup flags");
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolLookupFlags &LookupFlags) {
+  switch (LookupFlags) {
+  case SymbolLookupFlags::RequiredSymbol:
+    return OS << "RequiredSymbol";
+  case SymbolLookupFlags::WeaklyReferencedSymbol:
+    return OS << "WeaklyReferencedSymbol";
+  }
+  llvm_unreachable("Invalid symbol lookup flags");
+}
+
+raw_ostream &operator<<(raw_ostream &OS,
+                        const SymbolLookupSet::value_type &KV) {
+  return OS << "(" << KV.first << ", " << KV.second << ")";
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolLookupSet &LookupSet) {
+  return OS << printSequence(LookupSet, '{', '}',
+                             PrintAll<SymbolLookupSet::value_type>());
+}
+
+raw_ostream &operator<<(raw_ostream &OS,
+                        const JITDylibSearchOrder &SearchOrder) {
+  OS << "[";
+  if (!SearchOrder.empty()) {
+    assert(SearchOrder.front().first &&
+           "JITDylibList entries must not be null");
+    OS << " (\"" << SearchOrder.front().first->getName() << "\", "
+       << SearchOrder.begin()->second << ")";
+    for (auto &KV :
+         make_range(std::next(SearchOrder.begin(), 1), SearchOrder.end())) {
+      assert(KV.first && "JITDylibList entries must not be null");
+      OS << ", (\"" << KV.first->getName() << "\", " << KV.second << ")";
+    }
+  }
+  OS << " ]";
+  return OS;
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolAliasMap &Aliases) {
+  OS << "{";
+  for (auto &KV : Aliases)
+    OS << " " << *KV.first << ": " << KV.second.Aliasee << " "
+       << KV.second.AliasFlags;
+  OS << " }";
+  return OS;
+}
+
+raw_ostream &operator<<(raw_ostream &OS, const SymbolState &S) {
+  switch (S) {
+  case SymbolState::Invalid:
+    return OS << "Invalid";
+  case SymbolState::NeverSearched:
+    return OS << "Never-Searched";
+  case SymbolState::Materializing:
+    return OS << "Materializing";
+  case SymbolState::Resolved:
+    return OS << "Resolved";
+  case SymbolState::Emitted:
+    return OS << "Emitted";
+  case SymbolState::Ready:
+    return OS << "Ready";
+  }
+  llvm_unreachable("Invalid state");
+}
+
 DumpObjects::DumpObjects(std::string DumpDir, std::string IdentifierOverride)
     : DumpDir(std::move(DumpDir)),
       IdentifierOverride(std::move(IdentifierOverride)) {
diff --git a/llvm/lib/ExecutionEngine/Orc/ExecutionUtils.cpp b/llvm/lib/ExecutionEngine/Orc/ExecutionUtils.cpp
index 3d97fe9eeab1..4d255cd66c1b 100644
--- a/llvm/lib/ExecutionEngine/Orc/ExecutionUtils.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/ExecutionUtils.cpp
@@ -13,6 +13,8 @@
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalVariable.h"
 #include "llvm/IR/Module.h"
+#include "llvm/Object/MachOUniversal.h"
+#include "llvm/Support/FormatVariadic.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Target/TargetMachine.h"
 
@@ -113,6 +115,26 @@ iterator_range<CtorDtorIterator> getDestructors(const Module &M) {
                     CtorDtorIterator(DtorsList, true));
 }
 
+bool StaticInitGVIterator::isStaticInitGlobal(GlobalValue &GV) {
+  if (GV.isDeclaration())
+    return false;
+
+  if (GV.hasName() && (GV.getName() == "llvm.global_ctors" ||
+                       GV.getName() == "llvm.global_dtors"))
+    return true;
+
+  if (ObjFmt == Triple::MachO) {
+    // FIXME: These section checks are too strict: We should match first and
+    // second word split by comma.
+    if (GV.hasSection() &&
+        (GV.getSection().startswith("__DATA,__objc_classlist") ||
+         GV.getSection().startswith("__DATA,__objc_selrefs")))
+      return true;
+  }
+
+  return false;
+}
+
 void CtorDtorRunner::add(iterator_range<CtorDtorIterator> CtorDtors) {
   if (CtorDtors.empty())
     return;
@@ -198,6 +220,30 @@ Error LocalCXXRuntimeOverrides::enable(JITDylib &JD,
   return JD.define(absoluteSymbols(std::move(RuntimeInterposes)));
 }
 
+void ItaniumCXAAtExitSupport::registerAtExit(void (*F)(void *), void *Ctx,
+                                             void *DSOHandle) {
+  std::lock_guard<std::mutex> Lock(AtExitsMutex);
+  AtExitRecords[DSOHandle].push_back({F, Ctx});
+}
+
+void ItaniumCXAAtExitSupport::runAtExits(void *DSOHandle) {
+  std::vector<AtExitRecord> AtExitsToRun;
+
+  {
+    std::lock_guard<std::mutex> Lock(AtExitsMutex);
+    auto I = AtExitRecords.find(DSOHandle);
+    if (I != AtExitRecords.end()) {
+      AtExitsToRun = std::move(I->second);
+      AtExitRecords.erase(I);
+    }
+  }
+
+  while (!AtExitsToRun.empty()) {
+    AtExitsToRun.back().F(AtExitsToRun.back().Ctx);
+    AtExitsToRun.pop_back();
+  }
+}
+
 DynamicLibrarySearchGenerator::DynamicLibrarySearchGenerator(
     sys::DynamicLibrary Dylib, char GlobalPrefix, SymbolPredicate Allow)
     : Dylib(std::move(Dylib)), Allow(std::move(Allow)),
@@ -259,6 +305,51 @@ StaticLibraryDefinitionGenerator::Load(ObjectLayer &L, const char *FileName) {
 }
 
 Expected<std::unique_ptr<StaticLibraryDefinitionGenerator>>
+StaticLibraryDefinitionGenerator::Load(ObjectLayer &L, const char *FileName,
+                                       const Triple &TT) {
+  auto B = object::createBinary(FileName);
+  if (!B)
+    return B.takeError();
+
+  // If this is a regular archive then create an instance from it.
+  if (isa<object::Archive>(B->getBinary()))
+    return Create(L, std::move(B->takeBinary().second));
+
+  // If this is a universal binary then search for a slice matching the given
+  // Triple.
+  if (auto *UB = cast<object::MachOUniversalBinary>(B->getBinary())) {
+    for (const auto &Obj : UB->objects()) {
+      auto ObjTT = Obj.getTriple();
+      if (ObjTT.getArch() == TT.getArch() &&
+          ObjTT.getSubArch() == TT.getSubArch() &&
+          ObjTT.getVendor() == TT.getVendor()) {
+        // We found a match. Create an instance from a buffer covering this
+        // slice.
+        auto SliceBuffer = MemoryBuffer::getFileSlice(FileName, Obj.getSize(),
+                                                      Obj.getOffset());
+        if (!SliceBuffer)
+          return make_error<StringError>(
+              Twine("Could not create buffer for ") + TT.str() + " slice of " +
+                  FileName + ": [ " + formatv("{0:x}", Obj.getOffset()) +
+                  " .. " + formatv("{0:x}", Obj.getOffset() + Obj.getSize()) +
+                  ": " + SliceBuffer.getError().message(),
+              SliceBuffer.getError());
+        return Create(L, std::move(*SliceBuffer));
+      }
+    }
+
+    return make_error<StringError>(Twine("Universal binary ") + FileName +
+                                       " does not contain a slice for " +
+                                       TT.str(),
+                                   inconvertibleErrorCode());
+  }
+
+  return make_error<StringError>(Twine("Unrecognized file type for ") +
+                                     FileName,
+                                 inconvertibleErrorCode());
+}
+
+Expected<std::unique_ptr<StaticLibraryDefinitionGenerator>>
 StaticLibraryDefinitionGenerator::Create(
     ObjectLayer &L, std::unique_ptr<MemoryBuffer> ArchiveBuffer) {
   Error Err = Error::success();
@@ -305,8 +396,8 @@ Error StaticLibraryDefinitionGenerator::tryToGenerate(
     MemoryBufferRef ChildBufferRef(ChildBufferInfo.first,
                                    ChildBufferInfo.second);
 
-    if (auto Err =
-            L.add(JD, MemoryBuffer::getMemBuffer(ChildBufferRef), VModuleKey()))
+    if (auto Err = L.add(JD, MemoryBuffer::getMemBuffer(ChildBufferRef, false),
+                         VModuleKey()))
       return Err;
   }
 
diff --git a/llvm/lib/ExecutionEngine/Orc/IRCompileLayer.cpp b/llvm/lib/ExecutionEngine/Orc/IRCompileLayer.cpp
index d311f34179c7..023940dc8298 100644
--- a/llvm/lib/ExecutionEngine/Orc/IRCompileLayer.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/IRCompileLayer.cpp
@@ -11,9 +11,14 @@
 namespace llvm {
 namespace orc {
 
+IRCompileLayer::IRCompiler::~IRCompiler() {}
+
 IRCompileLayer::IRCompileLayer(ExecutionSession &ES, ObjectLayer &BaseLayer,
-                                 CompileFunction Compile)
-    : IRLayer(ES), BaseLayer(BaseLayer), Compile(std::move(Compile)) {}
+                               std::unique_ptr<IRCompiler> Compile)
+    : IRLayer(ES, ManglingOpts), BaseLayer(BaseLayer),
+      Compile(std::move(Compile)) {
+  ManglingOpts = &this->Compile->getManglingOptions();
+}
 
 void IRCompileLayer::setNotifyCompiled(NotifyCompiledFunction NotifyCompiled) {
   std::lock_guard<std::mutex> Lock(IRLayerMutex);
@@ -24,7 +29,7 @@ void IRCompileLayer::emit(MaterializationResponsibility R,
                           ThreadSafeModule TSM) {
   assert(TSM && "Module must not be null");
 
-  if (auto Obj = TSM.withModuleDo(Compile)) {
+  if (auto Obj = TSM.withModuleDo(*Compile)) {
     {
       std::lock_guard<std::mutex> Lock(IRLayerMutex);
       if (NotifyCompiled)
diff --git a/llvm/lib/ExecutionEngine/Orc/IRTransformLayer.cpp b/llvm/lib/ExecutionEngine/Orc/IRTransformLayer.cpp
index 845ecc71eb87..511248f83b25 100644
--- a/llvm/lib/ExecutionEngine/Orc/IRTransformLayer.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/IRTransformLayer.cpp
@@ -12,10 +12,10 @@
 namespace llvm {
 namespace orc {
 
-IRTransformLayer::IRTransformLayer(ExecutionSession &ES,
-                                     IRLayer &BaseLayer,
-                                     TransformFunction Transform)
-    : IRLayer(ES), BaseLayer(BaseLayer), Transform(std::move(Transform)) {}
+IRTransformLayer::IRTransformLayer(ExecutionSession &ES, IRLayer &BaseLayer,
+                                   TransformFunction Transform)
+    : IRLayer(ES, BaseLayer.getManglingOptions()), BaseLayer(BaseLayer),
+      Transform(std::move(Transform)) {}
 
 void IRTransformLayer::emit(MaterializationResponsibility R,
                             ThreadSafeModule TSM) {
diff --git a/llvm/lib/ExecutionEngine/Orc/IndirectionUtils.cpp b/llvm/lib/ExecutionEngine/Orc/IndirectionUtils.cpp
index 1ac9a58aeaef..031b1afefc9d 100644
--- a/llvm/lib/ExecutionEngine/Orc/IndirectionUtils.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/IndirectionUtils.cpp
@@ -10,7 +10,6 @@
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/Triple.h"
 #include "llvm/ExecutionEngine/Orc/OrcABISupport.h"
-#include "llvm/IR/CallSite.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Transforms/Utils/Cloning.h"
@@ -28,7 +27,7 @@ public:
   CompileCallbackMaterializationUnit(SymbolStringPtr Name,
                                      CompileFunction Compile, VModuleKey K)
       : MaterializationUnit(SymbolFlagsMap({{Name, JITSymbolFlags::Exported}}),
-                            std::move(K)),
+                            nullptr, std::move(K)),
         Name(std::move(Name)), Compile(std::move(Compile)) {}
 
   StringRef getName() const override { return "<Compile Callbacks>"; }
diff --git a/llvm/lib/ExecutionEngine/Orc/JITTargetMachineBuilder.cpp b/llvm/lib/ExecutionEngine/Orc/JITTargetMachineBuilder.cpp
index 114e81e41771..8cf66c9e759a 100644
--- a/llvm/lib/ExecutionEngine/Orc/JITTargetMachineBuilder.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/JITTargetMachineBuilder.cpp
@@ -10,6 +10,7 @@
 
 #include "llvm/Support/Host.h"
 #include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/raw_ostream.h"
 
 namespace llvm {
 namespace orc {
@@ -33,7 +34,7 @@ Expected<JITTargetMachineBuilder> JITTargetMachineBuilder::detectHost() {
   for (auto &Feature : FeatureMap)
     TMBuilder.getFeatures().AddFeature(Feature.first(), Feature.second);
 
-  TMBuilder.setCPU(llvm::sys::getHostCPUName());
+  TMBuilder.setCPU(std::string(llvm::sys::getHostCPUName()));
 
   return TMBuilder;
 }
@@ -63,5 +64,78 @@ JITTargetMachineBuilder &JITTargetMachineBuilder::addFeatures(
   return *this;
 }
 
+#ifndef NDEBUG
+raw_ostream &operator<<(raw_ostream &OS, const JITTargetMachineBuilder &JTMB) {
+  OS << "{ Triple = \"" << JTMB.TT.str() << "\", CPU = \"" << JTMB.CPU
+     << "\", Options = <not-printable>, Relocation Model = ";
+
+  if (JTMB.RM) {
+    switch (*JTMB.RM) {
+    case Reloc::Static:
+      OS << "Static";
+      break;
+    case Reloc::PIC_:
+      OS << "PIC_";
+      break;
+    case Reloc::DynamicNoPIC:
+      OS << "DynamicNoPIC";
+      break;
+    case Reloc::ROPI:
+      OS << "ROPI";
+      break;
+    case Reloc::RWPI:
+      OS << "RWPI";
+      break;
+    case Reloc::ROPI_RWPI:
+      OS << "ROPI_RWPI";
+      break;
+    }
+  } else
+    OS << "unspecified";
+
+  OS << ", Code Model = ";
+
+  if (JTMB.CM) {
+    switch (*JTMB.CM) {
+    case CodeModel::Tiny:
+      OS << "Tiny";
+      break;
+    case CodeModel::Small:
+      OS << "Small";
+      break;
+    case CodeModel::Kernel:
+      OS << "Kernel";
+      break;
+    case CodeModel::Medium:
+      OS << "Medium";
+      break;
+    case CodeModel::Large:
+      OS << "Large";
+      break;
+    }
+  } else
+    OS << "unspecified";
+
+  OS << ", Optimization Level = ";
+  switch (JTMB.OptLevel) {
+  case CodeGenOpt::None:
+    OS << "None";
+    break;
+  case CodeGenOpt::Less:
+    OS << "Less";
+    break;
+  case CodeGenOpt::Default:
+    OS << "Default";
+    break;
+  case CodeGenOpt::Aggressive:
+    OS << "Aggressive";
+    break;
+  }
+
+  OS << " }";
+  return OS;
+}
+#endif // NDEBUG
+
 } // End namespace orc.
 } // End namespace llvm.
diff --git a/llvm/lib/ExecutionEngine/Orc/LLJIT.cpp b/llvm/lib/ExecutionEngine/Orc/LLJIT.cpp
index 54473ab46423..713a48fbf3eb 100644
--- a/llvm/lib/ExecutionEngine/Orc/LLJIT.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/LLJIT.cpp
@@ -7,30 +7,965 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ExecutionEngine/Orc/LLJIT.h"
+#include "llvm/ExecutionEngine/JITLink/EHFrameSupport.h"
 #include "llvm/ExecutionEngine/JITLink/JITLinkMemoryManager.h"
+#include "llvm/ExecutionEngine/Orc/MachOPlatform.h"
 #include "llvm/ExecutionEngine/Orc/ObjectLinkingLayer.h"
 #include "llvm/ExecutionEngine/Orc/OrcError.h"
 #include "llvm/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.h"
 #include "llvm/ExecutionEngine/SectionMemoryManager.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/Mangler.h"
+#include "llvm/IR/Module.h"
+#include "llvm/Support/DynamicLibrary.h"
+
+#include <map>
+
+#define DEBUG_TYPE "orc"
+
+using namespace llvm;
+using namespace llvm::orc;
+
+namespace {
+
+/// Adds helper function decls and wrapper functions that call the helper with
+/// some additional prefix arguments.
+///
+/// E.g. For wrapper "foo" with type i8(i8, i64), helper "bar", and prefix
+/// args i32 4 and i16 12345, this function will add:
+///
+/// declare i8 @bar(i32, i16, i8, i64)
+///
+/// define i8 @foo(i8, i64) {
+/// entry:
+///   %2 = call i8 @bar(i32 4, i16 12345, i8 %0, i64 %1)
+///   ret i8 %2
+/// }
+///
+Function *addHelperAndWrapper(Module &M, StringRef WrapperName,
+                              FunctionType *WrapperFnType,
+                              GlobalValue::VisibilityTypes WrapperVisibility,
+                              StringRef HelperName,
+                              ArrayRef<Value *> HelperPrefixArgs) {
+  std::vector<Type *> HelperArgTypes;
+  for (auto *Arg : HelperPrefixArgs)
+    HelperArgTypes.push_back(Arg->getType());
+  for (auto *T : WrapperFnType->params())
+    HelperArgTypes.push_back(T);
+  auto *HelperFnType =
+      FunctionType::get(WrapperFnType->getReturnType(), HelperArgTypes, false);
+  auto *HelperFn = Function::Create(HelperFnType, GlobalValue::ExternalLinkage,
+                                    HelperName, M);
+
+  auto *WrapperFn = Function::Create(
+      WrapperFnType, GlobalValue::ExternalLinkage, WrapperName, M);
+  WrapperFn->setVisibility(WrapperVisibility);
+
+  auto *EntryBlock = BasicBlock::Create(M.getContext(), "entry", WrapperFn);
+  IRBuilder<> IB(EntryBlock);
+
+  std::vector<Value *> HelperArgs;
+  for (auto *Arg : HelperPrefixArgs)
+    HelperArgs.push_back(Arg);
+  for (auto &Arg : WrapperFn->args())
+    HelperArgs.push_back(&Arg);
+  auto *HelperResult = IB.CreateCall(HelperFn, HelperArgs);
+  if (HelperFn->getReturnType()->isVoidTy())
+    IB.CreateRetVoid();
+  else
+    IB.CreateRet(HelperResult);
+
+  return WrapperFn;
+}
+
+class GenericLLVMIRPlatformSupport;
+
+/// orc::Platform component of Generic LLVM IR Platform support.
+/// Just forwards calls to the GenericLLVMIRPlatformSupport class below.
+class GenericLLVMIRPlatform : public Platform {
+public:
+  GenericLLVMIRPlatform(GenericLLVMIRPlatformSupport &S) : S(S) {}
+  Error setupJITDylib(JITDylib &JD) override;
+  Error notifyAdding(JITDylib &JD, const MaterializationUnit &MU) override;
+  Error notifyRemoving(JITDylib &JD, VModuleKey K) override {
+    // Noop -- Nothing to do (yet).
+    return Error::success();
+  }
+
+private:
+  GenericLLVMIRPlatformSupport &S;
+};
+
+/// This transform parses llvm.global_ctors to produce a single initialization
+/// function for the module, records the function, then deletes
+/// llvm.global_ctors.
+class GlobalCtorDtorScraper {
+public:
+
+  GlobalCtorDtorScraper(GenericLLVMIRPlatformSupport &PS,
+                        StringRef InitFunctionPrefix)
+    : PS(PS), InitFunctionPrefix(InitFunctionPrefix) {}
+  Expected<ThreadSafeModule> operator()(ThreadSafeModule TSM,
+                                        MaterializationResponsibility &R);
+
+private:
+  GenericLLVMIRPlatformSupport &PS;
+  StringRef InitFunctionPrefix;
+};
+
+/// Generic IR Platform Support
+///
+/// Scrapes llvm.global_ctors and llvm.global_dtors and replaces them with
+/// specially named 'init' and 'deinit'. Injects definitions / interposes for
+/// some runtime API, including __cxa_atexit, dlopen, and dlclose.
+class GenericLLVMIRPlatformSupport : public LLJIT::PlatformSupport {
+public:
+  // GenericLLVMIRPlatform &P) : P(P) {
+  GenericLLVMIRPlatformSupport(LLJIT &J)
+      : J(J), InitFunctionPrefix(J.mangle("__orc_init_func.")) {
+
+    getExecutionSession().setPlatform(
+        std::make_unique<GenericLLVMIRPlatform>(*this));
+
+    setInitTransform(J, GlobalCtorDtorScraper(*this, InitFunctionPrefix));
+
+    SymbolMap StdInterposes;
+
+    StdInterposes[J.mangleAndIntern("__lljit.platform_support_instance")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(this),
+                           JITSymbolFlags::Exported);
+    StdInterposes[J.mangleAndIntern("__lljit.cxa_atexit_helper")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(registerAtExitHelper),
+                           JITSymbolFlags());
+
+    cantFail(
+        J.getMainJITDylib().define(absoluteSymbols(std::move(StdInterposes))));
+    cantFail(setupJITDylib(J.getMainJITDylib()));
+    cantFail(J.addIRModule(J.getMainJITDylib(), createPlatformRuntimeModule()));
+  }
+
+  ExecutionSession &getExecutionSession() { return J.getExecutionSession(); }
+
+  /// Adds a module that defines the __dso_handle global.
+  Error setupJITDylib(JITDylib &JD) {
+
+    // Add per-jitdylib standard interposes.
+    SymbolMap PerJDInterposes;
+    PerJDInterposes[J.mangleAndIntern("__lljit.run_atexits_helper")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(runAtExitsHelper),
+                           JITSymbolFlags());
+    cantFail(JD.define(absoluteSymbols(std::move(PerJDInterposes))));
+
+    auto Ctx = std::make_unique<LLVMContext>();
+    auto M = std::make_unique<Module>("__standard_lib", *Ctx);
+    M->setDataLayout(J.getDataLayout());
+
+    auto *Int64Ty = Type::getInt64Ty(*Ctx);
+    auto *DSOHandle = new GlobalVariable(
+        *M, Int64Ty, true, GlobalValue::ExternalLinkage,
+        ConstantInt::get(Int64Ty, reinterpret_cast<uintptr_t>(&JD)),
+        "__dso_handle");
+    DSOHandle->setVisibility(GlobalValue::DefaultVisibility);
+    DSOHandle->setInitializer(
+        ConstantInt::get(Int64Ty, pointerToJITTargetAddress(&JD)));
+
+    auto *GenericIRPlatformSupportTy =
+        StructType::create(*Ctx, "lljit.GenericLLJITIRPlatformSupport");
+
+    auto *PlatformInstanceDecl = new GlobalVariable(
+        *M, GenericIRPlatformSupportTy, true, GlobalValue::ExternalLinkage,
+        nullptr, "__lljit.platform_support_instance");
+
+    auto *VoidTy = Type::getVoidTy(*Ctx);
+    addHelperAndWrapper(
+        *M, "__lljit_run_atexits", FunctionType::get(VoidTy, {}, false),
+        GlobalValue::HiddenVisibility, "__lljit.run_atexits_helper",
+        {PlatformInstanceDecl, DSOHandle});
+
+    return J.addIRModule(JD, ThreadSafeModule(std::move(M), std::move(Ctx)));
+  }
+
+  Error notifyAdding(JITDylib &JD, const MaterializationUnit &MU) {
+    if (auto &InitSym = MU.getInitializerSymbol())
+      InitSymbols[&JD].add(InitSym, SymbolLookupFlags::WeaklyReferencedSymbol);
+    else {
+      // If there's no identified init symbol attached, but there is a symbol
+      // with the GenericIRPlatform::InitFunctionPrefix, then treat that as
+      // an init function. Add the symbol to both the InitSymbols map (which
+      // will trigger a lookup to materialize the module) and the InitFunctions
+      // map (which holds the names of the symbols to execute).
+      for (auto &KV : MU.getSymbols())
+        if ((*KV.first).startswith(InitFunctionPrefix)) {
+          InitSymbols[&JD].add(KV.first,
+                               SymbolLookupFlags::WeaklyReferencedSymbol);
+          InitFunctions[&JD].add(KV.first);
+        }
+    }
+    return Error::success();
+  }
+
+  Error initialize(JITDylib &JD) override {
+    LLVM_DEBUG({
+      dbgs() << "GenericLLVMIRPlatformSupport getting initializers to run\n";
+    });
+    if (auto Initializers = getInitializers(JD)) {
+      LLVM_DEBUG(
+          { dbgs() << "GenericLLVMIRPlatformSupport running initializers\n"; });
+      for (auto InitFnAddr : *Initializers) {
+        LLVM_DEBUG({
+          dbgs() << "  Running init " << formatv("{0:x16}", InitFnAddr)
+                 << "...\n";
+        });
+        auto *InitFn = jitTargetAddressToFunction<void (*)()>(InitFnAddr);
+        InitFn();
+      }
+    } else
+      return Initializers.takeError();
+    return Error::success();
+  }
+
+  Error deinitialize(JITDylib &JD) override {
+    LLVM_DEBUG({
+      dbgs() << "GenericLLVMIRPlatformSupport getting deinitializers to run\n";
+    });
+    if (auto Deinitializers = getDeinitializers(JD)) {
+      LLVM_DEBUG({
+        dbgs() << "GenericLLVMIRPlatformSupport running deinitializers\n";
+      });
+      for (auto DeinitFnAddr : *Deinitializers) {
+        LLVM_DEBUG({
+          dbgs() << "  Running init " << formatv("{0:x16}", DeinitFnAddr)
+                 << "...\n";
+        });
+        auto *DeinitFn = jitTargetAddressToFunction<void (*)()>(DeinitFnAddr);
+        DeinitFn();
+      }
+    } else
+      return Deinitializers.takeError();
+
+    return Error::success();
+  }
+
+  void registerInitFunc(JITDylib &JD, SymbolStringPtr InitName) {
+    getExecutionSession().runSessionLocked([&]() {
+        InitFunctions[&JD].add(InitName);
+      });
+  }
+
+private:
+
+  Expected<std::vector<JITTargetAddress>> getInitializers(JITDylib &JD) {
+    if (auto Err = issueInitLookups(JD))
+      return std::move(Err);
+
+    DenseMap<JITDylib *, SymbolLookupSet> LookupSymbols;
+    std::vector<JITDylib *> DFSLinkOrder;
+
+    getExecutionSession().runSessionLocked([&]() {
+        DFSLinkOrder = getDFSLinkOrder(JD);
+
+        for (auto *NextJD : DFSLinkOrder) {
+          auto IFItr = InitFunctions.find(NextJD);
+          if (IFItr != InitFunctions.end()) {
+            LookupSymbols[NextJD] = std::move(IFItr->second);
+            InitFunctions.erase(IFItr);
+          }
+        }
+      });
+
+    LLVM_DEBUG({
+      dbgs() << "JITDylib init order is [ ";
+      for (auto *JD : llvm::reverse(DFSLinkOrder))
+        dbgs() << "\"" << JD->getName() << "\" ";
+      dbgs() << "]\n";
+      dbgs() << "Looking up init functions:\n";
+      for (auto &KV : LookupSymbols)
+        dbgs() << "  \"" << KV.first->getName() << "\": " << KV.second << "\n";
+    });
+
+    auto &ES = getExecutionSession();
+    auto LookupResult = Platform::lookupInitSymbols(ES, LookupSymbols);
+
+    if (!LookupResult)
+      return LookupResult.takeError();
+
+    std::vector<JITTargetAddress> Initializers;
+    while (!DFSLinkOrder.empty()) {
+      auto &NextJD = *DFSLinkOrder.back();
+      DFSLinkOrder.pop_back();
+      auto InitsItr = LookupResult->find(&NextJD);
+      if (InitsItr == LookupResult->end())
+        continue;
+      for (auto &KV : InitsItr->second)
+        Initializers.push_back(KV.second.getAddress());
+    }
+
+    return Initializers;
+  }
+
+  Expected<std::vector<JITTargetAddress>> getDeinitializers(JITDylib &JD) {
+    auto &ES = getExecutionSession();
+
+    auto LLJITRunAtExits = J.mangleAndIntern("__lljit_run_atexits");
+
+    DenseMap<JITDylib *, SymbolLookupSet> LookupSymbols;
+    std::vector<JITDylib *> DFSLinkOrder;
+
+    ES.runSessionLocked([&]() {
+        DFSLinkOrder = getDFSLinkOrder(JD);
+
+        for (auto *NextJD : DFSLinkOrder) {
+          auto &JDLookupSymbols = LookupSymbols[NextJD];
+          auto DIFItr = DeInitFunctions.find(NextJD);
+          if (DIFItr != DeInitFunctions.end()) {
+            LookupSymbols[NextJD] = std::move(DIFItr->second);
+            DeInitFunctions.erase(DIFItr);
+          }
+          JDLookupSymbols.add(LLJITRunAtExits,
+                            SymbolLookupFlags::WeaklyReferencedSymbol);
+      }
+    });
+
+    LLVM_DEBUG({
+      dbgs() << "JITDylib deinit order is [ ";
+      for (auto *JD : DFSLinkOrder)
+        dbgs() << "\"" << JD->getName() << "\" ";
+      dbgs() << "]\n";
+      dbgs() << "Looking up deinit functions:\n";
+      for (auto &KV : LookupSymbols)
+        dbgs() << "  \"" << KV.first->getName() << "\": " << KV.second << "\n";
+    });
+
+    auto LookupResult = Platform::lookupInitSymbols(ES, LookupSymbols);
+
+    if (!LookupResult)
+      return LookupResult.takeError();
+
+    std::vector<JITTargetAddress> DeInitializers;
+    for (auto *NextJD : DFSLinkOrder) {
+      auto DeInitsItr = LookupResult->find(NextJD);
+      assert(DeInitsItr != LookupResult->end() &&
+             "Every JD should have at least __lljit_run_atexits");
+
+      auto RunAtExitsItr = DeInitsItr->second.find(LLJITRunAtExits);
+      if (RunAtExitsItr != DeInitsItr->second.end())
+        DeInitializers.push_back(RunAtExitsItr->second.getAddress());
+
+      for (auto &KV : DeInitsItr->second)
+        if (KV.first != LLJITRunAtExits)
+          DeInitializers.push_back(KV.second.getAddress());
+    }
+
+    return DeInitializers;
+  }
+
+  // Returns a DFS traversal order of the JITDylibs reachable (via
+  // links-against edges) from JD, starting with JD itself.
+  static std::vector<JITDylib *> getDFSLinkOrder(JITDylib &JD) {
+    std::vector<JITDylib *> DFSLinkOrder;
+    std::vector<JITDylib *> WorkStack({&JD});
+    DenseSet<JITDylib *> Visited;
+
+    while (!WorkStack.empty()) {
+      auto &NextJD = *WorkStack.back();
+      WorkStack.pop_back();
+      if (Visited.count(&NextJD))
+        continue;
+      Visited.insert(&NextJD);
+      DFSLinkOrder.push_back(&NextJD);
+      NextJD.withLinkOrderDo([&](const JITDylibSearchOrder &LinkOrder) {
+        for (auto &KV : LinkOrder)
+          WorkStack.push_back(KV.first);
+      });
+    }
+
+    return DFSLinkOrder;
+  }
+
+  /// Issue lookups for all init symbols required to initialize JD (and any
+  /// JITDylibs that it depends on).
+  Error issueInitLookups(JITDylib &JD) {
+    DenseMap<JITDylib *, SymbolLookupSet> RequiredInitSymbols;
+    std::vector<JITDylib *> DFSLinkOrder;
+
+    getExecutionSession().runSessionLocked([&]() {
+        DFSLinkOrder = getDFSLinkOrder(JD);
+
+        for (auto *NextJD : DFSLinkOrder) {
+          auto ISItr = InitSymbols.find(NextJD);
+          if (ISItr != InitSymbols.end()) {
+            RequiredInitSymbols[NextJD] = std::move(ISItr->second);
+            InitSymbols.erase(ISItr);
+          }
+        }
+      });
+
+    return Platform::lookupInitSymbols(getExecutionSession(),
+                                       RequiredInitSymbols)
+        .takeError();
+  }
+
+  static void registerAtExitHelper(void *Self, void (*F)(void *), void *Ctx,
+                                   void *DSOHandle) {
+    LLVM_DEBUG({
+      dbgs() << "Registering atexit function " << (void *)F << " for JD "
+             << (*static_cast<JITDylib **>(DSOHandle))->getName() << "\n";
+    });
+    static_cast<GenericLLVMIRPlatformSupport *>(Self)->AtExitMgr.registerAtExit(
+        F, Ctx, DSOHandle);
+  }
+
+  static void runAtExitsHelper(void *Self, void *DSOHandle) {
+    LLVM_DEBUG({
+      dbgs() << "Running atexit functions for JD "
+             << (*static_cast<JITDylib **>(DSOHandle))->getName() << "\n";
+    });
+    static_cast<GenericLLVMIRPlatformSupport *>(Self)->AtExitMgr.runAtExits(
+        DSOHandle);
+  }
+
+  // Constructs an LLVM IR module containing platform runtime globals,
+  // functions, and interposes.
+  ThreadSafeModule createPlatformRuntimeModule() {
+    auto Ctx = std::make_unique<LLVMContext>();
+    auto M = std::make_unique<Module>("__standard_lib", *Ctx);
+    M->setDataLayout(J.getDataLayout());
+
+    auto *GenericIRPlatformSupportTy =
+        StructType::create(*Ctx, "lljit.GenericLLJITIRPlatformSupport");
+
+    auto *PlatformInstanceDecl = new GlobalVariable(
+        *M, GenericIRPlatformSupportTy, true, GlobalValue::ExternalLinkage,
+        nullptr, "__lljit.platform_support_instance");
+
+    auto *Int8Ty = Type::getInt8Ty(*Ctx);
+    auto *IntTy = Type::getIntNTy(*Ctx, sizeof(int) * CHAR_BIT);
+    auto *VoidTy = Type::getVoidTy(*Ctx);
+    auto *BytePtrTy = PointerType::getUnqual(Int8Ty);
+    auto *AtExitCallbackTy = FunctionType::get(VoidTy, {BytePtrTy}, false);
+    auto *AtExitCallbackPtrTy = PointerType::getUnqual(AtExitCallbackTy);
+
+    addHelperAndWrapper(
+        *M, "__cxa_atexit",
+        FunctionType::get(IntTy, {AtExitCallbackPtrTy, BytePtrTy, BytePtrTy},
+                          false),
+        GlobalValue::DefaultVisibility, "__lljit.cxa_atexit_helper",
+        {PlatformInstanceDecl});
+
+    return ThreadSafeModule(std::move(M), std::move(Ctx));
+  }
+
+  LLJIT &J;
+  std::string InitFunctionPrefix;
+  DenseMap<JITDylib *, SymbolLookupSet> InitSymbols;
+  DenseMap<JITDylib *, SymbolLookupSet> InitFunctions;
+  DenseMap<JITDylib *, SymbolLookupSet> DeInitFunctions;
+  ItaniumCXAAtExitSupport AtExitMgr;
+};
+
+Error GenericLLVMIRPlatform::setupJITDylib(JITDylib &JD) {
+  return S.setupJITDylib(JD);
+}
+
+Error GenericLLVMIRPlatform::notifyAdding(JITDylib &JD,
+                                          const MaterializationUnit &MU) {
+  return S.notifyAdding(JD, MU);
+}
+
+Expected<ThreadSafeModule>
+GlobalCtorDtorScraper::operator()(ThreadSafeModule TSM,
+                                  MaterializationResponsibility &R) {
+  auto Err = TSM.withModuleDo([&](Module &M) -> Error {
+    auto &Ctx = M.getContext();
+    auto *GlobalCtors = M.getNamedGlobal("llvm.global_ctors");
+
+    // If there's no llvm.global_ctors or it's just a decl then skip.
+    if (!GlobalCtors || GlobalCtors->isDeclaration())
+      return Error::success();
+
+    std::string InitFunctionName;
+    raw_string_ostream(InitFunctionName)
+        << InitFunctionPrefix << M.getModuleIdentifier();
+
+    MangleAndInterner Mangle(PS.getExecutionSession(), M.getDataLayout());
+    auto InternedName = Mangle(InitFunctionName);
+    if (auto Err =
+            R.defineMaterializing({{InternedName, JITSymbolFlags::Callable}}))
+      return Err;
+
+    auto *InitFunc =
+        Function::Create(FunctionType::get(Type::getVoidTy(Ctx), {}, false),
+                         GlobalValue::ExternalLinkage, InitFunctionName, &M);
+    InitFunc->setVisibility(GlobalValue::HiddenVisibility);
+    std::vector<std::pair<Function *, unsigned>> Inits;
+    for (auto E : getConstructors(M))
+      Inits.push_back(std::make_pair(E.Func, E.Priority));
+    llvm::sort(Inits, [](const std::pair<Function *, unsigned> &LHS,
+                         const std::pair<Function *, unsigned> &RHS) {
+      return LHS.first < RHS.first;
+    });
+    auto *EntryBlock = BasicBlock::Create(Ctx, "entry", InitFunc);
+    IRBuilder<> IB(EntryBlock);
+    for (auto &KV : Inits)
+      IB.CreateCall(KV.first);
+    IB.CreateRetVoid();
+
+    PS.registerInitFunc(R.getTargetJITDylib(), InternedName);
+    GlobalCtors->eraseFromParent();
+    return Error::success();
+  });
+
+  if (Err)
+    return std::move(Err);
+
+  return std::move(TSM);
+}
+
+class MachOPlatformSupport : public LLJIT::PlatformSupport {
+public:
+  using DLOpenType = void *(*)(const char *Name, int Mode);
+  using DLCloseType = int (*)(void *Handle);
+  using DLSymType = void *(*)(void *Handle, const char *Name);
+  using DLErrorType = const char *(*)();
+
+  struct DlFcnValues {
+    Optional<void *> RTLDDefault;
+    DLOpenType dlopen = nullptr;
+    DLCloseType dlclose = nullptr;
+    DLSymType dlsym = nullptr;
+    DLErrorType dlerror = nullptr;
+  };
+
+  static Expected<std::unique_ptr<MachOPlatformSupport>>
+  Create(LLJIT &J, JITDylib &PlatformJITDylib) {
+
+    // Make process symbols visible.
+    {
+      std::string ErrMsg;
+      auto Lib = sys::DynamicLibrary::getPermanentLibrary(nullptr, &ErrMsg);
+      if (!Lib.isValid())
+        return make_error<StringError>(std::move(ErrMsg),
+                                       inconvertibleErrorCode());
+    }
+
+    DlFcnValues DlFcn;
+
+    // Add support for RTLDDefault on known platforms.
+#ifdef __APPLE__
+    DlFcn.RTLDDefault = reinterpret_cast<void *>(-2);
+#endif // __APPLE__
+
+    if (auto Err = hookUpFunction(DlFcn.dlopen, "dlopen"))
+      return std::move(Err);
+    if (auto Err = hookUpFunction(DlFcn.dlclose, "dlclose"))
+      return std::move(Err);
+    if (auto Err = hookUpFunction(DlFcn.dlsym, "dlsym"))
+      return std::move(Err);
+    if (auto Err = hookUpFunction(DlFcn.dlerror, "dlerror"))
+      return std::move(Err);
+
+    std::unique_ptr<MachOPlatformSupport> MP(
+        new MachOPlatformSupport(J, PlatformJITDylib, DlFcn));
+    return std::move(MP);
+  }
+
+  Error initialize(JITDylib &JD) override {
+    LLVM_DEBUG({
+      dbgs() << "MachOPlatformSupport initializing \"" << JD.getName()
+             << "\"\n";
+    });
+
+    auto InitSeq = MP.getInitializerSequence(JD);
+    if (!InitSeq)
+      return InitSeq.takeError();
+
+    // If ObjC is not enabled but there are JIT'd ObjC inits then return
+    // an error.
+    if (!objCRegistrationEnabled())
+      for (auto &KV : *InitSeq) {
+        if (!KV.second.getObjCSelRefsSections().empty() ||
+            !KV.second.getObjCClassListSections().empty())
+          return make_error<StringError>("JITDylib " + KV.first->getName() +
+                                             " contains objc metadata but objc"
+                                             " is not enabled",
+                                         inconvertibleErrorCode());
+      }
+
+    // Run the initializers.
+    for (auto &KV : *InitSeq) {
+      if (objCRegistrationEnabled()) {
+        KV.second.registerObjCSelectors();
+        if (auto Err = KV.second.registerObjCClasses()) {
+          // FIXME: Roll back registrations on error?
+          return Err;
+        }
+      }
+      KV.second.runModInits();
+    }
+
+    return Error::success();
+  }
+
+  Error deinitialize(JITDylib &JD) override {
+    auto &ES = J.getExecutionSession();
+    if (auto DeinitSeq = MP.getDeinitializerSequence(JD)) {
+      for (auto &KV : *DeinitSeq) {
+        auto DSOHandleName = ES.intern("___dso_handle");
+
+        // FIXME: Run DeInits here.
+        auto Result = ES.lookup(
+            {{KV.first, JITDylibLookupFlags::MatchAllSymbols}},
+            SymbolLookupSet(DSOHandleName,
+                            SymbolLookupFlags::WeaklyReferencedSymbol));
+        if (!Result)
+          return Result.takeError();
+        if (Result->empty())
+          continue;
+        assert(Result->count(DSOHandleName) &&
+               "Result does not contain __dso_handle");
+        auto *DSOHandle = jitTargetAddressToPointer<void *>(
+            Result->begin()->second.getAddress());
+        AtExitMgr.runAtExits(DSOHandle);
+      }
+    } else
+      return DeinitSeq.takeError();
+    return Error::success();
+  }
+
+private:
+  template <typename FunctionPtrTy>
+  static Error hookUpFunction(FunctionPtrTy &Fn, const char *Name) {
+    if (auto *FnAddr = sys::DynamicLibrary::SearchForAddressOfSymbol(Name)) {
+      Fn = reinterpret_cast<FunctionPtrTy>(Fn);
+      return Error::success();
+    }
+
+    return make_error<StringError>((Twine("Can not enable MachO JIT Platform: "
+                                          "missing function: ") +
+                                    Name)
+                                       .str(),
+                                   inconvertibleErrorCode());
+  }
+
+  MachOPlatformSupport(LLJIT &J, JITDylib &PlatformJITDylib, DlFcnValues DlFcn)
+      : J(J), MP(setupPlatform(J)), DlFcn(std::move(DlFcn)) {
+
+    SymbolMap HelperSymbols;
+
+    // platform and atexit helpers.
+    HelperSymbols[J.mangleAndIntern("__lljit.platform_support_instance")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(this), JITSymbolFlags());
+    HelperSymbols[J.mangleAndIntern("__lljit.cxa_atexit_helper")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(registerAtExitHelper),
+                           JITSymbolFlags());
+    HelperSymbols[J.mangleAndIntern("__lljit.run_atexits_helper")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(runAtExitsHelper),
+                           JITSymbolFlags());
+
+    // dlfcn helpers.
+    HelperSymbols[J.mangleAndIntern("__lljit.dlopen_helper")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(dlopenHelper),
+                           JITSymbolFlags());
+    HelperSymbols[J.mangleAndIntern("__lljit.dlclose_helper")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(dlcloseHelper),
+                           JITSymbolFlags());
+    HelperSymbols[J.mangleAndIntern("__lljit.dlsym_helper")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(dlsymHelper),
+                           JITSymbolFlags());
+    HelperSymbols[J.mangleAndIntern("__lljit.dlerror_helper")] =
+        JITEvaluatedSymbol(pointerToJITTargetAddress(dlerrorHelper),
+                           JITSymbolFlags());
+
+    cantFail(
+        PlatformJITDylib.define(absoluteSymbols(std::move(HelperSymbols))));
+    cantFail(MP.setupJITDylib(J.getMainJITDylib()));
+    cantFail(J.addIRModule(PlatformJITDylib, createPlatformRuntimeModule()));
+  }
+
+  static MachOPlatform &setupPlatform(LLJIT &J) {
+    auto Tmp = std::make_unique<MachOPlatform>(
+        J.getExecutionSession(),
+        static_cast<ObjectLinkingLayer &>(J.getObjLinkingLayer()),
+        createStandardSymbolsObject(J));
+    auto &MP = *Tmp;
+    J.getExecutionSession().setPlatform(std::move(Tmp));
+    return MP;
+  }
+
+  static std::unique_ptr<MemoryBuffer> createStandardSymbolsObject(LLJIT &J) {
+    LLVMContext Ctx;
+    Module M("__standard_symbols", Ctx);
+    M.setDataLayout(J.getDataLayout());
+
+    auto *Int64Ty = Type::getInt64Ty(Ctx);
+
+    auto *DSOHandle =
+        new GlobalVariable(M, Int64Ty, true, GlobalValue::ExternalLinkage,
+                           ConstantInt::get(Int64Ty, 0), "__dso_handle");
+    DSOHandle->setVisibility(GlobalValue::DefaultVisibility);
+
+    return cantFail(J.getIRCompileLayer().getCompiler()(M));
+  }
+
+  ThreadSafeModule createPlatformRuntimeModule() {
+    auto Ctx = std::make_unique<LLVMContext>();
+    auto M = std::make_unique<Module>("__standard_lib", *Ctx);
+    M->setDataLayout(J.getDataLayout());
+
+    auto *MachOPlatformSupportTy =
+        StructType::create(*Ctx, "lljit.MachOPlatformSupport");
+
+    auto *PlatformInstanceDecl = new GlobalVariable(
+        *M, MachOPlatformSupportTy, true, GlobalValue::ExternalLinkage, nullptr,
+        "__lljit.platform_support_instance");
+
+    auto *Int8Ty = Type::getInt8Ty(*Ctx);
+    auto *IntTy = Type::getIntNTy(*Ctx, sizeof(int) * CHAR_BIT);
+    auto *VoidTy = Type::getVoidTy(*Ctx);
+    auto *BytePtrTy = PointerType::getUnqual(Int8Ty);
+    auto *AtExitCallbackTy = FunctionType::get(VoidTy, {BytePtrTy}, false);
+    auto *AtExitCallbackPtrTy = PointerType::getUnqual(AtExitCallbackTy);
+
+    addHelperAndWrapper(
+        *M, "__cxa_atexit",
+        FunctionType::get(IntTy, {AtExitCallbackPtrTy, BytePtrTy, BytePtrTy},
+                          false),
+        GlobalValue::DefaultVisibility, "__lljit.cxa_atexit_helper",
+        {PlatformInstanceDecl});
+
+    addHelperAndWrapper(*M, "dlopen",
+                        FunctionType::get(BytePtrTy, {BytePtrTy, IntTy}, false),
+                        GlobalValue::DefaultVisibility, "__lljit.dlopen_helper",
+                        {PlatformInstanceDecl});
+
+    addHelperAndWrapper(*M, "dlclose",
+                        FunctionType::get(IntTy, {BytePtrTy}, false),
+                        GlobalValue::DefaultVisibility,
+                        "__lljit.dlclose_helper", {PlatformInstanceDecl});
+
+    addHelperAndWrapper(
+        *M, "dlsym",
+        FunctionType::get(BytePtrTy, {BytePtrTy, BytePtrTy}, false),
+        GlobalValue::DefaultVisibility, "__lljit.dlsym_helper",
+        {PlatformInstanceDecl});
+
+    addHelperAndWrapper(*M, "dlerror", FunctionType::get(BytePtrTy, {}, false),
+                        GlobalValue::DefaultVisibility,
+                        "__lljit.dlerror_helper", {PlatformInstanceDecl});
+
+    return ThreadSafeModule(std::move(M), std::move(Ctx));
+  }
+
+  static void registerAtExitHelper(void *Self, void (*F)(void *), void *Ctx,
+                                   void *DSOHandle) {
+    static_cast<MachOPlatformSupport *>(Self)->AtExitMgr.registerAtExit(
+        F, Ctx, DSOHandle);
+  }
+
+  static void runAtExitsHelper(void *Self, void *DSOHandle) {
+    static_cast<MachOPlatformSupport *>(Self)->AtExitMgr.runAtExits(DSOHandle);
+  }
+
+  void *jit_dlopen(const char *Path, int Mode) {
+    JITDylib *JDToOpen = nullptr;
+    // FIXME: Do the right thing with Mode flags.
+    {
+      std::lock_guard<std::mutex> Lock(PlatformSupportMutex);
+
+      // Clear any existing error messages.
+      dlErrorMsgs.erase(std::this_thread::get_id());
+
+      if (auto *JD = J.getExecutionSession().getJITDylibByName(Path)) {
+        auto I = JDRefCounts.find(JD);
+        if (I != JDRefCounts.end()) {
+          ++I->second;
+          return JD;
+        }
+
+        JDRefCounts[JD] = 1;
+        JDToOpen = JD;
+      }
+    }
+
+    if (JDToOpen) {
+      if (auto Err = initialize(*JDToOpen)) {
+        recordError(std::move(Err));
+        return 0;
+      }
+    }
+
+    // Fall through to dlopen if no JITDylib found for Path.
+    return DlFcn.dlopen(Path, Mode);
+  }
+
+  static void *dlopenHelper(void *Self, const char *Path, int Mode) {
+    return static_cast<MachOPlatformSupport *>(Self)->jit_dlopen(Path, Mode);
+  }
+
+  int jit_dlclose(void *Handle) {
+    JITDylib *JDToClose = nullptr;
+
+    {
+      std::lock_guard<std::mutex> Lock(PlatformSupportMutex);
+
+      // Clear any existing error messages.
+      dlErrorMsgs.erase(std::this_thread::get_id());
+
+      auto I = JDRefCounts.find(Handle);
+      if (I != JDRefCounts.end()) {
+        --I->second;
+        if (I->second == 0) {
+          JDRefCounts.erase(I);
+          JDToClose = static_cast<JITDylib *>(Handle);
+        } else
+          return 0;
+      }
+    }
+
+    if (JDToClose) {
+      if (auto Err = deinitialize(*JDToClose)) {
+        recordError(std::move(Err));
+        return -1;
+      }
+      return 0;
+    }
+
+    // Fall through to dlclose if no JITDylib found for Path.
+    return DlFcn.dlclose(Handle);
+  }
+
+  static int dlcloseHelper(void *Self, void *Handle) {
+    return static_cast<MachOPlatformSupport *>(Self)->jit_dlclose(Handle);
+  }
+
+  void *jit_dlsym(void *Handle, const char *Name) {
+    JITDylibSearchOrder JITSymSearchOrder;
+
+    // FIXME: RTLD_NEXT, RTLD_SELF not supported.
+    {
+      std::lock_guard<std::mutex> Lock(PlatformSupportMutex);
+
+      // Clear any existing error messages.
+      dlErrorMsgs.erase(std::this_thread::get_id());
+
+      if (JDRefCounts.count(Handle)) {
+        JITSymSearchOrder.push_back(
+            {static_cast<JITDylib *>(Handle),
+             JITDylibLookupFlags::MatchExportedSymbolsOnly});
+      } else if (Handle == DlFcn.RTLDDefault) {
+        for (auto &KV : JDRefCounts)
+          JITSymSearchOrder.push_back(
+              {static_cast<JITDylib *>(KV.first),
+               JITDylibLookupFlags::MatchExportedSymbolsOnly});
+      }
+    }
+
+    if (!JITSymSearchOrder.empty()) {
+      auto MangledName = J.mangleAndIntern(Name);
+      SymbolLookupSet Syms(MangledName,
+                           SymbolLookupFlags::WeaklyReferencedSymbol);
+      if (auto Result = J.getExecutionSession().lookup(JITSymSearchOrder, Syms,
+                                                       LookupKind::DLSym)) {
+        auto I = Result->find(MangledName);
+        if (I != Result->end())
+          return jitTargetAddressToPointer<void *>(I->second.getAddress());
+      } else {
+        recordError(Result.takeError());
+        return 0;
+      }
+    }
+
+    // Fall through to dlsym.
+    return DlFcn.dlsym(Handle, Name);
+  }
+
+  static void *dlsymHelper(void *Self, void *Handle, const char *Name) {
+    return static_cast<MachOPlatformSupport *>(Self)->jit_dlsym(Handle, Name);
+  }
+
+  const char *jit_dlerror() {
+    {
+      std::lock_guard<std::mutex> Lock(PlatformSupportMutex);
+      auto I = dlErrorMsgs.find(std::this_thread::get_id());
+      if (I != dlErrorMsgs.end())
+        return I->second->c_str();
+    }
+    return DlFcn.dlerror();
+  }
+
+  static const char *dlerrorHelper(void *Self) {
+    return static_cast<MachOPlatformSupport *>(Self)->jit_dlerror();
+  }
+
+  void recordError(Error Err) {
+    std::lock_guard<std::mutex> Lock(PlatformSupportMutex);
+    dlErrorMsgs[std::this_thread::get_id()] =
+        std::make_unique<std::string>(toString(std::move(Err)));
+  }
+
+  std::mutex PlatformSupportMutex;
+  LLJIT &J;
+  MachOPlatform &MP;
+  DlFcnValues DlFcn;
+  ItaniumCXAAtExitSupport AtExitMgr;
+  DenseMap<void *, unsigned> JDRefCounts;
+  std::map<std::thread::id, std::unique_ptr<std::string>> dlErrorMsgs;
+};
+
+} // end anonymous namespace
 
 namespace llvm {
 namespace orc {
 
+void LLJIT::PlatformSupport::setInitTransform(
+    LLJIT &J, IRTransformLayer::TransformFunction T) {
+  J.InitHelperTransformLayer->setTransform(std::move(T));
+}
+
+LLJIT::PlatformSupport::~PlatformSupport() {}
+
 Error LLJITBuilderState::prepareForConstruction() {
 
+  LLVM_DEBUG(dbgs() << "Preparing to create LLIT instance...\n");
+
   if (!JTMB) {
+    LLVM_DEBUG({
+      dbgs() << "  No explicitly set JITTargetMachineBuilder. "
+                "Detecting host...\n";
+    });
     if (auto JTMBOrErr = JITTargetMachineBuilder::detectHost())
       JTMB = std::move(*JTMBOrErr);
     else
       return JTMBOrErr.takeError();
   }
 
+  LLVM_DEBUG({
+    dbgs() << "  JITTargetMachineBuilder is " << JTMB << "\n"
+           << "  Pre-constructed ExecutionSession: " << (ES ? "Yes" : "No")
+           << "\n"
+           << "  DataLayout: ";
+    if (DL)
+      dbgs() << DL->getStringRepresentation() << "\n";
+    else
+      dbgs() << "None (will be created by JITTargetMachineBuilder)\n";
+
+    dbgs() << "  Custom object-linking-layer creator: "
+           << (CreateObjectLinkingLayer ? "Yes" : "No") << "\n"
+           << "  Custom compile-function creator: "
+           << (CreateCompileFunction ? "Yes" : "No") << "\n"
+           << "  Custom platform-setup function: "
+           << (SetUpPlatform ? "Yes" : "No") << "\n"
+           << "  Number of compile threads: " << NumCompileThreads;
+    if (!NumCompileThreads)
+      dbgs() << " (code will be compiled on the execution thread)\n";
+    else
+      dbgs() << "\n";
+  });
+
   // If the client didn't configure any linker options then auto-configure the
   // JIT linker.
-  if (!CreateObjectLinkingLayer && JTMB->getCodeModel() == None &&
-      JTMB->getRelocationModel() == None) {
-
+  if (!CreateObjectLinkingLayer) {
     auto &TT = JTMB->getTargetTriple();
     if (TT.isOSBinFormatMachO() &&
         (TT.getArch() == Triple::aarch64 || TT.getArch() == Triple::x86_64)) {
@@ -40,8 +975,11 @@ Error LLJITBuilderState::prepareForConstruction() {
       CreateObjectLinkingLayer =
           [](ExecutionSession &ES,
              const Triple &) -> std::unique_ptr<ObjectLayer> {
-        return std::make_unique<ObjectLinkingLayer>(
+        auto ObjLinkingLayer = std::make_unique<ObjectLinkingLayer>(
             ES, std::make_unique<jitlink::InProcessMemoryManager>());
+        ObjLinkingLayer->addPlugin(std::make_unique<EHFrameRegistrationPlugin>(
+            jitlink::InProcessEHFrameRegistrar::getInstance()));
+        return std::move(ObjLinkingLayer);
       };
     }
   }
@@ -54,12 +992,6 @@ LLJIT::~LLJIT() {
     CompileThreads->wait();
 }
 
-Error LLJIT::defineAbsolute(StringRef Name, JITEvaluatedSymbol Sym) {
-  auto InternedName = ES->intern(Name);
-  SymbolMap Symbols({{InternedName, Sym}});
-  return Main.define(absoluteSymbols(std::move(Symbols)));
-}
-
 Error LLJIT::addIRModule(JITDylib &JD, ThreadSafeModule TSM) {
   assert(TSM && "Can not add null module");
 
@@ -67,7 +999,8 @@ Error LLJIT::addIRModule(JITDylib &JD, ThreadSafeModule TSM) {
           TSM.withModuleDo([&](Module &M) { return applyDataLayout(M); }))
     return Err;
 
-  return CompileLayer->add(JD, std::move(TSM), ES->allocateVModule());
+  return InitHelperTransformLayer->add(JD, std::move(TSM),
+                                       ES->allocateVModule());
 }
 
 Error LLJIT::addObjectFile(JITDylib &JD, std::unique_ptr<MemoryBuffer> Obj) {
@@ -77,10 +1010,9 @@ Error LLJIT::addObjectFile(JITDylib &JD, std::unique_ptr<MemoryBuffer> Obj) {
 }
 
 Expected<JITEvaluatedSymbol> LLJIT::lookupLinkerMangled(JITDylib &JD,
-                                                        StringRef Name) {
+                                                        SymbolStringPtr Name) {
   return ES->lookup(
-      makeJITDylibSearchOrder(&JD, JITDylibLookupFlags::MatchAllSymbols),
-      ES->intern(Name));
+      makeJITDylibSearchOrder(&JD, JITDylibLookupFlags::MatchAllSymbols), Name);
 }
 
 std::unique_ptr<ObjectLayer>
@@ -96,8 +1028,10 @@ LLJIT::createObjectLinkingLayer(LLJITBuilderState &S, ExecutionSession &ES) {
   auto ObjLinkingLayer =
       std::make_unique<RTDyldObjectLinkingLayer>(ES, std::move(GetMemMgr));
 
-  if (S.JTMB->getTargetTriple().isOSBinFormatCOFF())
+  if (S.JTMB->getTargetTriple().isOSBinFormatCOFF()) {
     ObjLinkingLayer->setOverrideObjectFlagsWithResponsibilityFlags(true);
+    ObjLinkingLayer->setAutoClaimResponsibilityForObjectSymbols(true);
+  }
 
   // FIXME: Explicit conversion to std::unique_ptr<ObjectLayer> added to silence
   //        errors from some GCC / libstdc++ bots. Remove this conversion (i.e.
@@ -105,7 +1039,7 @@ LLJIT::createObjectLinkingLayer(LLJITBuilderState &S, ExecutionSession &ES) {
   return std::unique_ptr<ObjectLayer>(std::move(ObjLinkingLayer));
 }
 
-Expected<IRCompileLayer::CompileFunction>
+Expected<std::unique_ptr<IRCompileLayer::IRCompiler>>
 LLJIT::createCompileFunction(LLJITBuilderState &S,
                              JITTargetMachineBuilder JTMB) {
 
@@ -116,25 +1050,33 @@ LLJIT::createCompileFunction(LLJITBuilderState &S,
   // Otherwise default to creating a SimpleCompiler, or ConcurrentIRCompiler,
   // depending on the number of threads requested.
   if (S.NumCompileThreads > 0)
-    return ConcurrentIRCompiler(std::move(JTMB));
+    return std::make_unique<ConcurrentIRCompiler>(std::move(JTMB));
 
   auto TM = JTMB.createTargetMachine();
   if (!TM)
     return TM.takeError();
 
-  return TMOwningSimpleCompiler(std::move(*TM));
+  return std::make_unique<TMOwningSimpleCompiler>(std::move(*TM));
 }
 
 LLJIT::LLJIT(LLJITBuilderState &S, Error &Err)
-    : ES(S.ES ? std::move(S.ES) : std::make_unique<ExecutionSession>()),
-      Main(this->ES->createJITDylib("<main>")), DL(""),
+    : ES(S.ES ? std::move(S.ES) : std::make_unique<ExecutionSession>()), Main(),
+      DL(""), TT(S.JTMB->getTargetTriple()),
       ObjLinkingLayer(createObjectLinkingLayer(S, *ES)),
-      ObjTransformLayer(*this->ES, *ObjLinkingLayer), CtorRunner(Main),
-      DtorRunner(Main) {
+      ObjTransformLayer(*this->ES, *ObjLinkingLayer) {
 
   ErrorAsOutParameter _(&Err);
 
-  if (auto DLOrErr = S.JTMB->getDefaultDataLayoutForTarget())
+  if (auto MainOrErr = this->ES->createJITDylib("main"))
+    Main = &*MainOrErr;
+  else {
+    Err = MainOrErr.takeError();
+    return;
+  }
+
+  if (S.DL)
+    DL = std::move(*S.DL);
+  else if (auto DLOrErr = S.JTMB->getDefaultDataLayoutForTarget())
     DL = std::move(*DLOrErr);
   else {
     Err = DLOrErr.takeError();
@@ -149,22 +1091,36 @@ LLJIT::LLJIT(LLJITBuilderState &S, Error &Err)
     }
     CompileLayer = std::make_unique<IRCompileLayer>(
         *ES, ObjTransformLayer, std::move(*CompileFunction));
+    TransformLayer = std::make_unique<IRTransformLayer>(*ES, *CompileLayer);
+    InitHelperTransformLayer =
+        std::make_unique<IRTransformLayer>(*ES, *TransformLayer);
   }
 
   if (S.NumCompileThreads > 0) {
-    CompileLayer->setCloneToNewContextOnEmit(true);
-    CompileThreads = std::make_unique<ThreadPool>(S.NumCompileThreads);
+    InitHelperTransformLayer->setCloneToNewContextOnEmit(true);
+    CompileThreads =
+        std::make_unique<ThreadPool>(hardware_concurrency(S.NumCompileThreads));
     ES->setDispatchMaterialization(
-        [this](JITDylib &JD, std::unique_ptr<MaterializationUnit> MU) {
-          // FIXME: Switch to move capture once we have c++14.
+        [this](std::unique_ptr<MaterializationUnit> MU,
+               MaterializationResponsibility MR) {
+          // FIXME: Switch to move capture once ThreadPool uses unique_function.
           auto SharedMU = std::shared_ptr<MaterializationUnit>(std::move(MU));
-          auto Work = [SharedMU, &JD]() { SharedMU->doMaterialize(JD); };
+          auto SharedMR =
+              std::make_shared<MaterializationResponsibility>(std::move(MR));
+          auto Work = [SharedMU, SharedMR]() mutable {
+            SharedMU->materialize(std::move(*SharedMR));
+          };
           CompileThreads->async(std::move(Work));
         });
   }
+
+  if (S.SetUpPlatform)
+    Err = S.SetUpPlatform(*this);
+  else
+    setUpGenericLLVMIRPlatform(*this);
 }
 
-std::string LLJIT::mangle(StringRef UnmangledName) {
+std::string LLJIT::mangle(StringRef UnmangledName) const {
   std::string MangledName;
   {
     raw_string_ostream MangledNameStream(MangledName);
@@ -179,15 +1135,27 @@ Error LLJIT::applyDataLayout(Module &M) {
 
   if (M.getDataLayout() != DL)
     return make_error<StringError>(
-        "Added modules have incompatible data layouts",
+        "Added modules have incompatible data layouts: " +
+            M.getDataLayout().getStringRepresentation() + " (module) vs " +
+            DL.getStringRepresentation() + " (jit)",
         inconvertibleErrorCode());
 
   return Error::success();
 }
 
-void LLJIT::recordCtorDtors(Module &M) {
-  CtorRunner.add(getConstructors(M));
-  DtorRunner.add(getDestructors(M));
+void setUpGenericLLVMIRPlatform(LLJIT &J) {
+  LLVM_DEBUG(
+      { dbgs() << "Setting up GenericLLVMIRPlatform support for LLJIT\n"; });
+  J.setPlatformSupport(std::make_unique<GenericLLVMIRPlatformSupport>(J));
+}
+
+Error setUpMachOPlatform(LLJIT &J) {
+  LLVM_DEBUG({ dbgs() << "Setting up MachOPlatform support for LLJIT\n"; });
+  auto MP = MachOPlatformSupport::Create(J, J.getMainJITDylib());
+  if (!MP)
+    return MP.takeError();
+  J.setPlatformSupport(std::move(*MP));
+  return Error::success();
 }
 
 Error LLLazyJITBuilderState::prepareForConstruction() {
@@ -200,13 +1168,8 @@ Error LLLazyJITBuilderState::prepareForConstruction() {
 Error LLLazyJIT::addLazyIRModule(JITDylib &JD, ThreadSafeModule TSM) {
   assert(TSM && "Can not add null module");
 
-  if (auto Err = TSM.withModuleDo([&](Module &M) -> Error {
-        if (auto Err = applyDataLayout(M))
-          return Err;
-
-        recordCtorDtors(M);
-        return Error::success();
-      }))
+  if (auto Err = TSM.withModuleDo(
+          [&](Module &M) -> Error { return applyDataLayout(M); }))
     return Err;
 
   return CODLayer->add(JD, std::move(TSM), ES->allocateVModule());
@@ -249,12 +1212,9 @@ LLLazyJIT::LLLazyJIT(LLLazyJITBuilderState &S, Error &Err) : LLJIT(S, Err) {
     return;
   }
 
-  // Create the transform layer.
-  TransformLayer = std::make_unique<IRTransformLayer>(*ES, *CompileLayer);
-
   // Create the COD layer.
   CODLayer = std::make_unique<CompileOnDemandLayer>(
-      *ES, *TransformLayer, *LCTMgr, std::move(ISMBuilder));
+      *ES, *InitHelperTransformLayer, *LCTMgr, std::move(ISMBuilder));
 
   if (S.NumCompileThreads > 0)
     CODLayer->setCloneToNewContextOnEmit(true);
diff --git a/llvm/lib/ExecutionEngine/Orc/Layer.cpp b/llvm/lib/ExecutionEngine/Orc/Layer.cpp
index 580e2682ec8c..61e7ab5ae68b 100644
--- a/llvm/lib/ExecutionEngine/Orc/Layer.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/Layer.cpp
@@ -7,6 +7,11 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ExecutionEngine/Orc/Layer.h"
+
+#include "llvm/ExecutionEngine/Orc/DebugUtils.h"
+#include "llvm/ExecutionEngine/Orc/ExecutionUtils.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/Object/MachO.h"
 #include "llvm/Object/ObjectFile.h"
 #include "llvm/Support/Debug.h"
 
@@ -15,28 +20,79 @@
 namespace llvm {
 namespace orc {
 
-IRLayer::IRLayer(ExecutionSession &ES) : ES(ES) {}
 IRLayer::~IRLayer() {}
 
 Error IRLayer::add(JITDylib &JD, ThreadSafeModule TSM, VModuleKey K) {
   return JD.define(std::make_unique<BasicIRLayerMaterializationUnit>(
-      *this, std::move(K), std::move(TSM)));
+      *this, *getManglingOptions(), std::move(TSM), std::move(K)));
 }
 
-IRMaterializationUnit::IRMaterializationUnit(ExecutionSession &ES,
-                                             ThreadSafeModule TSM, VModuleKey K)
-    : MaterializationUnit(SymbolFlagsMap(), std::move(K)), TSM(std::move(TSM)) {
+IRMaterializationUnit::IRMaterializationUnit(
+    ExecutionSession &ES, const IRSymbolMapper::ManglingOptions &MO,
+    ThreadSafeModule TSM, VModuleKey K)
+    : MaterializationUnit(SymbolFlagsMap(), nullptr, std::move(K)),
+      TSM(std::move(TSM)) {
 
   assert(this->TSM && "Module must not be null");
 
   MangleAndInterner Mangle(ES, this->TSM.getModuleUnlocked()->getDataLayout());
   this->TSM.withModuleDo([&](Module &M) {
     for (auto &G : M.global_values()) {
-      if (G.hasName() && !G.isDeclaration() && !G.hasLocalLinkage() &&
-          !G.hasAvailableExternallyLinkage() && !G.hasAppendingLinkage()) {
-        auto MangledName = Mangle(G.getName());
-        SymbolFlags[MangledName] = JITSymbolFlags::fromGlobalValue(G);
-        SymbolToDefinition[MangledName] = &G;
+      // Skip globals that don't generate symbols.
+
+      if (!G.hasName() || G.isDeclaration() || G.hasLocalLinkage() ||
+          G.hasAvailableExternallyLinkage() || G.hasAppendingLinkage())
+        continue;
+
+      // thread locals generate different symbols depending on whether or not
+      // emulated TLS is enabled.
+      if (G.isThreadLocal() && MO.EmulatedTLS) {
+        auto &GV = cast<GlobalVariable>(G);
+
+        auto Flags = JITSymbolFlags::fromGlobalValue(GV);
+
+        auto EmuTLSV = Mangle(("__emutls_v." + GV.getName()).str());
+        SymbolFlags[EmuTLSV] = Flags;
+        SymbolToDefinition[EmuTLSV] = &GV;
+
+        // If this GV has a non-zero initializer we'll need to emit an
+        // __emutls.t symbol too.
+        if (GV.hasInitializer()) {
+          const auto *InitVal = GV.getInitializer();
+
+          // Skip zero-initializers.
+          if (isa<ConstantAggregateZero>(InitVal))
+            continue;
+          const auto *InitIntValue = dyn_cast<ConstantInt>(InitVal);
+          if (InitIntValue && InitIntValue->isZero())
+            continue;
+
+          auto EmuTLST = Mangle(("__emutls_t." + GV.getName()).str());
+          SymbolFlags[EmuTLST] = Flags;
+        }
+        continue;
+      }
+
+      // Otherwise we just need a normal linker mangling.
+      auto MangledName = Mangle(G.getName());
+      SymbolFlags[MangledName] = JITSymbolFlags::fromGlobalValue(G);
+      SymbolToDefinition[MangledName] = &G;
+    }
+
+    // If we need an init symbol for this module then create one.
+    if (!llvm::empty(getStaticInitGVs(M))) {
+      size_t Counter = 0;
+
+      while (true) {
+        std::string InitSymbolName;
+        raw_string_ostream(InitSymbolName)
+            << "$." << M.getModuleIdentifier() << ".__inits." << Counter++;
+        InitSymbol = ES.intern(InitSymbolName);
+        if (SymbolFlags.count(InitSymbol))
+          continue;
+        SymbolFlags[InitSymbol] =
+            JITSymbolFlags::MaterializationSideEffectsOnly;
+        break;
       }
     }
   });
@@ -44,8 +100,9 @@ IRMaterializationUnit::IRMaterializationUnit(ExecutionSession &ES,
 
 IRMaterializationUnit::IRMaterializationUnit(
     ThreadSafeModule TSM, VModuleKey K, SymbolFlagsMap SymbolFlags,
-    SymbolNameToDefinitionMap SymbolToDefinition)
-    : MaterializationUnit(std::move(SymbolFlags), std::move(K)),
+    SymbolStringPtr InitSymbol, SymbolNameToDefinitionMap SymbolToDefinition)
+    : MaterializationUnit(std::move(SymbolFlags), std::move(InitSymbol),
+                          std::move(K)),
       TSM(std::move(TSM)), SymbolToDefinition(std::move(SymbolToDefinition)) {}
 
 StringRef IRMaterializationUnit::getName() const {
@@ -72,8 +129,9 @@ void IRMaterializationUnit::discard(const JITDylib &JD,
 }
 
 BasicIRLayerMaterializationUnit::BasicIRLayerMaterializationUnit(
-    IRLayer &L, VModuleKey K, ThreadSafeModule TSM)
-    : IRMaterializationUnit(L.getExecutionSession(), std::move(TSM),
+    IRLayer &L, const IRSymbolMapper::ManglingOptions &MO, ThreadSafeModule TSM,
+    VModuleKey K)
+    : IRMaterializationUnit(L.getExecutionSession(), MO, std::move(TSM),
                             std::move(K)),
       L(L), K(std::move(K)) {}
 
@@ -117,22 +175,26 @@ Error ObjectLayer::add(JITDylib &JD, std::unique_ptr<MemoryBuffer> O,
 Expected<std::unique_ptr<BasicObjectLayerMaterializationUnit>>
 BasicObjectLayerMaterializationUnit::Create(ObjectLayer &L, VModuleKey K,
                                             std::unique_ptr<MemoryBuffer> O) {
-  auto SymbolFlags =
-      getObjectSymbolFlags(L.getExecutionSession(), O->getMemBufferRef());
+  auto ObjSymInfo =
+      getObjectSymbolInfo(L.getExecutionSession(), O->getMemBufferRef());
+
+  if (!ObjSymInfo)
+    return ObjSymInfo.takeError();
 
-  if (!SymbolFlags)
-    return SymbolFlags.takeError();
+  auto &SymbolFlags = ObjSymInfo->first;
+  auto &InitSymbol = ObjSymInfo->second;
 
   return std::unique_ptr<BasicObjectLayerMaterializationUnit>(
-      new BasicObjectLayerMaterializationUnit(L, K, std::move(O),
-                                              std::move(*SymbolFlags)));
+      new BasicObjectLayerMaterializationUnit(
+          L, K, std::move(O), std::move(SymbolFlags), std::move(InitSymbol)));
 }
 
 BasicObjectLayerMaterializationUnit::BasicObjectLayerMaterializationUnit(
     ObjectLayer &L, VModuleKey K, std::unique_ptr<MemoryBuffer> O,
-    SymbolFlagsMap SymbolFlags)
-    : MaterializationUnit(std::move(SymbolFlags), std::move(K)), L(L),
-      O(std::move(O)) {}
+    SymbolFlagsMap SymbolFlags, SymbolStringPtr InitSymbol)
+    : MaterializationUnit(std::move(SymbolFlags), std::move(InitSymbol),
+                          std::move(K)),
+      L(L), O(std::move(O)) {}
 
 StringRef BasicObjectLayerMaterializationUnit::getName() const {
   if (O)
@@ -147,38 +209,8 @@ void BasicObjectLayerMaterializationUnit::materialize(
 
 void BasicObjectLayerMaterializationUnit::discard(const JITDylib &JD,
                                                   const SymbolStringPtr &Name) {
-  // FIXME: Support object file level discard. This could be done by building a
-  //        filter to pass to the object layer along with the object itself.
-}
-
-Expected<SymbolFlagsMap> getObjectSymbolFlags(ExecutionSession &ES,
-                                              MemoryBufferRef ObjBuffer) {
-  auto Obj = object::ObjectFile::createObjectFile(ObjBuffer);
-
-  if (!Obj)
-    return Obj.takeError();
-
-  SymbolFlagsMap SymbolFlags;
-  for (auto &Sym : (*Obj)->symbols()) {
-    // Skip symbols not defined in this object file.
-    if (Sym.getFlags() & object::BasicSymbolRef::SF_Undefined)
-      continue;
-
-    // Skip symbols that are not global.
-    if (!(Sym.getFlags() & object::BasicSymbolRef::SF_Global))
-      continue;
-
-    auto Name = Sym.getName();
-    if (!Name)
-      return Name.takeError();
-    auto InternedName = ES.intern(*Name);
-    auto SymFlags = JITSymbolFlags::fromObjectSymbol(Sym);
-    if (!SymFlags)
-      return SymFlags.takeError();
-    SymbolFlags[InternedName] = std::move(*SymFlags);
-  }
-
-  return SymbolFlags;
+  // This is a no-op for object files: Having removed 'Name' from SymbolFlags
+  // the symbol will be dead-stripped by the JIT linker.
 }
 
 } // End namespace orc.
diff --git a/llvm/lib/ExecutionEngine/Orc/LazyReexports.cpp b/llvm/lib/ExecutionEngine/Orc/LazyReexports.cpp
index aab490feb8ea..153f6b80784f 100644
--- a/llvm/lib/ExecutionEngine/Orc/LazyReexports.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/LazyReexports.cpp
@@ -16,70 +16,86 @@
 namespace llvm {
 namespace orc {
 
-void LazyCallThroughManager::NotifyResolvedFunction::anchor() {}
-
 LazyCallThroughManager::LazyCallThroughManager(
-    ExecutionSession &ES, JITTargetAddress ErrorHandlerAddr,
-    std::unique_ptr<TrampolinePool> TP)
-    : ES(ES), ErrorHandlerAddr(ErrorHandlerAddr), TP(std::move(TP)) {}
+    ExecutionSession &ES, JITTargetAddress ErrorHandlerAddr, TrampolinePool *TP)
+    : ES(ES), ErrorHandlerAddr(ErrorHandlerAddr), TP(TP) {}
 
 Expected<JITTargetAddress> LazyCallThroughManager::getCallThroughTrampoline(
     JITDylib &SourceJD, SymbolStringPtr SymbolName,
-    std::shared_ptr<NotifyResolvedFunction> NotifyResolved) {
+    NotifyResolvedFunction NotifyResolved) {
+  assert(TP && "TrampolinePool not set");
+
   std::lock_guard<std::mutex> Lock(LCTMMutex);
   auto Trampoline = TP->getTrampoline();
 
   if (!Trampoline)
     return Trampoline.takeError();
 
-  Reexports[*Trampoline] = std::make_pair(&SourceJD, std::move(SymbolName));
+  Reexports[*Trampoline] = ReexportsEntry{&SourceJD, std::move(SymbolName)};
   Notifiers[*Trampoline] = std::move(NotifyResolved);
   return *Trampoline;
 }
 
-JITTargetAddress
-LazyCallThroughManager::callThroughToSymbol(JITTargetAddress TrampolineAddr) {
-  JITDylib *SourceJD = nullptr;
-  SymbolStringPtr SymbolName;
-
-  {
-    std::lock_guard<std::mutex> Lock(LCTMMutex);
-    auto I = Reexports.find(TrampolineAddr);
-    if (I == Reexports.end())
-      return ErrorHandlerAddr;
-    SourceJD = I->second.first;
-    SymbolName = I->second.second;
-  }
-
-  auto LookupResult = ES.lookup(
-      makeJITDylibSearchOrder(SourceJD, JITDylibLookupFlags::MatchAllSymbols),
-      SymbolName);
-
-  if (!LookupResult) {
-    ES.reportError(LookupResult.takeError());
-    return ErrorHandlerAddr;
-  }
+JITTargetAddress LazyCallThroughManager::reportCallThroughError(Error Err) {
+  ES.reportError(std::move(Err));
+  return ErrorHandlerAddr;
+}
 
-  auto ResolvedAddr = LookupResult->getAddress();
+Expected<LazyCallThroughManager::ReexportsEntry>
+LazyCallThroughManager::findReexport(JITTargetAddress TrampolineAddr) {
+  std::lock_guard<std::mutex> Lock(LCTMMutex);
+  auto I = Reexports.find(TrampolineAddr);
+  if (I == Reexports.end())
+    return createStringError(inconvertibleErrorCode(),
+                             "Missing reexport for trampoline address %p",
+                             TrampolineAddr);
+  return I->second;
+}
 
-  std::shared_ptr<NotifyResolvedFunction> NotifyResolved = nullptr;
+Error LazyCallThroughManager::notifyResolved(JITTargetAddress TrampolineAddr,
+                                             JITTargetAddress ResolvedAddr) {
+  NotifyResolvedFunction NotifyResolved;
   {
     std::lock_guard<std::mutex> Lock(LCTMMutex);
     auto I = Notifiers.find(TrampolineAddr);
     if (I != Notifiers.end()) {
-      NotifyResolved = I->second;
+      NotifyResolved = std::move(I->second);
       Notifiers.erase(I);
     }
   }
 
-  if (NotifyResolved) {
-    if (auto Err = (*NotifyResolved)(*SourceJD, SymbolName, ResolvedAddr)) {
-      ES.reportError(std::move(Err));
-      return ErrorHandlerAddr;
-    }
-  }
+  return NotifyResolved ? NotifyResolved(ResolvedAddr) : Error::success();
+}
 
-  return ResolvedAddr;
+void LazyCallThroughManager::resolveTrampolineLandingAddress(
+    JITTargetAddress TrampolineAddr,
+    NotifyLandingResolvedFunction NotifyLandingResolved) {
+
+  auto Entry = findReexport(TrampolineAddr);
+  if (!Entry)
+    return NotifyLandingResolved(reportCallThroughError(Entry.takeError()));
+
+  ES.lookup(
+      LookupKind::Static,
+      makeJITDylibSearchOrder(Entry->SourceJD,
+                              JITDylibLookupFlags::MatchAllSymbols),
+      SymbolLookupSet({Entry->SymbolName}), SymbolState::Ready,
+      [this, TrampolineAddr, SymbolName = Entry->SymbolName,
+       NotifyLandingResolved = std::move(NotifyLandingResolved)](
+          Expected<SymbolMap> Result) mutable {
+        if (Result) {
+          assert(Result->size() == 1 && "Unexpected result size");
+          assert(Result->count(SymbolName) && "Unexpected result value");
+          JITTargetAddress LandingAddr = (*Result)[SymbolName].getAddress();
+
+          if (auto Err = notifyResolved(TrampolineAddr, LandingAddr))
+            NotifyLandingResolved(reportCallThroughError(std::move(Err)));
+          else
+            NotifyLandingResolved(LandingAddr);
+        } else
+          NotifyLandingResolved(reportCallThroughError(Result.takeError()));
+      },
+      NoDependenciesToRegister);
 }
 
 Expected<std::unique_ptr<LazyCallThroughManager>>
@@ -125,15 +141,9 @@ LazyReexportsMaterializationUnit::LazyReexportsMaterializationUnit(
     LazyCallThroughManager &LCTManager, IndirectStubsManager &ISManager,
     JITDylib &SourceJD, SymbolAliasMap CallableAliases, ImplSymbolMap *SrcJDLoc,
     VModuleKey K)
-    : MaterializationUnit(extractFlags(CallableAliases), std::move(K)),
+    : MaterializationUnit(extractFlags(CallableAliases), nullptr, std::move(K)),
       LCTManager(LCTManager), ISManager(ISManager), SourceJD(SourceJD),
-      CallableAliases(std::move(CallableAliases)),
-      NotifyResolved(LazyCallThroughManager::createNotifyResolvedFunction(
-          [&ISManager](JITDylib &JD, const SymbolStringPtr &SymbolName,
-                       JITTargetAddress ResolvedAddr) {
-            return ISManager.updatePointer(*SymbolName, ResolvedAddr);
-          })),
-      AliaseeTable(SrcJDLoc) {}
+      CallableAliases(std::move(CallableAliases)), AliaseeTable(SrcJDLoc) {}
 
 StringRef LazyReexportsMaterializationUnit::getName() const {
   return "<Lazy Reexports>";
@@ -159,7 +169,11 @@ void LazyReexportsMaterializationUnit::materialize(
   for (auto &Alias : RequestedAliases) {
 
     auto CallThroughTrampoline = LCTManager.getCallThroughTrampoline(
-        SourceJD, Alias.second.Aliasee, NotifyResolved);
+        SourceJD, Alias.second.Aliasee,
+        [&ISManager = this->ISManager,
+         StubSym = Alias.first](JITTargetAddress ResolvedAddr) -> Error {
+          return ISManager.updatePointer(*StubSym, ResolvedAddr);
+        });
 
     if (!CallThroughTrampoline) {
       SourceJD.getExecutionSession().reportError(
diff --git a/llvm/lib/ExecutionEngine/Orc/MachOPlatform.cpp b/llvm/lib/ExecutionEngine/Orc/MachOPlatform.cpp
new file mode 100644
index 000000000000..15c3aa79a2a8
--- /dev/null
+++ b/llvm/lib/ExecutionEngine/Orc/MachOPlatform.cpp
@@ -0,0 +1,506 @@
+//===------ MachOPlatform.cpp - Utilities for executing MachO in Orc ------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ExecutionEngine/Orc/MachOPlatform.h"
+
+#include "llvm/BinaryFormat/MachO.h"
+#include "llvm/ExecutionEngine/Orc/DebugUtils.h"
+#include "llvm/Support/BinaryByteStream.h"
+#include "llvm/Support/Debug.h"
+
+#define DEBUG_TYPE "orc"
+
+namespace {
+
+struct objc_class;
+struct objc_image_info;
+struct objc_object;
+struct objc_selector;
+
+using Class = objc_class *;
+using id = objc_object *;
+using SEL = objc_selector *;
+
+using ObjCMsgSendTy = id (*)(id, SEL, ...);
+using ObjCReadClassPairTy = Class (*)(Class, const objc_image_info *);
+using SelRegisterNameTy = SEL (*)(const char *);
+
+enum class ObjCRegistrationAPI { Uninitialized, Unavailable, Initialized };
+
+ObjCRegistrationAPI ObjCRegistrationAPIState =
+    ObjCRegistrationAPI::Uninitialized;
+ObjCMsgSendTy objc_msgSend = nullptr;
+ObjCReadClassPairTy objc_readClassPair = nullptr;
+SelRegisterNameTy sel_registerName = nullptr;
+
+} // end anonymous namespace
+
+namespace llvm {
+namespace orc {
+
+template <typename FnTy>
+static Error setUpObjCRegAPIFunc(FnTy &Target, sys::DynamicLibrary &LibObjC,
+                                 const char *Name) {
+  if (void *Addr = LibObjC.getAddressOfSymbol(Name))
+    Target = reinterpret_cast<FnTy>(Addr);
+  else
+    return make_error<StringError>(
+        (Twine("Could not find address for ") + Name).str(),
+        inconvertibleErrorCode());
+  return Error::success();
+}
+
+Error enableObjCRegistration(const char *PathToLibObjC) {
+  // If we've already tried to initialize then just bail out.
+  if (ObjCRegistrationAPIState != ObjCRegistrationAPI::Uninitialized)
+    return Error::success();
+
+  ObjCRegistrationAPIState = ObjCRegistrationAPI::Unavailable;
+
+  std::string ErrMsg;
+  auto LibObjC =
+      sys::DynamicLibrary::getPermanentLibrary(PathToLibObjC, &ErrMsg);
+
+  if (!LibObjC.isValid())
+    return make_error<StringError>(std::move(ErrMsg), inconvertibleErrorCode());
+
+  if (auto Err = setUpObjCRegAPIFunc(objc_msgSend, LibObjC, "objc_msgSend"))
+    return Err;
+  if (auto Err = setUpObjCRegAPIFunc(objc_readClassPair, LibObjC,
+                                     "objc_readClassPair"))
+    return Err;
+  if (auto Err =
+          setUpObjCRegAPIFunc(sel_registerName, LibObjC, "sel_registerName"))
+    return Err;
+
+  ObjCRegistrationAPIState = ObjCRegistrationAPI::Initialized;
+  return Error::success();
+}
+
+bool objCRegistrationEnabled() {
+  return ObjCRegistrationAPIState == ObjCRegistrationAPI::Initialized;
+}
+
+void MachOJITDylibInitializers::runModInits() const {
+  for (const auto &ModInit : ModInitSections) {
+    for (uint64_t I = 0; I != ModInit.NumPtrs; ++I) {
+      auto *InitializerAddr = jitTargetAddressToPointer<uintptr_t *>(
+          ModInit.Address + (I * sizeof(uintptr_t)));
+      auto *Initializer =
+          jitTargetAddressToFunction<void (*)()>(*InitializerAddr);
+      Initializer();
+    }
+  }
+}
+
+void MachOJITDylibInitializers::registerObjCSelectors() const {
+  assert(objCRegistrationEnabled() && "ObjC registration not enabled.");
+
+  for (const auto &ObjCSelRefs : ObjCSelRefsSections) {
+    for (uint64_t I = 0; I != ObjCSelRefs.NumPtrs; ++I) {
+      auto SelEntryAddr = ObjCSelRefs.Address + (I * sizeof(uintptr_t));
+      const auto *SelName =
+          *jitTargetAddressToPointer<const char **>(SelEntryAddr);
+      auto Sel = sel_registerName(SelName);
+      *jitTargetAddressToPointer<SEL *>(SelEntryAddr) = Sel;
+    }
+  }
+}
+
+Error MachOJITDylibInitializers::registerObjCClasses() const {
+  assert(objCRegistrationEnabled() && "ObjC registration not enabled.");
+
+  struct ObjCClassCompiled {
+    void *Metaclass;
+    void *Parent;
+    void *Cache1;
+    void *Cache2;
+    void *Data;
+  };
+
+  auto *ImageInfo =
+      jitTargetAddressToPointer<const objc_image_info *>(ObjCImageInfoAddr);
+  auto ClassSelector = sel_registerName("class");
+
+  for (const auto &ObjCClassList : ObjCClassListSections) {
+    for (uint64_t I = 0; I != ObjCClassList.NumPtrs; ++I) {
+      auto ClassPtrAddr = ObjCClassList.Address + (I * sizeof(uintptr_t));
+      auto Cls = *jitTargetAddressToPointer<Class *>(ClassPtrAddr);
+      auto *ClassCompiled =
+          *jitTargetAddressToPointer<ObjCClassCompiled **>(ClassPtrAddr);
+      objc_msgSend(reinterpret_cast<id>(ClassCompiled->Parent), ClassSelector);
+      auto Registered = objc_readClassPair(Cls, ImageInfo);
+
+      // FIXME: Improve diagnostic by reporting the failed class's name.
+      if (Registered != Cls)
+        return make_error<StringError>("Unable to register Objective-C class",
+                                       inconvertibleErrorCode());
+    }
+  }
+  return Error::success();
+}
+
+MachOPlatform::MachOPlatform(
+    ExecutionSession &ES, ObjectLinkingLayer &ObjLinkingLayer,
+    std::unique_ptr<MemoryBuffer> StandardSymbolsObject)
+    : ES(ES), ObjLinkingLayer(ObjLinkingLayer),
+      StandardSymbolsObject(std::move(StandardSymbolsObject)) {
+  ObjLinkingLayer.addPlugin(std::make_unique<InitScraperPlugin>(*this));
+}
+
+Error MachOPlatform::setupJITDylib(JITDylib &JD) {
+  auto ObjBuffer = MemoryBuffer::getMemBuffer(
+      StandardSymbolsObject->getMemBufferRef(), false);
+  return ObjLinkingLayer.add(JD, std::move(ObjBuffer));
+}
+
+Error MachOPlatform::notifyAdding(JITDylib &JD, const MaterializationUnit &MU) {
+  const auto &InitSym = MU.getInitializerSymbol();
+  if (!InitSym)
+    return Error::success();
+
+  RegisteredInitSymbols[&JD].add(InitSym,
+                                 SymbolLookupFlags::WeaklyReferencedSymbol);
+  LLVM_DEBUG({
+    dbgs() << "MachOPlatform: Registered init symbol " << *InitSym << " for MU "
+           << MU.getName() << "\n";
+  });
+  return Error::success();
+}
+
+Error MachOPlatform::notifyRemoving(JITDylib &JD, VModuleKey K) {
+  llvm_unreachable("Not supported yet");
+}
+
+Expected<MachOPlatform::InitializerSequence>
+MachOPlatform::getInitializerSequence(JITDylib &JD) {
+
+  LLVM_DEBUG({
+    dbgs() << "MachOPlatform: Building initializer sequence for "
+           << JD.getName() << "\n";
+  });
+
+  std::vector<JITDylib *> DFSLinkOrder;
+
+  while (true) {
+
+    DenseMap<JITDylib *, SymbolLookupSet> NewInitSymbols;
+
+    ES.runSessionLocked([&]() {
+      DFSLinkOrder = getDFSLinkOrder(JD);
+
+      for (auto *InitJD : DFSLinkOrder) {
+        auto RISItr = RegisteredInitSymbols.find(InitJD);
+        if (RISItr != RegisteredInitSymbols.end()) {
+          NewInitSymbols[InitJD] = std::move(RISItr->second);
+          RegisteredInitSymbols.erase(RISItr);
+        }
+      }
+    });
+
+    if (NewInitSymbols.empty())
+      break;
+
+    LLVM_DEBUG({
+      dbgs() << "MachOPlatform: Issuing lookups for new init symbols: "
+                "(lookup may require multiple rounds)\n";
+      for (auto &KV : NewInitSymbols)
+        dbgs() << "  \"" << KV.first->getName() << "\": " << KV.second << "\n";
+    });
+
+    // Outside the lock, issue the lookup.
+    if (auto R = lookupInitSymbols(JD.getExecutionSession(), NewInitSymbols))
+      ; // Nothing to do in the success case.
+    else
+      return R.takeError();
+  }
+
+  LLVM_DEBUG({
+    dbgs() << "MachOPlatform: Init symbol lookup complete, building init "
+              "sequence\n";
+  });
+
+  // Lock again to collect the initializers.
+  InitializerSequence FullInitSeq;
+  {
+    std::lock_guard<std::mutex> Lock(InitSeqsMutex);
+    for (auto *InitJD : reverse(DFSLinkOrder)) {
+      LLVM_DEBUG({
+        dbgs() << "MachOPlatform: Appending inits for \"" << InitJD->getName()
+               << "\" to sequence\n";
+      });
+      auto ISItr = InitSeqs.find(InitJD);
+      if (ISItr != InitSeqs.end()) {
+        FullInitSeq.emplace_back(InitJD, std::move(ISItr->second));
+        InitSeqs.erase(ISItr);
+      }
+    }
+  }
+
+  return FullInitSeq;
+}
+
+Expected<MachOPlatform::DeinitializerSequence>
+MachOPlatform::getDeinitializerSequence(JITDylib &JD) {
+  std::vector<JITDylib *> DFSLinkOrder = getDFSLinkOrder(JD);
+
+  DeinitializerSequence FullDeinitSeq;
+  {
+    std::lock_guard<std::mutex> Lock(InitSeqsMutex);
+    for (auto *DeinitJD : DFSLinkOrder) {
+      FullDeinitSeq.emplace_back(DeinitJD, MachOJITDylibDeinitializers());
+    }
+  }
+
+  return FullDeinitSeq;
+}
+
+std::vector<JITDylib *> MachOPlatform::getDFSLinkOrder(JITDylib &JD) {
+  std::vector<JITDylib *> Result, WorkStack({&JD});
+  DenseSet<JITDylib *> Visited;
+
+  while (!WorkStack.empty()) {
+    auto *NextJD = WorkStack.back();
+    WorkStack.pop_back();
+    if (Visited.count(NextJD))
+      continue;
+    Visited.insert(NextJD);
+    Result.push_back(NextJD);
+    NextJD->withLinkOrderDo([&](const JITDylibSearchOrder &LO) {
+      for (auto &KV : LO)
+        WorkStack.push_back(KV.first);
+    });
+  }
+
+  return Result;
+}
+
+void MachOPlatform::registerInitInfo(
+    JITDylib &JD, JITTargetAddress ObjCImageInfoAddr,
+    MachOJITDylibInitializers::SectionExtent ModInits,
+    MachOJITDylibInitializers::SectionExtent ObjCSelRefs,
+    MachOJITDylibInitializers::SectionExtent ObjCClassList) {
+  std::lock_guard<std::mutex> Lock(InitSeqsMutex);
+
+  auto &InitSeq = InitSeqs[&JD];
+
+  InitSeq.setObjCImageInfoAddr(ObjCImageInfoAddr);
+
+  if (ModInits.Address)
+    InitSeq.addModInitsSection(std::move(ModInits));
+
+  if (ObjCSelRefs.Address)
+    InitSeq.addObjCSelRefsSection(std::move(ObjCSelRefs));
+
+  if (ObjCClassList.Address)
+    InitSeq.addObjCClassListSection(std::move(ObjCClassList));
+}
+
+static Expected<MachOJITDylibInitializers::SectionExtent>
+getSectionExtent(jitlink::LinkGraph &G, StringRef SectionName) {
+  auto *Sec = G.findSectionByName(SectionName);
+  if (!Sec)
+    return MachOJITDylibInitializers::SectionExtent();
+  jitlink::SectionRange R(*Sec);
+  if (R.getSize() % G.getPointerSize() != 0)
+    return make_error<StringError>(SectionName + " section size is not a "
+                                                 "multiple of the pointer size",
+                                   inconvertibleErrorCode());
+  return MachOJITDylibInitializers::SectionExtent(
+      R.getStart(), R.getSize() / G.getPointerSize());
+}
+
+void MachOPlatform::InitScraperPlugin::modifyPassConfig(
+    MaterializationResponsibility &MR, const Triple &TT,
+    jitlink::PassConfiguration &Config) {
+
+  Config.PrePrunePasses.push_back([this, &MR](jitlink::LinkGraph &G) -> Error {
+    JITLinkSymbolVector InitSectionSymbols;
+    preserveInitSectionIfPresent(InitSectionSymbols, G, "__mod_init_func");
+    preserveInitSectionIfPresent(InitSectionSymbols, G, "__objc_selrefs");
+    preserveInitSectionIfPresent(InitSectionSymbols, G, "__objc_classlist");
+
+    if (!InitSymbolDeps.empty()) {
+      std::lock_guard<std::mutex> Lock(InitScraperMutex);
+      InitSymbolDeps[&MR] = std::move(InitSectionSymbols);
+    }
+
+    if (auto Err = processObjCImageInfo(G, MR))
+      return Err;
+
+    return Error::success();
+  });
+
+  Config.PostFixupPasses.push_back([this, &JD = MR.getTargetJITDylib()](
+                                       jitlink::LinkGraph &G) -> Error {
+    MachOJITDylibInitializers::SectionExtent ModInits, ObjCSelRefs,
+        ObjCClassList;
+
+    JITTargetAddress ObjCImageInfoAddr = 0;
+    if (auto *ObjCImageInfoSec = G.findSectionByName("__objc_image_info")) {
+      if (auto Addr = jitlink::SectionRange(*ObjCImageInfoSec).getStart()) {
+        ObjCImageInfoAddr = Addr;
+        dbgs() << "Recorded __objc_imageinfo @ " << formatv("{0:x16}", Addr);
+      }
+    }
+
+    // Record __mod_init_func.
+    if (auto ModInitsOrErr = getSectionExtent(G, "__mod_init_func"))
+      ModInits = std::move(*ModInitsOrErr);
+    else
+      return ModInitsOrErr.takeError();
+
+    // Record __objc_selrefs.
+    if (auto ObjCSelRefsOrErr = getSectionExtent(G, "__objc_selrefs"))
+      ObjCSelRefs = std::move(*ObjCSelRefsOrErr);
+    else
+      return ObjCSelRefsOrErr.takeError();
+
+    // Record __objc_classlist.
+    if (auto ObjCClassListOrErr = getSectionExtent(G, "__objc_classlist"))
+      ObjCClassList = std::move(*ObjCClassListOrErr);
+    else
+      return ObjCClassListOrErr.takeError();
+
+    // Dump the scraped inits.
+    LLVM_DEBUG({
+      dbgs() << "MachOPlatform: Scraped " << G.getName() << " init sections:\n";
+      dbgs() << "  __objc_selrefs: ";
+      if (ObjCSelRefs.NumPtrs)
+        dbgs() << ObjCSelRefs.NumPtrs << " pointer(s) at "
+               << formatv("{0:x16}", ObjCSelRefs.Address) << "\n";
+      else
+        dbgs() << "none\n";
+
+      dbgs() << "  __objc_classlist: ";
+      if (ObjCClassList.NumPtrs)
+        dbgs() << ObjCClassList.NumPtrs << " pointer(s) at "
+               << formatv("{0:x16}", ObjCClassList.Address) << "\n";
+      else
+        dbgs() << "none\n";
+
+      dbgs() << "  __mod_init_func: ";
+      if (ModInits.NumPtrs)
+        dbgs() << ModInits.NumPtrs << " pointer(s) at "
+               << formatv("{0:x16}", ModInits.Address) << "\n";
+      else
+        dbgs() << "none\n";
+    });
+
+    MP.registerInitInfo(JD, ObjCImageInfoAddr, std::move(ModInits),
+                        std::move(ObjCSelRefs), std::move(ObjCClassList));
+
+    return Error::success();
+  });
+}
+
+ObjectLinkingLayer::Plugin::LocalDependenciesMap
+MachOPlatform::InitScraperPlugin::getSyntheticSymbolLocalDependencies(
+    MaterializationResponsibility &MR) {
+  std::lock_guard<std::mutex> Lock(InitScraperMutex);
+  auto I = InitSymbolDeps.find(&MR);
+  if (I != InitSymbolDeps.end()) {
+    LocalDependenciesMap Result;
+    Result[MR.getInitializerSymbol()] = std::move(I->second);
+    InitSymbolDeps.erase(&MR);
+    return Result;
+  }
+  return LocalDependenciesMap();
+}
+
+void MachOPlatform::InitScraperPlugin::preserveInitSectionIfPresent(
+    JITLinkSymbolVector &Symbols, jitlink::LinkGraph &G,
+    StringRef SectionName) {
+  if (auto *Sec = G.findSectionByName(SectionName)) {
+    auto SecBlocks = Sec->blocks();
+    if (!llvm::empty(SecBlocks))
+      Symbols.push_back(
+          &G.addAnonymousSymbol(**SecBlocks.begin(), 0, 0, false, true));
+  }
+}
+
+Error MachOPlatform::InitScraperPlugin::processObjCImageInfo(
+    jitlink::LinkGraph &G, MaterializationResponsibility &MR) {
+
+  // If there's an ObjC imagine info then either
+  //   (1) It's the first __objc_imageinfo we've seen in this JITDylib. In
+  //       this case we name and record it.
+  // OR
+  //   (2) We already have a recorded __objc_imageinfo for this JITDylib,
+  //       in which case we just verify it.
+  auto *ObjCImageInfo = G.findSectionByName("__objc_imageinfo");
+  if (!ObjCImageInfo)
+    return Error::success();
+
+  auto ObjCImageInfoBlocks = ObjCImageInfo->blocks();
+
+  // Check that the section is not empty if present.
+  if (llvm::empty(ObjCImageInfoBlocks))
+    return make_error<StringError>("Empty __objc_imageinfo section in " +
+                                       G.getName(),
+                                   inconvertibleErrorCode());
+
+  // Check that there's only one block in the section.
+  if (std::next(ObjCImageInfoBlocks.begin()) != ObjCImageInfoBlocks.end())
+    return make_error<StringError>("Multiple blocks in __objc_imageinfo "
+                                   "section in " +
+                                       G.getName(),
+                                   inconvertibleErrorCode());
+
+  // Check that the __objc_imageinfo section is unreferenced.
+  // FIXME: We could optimize this check if Symbols had a ref-count.
+  for (auto &Sec : G.sections()) {
+    if (&Sec != ObjCImageInfo)
+      for (auto *B : Sec.blocks())
+        for (auto &E : B->edges())
+          if (E.getTarget().isDefined() &&
+              &E.getTarget().getBlock().getSection() == ObjCImageInfo)
+            return make_error<StringError>("__objc_imageinfo is referenced "
+                                           "within file " +
+                                               G.getName(),
+                                           inconvertibleErrorCode());
+  }
+
+  auto &ObjCImageInfoBlock = **ObjCImageInfoBlocks.begin();
+  auto *ObjCImageInfoData = ObjCImageInfoBlock.getContent().data();
+  auto Version = support::endian::read32(ObjCImageInfoData, G.getEndianness());
+  auto Flags =
+      support::endian::read32(ObjCImageInfoData + 4, G.getEndianness());
+
+  // Lock the mutex while we verify / update the ObjCImageInfos map.
+  std::lock_guard<std::mutex> Lock(InitScraperMutex);
+
+  auto ObjCImageInfoItr = ObjCImageInfos.find(&MR.getTargetJITDylib());
+  if (ObjCImageInfoItr != ObjCImageInfos.end()) {
+    // We've already registered an __objc_imageinfo section. Verify the
+    // content of this new section matches, then delete it.
+    if (ObjCImageInfoItr->second.first != Version)
+      return make_error<StringError>(
+          "ObjC version in " + G.getName() +
+              " does not match first registered version",
+          inconvertibleErrorCode());
+    if (ObjCImageInfoItr->second.second != Flags)
+      return make_error<StringError>("ObjC flags in " + G.getName() +
+                                         " do not match first registered flags",
+                                     inconvertibleErrorCode());
+
+    // __objc_imageinfo is valid. Delete the block.
+    for (auto *S : ObjCImageInfo->symbols())
+      G.removeDefinedSymbol(*S);
+    G.removeBlock(ObjCImageInfoBlock);
+  } else {
+    // We haven't registered an __objc_imageinfo section yet. Register and
+    // move on. The section should already be marked no-dead-strip.
+    ObjCImageInfos[&MR.getTargetJITDylib()] = std::make_pair(Version, Flags);
+  }
+
+  return Error::success();
+}
+
+} // End namespace orc.
+} // End namespace llvm.
diff --git a/llvm/lib/ExecutionEngine/Orc/Mangling.cpp b/llvm/lib/ExecutionEngine/Orc/Mangling.cpp
new file mode 100644
index 000000000000..606304741cf7
--- /dev/null
+++ b/llvm/lib/ExecutionEngine/Orc/Mangling.cpp
@@ -0,0 +1,160 @@
+//===----------- Mangling.cpp -- Name Mangling Utilities for ORC ----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ExecutionEngine/Orc/Mangling.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/Mangler.h"
+#include "llvm/Object/MachO.h"
+#include "llvm/Object/ObjectFile.h"
+#include "llvm/Support/Debug.h"
+
+#define DEBUG_TYPE "orc"
+
+namespace llvm {
+namespace orc {
+
+MangleAndInterner::MangleAndInterner(ExecutionSession &ES, const DataLayout &DL)
+    : ES(ES), DL(DL) {}
+
+SymbolStringPtr MangleAndInterner::operator()(StringRef Name) {
+  std::string MangledName;
+  {
+    raw_string_ostream MangledNameStream(MangledName);
+    Mangler::getNameWithPrefix(MangledNameStream, Name, DL);
+  }
+  return ES.intern(MangledName);
+}
+
+void IRSymbolMapper::add(ExecutionSession &ES, const ManglingOptions &MO,
+                         ArrayRef<GlobalValue *> GVs,
+                         SymbolFlagsMap &SymbolFlags,
+                         SymbolNameToDefinitionMap *SymbolToDefinition) {
+  if (GVs.empty())
+    return;
+
+  MangleAndInterner Mangle(ES, GVs[0]->getParent()->getDataLayout());
+  for (auto *G : GVs) {
+    assert(G && "GVs cannot contain null elements");
+    if (!G->hasName() || G->isDeclaration() || G->hasLocalLinkage() ||
+        G->hasAvailableExternallyLinkage() || G->hasAppendingLinkage())
+      continue;
+
+    if (G->isThreadLocal() && MO.EmulatedTLS) {
+      auto *GV = cast<GlobalVariable>(G);
+
+      auto Flags = JITSymbolFlags::fromGlobalValue(*GV);
+
+      auto EmuTLSV = Mangle(("__emutls_v." + GV->getName()).str());
+      SymbolFlags[EmuTLSV] = Flags;
+      if (SymbolToDefinition)
+        (*SymbolToDefinition)[EmuTLSV] = GV;
+
+      // If this GV has a non-zero initializer we'll need to emit an
+      // __emutls.t symbol too.
+      if (GV->hasInitializer()) {
+        const auto *InitVal = GV->getInitializer();
+
+        // Skip zero-initializers.
+        if (isa<ConstantAggregateZero>(InitVal))
+          continue;
+        const auto *InitIntValue = dyn_cast<ConstantInt>(InitVal);
+        if (InitIntValue && InitIntValue->isZero())
+          continue;
+
+        auto EmuTLST = Mangle(("__emutls_t." + GV->getName()).str());
+        SymbolFlags[EmuTLST] = Flags;
+        if (SymbolToDefinition)
+          (*SymbolToDefinition)[EmuTLST] = GV;
+      }
+      continue;
+    }
+
+    // Otherwise we just need a normal linker mangling.
+    auto MangledName = Mangle(G->getName());
+    SymbolFlags[MangledName] = JITSymbolFlags::fromGlobalValue(*G);
+    if (SymbolToDefinition)
+      (*SymbolToDefinition)[MangledName] = G;
+  }
+}
+
+Expected<std::pair<SymbolFlagsMap, SymbolStringPtr>>
+getObjectSymbolInfo(ExecutionSession &ES, MemoryBufferRef ObjBuffer) {
+  auto Obj = object::ObjectFile::createObjectFile(ObjBuffer);
+
+  if (!Obj)
+    return Obj.takeError();
+
+  bool IsMachO = isa<object::MachOObjectFile>(Obj->get());
+
+  SymbolFlagsMap SymbolFlags;
+  for (auto &Sym : (*Obj)->symbols()) {
+    Expected<uint32_t> SymFlagsOrErr = Sym.getFlags();
+    if (!SymFlagsOrErr)
+      // TODO: Test this error.
+      return SymFlagsOrErr.takeError();
+
+    // Skip symbols not defined in this object file.
+    if (*SymFlagsOrErr & object::BasicSymbolRef::SF_Undefined)
+      continue;
+
+    // Skip symbols that are not global.
+    if (!(*SymFlagsOrErr & object::BasicSymbolRef::SF_Global))
+      continue;
+
+    // Skip symbols that have type SF_File.
+    if (auto SymType = Sym.getType()) {
+      if (*SymType == object::SymbolRef::ST_File)
+        continue;
+    } else
+      return SymType.takeError();
+
+    auto Name = Sym.getName();
+    if (!Name)
+      return Name.takeError();
+    auto InternedName = ES.intern(*Name);
+    auto SymFlags = JITSymbolFlags::fromObjectSymbol(Sym);
+    if (!SymFlags)
+      return SymFlags.takeError();
+
+    // Strip the 'exported' flag from MachO linker-private symbols.
+    if (IsMachO && Name->startswith("l"))
+      *SymFlags &= ~JITSymbolFlags::Exported;
+
+    SymbolFlags[InternedName] = std::move(*SymFlags);
+  }
+
+  SymbolStringPtr InitSymbol;
+
+  if (IsMachO) {
+    auto &MachOObj = cast<object::MachOObjectFile>(*Obj->get());
+    for (auto &Sec : MachOObj.sections()) {
+      auto SecType = MachOObj.getSectionType(Sec);
+      if ((SecType & MachO::SECTION_TYPE) == MachO::S_MOD_INIT_FUNC_POINTERS) {
+        size_t Counter = 0;
+        while (true) {
+          std::string InitSymString;
+          raw_string_ostream(InitSymString)
+              << "$." << ObjBuffer.getBufferIdentifier() << ".__inits."
+              << Counter++;
+          InitSymbol = ES.intern(InitSymString);
+          if (SymbolFlags.count(InitSymbol))
+            continue;
+          SymbolFlags[InitSymbol] =
+              JITSymbolFlags::MaterializationSideEffectsOnly;
+          break;
+        }
+        break;
+      }
+    }
+  }
+
+  return std::make_pair(std::move(SymbolFlags), std::move(InitSymbol));
+}
+
+} // End namespace orc.
+} // End namespace llvm.
diff --git a/llvm/lib/ExecutionEngine/Orc/ObjectLinkingLayer.cpp b/llvm/lib/ExecutionEngine/Orc/ObjectLinkingLayer.cpp
index 2572b7f4878d..02066b458dfc 100644
--- a/llvm/lib/ExecutionEngine/Orc/ObjectLinkingLayer.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/ObjectLinkingLayer.cpp
@@ -50,9 +50,9 @@ public:
   void lookup(const LookupMap &Symbols,
               std::unique_ptr<JITLinkAsyncLookupContinuation> LC) override {
 
-    JITDylibSearchOrder SearchOrder;
-    MR.getTargetJITDylib().withSearchOrderDo(
-        [&](const JITDylibSearchOrder &O) { SearchOrder = O; });
+    JITDylibSearchOrder LinkOrder;
+    MR.getTargetJITDylib().withLinkOrderDo(
+        [&](const JITDylibSearchOrder &LO) { LinkOrder = LO; });
 
     auto &ES = Layer.getExecutionSession();
 
@@ -84,7 +84,13 @@ public:
       }
     };
 
-    ES.lookup(LookupKind::Static, SearchOrder, std::move(LookupSet),
+    for (auto &KV : InternalNamedSymbolDeps) {
+      SymbolDependenceMap InternalDeps;
+      InternalDeps[&MR.getTargetJITDylib()] = std::move(KV.second);
+      MR.addDependencies(KV.first, InternalDeps);
+    }
+
+    ES.lookup(LookupKind::Static, LinkOrder, std::move(LookupSet),
               SymbolState::Resolved, std::move(OnResolve),
               [this](const SymbolDependenceMap &Deps) {
                 registerDependencies(Deps);
@@ -138,6 +144,56 @@ public:
     if (!ExtraSymbolsToClaim.empty())
       if (auto Err = MR.defineMaterializing(ExtraSymbolsToClaim))
         return notifyFailed(std::move(Err));
+
+    {
+
+      // Check that InternedResult matches up with MR.getSymbols().
+      // This guards against faulty transformations / compilers / object caches.
+
+      // First check that there aren't any missing symbols.
+      size_t NumMaterializationSideEffectsOnlySymbols = 0;
+      SymbolNameVector ExtraSymbols;
+      SymbolNameVector MissingSymbols;
+      for (auto &KV : MR.getSymbols()) {
+
+        // If this is a materialization-side-effects only symbol then bump
+        // the counter and make sure it's *not* defined, otherwise make
+        // sure that it is defined.
+        if (KV.second.hasMaterializationSideEffectsOnly()) {
+          ++NumMaterializationSideEffectsOnlySymbols;
+          if (InternedResult.count(KV.first))
+            ExtraSymbols.push_back(KV.first);
+          continue;
+        } else if (!InternedResult.count(KV.first))
+          MissingSymbols.push_back(KV.first);
+      }
+
+      // If there were missing symbols then report the error.
+      if (!MissingSymbols.empty()) {
+        ES.reportError(make_error<MissingSymbolDefinitions>(
+            G.getName(), std::move(MissingSymbols)));
+        MR.failMaterialization();
+        return;
+      }
+
+      // If there are more definitions than expected, add them to the
+      // ExtraSymbols vector.
+      if (InternedResult.size() >
+          MR.getSymbols().size() - NumMaterializationSideEffectsOnlySymbols) {
+        for (auto &KV : InternedResult)
+          if (!MR.getSymbols().count(KV.first))
+            ExtraSymbols.push_back(KV.first);
+      }
+
+      // If there were extra definitions then report the error.
+      if (!ExtraSymbols.empty()) {
+        ES.reportError(make_error<UnexpectedSymbolDefinitions>(
+            G.getName(), std::move(ExtraSymbols)));
+        MR.failMaterialization();
+        return;
+      }
+    }
+
     if (auto Err = MR.notifyResolved(InternedResult)) {
       Layer.getExecutionSession().reportError(std::move(Err));
       MR.failMaterialization();
@@ -168,16 +224,22 @@ public:
     // link graph to build the symbol dependence graph.
     Config.PrePrunePasses.push_back(
         [this](LinkGraph &G) { return externalizeWeakAndCommonSymbols(G); });
-    Config.PostPrunePasses.push_back(
-        [this](LinkGraph &G) { return computeNamedSymbolDependencies(G); });
 
     Layer.modifyPassConfig(MR, TT, Config);
 
+    Config.PostPrunePasses.push_back(
+        [this](LinkGraph &G) { return computeNamedSymbolDependencies(G); });
+
     return Error::success();
   }
 
 private:
-  using AnonToNamedDependenciesMap = DenseMap<const Symbol *, SymbolNameSet>;
+  struct LocalSymbolNamedDependencies {
+    SymbolNameSet Internal, External;
+  };
+
+  using LocalSymbolNamedDependenciesMap =
+      DenseMap<const Symbol *, LocalSymbolNamedDependencies>;
 
   Error externalizeWeakAndCommonSymbols(LinkGraph &G) {
     auto &ES = Layer.getExecutionSession();
@@ -206,30 +268,69 @@ private:
 
   Error computeNamedSymbolDependencies(LinkGraph &G) {
     auto &ES = MR.getTargetJITDylib().getExecutionSession();
-    auto AnonDeps = computeAnonDeps(G);
+    auto LocalDeps = computeLocalDeps(G);
 
+    // Compute dependencies for symbols defined in the JITLink graph.
     for (auto *Sym : G.defined_symbols()) {
 
-      // Skip anonymous and non-global atoms: we do not need dependencies for
-      // these.
+      // Skip local symbols: we do not track dependencies for these.
       if (Sym->getScope() == Scope::Local)
         continue;
+      assert(Sym->hasName() &&
+             "Defined non-local jitlink::Symbol should have a name");
 
-      auto SymName = ES.intern(Sym->getName());
-      SymbolNameSet &SymDeps = NamedSymbolDeps[SymName];
+      SymbolNameSet ExternalSymDeps, InternalSymDeps;
 
+      // Find internal and external named symbol dependencies.
       for (auto &E : Sym->getBlock().edges()) {
         auto &TargetSym = E.getTarget();
 
-        if (TargetSym.getScope() != Scope::Local)
-          SymDeps.insert(ES.intern(TargetSym.getName()));
-        else {
+        if (TargetSym.getScope() != Scope::Local) {
+          if (TargetSym.isExternal())
+            ExternalSymDeps.insert(ES.intern(TargetSym.getName()));
+          else if (&TargetSym != Sym)
+            InternalSymDeps.insert(ES.intern(TargetSym.getName()));
+        } else {
           assert(TargetSym.isDefined() &&
-                 "Anonymous/local symbols must be defined");
-          auto I = AnonDeps.find(&TargetSym);
-          if (I != AnonDeps.end())
-            for (auto &S : I->second)
-              SymDeps.insert(S);
+                 "local symbols must be defined");
+          auto I = LocalDeps.find(&TargetSym);
+          if (I != LocalDeps.end()) {
+            for (auto &S : I->second.External)
+              ExternalSymDeps.insert(S);
+            for (auto &S : I->second.Internal)
+              InternalSymDeps.insert(S);
+          }
+        }
+      }
+
+      if (ExternalSymDeps.empty() && InternalSymDeps.empty())
+        continue;
+
+      auto SymName = ES.intern(Sym->getName());
+      if (!ExternalSymDeps.empty())
+        ExternalNamedSymbolDeps[SymName] = std::move(ExternalSymDeps);
+      if (!InternalSymDeps.empty())
+        InternalNamedSymbolDeps[SymName] = std::move(InternalSymDeps);
+    }
+
+    for (auto &P : Layer.Plugins) {
+      auto SyntheticLocalDeps = P->getSyntheticSymbolLocalDependencies(MR);
+      if (SyntheticLocalDeps.empty())
+        continue;
+
+      for (auto &KV : SyntheticLocalDeps) {
+        auto &Name = KV.first;
+        auto &LocalDepsForName = KV.second;
+        for (auto *Local : LocalDepsForName) {
+          assert(Local->getScope() == Scope::Local &&
+                 "Dependence on non-local symbol");
+          auto LocalNamedDepsItr = LocalDeps.find(Local);
+          if (LocalNamedDepsItr == LocalDeps.end())
+            continue;
+          for (auto &S : LocalNamedDepsItr->second.Internal)
+            InternalNamedSymbolDeps[Name].insert(S);
+          for (auto &S : LocalNamedDepsItr->second.External)
+            ExternalNamedSymbolDeps[Name].insert(S);
         }
       }
     }
@@ -237,68 +338,85 @@ private:
     return Error::success();
   }
 
-  AnonToNamedDependenciesMap computeAnonDeps(LinkGraph &G) {
+  LocalSymbolNamedDependenciesMap computeLocalDeps(LinkGraph &G) {
+    DenseMap<jitlink::Symbol *, DenseSet<jitlink::Symbol *>> DepMap;
 
-    auto &ES = MR.getTargetJITDylib().getExecutionSession();
-    AnonToNamedDependenciesMap DepMap;
-
-    // For all anonymous symbols:
+    // For all local symbols:
     // (1) Add their named dependencies.
     // (2) Add them to the worklist for further iteration if they have any
-    //     depend on any other anonymous symbols.
+    //     depend on any other local symbols.
     struct WorklistEntry {
-      WorklistEntry(Symbol *Sym, DenseSet<Symbol *> SymAnonDeps)
-          : Sym(Sym), SymAnonDeps(std::move(SymAnonDeps)) {}
+      WorklistEntry(Symbol *Sym, DenseSet<Symbol *> LocalDeps)
+          : Sym(Sym), LocalDeps(std::move(LocalDeps)) {}
 
       Symbol *Sym = nullptr;
-      DenseSet<Symbol *> SymAnonDeps;
+      DenseSet<Symbol *> LocalDeps;
     };
     std::vector<WorklistEntry> Worklist;
     for (auto *Sym : G.defined_symbols())
-      if (!Sym->hasName()) {
+      if (Sym->getScope() == Scope::Local) {
         auto &SymNamedDeps = DepMap[Sym];
-        DenseSet<Symbol *> SymAnonDeps;
+        DenseSet<Symbol *> LocalDeps;
 
         for (auto &E : Sym->getBlock().edges()) {
           auto &TargetSym = E.getTarget();
-          if (TargetSym.hasName())
-            SymNamedDeps.insert(ES.intern(TargetSym.getName()));
+          if (TargetSym.getScope() != Scope::Local)
+            SymNamedDeps.insert(&TargetSym);
           else {
             assert(TargetSym.isDefined() &&
-                   "Anonymous symbols must be defined");
-            SymAnonDeps.insert(&TargetSym);
+                   "local symbols must be defined");
+            LocalDeps.insert(&TargetSym);
           }
         }
 
-        if (!SymAnonDeps.empty())
-          Worklist.push_back(WorklistEntry(Sym, std::move(SymAnonDeps)));
+        if (!LocalDeps.empty())
+          Worklist.push_back(WorklistEntry(Sym, std::move(LocalDeps)));
       }
 
-    // Loop over all anonymous symbols with anonymous dependencies, propagating
-    // their respective *named* dependencies. Iterate until we hit a stable
+    // Loop over all local symbols with local dependencies, propagating
+    // their respective non-local dependencies. Iterate until we hit a stable
     // state.
     bool Changed;
     do {
       Changed = false;
       for (auto &WLEntry : Worklist) {
         auto *Sym = WLEntry.Sym;
-        auto &SymNamedDeps = DepMap[Sym];
-        auto &SymAnonDeps = WLEntry.SymAnonDeps;
+        auto &NamedDeps = DepMap[Sym];
+        auto &LocalDeps = WLEntry.LocalDeps;
 
-        for (auto *TargetSym : SymAnonDeps) {
+        for (auto *TargetSym : LocalDeps) {
           auto I = DepMap.find(TargetSym);
           if (I != DepMap.end())
             for (const auto &S : I->second)
-              Changed |= SymNamedDeps.insert(S).second;
+              Changed |= NamedDeps.insert(S).second;
         }
       }
     } while (Changed);
 
-    return DepMap;
+    // Intern the results to produce a mapping of jitlink::Symbol* to internal
+    // and external symbol names.
+    auto &ES = Layer.getExecutionSession();
+    LocalSymbolNamedDependenciesMap Result;
+    for (auto &KV : DepMap) {
+      auto *Local = KV.first;
+      assert(Local->getScope() == Scope::Local &&
+             "DepMap keys should all be local symbols");
+      auto &LocalNamedDeps = Result[Local];
+      for (auto *Named : KV.second) {
+        assert(Named->getScope() != Scope::Local &&
+               "DepMap values should all be non-local symbol sets");
+        if (Named->isExternal())
+          LocalNamedDeps.External.insert(ES.intern(Named->getName()));
+        else
+          LocalNamedDeps.Internal.insert(ES.intern(Named->getName()));
+      }
+    }
+
+    return Result;
   }
 
   void registerDependencies(const SymbolDependenceMap &QueryDeps) {
-    for (auto &NamedDepsEntry : NamedSymbolDeps) {
+    for (auto &NamedDepsEntry : ExternalNamedSymbolDeps) {
       auto &Name = NamedDepsEntry.first;
       auto &NameDeps = NamedDepsEntry.second;
       SymbolDependenceMap SymbolDeps;
@@ -323,7 +441,8 @@ private:
   ObjectLinkingLayer &Layer;
   MaterializationResponsibility MR;
   std::unique_ptr<MemoryBuffer> ObjBuffer;
-  DenseMap<SymbolStringPtr, SymbolNameSet> NamedSymbolDeps;
+  DenseMap<SymbolStringPtr, SymbolNameSet> ExternalNamedSymbolDeps;
+  DenseMap<SymbolStringPtr, SymbolNameSet> InternalNamedSymbolDeps;
 };
 
 ObjectLinkingLayer::Plugin::~Plugin() {}
@@ -426,18 +545,21 @@ EHFrameRegistrationPlugin::EHFrameRegistrationPlugin(
 void EHFrameRegistrationPlugin::modifyPassConfig(
     MaterializationResponsibility &MR, const Triple &TT,
     PassConfiguration &PassConfig) {
-  assert(!InProcessLinks.count(&MR) && "Link for MR already being tracked?");
 
-  PassConfig.PostFixupPasses.push_back(
-      createEHFrameRecorderPass(TT, [this, &MR](JITTargetAddress Addr,
-                                                size_t Size) {
-        if (Addr)
-          InProcessLinks[&MR] = { Addr, Size };
+  PassConfig.PostFixupPasses.push_back(createEHFrameRecorderPass(
+      TT, [this, &MR](JITTargetAddress Addr, size_t Size) {
+        if (Addr) {
+          std::lock_guard<std::mutex> Lock(EHFramePluginMutex);
+          assert(!InProcessLinks.count(&MR) &&
+                 "Link for MR already being tracked?");
+          InProcessLinks[&MR] = {Addr, Size};
+        }
       }));
 }
 
 Error EHFrameRegistrationPlugin::notifyEmitted(
     MaterializationResponsibility &MR) {
+  std::lock_guard<std::mutex> Lock(EHFramePluginMutex);
 
   auto EHFrameRangeItr = InProcessLinks.find(&MR);
   if (EHFrameRangeItr == InProcessLinks.end())
@@ -457,6 +579,8 @@ Error EHFrameRegistrationPlugin::notifyEmitted(
 }
 
 Error EHFrameRegistrationPlugin::notifyRemovingModule(VModuleKey K) {
+  std::lock_guard<std::mutex> Lock(EHFramePluginMutex);
+
   auto EHFrameRangeItr = TrackedEHFrameRanges.find(K);
   if (EHFrameRangeItr == TrackedEHFrameRanges.end())
     return Error::success();
@@ -470,6 +594,7 @@ Error EHFrameRegistrationPlugin::notifyRemovingModule(VModuleKey K) {
 }
 
 Error EHFrameRegistrationPlugin::notifyRemovingAllModules() {
+  std::lock_guard<std::mutex> Lock(EHFramePluginMutex);
 
   std::vector<EHFrameRange> EHFrameRanges =
     std::move(UntrackedEHFrameRanges);
diff --git a/llvm/lib/ExecutionEngine/Orc/OrcABISupport.cpp b/llvm/lib/ExecutionEngine/Orc/OrcABISupport.cpp
index 8ed23de419d1..18b3c5e12b1c 100644
--- a/llvm/lib/ExecutionEngine/Orc/OrcABISupport.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/OrcABISupport.cpp
@@ -7,13 +7,46 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ExecutionEngine/Orc/OrcABISupport.h"
+#include "llvm/Support/FormatVariadic.h"
 #include "llvm/Support/Process.h"
+#include "llvm/Support/raw_ostream.h"
+
+#define DEBUG_TYPE "orc"
+
+using namespace llvm;
+
+template <typename ORCABI>
+bool stubAndPointerRangesOk(JITTargetAddress StubBlockAddr,
+                            JITTargetAddress PointerBlockAddr,
+                            unsigned NumStubs) {
+  constexpr unsigned MaxDisp = ORCABI::StubToPointerMaxDisplacement;
+  JITTargetAddress FirstStub = StubBlockAddr;
+  JITTargetAddress LastStub = FirstStub + ((NumStubs - 1) * ORCABI::StubSize);
+  JITTargetAddress FirstPointer = PointerBlockAddr;
+  JITTargetAddress LastPointer =
+      FirstPointer + ((NumStubs - 1) * ORCABI::StubSize);
+
+  if (FirstStub < FirstPointer) {
+    if (LastStub >= FirstPointer)
+      return false; // Ranges overlap.
+    return (FirstPointer - FirstStub <= MaxDisp) &&
+           (LastPointer - LastStub <= MaxDisp); // out-of-range.
+  }
+
+  if (LastPointer >= FirstStub)
+    return false; // Ranges overlap.
+
+  return (FirstStub - FirstPointer <= MaxDisp) &&
+         (LastStub - LastPointer <= MaxDisp);
+}
 
 namespace llvm {
 namespace orc {
 
-void OrcAArch64::writeResolverCode(uint8_t *ResolverMem, JITReentryFn ReentryFn,
-                                   void *CallbackMgr) {
+void OrcAArch64::writeResolverCode(char *ResolverWorkingMem,
+                                   JITTargetAddress ResolverTargetAddress,
+                                   JITTargetAddress ReentryFnAddr,
+                                   JITTargetAddress ReentryCtxAddr) {
 
   const uint32_t ResolverCode[] = {
     // resolver_entry:
@@ -48,7 +81,7 @@ void OrcAArch64::writeResolverCode(uint8_t *ResolverMem, JITReentryFn ReentryFn,
     0xadbf17e4,        // 0x070:  stp   q4,  q5, [sp, #-32]!
     0xadbf0fe2,        // 0x074:  stp   q2,  q3, [sp, #-32]!
     0xadbf07e0,        // 0x078:  stp   q0,  q1, [sp, #-32]!
-    0x580004e0,        // 0x07c:  ldr   x0, Lcallbackmgr
+    0x580004e0,        // 0x07c:  ldr   x0, Lreentry_ctx_ptr
     0xaa1e03e1,        // 0x080:  mov   x1, x30
     0xd1003021,        // 0x084:  sub   x1,  x1, #12
     0x58000442,        // 0x088:  ldr   x2, Lreentry_fn_ptr
@@ -87,43 +120,47 @@ void OrcAArch64::writeResolverCode(uint8_t *ResolverMem, JITReentryFn ReentryFn,
     0xd65f0220,        // 0x10c:  ret  x17
     0x01234567,        // 0x110:  Lreentry_fn_ptr:
     0xdeadbeef,        // 0x114:      .quad 0
-    0x98765432,        // 0x118:  Lcallbackmgr:
+    0x98765432,        // 0x118:  Lreentry_ctx_ptr:
     0xcafef00d         // 0x11c:      .quad 0
   };
 
   const unsigned ReentryFnAddrOffset = 0x110;
-  const unsigned CallbackMgrAddrOffset = 0x118;
+  const unsigned ReentryCtxAddrOffset = 0x118;
 
-  memcpy(ResolverMem, ResolverCode, sizeof(ResolverCode));
-  memcpy(ResolverMem + ReentryFnAddrOffset, &ReentryFn, sizeof(ReentryFn));
-  memcpy(ResolverMem + CallbackMgrAddrOffset, &CallbackMgr,
-         sizeof(CallbackMgr));
+  memcpy(ResolverWorkingMem, ResolverCode, sizeof(ResolverCode));
+  memcpy(ResolverWorkingMem + ReentryFnAddrOffset, &ReentryFnAddr,
+         sizeof(uint64_t));
+  memcpy(ResolverWorkingMem + ReentryCtxAddrOffset, &ReentryCtxAddr,
+         sizeof(uint64_t));
 }
 
-void OrcAArch64::writeTrampolines(uint8_t *TrampolineMem, void *ResolverAddr,
+void OrcAArch64::writeTrampolines(char *TrampolineBlockWorkingMem,
+                                  JITTargetAddress TrampolineBlockTargetAddress,
+                                  JITTargetAddress ResolverAddr,
                                   unsigned NumTrampolines) {
 
   unsigned OffsetToPtr = alignTo(NumTrampolines * TrampolineSize, 8);
 
-  memcpy(TrampolineMem + OffsetToPtr, &ResolverAddr, sizeof(void *));
+  memcpy(TrampolineBlockWorkingMem + OffsetToPtr, &ResolverAddr,
+         sizeof(uint64_t));
 
   // OffsetToPtr is actually the offset from the PC for the 2nd instruction, so
   // subtract 32-bits.
   OffsetToPtr -= 4;
 
-  uint32_t *Trampolines = reinterpret_cast<uint32_t *>(TrampolineMem);
+  uint32_t *Trampolines =
+      reinterpret_cast<uint32_t *>(TrampolineBlockWorkingMem);
 
   for (unsigned I = 0; I < NumTrampolines; ++I, OffsetToPtr -= TrampolineSize) {
     Trampolines[3 * I + 0] = 0xaa1e03f1;                      // mov x17, x30
     Trampolines[3 * I + 1] = 0x58000010 | (OffsetToPtr << 3); // adr x16, Lptr
     Trampolines[3 * I + 2] = 0xd63f0200;                      // blr x16
   }
-
 }
 
-Error OrcAArch64::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
-                                         unsigned MinStubs,
-                                         void *InitialPtrVal) {
+void OrcAArch64::writeIndirectStubsBlock(
+    char *StubsBlockWorkingMem, JITTargetAddress StubsBlockTargetAddress,
+    JITTargetAddress PointersBlockTargetAddress, unsigned NumStubs) {
   // Stub format is:
   //
   // .section __orc_stubs
@@ -144,68 +181,41 @@ Error OrcAArch64::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
   //
   // ...
 
-  const unsigned StubSize = IndirectStubsInfo::StubSize;
-
-  // Emit at least MinStubs, rounded up to fill the pages allocated.
-  static const unsigned PageSize = sys::Process::getPageSizeEstimate();
-  unsigned NumPages = ((MinStubs * StubSize) + (PageSize - 1)) / PageSize;
-  unsigned NumStubs = (NumPages * PageSize) / StubSize;
-
-  // Allocate memory for stubs and pointers in one call.
-  std::error_code EC;
-  auto StubsMem = sys::OwningMemoryBlock(sys::Memory::allocateMappedMemory(
-      2 * NumPages * PageSize, nullptr,
-      sys::Memory::MF_READ | sys::Memory::MF_WRITE, EC));
-
-  if (EC)
-    return errorCodeToError(EC);
-
-  // Create separate MemoryBlocks representing the stubs and pointers.
-  sys::MemoryBlock StubsBlock(StubsMem.base(), NumPages * PageSize);
-  sys::MemoryBlock PtrsBlock(static_cast<char *>(StubsMem.base()) +
-                                 NumPages * PageSize,
-                             NumPages * PageSize);
-
-  // Populate the stubs page stubs and mark it executable.
-  uint64_t *Stub = reinterpret_cast<uint64_t *>(StubsBlock.base());
-  uint64_t PtrOffsetField = static_cast<uint64_t>(NumPages * PageSize)
-                            << 3;
+  static_assert(StubSize == PointerSize,
+                "Pointer and stub size must match for algorithm below");
+  assert(stubAndPointerRangesOk<OrcAArch64>(
+             StubsBlockTargetAddress, PointersBlockTargetAddress, NumStubs) &&
+         "PointersBlock is out of range");
+  uint64_t PtrDisplacement =
+      PointersBlockTargetAddress - StubsBlockTargetAddress;
+  uint64_t *Stub = reinterpret_cast<uint64_t *>(StubsBlockWorkingMem);
+  uint64_t PtrOffsetField = PtrDisplacement << 3;
 
   for (unsigned I = 0; I < NumStubs; ++I)
     Stub[I] = 0xd61f020058000010 | PtrOffsetField;
-
-  if (auto EC = sys::Memory::protectMappedMemory(
-          StubsBlock, sys::Memory::MF_READ | sys::Memory::MF_EXEC))
-    return errorCodeToError(EC);
-
-  // Initialize all pointers to point at FailureAddress.
-  void **Ptr = reinterpret_cast<void **>(PtrsBlock.base());
-  for (unsigned I = 0; I < NumStubs; ++I)
-    Ptr[I] = InitialPtrVal;
-
-  StubsInfo = IndirectStubsInfo(NumStubs, std::move(StubsMem));
-
-  return Error::success();
 }
 
-void OrcX86_64_Base::writeTrampolines(uint8_t *TrampolineMem,
-                                      void *ResolverAddr,
-                                      unsigned NumTrampolines) {
+void OrcX86_64_Base::writeTrampolines(
+    char *TrampolineBlockWorkingMem,
+    JITTargetAddress TrampolineBlockTargetAddress,
+    JITTargetAddress ResolverAddr, unsigned NumTrampolines) {
 
   unsigned OffsetToPtr = NumTrampolines * TrampolineSize;
 
-  memcpy(TrampolineMem + OffsetToPtr, &ResolverAddr, sizeof(void *));
+  memcpy(TrampolineBlockWorkingMem + OffsetToPtr, &ResolverAddr,
+         sizeof(uint64_t));
 
-  uint64_t *Trampolines = reinterpret_cast<uint64_t *>(TrampolineMem);
+  uint64_t *Trampolines =
+      reinterpret_cast<uint64_t *>(TrampolineBlockWorkingMem);
   uint64_t CallIndirPCRel = 0xf1c40000000015ff;
 
   for (unsigned I = 0; I < NumTrampolines; ++I, OffsetToPtr -= TrampolineSize)
     Trampolines[I] = CallIndirPCRel | ((OffsetToPtr - 6) << 16);
 }
 
-Error OrcX86_64_Base::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
-                                             unsigned MinStubs,
-                                             void *InitialPtrVal) {
+void OrcX86_64_Base::writeIndirectStubsBlock(
+    char *StubsBlockWorkingMem, JITTargetAddress StubsBlockTargetAddress,
+    JITTargetAddress PointersBlockTargetAddress, unsigned NumStubs) {
   // Stub format is:
   //
   // .section __orc_stubs
@@ -226,52 +236,28 @@ Error OrcX86_64_Base::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
   //
   // ...
 
-  const unsigned StubSize = IndirectStubsInfo::StubSize;
-
-  // Emit at least MinStubs, rounded up to fill the pages allocated.
-  static const unsigned PageSize = sys::Process::getPageSizeEstimate();
-  unsigned NumPages = ((MinStubs * StubSize) + (PageSize - 1)) / PageSize;
-  unsigned NumStubs = (NumPages * PageSize) / StubSize;
-
-  // Allocate memory for stubs and pointers in one call.
-  std::error_code EC;
-  auto StubsMem = sys::OwningMemoryBlock(sys::Memory::allocateMappedMemory(
-      2 * NumPages * PageSize, nullptr,
-      sys::Memory::MF_READ | sys::Memory::MF_WRITE, EC));
-
-  if (EC)
-    return errorCodeToError(EC);
-
-  // Create separate MemoryBlocks representing the stubs and pointers.
-  sys::MemoryBlock StubsBlock(StubsMem.base(), NumPages * PageSize);
-  sys::MemoryBlock PtrsBlock(static_cast<char *>(StubsMem.base()) +
-                                 NumPages * PageSize,
-                             NumPages * PageSize);
-
   // Populate the stubs page stubs and mark it executable.
-  uint64_t *Stub = reinterpret_cast<uint64_t *>(StubsBlock.base());
-  uint64_t PtrOffsetField = static_cast<uint64_t>(NumPages * PageSize - 6)
-                            << 16;
+  static_assert(StubSize == PointerSize,
+                "Pointer and stub size must match for algorithm below");
+  assert(stubAndPointerRangesOk<OrcX86_64_Base>(
+             StubsBlockTargetAddress, PointersBlockTargetAddress, NumStubs) &&
+         "PointersBlock is out of range");
+  uint64_t *Stub = reinterpret_cast<uint64_t *>(StubsBlockWorkingMem);
+  uint64_t PtrOffsetField =
+      (PointersBlockTargetAddress - StubsBlockTargetAddress - 6) << 16;
   for (unsigned I = 0; I < NumStubs; ++I)
     Stub[I] = 0xF1C40000000025ff | PtrOffsetField;
-
-  if (auto EC = sys::Memory::protectMappedMemory(
-          StubsBlock, sys::Memory::MF_READ | sys::Memory::MF_EXEC))
-    return errorCodeToError(EC);
-
-  // Initialize all pointers to point at FailureAddress.
-  void **Ptr = reinterpret_cast<void **>(PtrsBlock.base());
-  for (unsigned I = 0; I < NumStubs; ++I)
-    Ptr[I] = InitialPtrVal;
-
-  StubsInfo = IndirectStubsInfo(NumStubs, std::move(StubsMem));
-
-  return Error::success();
 }
 
-void OrcX86_64_SysV::writeResolverCode(uint8_t *ResolverMem,
-                                       JITReentryFn ReentryFn,
-                                       void *CallbackMgr) {
+void OrcX86_64_SysV::writeResolverCode(char *ResolverWorkingMem,
+                                       JITTargetAddress ResolverTargetAddress,
+                                       JITTargetAddress ReentryFnAddr,
+                                       JITTargetAddress ReentryCtxAddr) {
+
+  LLVM_DEBUG({
+    dbgs() << "Writing resolver code to "
+           << formatv("{0:x16}", ResolverTargetAddress) << "\n";
+  });
 
   const uint8_t ResolverCode[] = {
       // resolver_entry:
@@ -295,7 +281,7 @@ void OrcX86_64_SysV::writeResolverCode(uint8_t *ResolverMem,
       0x48, 0x0f, 0xae, 0x04, 0x24,             // 0x21: fxsave64  (%rsp)
       0x48, 0xbf,                               // 0x26: movabsq   <CBMgr>, %rdi
 
-      // 0x28: Callback manager addr.
+      // 0x28: JIT re-entry ctx addr.
       0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 
       0x48, 0x8b, 0x75, 0x08,                   // 0x30: movq      8(%rbp), %rsi
@@ -325,23 +311,26 @@ void OrcX86_64_SysV::writeResolverCode(uint8_t *ResolverMem,
       0x58,                                     // 0x69: popq      %rax
       0x5d,                                     // 0x6a: popq      %rbp
       0xc3,                                     // 0x6b: retq
-  };
+ };
 
   const unsigned ReentryFnAddrOffset = 0x3a;
-  const unsigned CallbackMgrAddrOffset = 0x28;
+  const unsigned ReentryCtxAddrOffset = 0x28;
 
-  memcpy(ResolverMem, ResolverCode, sizeof(ResolverCode));
-  memcpy(ResolverMem + ReentryFnAddrOffset, &ReentryFn, sizeof(ReentryFn));
-  memcpy(ResolverMem + CallbackMgrAddrOffset, &CallbackMgr,
-         sizeof(CallbackMgr));
+  memcpy(ResolverWorkingMem, ResolverCode, sizeof(ResolverCode));
+  memcpy(ResolverWorkingMem + ReentryFnAddrOffset, &ReentryFnAddr,
+         sizeof(uint64_t));
+  memcpy(ResolverWorkingMem + ReentryCtxAddrOffset, &ReentryCtxAddr,
+         sizeof(uint64_t));
 }
 
-void OrcX86_64_Win32::writeResolverCode(uint8_t *ResolverMem,
-                                        JITReentryFn ReentryFn,
-                                        void *CallbackMgr) {
+void OrcX86_64_Win32::writeResolverCode(char *ResolverWorkingMem,
+                                        JITTargetAddress ResolverTargetAddress,
+                                        JITTargetAddress ReentryFnAddr,
+                                        JITTargetAddress ReentryCtxAddr) {
 
-  // resolverCode is similar to OrcX86_64 with differences specific to windows x64 calling convention:
-  // arguments go into rcx, rdx and come in reverse order, shadow space allocation on stack
+  // resolverCode is similar to OrcX86_64 with differences specific to windows
+  // x64 calling convention: arguments go into rcx, rdx and come in reverse
+  // order, shadow space allocation on stack
   const uint8_t ResolverCode[] = {
       // resolver_entry:
       0x55,                                      // 0x00: pushq     %rbp
@@ -364,7 +353,7 @@ void OrcX86_64_Win32::writeResolverCode(uint8_t *ResolverMem,
       0x48, 0x0f, 0xae, 0x04, 0x24,              // 0x21: fxsave64  (%rsp)
 
       0x48, 0xb9,                                // 0x26: movabsq   <CBMgr>, %rcx
-      // 0x28: Callback manager addr.
+      // 0x28: JIT re-entry ctx addr.
       0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 
       0x48, 0x8B, 0x55, 0x08,                    // 0x30: mov       rdx, [rbp+0x8]
@@ -402,18 +391,23 @@ void OrcX86_64_Win32::writeResolverCode(uint8_t *ResolverMem,
       0xc3,                                      // 0x73: retq
   };
 
-
   const unsigned ReentryFnAddrOffset = 0x3a;
-  const unsigned CallbackMgrAddrOffset = 0x28;
+  const unsigned ReentryCtxAddrOffset = 0x28;
 
-  memcpy(ResolverMem, ResolverCode, sizeof(ResolverCode));
-  memcpy(ResolverMem + ReentryFnAddrOffset, &ReentryFn, sizeof(ReentryFn));
-  memcpy(ResolverMem + CallbackMgrAddrOffset, &CallbackMgr,
-         sizeof(CallbackMgr));
+  memcpy(ResolverWorkingMem, ResolverCode, sizeof(ResolverCode));
+  memcpy(ResolverWorkingMem + ReentryFnAddrOffset, &ReentryFnAddr,
+         sizeof(uint64_t));
+  memcpy(ResolverWorkingMem + ReentryCtxAddrOffset, &ReentryCtxAddr,
+         sizeof(uint64_t));
 }
 
-void OrcI386::writeResolverCode(uint8_t *ResolverMem, JITReentryFn ReentryFn,
-                                void *CallbackMgr) {
+void OrcI386::writeResolverCode(char *ResolverWorkingMem,
+                                JITTargetAddress ResolverTargetAddress,
+                                JITTargetAddress ReentryFnAddr,
+                                JITTargetAddress ReentryCtxAddr) {
+
+  assert((ReentryFnAddr >> 32) == 0 && "ReentryFnAddr out of range");
+  assert((ReentryCtxAddr >> 32) == 0 && "ReentryCtxAddr out of range");
 
   const uint8_t ResolverCode[] = {
       // resolver_entry:
@@ -451,29 +445,37 @@ void OrcI386::writeResolverCode(uint8_t *ResolverMem, JITReentryFn ReentryFn,
   };
 
   const unsigned ReentryFnAddrOffset = 0x2a;
-  const unsigned CallbackMgrAddrOffset = 0x25;
+  const unsigned ReentryCtxAddrOffset = 0x25;
 
-  memcpy(ResolverMem, ResolverCode, sizeof(ResolverCode));
-  memcpy(ResolverMem + ReentryFnAddrOffset, &ReentryFn, sizeof(ReentryFn));
-  memcpy(ResolverMem + CallbackMgrAddrOffset, &CallbackMgr,
-         sizeof(CallbackMgr));
+  memcpy(ResolverWorkingMem, ResolverCode, sizeof(ResolverCode));
+  memcpy(ResolverWorkingMem + ReentryFnAddrOffset, &ReentryFnAddr,
+         sizeof(uint32_t));
+  memcpy(ResolverWorkingMem + ReentryCtxAddrOffset, &ReentryCtxAddr,
+         sizeof(uint32_t));
 }
 
-void OrcI386::writeTrampolines(uint8_t *TrampolineMem, void *ResolverAddr,
+void OrcI386::writeTrampolines(char *TrampolineWorkingMem,
+                               JITTargetAddress TrampolineBlockTargetAddress,
+                               JITTargetAddress ResolverAddr,
                                unsigned NumTrampolines) {
+  assert((ResolverAddr >> 32) == 0 && "ResolverAddr out of range");
 
   uint64_t CallRelImm = 0xF1C4C400000000e8;
-  uint64_t Resolver = reinterpret_cast<uint64_t>(ResolverAddr);
-  uint64_t ResolverRel =
-      Resolver - reinterpret_cast<uint64_t>(TrampolineMem) - 5;
+  uint64_t ResolverRel = ResolverAddr - TrampolineBlockTargetAddress - 5;
 
-  uint64_t *Trampolines = reinterpret_cast<uint64_t *>(TrampolineMem);
+  uint64_t *Trampolines = reinterpret_cast<uint64_t *>(TrampolineWorkingMem);
   for (unsigned I = 0; I < NumTrampolines; ++I, ResolverRel -= TrampolineSize)
     Trampolines[I] = CallRelImm | (ResolverRel << 8);
 }
 
-Error OrcI386::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
-                                      unsigned MinStubs, void *InitialPtrVal) {
+void OrcI386::writeIndirectStubsBlock(
+    char *StubsBlockWorkingMem, JITTargetAddress StubsBlockTargetAddress,
+    JITTargetAddress PointersBlockTargetAddress, unsigned NumStubs) {
+  assert((StubsBlockTargetAddress >> 32) == 0 &&
+         "StubsBlockTargetAddress is out of range");
+  assert((PointersBlockTargetAddress >> 32) == 0 &&
+         "PointersBlockTargetAddress is out of range");
+
   // Stub format is:
   //
   // .section __orc_stubs
@@ -494,51 +496,21 @@ Error OrcI386::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
   //
   // ...
 
-  const unsigned StubSize = IndirectStubsInfo::StubSize;
+  assert(stubAndPointerRangesOk<OrcI386>(
+             StubsBlockTargetAddress, PointersBlockTargetAddress, NumStubs) &&
+         "PointersBlock is out of range");
 
-  // Emit at least MinStubs, rounded up to fill the pages allocated.
-  static const unsigned PageSize = sys::Process::getPageSizeEstimate();
-  unsigned NumPages = ((MinStubs * StubSize) + (PageSize - 1)) / PageSize;
-  unsigned NumStubs = (NumPages * PageSize) / StubSize;
-
-  // Allocate memory for stubs and pointers in one call.
-  std::error_code EC;
-  auto StubsMem = sys::OwningMemoryBlock(sys::Memory::allocateMappedMemory(
-      2 * NumPages * PageSize, nullptr,
-      sys::Memory::MF_READ | sys::Memory::MF_WRITE, EC));
-
-  if (EC)
-    return errorCodeToError(EC);
-
-  // Create separate MemoryBlocks representing the stubs and pointers.
-  sys::MemoryBlock StubsBlock(StubsMem.base(), NumPages * PageSize);
-  sys::MemoryBlock PtrsBlock(static_cast<char *>(StubsMem.base()) +
-                                 NumPages * PageSize,
-                             NumPages * PageSize);
-
-  // Populate the stubs page stubs and mark it executable.
-  uint64_t *Stub = reinterpret_cast<uint64_t *>(StubsBlock.base());
-  uint64_t PtrAddr = reinterpret_cast<uint64_t>(PtrsBlock.base());
+  uint64_t *Stub = reinterpret_cast<uint64_t *>(StubsBlockWorkingMem);
+  uint64_t PtrAddr = PointersBlockTargetAddress;
   for (unsigned I = 0; I < NumStubs; ++I, PtrAddr += 4)
     Stub[I] = 0xF1C40000000025ff | (PtrAddr << 16);
-
-  if (auto EC = sys::Memory::protectMappedMemory(
-          StubsBlock, sys::Memory::MF_READ | sys::Memory::MF_EXEC))
-    return errorCodeToError(EC);
-
-  // Initialize all pointers to point at FailureAddress.
-  void **Ptr = reinterpret_cast<void **>(PtrsBlock.base());
-  for (unsigned I = 0; I < NumStubs; ++I)
-    Ptr[I] = InitialPtrVal;
-
-  StubsInfo = IndirectStubsInfo(NumStubs, std::move(StubsMem));
-
-  return Error::success();
 }
 
-void OrcMips32_Base::writeResolverCode(uint8_t *ResolverMem,
-                                       JITReentryFn ReentryFn,
-                                       void *CallbackMgr, bool isBigEndian) {
+void OrcMips32_Base::writeResolverCode(char *ResolverWorkingMem,
+                                       JITTargetAddress ResolverTargetAddress,
+                                       JITTargetAddress ReentryFnAddr,
+                                       JITTargetAddress ReentryCtxAddr,
+                                       bool isBigEndian) {
 
   const uint32_t ResolverCode[] = {
       // resolver_entry:
@@ -570,9 +542,9 @@ void OrcMips32_Base::writeResolverCode(uint8_t *ResolverMem,
       0xafbe0060,                    // 0x64: sw $fp,96($sp)
       0xafbf0064,                    // 0x68: sw $ra,100($sp)
 
-      // Callback manager addr.
-      0x00000000,                    // 0x6c: lui $a0,callbackmgr
-      0x00000000,                    // 0x70: addiu $a0,$a0,callbackmgr
+      // JIT re-entry ctx addr.
+      0x00000000,                    // 0x6c: lui $a0,ctx
+      0x00000000,                    // 0x70: addiu $a0,$a0,ctx
 
       0x03e02825,                    // 0x74: move $a1, $ra
       0x24a5ffec,                    // 0x78: addiu $a1,$a1,-20
@@ -614,50 +586,63 @@ void OrcMips32_Base::writeResolverCode(uint8_t *ResolverMem,
   };
 
   const unsigned ReentryFnAddrOffset = 0x7c;   // JIT re-entry fn addr lui
-  const unsigned CallbackMgrAddrOffset = 0x6c; // Callback manager addr lui
+  const unsigned ReentryCtxAddrOffset = 0x6c;  // JIT re-entry context addr lui
   const unsigned Offsett = 0xf8;
 
-  memcpy(ResolverMem, ResolverCode, sizeof(ResolverCode));
+  memcpy(ResolverWorkingMem, ResolverCode, sizeof(ResolverCode));
 
   // Depending on endian return value will be in v0 or v1.
   uint32_t MoveVxT9 = isBigEndian ? 0x0060c825 : 0x0040c825;
-  memcpy(ResolverMem + Offsett, &MoveVxT9, sizeof(MoveVxT9));
-
-  uint64_t CallMgrAddr = reinterpret_cast<uint64_t>(CallbackMgr);
-  uint32_t CallMgrLUi = 0x3c040000 | (((CallMgrAddr + 0x8000) >> 16) & 0xFFFF);
-  uint32_t CallMgrADDiu = 0x24840000 | ((CallMgrAddr) & 0xFFFF);
-  memcpy(ResolverMem + CallbackMgrAddrOffset, &CallMgrLUi, sizeof(CallMgrLUi));
-  memcpy(ResolverMem + CallbackMgrAddrOffset + 4, &CallMgrADDiu,
-         sizeof(CallMgrADDiu));
-
-  uint64_t ReentryAddr = reinterpret_cast<uint64_t>(ReentryFn);
-  uint32_t ReentryLUi = 0x3c190000 | (((ReentryAddr + 0x8000) >> 16) & 0xFFFF);
-  uint32_t ReentryADDiu = 0x27390000 | ((ReentryAddr) & 0xFFFF);
-  memcpy(ResolverMem + ReentryFnAddrOffset, &ReentryLUi, sizeof(ReentryLUi));
-  memcpy(ResolverMem + ReentryFnAddrOffset + 4, &ReentryADDiu,
-         sizeof(ReentryADDiu));
+  memcpy(ResolverWorkingMem + Offsett, &MoveVxT9, sizeof(MoveVxT9));
+
+  uint32_t ReentryCtxLUi =
+      0x3c040000 | (((ReentryCtxAddr + 0x8000) >> 16) & 0xFFFF);
+  uint32_t ReentryCtxADDiu = 0x24840000 | ((ReentryCtxAddr)&0xFFFF);
+  memcpy(ResolverWorkingMem + ReentryCtxAddrOffset, &ReentryCtxLUi,
+         sizeof(ReentryCtxLUi));
+  memcpy(ResolverWorkingMem + ReentryCtxAddrOffset + 4, &ReentryCtxADDiu,
+         sizeof(ReentryCtxADDiu));
+
+  uint32_t ReentryFnLUi =
+      0x3c190000 | (((ReentryFnAddr + 0x8000) >> 16) & 0xFFFF);
+  uint32_t ReentryFnADDiu = 0x27390000 | ((ReentryFnAddr)&0xFFFF);
+  memcpy(ResolverWorkingMem + ReentryFnAddrOffset, &ReentryFnLUi,
+         sizeof(ReentryFnLUi));
+  memcpy(ResolverWorkingMem + ReentryFnAddrOffset + 4, &ReentryFnADDiu,
+         sizeof(ReentryFnADDiu));
 }
 
-void OrcMips32_Base::writeTrampolines(uint8_t *TrampolineMem,
-                                      void *ResolverAddr,
-                                      unsigned NumTrampolines) {
+void OrcMips32_Base::writeTrampolines(
+    char *TrampolineBlockWorkingMem,
+    JITTargetAddress TrampolineBlockTargetAddress,
+    JITTargetAddress ResolverAddr, unsigned NumTrampolines) {
 
-  uint32_t *Trampolines = reinterpret_cast<uint32_t *>(TrampolineMem);
-  uint64_t ResolveAddr = reinterpret_cast<uint64_t>(ResolverAddr);
-  uint32_t RHiAddr = ((ResolveAddr + 0x8000) >> 16);
+  assert((ResolverAddr >> 32) == 0 && "ResolverAddr out of range");
+
+  uint32_t *Trampolines =
+      reinterpret_cast<uint32_t *>(TrampolineBlockWorkingMem);
+  uint32_t RHiAddr = ((ResolverAddr + 0x8000) >> 16);
 
   for (unsigned I = 0; I < NumTrampolines; ++I) {
-    Trampolines[5 * I + 0] = 0x03e0c025;                           // move $t8,$ra
-    Trampolines[5 * I + 1] = 0x3c190000 | (RHiAddr & 0xFFFF);      // lui $t9,resolveAddr
-    Trampolines[5 * I + 2] = 0x27390000 | (ResolveAddr & 0xFFFF);  // addiu $t9,$t9,resolveAddr
-    Trampolines[5 * I + 3] = 0x0320f809;                           // jalr $t9
-    Trampolines[5 * I + 4] = 0x00000000;                           // nop
+    // move $t8,$ra
+    // lui $t9,ResolverAddr
+    // addiu $t9,$t9,ResolverAddr
+    // jalr $t9
+    // nop
+    Trampolines[5 * I + 0] = 0x03e0c025;
+    Trampolines[5 * I + 1] = 0x3c190000 | (RHiAddr & 0xFFFF);
+    Trampolines[5 * I + 2] = 0x27390000 | (ResolverAddr & 0xFFFF);
+    Trampolines[5 * I + 3] = 0x0320f809;
+    Trampolines[5 * I + 4] = 0x00000000;
   }
 }
 
-Error OrcMips32_Base::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
-                                             unsigned MinStubs,
-                                             void *InitialPtrVal) {
+void OrcMips32_Base::writeIndirectStubsBlock(
+    char *StubsBlockWorkingMem, JITTargetAddress StubsBlockTargetAddress,
+    JITTargetAddress PointersBlockTargetAddress, unsigned NumStubs) {
+  assert((StubsBlockTargetAddress >> 32) == 0 &&
+         "InitialPtrVal is out of range");
+
   // Stub format is:
   //
   // .section __orc_stubs
@@ -678,33 +663,15 @@ Error OrcMips32_Base::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
   // ptr2:
   //                 .word 0x0
   //
-  // ...
-
-  const unsigned StubSize = IndirectStubsInfo::StubSize;
-
-  // Emit at least MinStubs, rounded up to fill the pages allocated.
-  static const unsigned PageSize = sys::Process::getPageSizeEstimate();
-  unsigned NumPages = ((MinStubs * StubSize) + (PageSize - 1)) / PageSize;
-  unsigned NumStubs = (NumPages * PageSize) / StubSize;
+  // i..
 
-  // Allocate memory for stubs and pointers in one call.
-  std::error_code EC;
-  auto StubsMem = sys::OwningMemoryBlock(sys::Memory::allocateMappedMemory(
-      2 * NumPages * PageSize, nullptr,
-      sys::Memory::MF_READ | sys::Memory::MF_WRITE, EC));
-
-  if (EC)
-    return errorCodeToError(EC);
-
-  // Create separate MemoryBlocks representing the stubs and pointers.
-  sys::MemoryBlock StubsBlock(StubsMem.base(), NumPages * PageSize);
-  sys::MemoryBlock PtrsBlock(static_cast<char *>(StubsMem.base()) +
-                                 NumPages * PageSize,
-                             NumPages * PageSize);
+  assert(stubAndPointerRangesOk<OrcAArch64>(
+             StubsBlockTargetAddress, PointersBlockTargetAddress, NumStubs) &&
+         "PointersBlock is out of range");
 
   // Populate the stubs page stubs and mark it executable.
-  uint32_t *Stub = reinterpret_cast<uint32_t *>(StubsBlock.base());
-  uint64_t PtrAddr = reinterpret_cast<uint64_t>(Stub) + NumPages * PageSize;
+  uint32_t *Stub = reinterpret_cast<uint32_t *>(StubsBlockWorkingMem);
+  uint64_t PtrAddr = PointersBlockTargetAddress;
 
   for (unsigned I = 0; I < NumStubs; ++I) {
     uint32_t HiAddr = ((PtrAddr + 0x8000) >> 16);
@@ -714,26 +681,15 @@ Error OrcMips32_Base::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
     Stub[4 * I + 3] = 0x00000000;                      // nop
     PtrAddr += 4;
   }
-
-  if (auto EC = sys::Memory::protectMappedMemory(
-          StubsBlock, sys::Memory::MF_READ | sys::Memory::MF_EXEC))
-    return errorCodeToError(EC);
-
-  // Initialize all pointers to point at FailureAddress.
-  void **Ptr = reinterpret_cast<void **>(PtrsBlock.base());
-  for (unsigned I = 0; I < NumStubs; ++I)
-    Ptr[I] = InitialPtrVal;
-
-  StubsInfo = IndirectStubsInfo(NumStubs, std::move(StubsMem));
-
-  return Error::success();
 }
 
-void OrcMips64::writeResolverCode(uint8_t *ResolverMem, JITReentryFn ReentryFn,
-                                  void *CallbackMgr) {
+void OrcMips64::writeResolverCode(char *ResolverWorkingMem,
+                                  JITTargetAddress ResolverTargetAddress,
+                                  JITTargetAddress ReentryFnAddr,
+                                  JITTargetAddress ReentryCtxAddr) {
 
   const uint32_t ResolverCode[] = {
-      //resolver_entry:
+       //resolver_entry:
       0x67bdff30,                     // 0x00: daddiu $sp,$sp,-208
       0xffa20000,                     // 0x04: sd v0,0(sp)
       0xffa30008,                     // 0x08: sd v1,8(sp)
@@ -762,13 +718,13 @@ void OrcMips64::writeResolverCode(uint8_t *ResolverMem, JITReentryFn ReentryFn,
       0xffbe00c0,                     // 0x64: sd fp,192(sp)
       0xffbf00c8,                     // 0x68: sd ra,200(sp)
 
-      // Callback manager addr.
-      0x00000000,                     // 0x6c: lui $a0,heighest(callbackmgr)
-      0x00000000,                     // 0x70: daddiu $a0,$a0,heigher(callbackmgr)
+      // JIT re-entry ctx addr.
+      0x00000000,                     // 0x6c: lui $a0,heighest(ctx)
+      0x00000000,                     // 0x70: daddiu $a0,$a0,heigher(ctx)
       0x00000000,                     // 0x74: dsll $a0,$a0,16
-      0x00000000,                     // 0x78: daddiu $a0,$a0,hi(callbackmgr)
+      0x00000000,                     // 0x78: daddiu $a0,$a0,hi(ctx)
       0x00000000,                     // 0x7c: dsll $a0,$a0,16
-      0x00000000,                     // 0x80: daddiu $a0,$a0,lo(callbackmgr)
+      0x00000000,                     // 0x80: daddiu $a0,$a0,lo(ctx)
 
       0x03e02825,                     // 0x84: move $a1, $ra
       0x64a5ffdc,                     // 0x88: daddiu $a1,$a1,-36
@@ -814,73 +770,73 @@ void OrcMips64::writeResolverCode(uint8_t *ResolverMem, JITReentryFn ReentryFn,
   };
 
   const unsigned ReentryFnAddrOffset = 0x8c;   // JIT re-entry fn addr lui
-  const unsigned CallbackMgrAddrOffset = 0x6c; // Callback manager addr lui
-
-  memcpy(ResolverMem, ResolverCode, sizeof(ResolverCode));
-
-  uint64_t CallMgrAddr = reinterpret_cast<uint64_t>(CallbackMgr);
-
-  uint32_t CallMgrLUi =
-      0x3c040000 | (((CallMgrAddr + 0x800080008000) >> 48) & 0xFFFF);
-  uint32_t CallMgrDADDiu =
-      0x64840000 | (((CallMgrAddr + 0x80008000) >> 32) & 0xFFFF);
-  uint32_t CallMgrDSLL = 0x00042438;
-  uint32_t CallMgrDADDiu2 =
-      0x64840000 | ((((CallMgrAddr + 0x8000) >> 16) & 0xFFFF));
-  uint32_t CallMgrDSLL2 = 0x00042438;
-  uint32_t CallMgrDADDiu3 = 0x64840000 | ((CallMgrAddr)&0xFFFF);
-
-  memcpy(ResolverMem + CallbackMgrAddrOffset, &CallMgrLUi, sizeof(CallMgrLUi));
-  memcpy(ResolverMem + (CallbackMgrAddrOffset + 4), &CallMgrDADDiu,
-         sizeof(CallMgrDADDiu));
-  memcpy(ResolverMem + (CallbackMgrAddrOffset + 8), &CallMgrDSLL,
-         sizeof(CallMgrDSLL));
-  memcpy(ResolverMem + (CallbackMgrAddrOffset + 12), &CallMgrDADDiu2,
-         sizeof(CallMgrDADDiu2));
-  memcpy(ResolverMem + (CallbackMgrAddrOffset + 16), &CallMgrDSLL2,
-         sizeof(CallMgrDSLL2));
-  memcpy(ResolverMem + (CallbackMgrAddrOffset + 20), &CallMgrDADDiu3,
-         sizeof(CallMgrDADDiu3));
-
-  uint64_t ReentryAddr = reinterpret_cast<uint64_t>(ReentryFn);
-
-  uint32_t ReentryLUi =
-      0x3c190000 | (((ReentryAddr + 0x800080008000) >> 48) & 0xFFFF);
-
-  uint32_t ReentryDADDiu =
-      0x67390000 | (((ReentryAddr + 0x80008000) >> 32) & 0xFFFF);
-
-  uint32_t ReentryDSLL = 0x0019cc38;
-
-  uint32_t ReentryDADDiu2 =
-      0x67390000 | (((ReentryAddr + 0x8000) >> 16) & 0xFFFF);
-
-  uint32_t ReentryDSLL2 = 0x0019cc38;
-
-  uint32_t ReentryDADDiu3 = 0x67390000 | ((ReentryAddr)&0xFFFF);
-
-  memcpy(ResolverMem + ReentryFnAddrOffset, &ReentryLUi, sizeof(ReentryLUi));
-  memcpy(ResolverMem + (ReentryFnAddrOffset + 4), &ReentryDADDiu,
-         sizeof(ReentryDADDiu));
-  memcpy(ResolverMem + (ReentryFnAddrOffset + 8), &ReentryDSLL,
-         sizeof(ReentryDSLL));
-  memcpy(ResolverMem + (ReentryFnAddrOffset + 12), &ReentryDADDiu2,
-         sizeof(ReentryDADDiu2));
-  memcpy(ResolverMem + (ReentryFnAddrOffset + 16), &ReentryDSLL2,
-         sizeof(ReentryDSLL2));
-  memcpy(ResolverMem + (ReentryFnAddrOffset + 20), &ReentryDADDiu3,
-         sizeof(ReentryDADDiu3));
+  const unsigned ReentryCtxAddrOffset = 0x6c;  // JIT re-entry ctx addr lui
+
+  memcpy(ResolverWorkingMem, ResolverCode, sizeof(ResolverCode));
+
+  uint32_t ReentryCtxLUi =
+      0x3c040000 | (((ReentryCtxAddr + 0x800080008000) >> 48) & 0xFFFF);
+  uint32_t ReentryCtxDADDiu =
+      0x64840000 | (((ReentryCtxAddr + 0x80008000) >> 32) & 0xFFFF);
+  uint32_t ReentryCtxDSLL = 0x00042438;
+  uint32_t ReentryCtxDADDiu2 =
+      0x64840000 | ((((ReentryCtxAddr + 0x8000) >> 16) & 0xFFFF));
+  uint32_t ReentryCtxDSLL2 = 0x00042438;
+  uint32_t ReentryCtxDADDiu3 = 0x64840000 | ((ReentryCtxAddr)&0xFFFF);
+
+  memcpy(ResolverWorkingMem + ReentryCtxAddrOffset, &ReentryCtxLUi,
+         sizeof(ReentryCtxLUi));
+  memcpy(ResolverWorkingMem + (ReentryCtxAddrOffset + 4), &ReentryCtxDADDiu,
+         sizeof(ReentryCtxDADDiu));
+  memcpy(ResolverWorkingMem + (ReentryCtxAddrOffset + 8), &ReentryCtxDSLL,
+         sizeof(ReentryCtxDSLL));
+  memcpy(ResolverWorkingMem + (ReentryCtxAddrOffset + 12), &ReentryCtxDADDiu2,
+         sizeof(ReentryCtxDADDiu2));
+  memcpy(ResolverWorkingMem + (ReentryCtxAddrOffset + 16), &ReentryCtxDSLL2,
+         sizeof(ReentryCtxDSLL2));
+  memcpy(ResolverWorkingMem + (ReentryCtxAddrOffset + 20), &ReentryCtxDADDiu3,
+         sizeof(ReentryCtxDADDiu3));
+
+  uint32_t ReentryFnLUi =
+      0x3c190000 | (((ReentryFnAddr + 0x800080008000) >> 48) & 0xFFFF);
+
+  uint32_t ReentryFnDADDiu =
+      0x67390000 | (((ReentryFnAddr + 0x80008000) >> 32) & 0xFFFF);
+
+  uint32_t ReentryFnDSLL = 0x0019cc38;
+
+  uint32_t ReentryFnDADDiu2 =
+      0x67390000 | (((ReentryFnAddr + 0x8000) >> 16) & 0xFFFF);
+
+  uint32_t ReentryFnDSLL2 = 0x0019cc38;
+
+  uint32_t ReentryFnDADDiu3 = 0x67390000 | ((ReentryFnAddr)&0xFFFF);
+
+  memcpy(ResolverWorkingMem + ReentryFnAddrOffset, &ReentryFnLUi,
+         sizeof(ReentryFnLUi));
+  memcpy(ResolverWorkingMem + (ReentryFnAddrOffset + 4), &ReentryFnDADDiu,
+         sizeof(ReentryFnDADDiu));
+  memcpy(ResolverWorkingMem + (ReentryFnAddrOffset + 8), &ReentryFnDSLL,
+         sizeof(ReentryFnDSLL));
+  memcpy(ResolverWorkingMem + (ReentryFnAddrOffset + 12), &ReentryFnDADDiu2,
+         sizeof(ReentryFnDADDiu2));
+  memcpy(ResolverWorkingMem + (ReentryFnAddrOffset + 16), &ReentryFnDSLL2,
+         sizeof(ReentryFnDSLL2));
+  memcpy(ResolverWorkingMem + (ReentryFnAddrOffset + 20), &ReentryFnDADDiu3,
+         sizeof(ReentryFnDADDiu3));
 }
 
-void OrcMips64::writeTrampolines(uint8_t *TrampolineMem, void *ResolverAddr,
+void OrcMips64::writeTrampolines(char *TrampolineBlockWorkingMem,
+                                 JITTargetAddress TrampolineBlockTargetAddress,
+                                 JITTargetAddress ResolverAddr,
                                  unsigned NumTrampolines) {
 
-  uint32_t *Trampolines = reinterpret_cast<uint32_t *>(TrampolineMem);
-  uint64_t ResolveAddr = reinterpret_cast<uint64_t>(ResolverAddr);
+  uint32_t *Trampolines =
+      reinterpret_cast<uint32_t *>(TrampolineBlockWorkingMem);
 
-  uint64_t HeighestAddr = ((ResolveAddr + 0x800080008000) >> 48);
-  uint64_t HeigherAddr = ((ResolveAddr + 0x80008000) >> 32);
-  uint64_t HiAddr = ((ResolveAddr + 0x8000) >> 16);
+  uint64_t HeighestAddr = ((ResolverAddr + 0x800080008000) >> 48);
+  uint64_t HeigherAddr = ((ResolverAddr + 0x80008000) >> 32);
+  uint64_t HiAddr = ((ResolverAddr + 0x8000) >> 16);
 
   for (unsigned I = 0; I < NumTrampolines; ++I) {
     Trampolines[10 * I + 0] = 0x03e0c025;                            // move $t8,$ra
@@ -889,16 +845,17 @@ void OrcMips64::writeTrampolines(uint8_t *TrampolineMem, void *ResolverAddr,
     Trampolines[10 * I + 3] = 0x0019cc38;                            // dsll $t9,$t9,16
     Trampolines[10 * I + 4] = 0x67390000 | (HiAddr & 0xFFFF);        // daddiu $t9,$t9,%hi(ptr)
     Trampolines[10 * I + 5] = 0x0019cc38;                            // dsll $t9,$t9,16
-    Trampolines[10 * I + 6] = 0x67390000 | (ResolveAddr & 0xFFFF);   // daddiu $t9,$t9,%lo(ptr)
+    Trampolines[10 * I + 6] =
+        0x67390000 | (ResolverAddr & 0xFFFF); // daddiu $t9,$t9,%lo(ptr)
     Trampolines[10 * I + 7] = 0x0320f809;                            // jalr $t9
     Trampolines[10 * I + 8] = 0x00000000;                            // nop
     Trampolines[10 * I + 9] = 0x00000000;                            // nop
   }
 }
 
-Error OrcMips64::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
-                                        unsigned MinStubs,
-                                        void *InitialPtrVal) {
+void OrcMips64::writeIndirectStubsBlock(
+    char *StubsBlockWorkingMem, JITTargetAddress StubsBlockTargetAddress,
+    JITTargetAddress PointersBlockTargetAddress, unsigned NumStubs) {
   // Stub format is:
   //
   // .section __orc_stubs
@@ -926,31 +883,14 @@ Error OrcMips64::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
   //                 .dword 0x0
   //
   // ...
-  const unsigned StubSize = IndirectStubsInfo::StubSize;
-
-  // Emit at least MinStubs, rounded up to fill the pages allocated.
-  static const unsigned PageSize = sys::Process::getPageSizeEstimate();
-  unsigned NumPages = ((MinStubs * StubSize) + (PageSize - 1)) / PageSize;
-  unsigned NumStubs = (NumPages * PageSize) / StubSize;
 
-  // Allocate memory for stubs and pointers in one call.
-  std::error_code EC;
-  auto StubsMem = sys::OwningMemoryBlock(sys::Memory::allocateMappedMemory(
-      2 * NumPages * PageSize, nullptr,
-      sys::Memory::MF_READ | sys::Memory::MF_WRITE, EC));
-
-  if (EC)
-    return errorCodeToError(EC);
-
-  // Create separate MemoryBlocks representing the stubs and pointers.
-  sys::MemoryBlock StubsBlock(StubsMem.base(), NumPages * PageSize);
-  sys::MemoryBlock PtrsBlock(static_cast<char *>(StubsMem.base()) +
-                                 NumPages * PageSize,
-                             NumPages * PageSize);
+  assert(stubAndPointerRangesOk<OrcAArch64>(
+             StubsBlockTargetAddress, PointersBlockTargetAddress, NumStubs) &&
+         "PointersBlock is out of range");
 
   // Populate the stubs page stubs and mark it executable.
-  uint32_t *Stub = reinterpret_cast<uint32_t *>(StubsBlock.base());
-  uint64_t PtrAddr = reinterpret_cast<uint64_t>(PtrsBlock.base());
+  uint32_t *Stub = reinterpret_cast<uint32_t *>(StubsBlockWorkingMem);
+  uint64_t PtrAddr = PointersBlockTargetAddress;
 
   for (unsigned I = 0; I < NumStubs; ++I, PtrAddr += 8) {
     uint64_t HeighestAddr = ((PtrAddr + 0x800080008000) >> 48);
@@ -965,19 +905,6 @@ Error OrcMips64::emitIndirectStubsBlock(IndirectStubsInfo &StubsInfo,
     Stub[8 * I + 6] = 0x03200008;                            // jr $t9
     Stub[8 * I + 7] = 0x00000000;                            // nop
   }
-
-  if (auto EC = sys::Memory::protectMappedMemory(
-          StubsBlock, sys::Memory::MF_READ | sys::Memory::MF_EXEC))
-    return errorCodeToError(EC);
-
-  // Initialize all pointers to point at FailureAddress.
-  void **Ptr = reinterpret_cast<void **>(PtrsBlock.base());
-  for (unsigned I = 0; I < NumStubs; ++I)
-    Ptr[I] = InitialPtrVal;
-
-  StubsInfo = IndirectStubsInfo(NumStubs, std::move(StubsMem));
-
-  return Error::success();
 }
 } // End namespace orc.
 } // End namespace llvm.
diff --git a/llvm/lib/ExecutionEngine/Orc/OrcCBindingsStack.h b/llvm/lib/ExecutionEngine/Orc/OrcCBindingsStack.h
index e0af3df9d010..87bb4398765d 100644
--- a/llvm/lib/ExecutionEngine/Orc/OrcCBindingsStack.h
+++ b/llvm/lib/ExecutionEngine/Orc/OrcCBindingsStack.h
@@ -133,7 +133,7 @@ private:
       orc::SymbolNameSet Result;
 
       for (auto &S : Symbols) {
-        if (auto Sym = findSymbol(*S)) {
+        if (auto Sym = findSymbol(std::string(*S))) {
           if (!Sym.getFlags().isStrong())
             Result.insert(S);
         } else if (auto Err = Sym.takeError()) {
@@ -151,7 +151,7 @@ private:
       orc::SymbolNameSet UnresolvedSymbols;
 
       for (auto &S : Symbols) {
-        if (auto Sym = findSymbol(*S)) {
+        if (auto Sym = findSymbol(std::string(*S))) {
           if (auto Addr = Sym.getAddress()) {
             Query->notifySymbolMetRequiredState(
                 S, JITEvaluatedSymbol(*Addr, Sym.getFlags()));
diff --git a/llvm/lib/ExecutionEngine/Orc/OrcMCJITReplacement.h b/llvm/lib/ExecutionEngine/Orc/OrcMCJITReplacement.h
index 169dc8f1d02b..139572bd6977 100644
--- a/llvm/lib/ExecutionEngine/Orc/OrcMCJITReplacement.h
+++ b/llvm/lib/ExecutionEngine/Orc/OrcMCJITReplacement.h
@@ -154,7 +154,8 @@ class OrcMCJITReplacement : public ExecutionEngine {
           M.reportError(std::move(Err));
           return SymbolNameSet();
         } else {
-          if (auto Sym2 = M.ClientResolver->findSymbolInLogicalDylib(*S)) {
+          if (auto Sym2 =
+                  M.ClientResolver->findSymbolInLogicalDylib(std::string(*S))) {
             if (!Sym2.getFlags().isStrong())
               Result.insert(S);
           } else if (auto Err = Sym2.takeError()) {
@@ -187,7 +188,7 @@ class OrcMCJITReplacement : public ExecutionEngine {
           M.ES.legacyFailQuery(*Query, std::move(Err));
           return SymbolNameSet();
         } else {
-          if (auto Sym2 = M.ClientResolver->findSymbol(*S)) {
+          if (auto Sym2 = M.ClientResolver->findSymbol(std::string(*S))) {
             if (auto Addr = Sym2.getAddress()) {
               Query->notifySymbolMetRequiredState(
                   S, JITEvaluatedSymbol(*Addr, Sym2.getFlags()));
@@ -378,9 +379,9 @@ public:
 
 private:
   JITSymbol findMangledSymbol(StringRef Name) {
-    if (auto Sym = LazyEmitLayer.findSymbol(Name, false))
+    if (auto Sym = LazyEmitLayer.findSymbol(std::string(Name), false))
       return Sym;
-    if (auto Sym = ClientResolver->findSymbol(Name))
+    if (auto Sym = ClientResolver->findSymbol(std::string(Name)))
       return Sym;
     if (auto Sym = scanArchives(Name))
       return Sym;
diff --git a/llvm/lib/ExecutionEngine/Orc/OrcV2CBindings.cpp b/llvm/lib/ExecutionEngine/Orc/OrcV2CBindings.cpp
new file mode 100644
index 000000000000..5933c2e666d1
--- /dev/null
+++ b/llvm/lib/ExecutionEngine/Orc/OrcV2CBindings.cpp
@@ -0,0 +1,254 @@
+//===--------------- OrcV2CBindings.cpp - C bindings OrcV2 APIs -----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm-c/Orc.h"
+#include "llvm-c/TargetMachine.h"
+
+#include "llvm/ExecutionEngine/Orc/JITTargetMachineBuilder.h"
+#include "llvm/ExecutionEngine/Orc/LLJIT.h"
+
+using namespace llvm;
+using namespace llvm::orc;
+
+namespace llvm {
+namespace orc {
+
+class OrcV2CAPIHelper {
+public:
+  using PoolEntry = SymbolStringPtr::PoolEntry;
+  using PoolEntryPtr = SymbolStringPtr::PoolEntryPtr;
+
+  static PoolEntryPtr releaseSymbolStringPtr(SymbolStringPtr S) {
+    PoolEntryPtr Result = nullptr;
+    std::swap(Result, S.S);
+    return Result;
+  }
+
+  static PoolEntryPtr getRawPoolEntryPtr(const SymbolStringPtr &S) {
+    return S.S;
+  }
+
+  static void releasePoolEntry(PoolEntryPtr P) {
+    SymbolStringPtr S;
+    S.S = P;
+  }
+};
+
+} // end namespace orc
+} // end namespace llvm
+
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(ExecutionSession, LLVMOrcExecutionSessionRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(OrcV2CAPIHelper::PoolEntry,
+                                   LLVMOrcSymbolStringPoolEntryRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(JITDylib, LLVMOrcJITDylibRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(JITDylib::DefinitionGenerator,
+                                   LLVMOrcJITDylibDefinitionGeneratorRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(ThreadSafeContext,
+                                   LLVMOrcThreadSafeContextRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(ThreadSafeModule, LLVMOrcThreadSafeModuleRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(JITTargetMachineBuilder,
+                                   LLVMOrcJITTargetMachineBuilderRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(LLJITBuilder, LLVMOrcLLJITBuilderRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(LLJIT, LLVMOrcLLJITRef)
+
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(TargetMachine, LLVMTargetMachineRef)
+
+LLVMOrcSymbolStringPoolEntryRef
+LLVMOrcExecutionSessionIntern(LLVMOrcExecutionSessionRef ES, const char *Name) {
+  return wrap(
+      OrcV2CAPIHelper::releaseSymbolStringPtr(unwrap(ES)->intern(Name)));
+}
+
+void LLVMOrcReleaseSymbolStringPoolEntry(LLVMOrcSymbolStringPoolEntryRef S) {
+  OrcV2CAPIHelper::releasePoolEntry(unwrap(S));
+}
+
+void LLVMOrcDisposeJITDylibDefinitionGenerator(
+    LLVMOrcJITDylibDefinitionGeneratorRef DG) {
+  delete unwrap(DG);
+}
+
+void LLVMOrcJITDylibAddGenerator(LLVMOrcJITDylibRef JD,
+                                 LLVMOrcJITDylibDefinitionGeneratorRef DG) {
+  unwrap(JD)->addGenerator(
+      std::unique_ptr<JITDylib::DefinitionGenerator>(unwrap(DG)));
+}
+
+LLVMErrorRef LLVMOrcCreateDynamicLibrarySearchGeneratorForProcess(
+    LLVMOrcJITDylibDefinitionGeneratorRef *Result, char GlobalPrefix,
+    LLVMOrcSymbolPredicate Filter, void *FilterCtx) {
+  assert(Result && "Result can not be null");
+  assert((Filter || !FilterCtx) &&
+         "if Filter is null then FilterCtx must also be null");
+
+  DynamicLibrarySearchGenerator::SymbolPredicate Pred;
+  if (Filter)
+    Pred = [=](const SymbolStringPtr &Name) -> bool {
+      return Filter(wrap(OrcV2CAPIHelper::getRawPoolEntryPtr(Name)), FilterCtx);
+    };
+
+  auto ProcessSymsGenerator =
+      DynamicLibrarySearchGenerator::GetForCurrentProcess(GlobalPrefix, Pred);
+
+  if (!ProcessSymsGenerator) {
+    *Result = 0;
+    return wrap(ProcessSymsGenerator.takeError());
+  }
+
+  *Result = wrap(ProcessSymsGenerator->release());
+  return LLVMErrorSuccess;
+}
+
+LLVMOrcThreadSafeContextRef LLVMOrcCreateNewThreadSafeContext(void) {
+  return wrap(new ThreadSafeContext(std::make_unique<LLVMContext>()));
+}
+
+LLVMContextRef
+LLVMOrcThreadSafeContextGetContext(LLVMOrcThreadSafeContextRef TSCtx) {
+  return wrap(unwrap(TSCtx)->getContext());
+}
+
+void LLVMOrcDisposeThreadSafeContext(LLVMOrcThreadSafeContextRef TSCtx) {
+  delete unwrap(TSCtx);
+}
+
+LLVMOrcThreadSafeModuleRef
+LLVMOrcCreateNewThreadSafeModule(LLVMModuleRef M,
+                                 LLVMOrcThreadSafeContextRef TSCtx) {
+  return wrap(
+      new ThreadSafeModule(std::unique_ptr<Module>(unwrap(M)), *unwrap(TSCtx)));
+}
+
+void LLVMOrcDisposeThreadSafeModule(LLVMOrcThreadSafeModuleRef TSM) {
+  delete unwrap(TSM);
+}
+
+LLVMErrorRef LLVMOrcJITTargetMachineBuilderDetectHost(
+    LLVMOrcJITTargetMachineBuilderRef *Result) {
+  assert(Result && "Result can not be null");
+
+  auto JTMB = JITTargetMachineBuilder::detectHost();
+  if (!JTMB) {
+    Result = 0;
+    return wrap(JTMB.takeError());
+  }
+
+  *Result = wrap(new JITTargetMachineBuilder(std::move(*JTMB)));
+  return LLVMErrorSuccess;
+}
+
+LLVMOrcJITTargetMachineBuilderRef
+LLVMOrcJITTargetMachineBuilderFromTargetMachine(LLVMTargetMachineRef TM) {
+  auto *TemplateTM = unwrap(TM);
+
+  auto JTMB =
+      std::make_unique<JITTargetMachineBuilder>(TemplateTM->getTargetTriple());
+
+  (*JTMB)
+      .setCPU(TemplateTM->getTargetCPU().str())
+      .setRelocationModel(TemplateTM->getRelocationModel())
+      .setCodeModel(TemplateTM->getCodeModel())
+      .setCodeGenOptLevel(TemplateTM->getOptLevel())
+      .setFeatures(TemplateTM->getTargetFeatureString())
+      .setOptions(TemplateTM->Options);
+
+  LLVMDisposeTargetMachine(TM);
+
+  return wrap(JTMB.release());
+}
+
+void LLVMOrcDisposeJITTargetMachineBuilder(
+    LLVMOrcJITTargetMachineBuilderRef JTMB) {
+  delete unwrap(JTMB);
+}
+
+LLVMOrcLLJITBuilderRef LLVMOrcCreateLLJITBuilder(void) {
+  return wrap(new LLJITBuilder());
+}
+
+void LLVMOrcDisposeLLJITBuilder(LLVMOrcLLJITBuilderRef Builder) {
+  delete unwrap(Builder);
+}
+
+void LLVMOrcLLJITBuilderSetJITTargetMachineBuilder(
+    LLVMOrcLLJITBuilderRef Builder, LLVMOrcJITTargetMachineBuilderRef JTMB) {
+  unwrap(Builder)->setJITTargetMachineBuilder(*unwrap(JTMB));
+}
+
+LLVMErrorRef LLVMOrcCreateLLJIT(LLVMOrcLLJITRef *Result,
+                                LLVMOrcLLJITBuilderRef Builder) {
+  assert(Result && "Result can not be null");
+
+  if (!Builder)
+    Builder = LLVMOrcCreateLLJITBuilder();
+
+  auto J = unwrap(Builder)->create();
+  LLVMOrcDisposeLLJITBuilder(Builder);
+
+  if (!J) {
+    Result = 0;
+    return wrap(J.takeError());
+  }
+
+  *Result = wrap(J->release());
+  return LLVMErrorSuccess;
+}
+
+LLVMErrorRef LLVMOrcDisposeLLJIT(LLVMOrcLLJITRef J) {
+  delete unwrap(J);
+  return LLVMErrorSuccess;
+}
+
+LLVMOrcExecutionSessionRef LLVMOrcLLJITGetExecutionSession(LLVMOrcLLJITRef J) {
+  return wrap(&unwrap(J)->getExecutionSession());
+}
+
+LLVMOrcJITDylibRef LLVMOrcLLJITGetMainJITDylib(LLVMOrcLLJITRef J) {
+  return wrap(&unwrap(J)->getMainJITDylib());
+}
+
+const char *LLVMOrcLLJITGetTripleString(LLVMOrcLLJITRef J) {
+  return unwrap(J)->getTargetTriple().str().c_str();
+}
+
+char LLVMOrcLLJITGetGlobalPrefix(LLVMOrcLLJITRef J) {
+  return unwrap(J)->getDataLayout().getGlobalPrefix();
+}
+
+LLVMOrcSymbolStringPoolEntryRef
+LLVMOrcLLJITMangleAndIntern(LLVMOrcLLJITRef J, const char *UnmangledName) {
+  return wrap(OrcV2CAPIHelper::releaseSymbolStringPtr(
+      unwrap(J)->mangleAndIntern(UnmangledName)));
+}
+
+LLVMErrorRef LLVMOrcLLJITAddObjectFile(LLVMOrcLLJITRef J, LLVMOrcJITDylibRef JD,
+                                       LLVMMemoryBufferRef ObjBuffer) {
+  return wrap(unwrap(J)->addObjectFile(
+      *unwrap(JD), std::unique_ptr<MemoryBuffer>(unwrap(ObjBuffer))));
+}
+
+LLVMErrorRef LLVMOrcLLJITAddLLVMIRModule(LLVMOrcLLJITRef J,
+                                         LLVMOrcJITDylibRef JD,
+                                         LLVMOrcThreadSafeModuleRef TSM) {
+  return wrap(unwrap(J)->addIRModule(*unwrap(JD), std::move(*unwrap(TSM))));
+}
+
+LLVMErrorRef LLVMOrcLLJITLookup(LLVMOrcLLJITRef J,
+                                LLVMOrcJITTargetAddress *Result,
+                                const char *Name) {
+  assert(Result && "Result can not be null");
+
+  auto Sym = unwrap(J)->lookup(Name);
+  if (!Sym) {
+    *Result = 0;
+    return wrap(Sym.takeError());
+  }
+
+  *Result = Sym->getAddress();
+  return LLVMErrorSuccess;
+}
diff --git a/llvm/lib/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.cpp b/llvm/lib/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.cpp
index a92264c0be14..21925726072e 100644
--- a/llvm/lib/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.cpp
@@ -7,6 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.h"
+#include "llvm/Object/COFF.h"
 
 namespace {
 
@@ -46,10 +47,10 @@ public:
       MR.addDependenciesForAll(Deps);
     };
 
-    JITDylibSearchOrder SearchOrder;
-    MR.getTargetJITDylib().withSearchOrderDo(
-        [&](const JITDylibSearchOrder &JDs) { SearchOrder = JDs; });
-    ES.lookup(LookupKind::Static, SearchOrder, InternedSymbols,
+    JITDylibSearchOrder LinkOrder;
+    MR.getTargetJITDylib().withLinkOrderDo(
+        [&](const JITDylibSearchOrder &LO) { LinkOrder = LO; });
+    ES.lookup(LookupKind::Static, LinkOrder, InternedSymbols,
               SymbolState::Resolved, std::move(OnResolvedWithUnwrap),
               RegisterDependencies);
   }
@@ -80,8 +81,12 @@ RTDyldObjectLinkingLayer::RTDyldObjectLinkingLayer(
 
 RTDyldObjectLinkingLayer::~RTDyldObjectLinkingLayer() {
   std::lock_guard<std::mutex> Lock(RTDyldLayerMutex);
-  for (auto &MemMgr : MemMgrs)
+  for (auto &MemMgr : MemMgrs) {
+    for (auto *L : EventListeners)
+      L->notifyFreeingObject(
+          static_cast<uint64_t>(reinterpret_cast<uintptr_t>(MemMgr.get())));
     MemMgr->deregisterEHFrames();
+  }
 }
 
 void RTDyldObjectLinkingLayer::emit(MaterializationResponsibility R,
@@ -96,13 +101,7 @@ void RTDyldObjectLinkingLayer::emit(MaterializationResponsibility R,
 
   auto &ES = getExecutionSession();
 
-  // Create a MemoryBufferRef backed MemoryBuffer (i.e. shallow) copy of the
-  // the underlying buffer to pass into RuntimeDyld. This allows us to hold
-  // ownership of the real underlying buffer and return it to the user once
-  // the object has been emitted.
-  auto ObjBuffer = MemoryBuffer::getMemBuffer(O->getMemBufferRef(), false);
-
-  auto Obj = object::ObjectFile::createObjectFile(*ObjBuffer);
+  auto Obj = object::ObjectFile::createObjectFile(*O);
 
   if (!Obj) {
     getExecutionSession().reportError(Obj.takeError());
@@ -115,7 +114,27 @@ void RTDyldObjectLinkingLayer::emit(MaterializationResponsibility R,
   auto InternalSymbols = std::make_shared<std::set<StringRef>>();
   {
     for (auto &Sym : (*Obj)->symbols()) {
-      if (!(Sym.getFlags() & object::BasicSymbolRef::SF_Global)) {
+
+      // Skip file symbols.
+      if (auto SymType = Sym.getType()) {
+        if (*SymType == object::SymbolRef::ST_File)
+          continue;
+      } else {
+        ES.reportError(SymType.takeError());
+        R.failMaterialization();
+        return;
+      }
+
+      Expected<uint32_t> SymFlagsOrErr = Sym.getFlags();
+      if (!SymFlagsOrErr) {
+        // TODO: Test this error.
+        ES.reportError(SymFlagsOrErr.takeError());
+        R.failMaterialization();
+        return;
+      }
+
+      // Don't include symbols that aren't global.
+      if (!(*SymFlagsOrErr & object::BasicSymbolRef::SF_Global)) {
         if (auto SymName = Sym.getName())
           InternalSymbols->insert(*SymName);
         else {
@@ -141,25 +160,79 @@ void RTDyldObjectLinkingLayer::emit(MaterializationResponsibility R,
   JITDylibSearchOrderResolver Resolver(*SharedR);
 
   jitLinkForORC(
-      **Obj, std::move(O), *MemMgr, Resolver, ProcessAllSections,
-      [this, K, SharedR, &Obj, InternalSymbols](
+      object::OwningBinary<object::ObjectFile>(std::move(*Obj), std::move(O)),
+      *MemMgr, Resolver, ProcessAllSections,
+      [this, K, SharedR, MemMgr, InternalSymbols](
+          const object::ObjectFile &Obj,
           std::unique_ptr<RuntimeDyld::LoadedObjectInfo> LoadedObjInfo,
           std::map<StringRef, JITEvaluatedSymbol> ResolvedSymbols) {
-        return onObjLoad(K, *SharedR, **Obj, std::move(LoadedObjInfo),
+        return onObjLoad(K, *SharedR, Obj, MemMgr, std::move(LoadedObjInfo),
                          ResolvedSymbols, *InternalSymbols);
       },
-      [this, K, SharedR, O = std::move(O)](Error Err) mutable {
-        onObjEmit(K, std::move(O), *SharedR, std::move(Err));
+      [this, K, SharedR, MemMgr](object::OwningBinary<object::ObjectFile> Obj,
+                                 Error Err) mutable {
+        onObjEmit(K, *SharedR, std::move(Obj), MemMgr, std::move(Err));
       });
 }
 
+void RTDyldObjectLinkingLayer::registerJITEventListener(JITEventListener &L) {
+  std::lock_guard<std::mutex> Lock(RTDyldLayerMutex);
+  assert(llvm::none_of(EventListeners,
+                       [&](JITEventListener *O) { return O == &L; }) &&
+         "Listener has already been registered");
+  EventListeners.push_back(&L);
+}
+
+void RTDyldObjectLinkingLayer::unregisterJITEventListener(JITEventListener &L) {
+  std::lock_guard<std::mutex> Lock(RTDyldLayerMutex);
+  auto I = llvm::find(EventListeners, &L);
+  assert(I != EventListeners.end() && "Listener not registered");
+  EventListeners.erase(I);
+}
+
 Error RTDyldObjectLinkingLayer::onObjLoad(
-    VModuleKey K, MaterializationResponsibility &R, object::ObjectFile &Obj,
+    VModuleKey K, MaterializationResponsibility &R,
+    const object::ObjectFile &Obj, RuntimeDyld::MemoryManager *MemMgr,
     std::unique_ptr<RuntimeDyld::LoadedObjectInfo> LoadedObjInfo,
     std::map<StringRef, JITEvaluatedSymbol> Resolved,
     std::set<StringRef> &InternalSymbols) {
   SymbolFlagsMap ExtraSymbolsToClaim;
   SymbolMap Symbols;
+
+  // Hack to support COFF constant pool comdats introduced during compilation:
+  // (See http://llvm.org/PR40074)
+  if (auto *COFFObj = dyn_cast<object::COFFObjectFile>(&Obj)) {
+    auto &ES = getExecutionSession();
+
+    // For all resolved symbols that are not already in the responsibilty set:
+    // check whether the symbol is in a comdat section and if so mark it as
+    // weak.
+    for (auto &Sym : COFFObj->symbols()) {
+      // getFlags() on COFF symbols can't fail.
+      uint32_t SymFlags = cantFail(Sym.getFlags());
+      if (SymFlags & object::BasicSymbolRef::SF_Undefined)
+        continue;
+      auto Name = Sym.getName();
+      if (!Name)
+        return Name.takeError();
+      auto I = Resolved.find(*Name);
+
+      // Skip unresolved symbols, internal symbols, and symbols that are
+      // already in the responsibility set.
+      if (I == Resolved.end() || InternalSymbols.count(*Name) ||
+          R.getSymbols().count(ES.intern(*Name)))
+        continue;
+      auto Sec = Sym.getSection();
+      if (!Sec)
+        return Sec.takeError();
+      if (*Sec == COFFObj->section_end())
+        continue;
+      auto &COFFSec = *COFFObj->getCOFFSection(**Sec);
+      if (COFFSec.Characteristics & COFF::IMAGE_SCN_LNK_COMDAT)
+        I->second.setFlags(I->second.getFlags() | JITSymbolFlags::Weak);
+    }
+  }
+
   for (auto &KV : Resolved) {
     // Scan the symbols and add them to the Symbols map for resolution.
 
@@ -184,10 +257,17 @@ Error RTDyldObjectLinkingLayer::onObjLoad(
     Symbols[InternedName] = JITEvaluatedSymbol(KV.second.getAddress(), Flags);
   }
 
-  if (!ExtraSymbolsToClaim.empty())
+  if (!ExtraSymbolsToClaim.empty()) {
     if (auto Err = R.defineMaterializing(ExtraSymbolsToClaim))
       return Err;
 
+    // If we claimed responsibility for any weak symbols but were rejected then
+    // we need to remove them from the resolved set.
+    for (auto &KV : ExtraSymbolsToClaim)
+      if (KV.second.isWeak() && !R.getSymbols().count(KV.first))
+        Symbols.erase(KV.first);
+  }
+
   if (auto Err = R.notifyResolved(Symbols)) {
     R.failMaterialization();
     return Err;
@@ -196,12 +276,17 @@ Error RTDyldObjectLinkingLayer::onObjLoad(
   if (NotifyLoaded)
     NotifyLoaded(K, Obj, *LoadedObjInfo);
 
+  std::lock_guard<std::mutex> Lock(RTDyldLayerMutex);
+  assert(!LoadedObjInfos.count(MemMgr) && "Duplicate loaded info for MemMgr");
+  LoadedObjInfos[MemMgr] = std::move(LoadedObjInfo);
+
   return Error::success();
 }
 
 void RTDyldObjectLinkingLayer::onObjEmit(
-    VModuleKey K, std::unique_ptr<MemoryBuffer> ObjBuffer,
-    MaterializationResponsibility &R, Error Err) {
+    VModuleKey K, MaterializationResponsibility &R,
+    object::OwningBinary<object::ObjectFile> O,
+    RuntimeDyld::MemoryManager *MemMgr, Error Err) {
   if (Err) {
     getExecutionSession().reportError(std::move(Err));
     R.failMaterialization();
@@ -214,6 +299,22 @@ void RTDyldObjectLinkingLayer::onObjEmit(
     return;
   }
 
+  std::unique_ptr<object::ObjectFile> Obj;
+  std::unique_ptr<MemoryBuffer> ObjBuffer;
+  std::tie(Obj, ObjBuffer) = O.takeBinary();
+
+  // Run EventListener notifyLoaded callbacks.
+  {
+    std::lock_guard<std::mutex> Lock(RTDyldLayerMutex);
+    auto LOIItr = LoadedObjInfos.find(MemMgr);
+    assert(LOIItr != LoadedObjInfos.end() && "LoadedObjInfo missing");
+    for (auto *L : EventListeners)
+      L->notifyObjectLoaded(
+          static_cast<uint64_t>(reinterpret_cast<uintptr_t>(MemMgr)), *Obj,
+          *LOIItr->second);
+    LoadedObjInfos.erase(MemMgr);
+  }
+
   if (NotifyEmitted)
     NotifyEmitted(K, std::move(ObjBuffer));
 }
diff --git a/llvm/lib/ExecutionEngine/Orc/SpeculateAnalyses.cpp b/llvm/lib/ExecutionEngine/Orc/SpeculateAnalyses.cpp
index f22acf50419d..7240c1ed0ce9 100644
--- a/llvm/lib/ExecutionEngine/Orc/SpeculateAnalyses.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/SpeculateAnalyses.cpp
@@ -209,7 +209,7 @@ void SequenceBBQuery::traverseToExitBlock(const BasicBlock *AtBB,
     VisitedBlocks.insert(std::make_pair(AtBB, BlockHint));
   }
 
-  succ_const_iterator PIt = succ_begin(AtBB), EIt = succ_end(AtBB);
+  const_succ_iterator PIt = succ_begin(AtBB), EIt = succ_end(AtBB);
   if (PIt == EIt) // No succs.
     return;
 
diff --git a/llvm/lib/ExecutionEngine/Orc/Speculation.cpp b/llvm/lib/ExecutionEngine/Orc/Speculation.cpp
index f29201c147a1..0530b1a97b67 100644
--- a/llvm/lib/ExecutionEngine/Orc/Speculation.cpp
+++ b/llvm/lib/ExecutionEngine/Orc/Speculation.cpp
@@ -96,7 +96,7 @@ void IRSpeculationLayer::emit(MaterializationResponsibility R,
               M, LoadValueTy, false, GlobalValue::LinkageTypes::InternalLinkage,
               ConstantInt::get(LoadValueTy, 0),
               "__orc_speculate.guard.for." + Fn.getName());
-          SpeculatorGuard->setAlignment(Align::None());
+          SpeculatorGuard->setAlignment(Align(1));
           SpeculatorGuard->setUnnamedAddr(GlobalValue::UnnamedAddr::Local);
 
           BasicBlock &ProgramEntry = Fn.getEntryBlock();
diff --git a/llvm/lib/ExecutionEngine/OrcError/OrcError.cpp b/llvm/lib/ExecutionEngine/OrcError/OrcError.cpp
index 5eab246d4b48..cc99e154fbec 100644
--- a/llvm/lib/ExecutionEngine/OrcError/OrcError.cpp
+++ b/llvm/lib/ExecutionEngine/OrcError/OrcError.cpp
@@ -61,6 +61,10 @@ public:
              "(Use StringError to get error message)";
     case OrcErrorCode::UnknownResourceHandle:
       return "Unknown resource handle";
+    case OrcErrorCode::MissingSymbolDefinitions:
+      return "MissingSymbolsDefinitions";
+    case OrcErrorCode::UnexpectedSymbolDefinitions:
+      return "UnexpectedSymbolDefinitions";
     }
     llvm_unreachable("Unhandled error code");
   }
diff --git a/llvm/lib/ExecutionEngine/PerfJITEvents/PerfJITEventListener.cpp b/llvm/lib/ExecutionEngine/PerfJITEvents/PerfJITEventListener.cpp
index cc196df3b2fa..d4c715cc59f6 100644
--- a/llvm/lib/ExecutionEngine/PerfJITEvents/PerfJITEventListener.cpp
+++ b/llvm/lib/ExecutionEngine/PerfJITEvents/PerfJITEventListener.cpp
@@ -34,9 +34,8 @@
 #include <mutex>
 
 #include <sys/mman.h>  // mmap()
-#include <sys/types.h> // getpid()
 #include <time.h>      // clock_gettime(), time(), localtime_r() */
-#include <unistd.h>    // for getpid(), read(), close()
+#include <unistd.h>    // for read(), close()
 
 using namespace llvm;
 using namespace llvm::object;
@@ -81,7 +80,7 @@ private:
   void NotifyDebug(uint64_t CodeAddr, DILineInfoTable Lines);
 
   // cache lookups
-  pid_t Pid;
+  sys::Process::Pid Pid;
 
   // base directory for output data
   std::string JitPath;
@@ -177,7 +176,8 @@ static inline uint64_t perf_get_timestamp(void) {
   return timespec_to_ns(&ts);
 }
 
-PerfJITEventListener::PerfJITEventListener() : Pid(::getpid()) {
+PerfJITEventListener::PerfJITEventListener()
+    : Pid(sys::Process::getProcessId()) {
   // check if clock-source is supported
   if (!perf_get_timestamp()) {
     errs() << "kernel does not support CLOCK_MONOTONIC\n";
@@ -328,7 +328,7 @@ bool PerfJITEventListener::InitDebuggingDir() {
     return false;
   }
 
-  JitPath = UniqueDebugDir.str();
+  JitPath = std::string(UniqueDebugDir.str());
 
   return true;
 }
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/JITSymbol.cpp b/llvm/lib/ExecutionEngine/RuntimeDyld/JITSymbol.cpp
index 4e2d0f422f39..0f6f9efe1102 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/JITSymbol.cpp
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/JITSymbol.cpp
@@ -14,11 +14,14 @@
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalAlias.h"
 #include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/ModuleSummaryIndex.h"
 #include "llvm/Object/ObjectFile.h"
 
 using namespace llvm;
 
 JITSymbolFlags llvm::JITSymbolFlags::fromGlobalValue(const GlobalValue &GV) {
+  assert(GV.hasName() && "Can't get flags for anonymous symbol");
+
   JITSymbolFlags Flags = JITSymbolFlags::None;
   if (GV.hasWeakLinkage() || GV.hasLinkOnceLinkage())
     Flags |= JITSymbolFlags::Weak;
@@ -33,17 +36,48 @@ JITSymbolFlags llvm::JITSymbolFlags::fromGlobalValue(const GlobalValue &GV) {
            isa<Function>(cast<GlobalAlias>(GV).getAliasee()))
     Flags |= JITSymbolFlags::Callable;
 
+  // Check for a linker-private-global-prefix on the symbol name, in which
+  // case it must be marked as non-exported.
+  if (auto *M = GV.getParent()) {
+    const auto &DL = M->getDataLayout();
+    StringRef LPGP = DL.getLinkerPrivateGlobalPrefix();
+    if (!LPGP.empty() && GV.getName().front() == '\01' &&
+        GV.getName().substr(1).startswith(LPGP))
+      Flags &= ~JITSymbolFlags::Exported;
+  }
+
+  return Flags;
+}
+
+JITSymbolFlags llvm::JITSymbolFlags::fromSummary(GlobalValueSummary *S) {
+  JITSymbolFlags Flags = JITSymbolFlags::None;
+  auto L = S->linkage();
+  if (GlobalValue::isWeakLinkage(L) || GlobalValue::isLinkOnceLinkage(L))
+    Flags |= JITSymbolFlags::Weak;
+  if (GlobalValue::isCommonLinkage(L))
+    Flags |= JITSymbolFlags::Common;
+  if (GlobalValue::isExternalLinkage(L) || GlobalValue::isExternalWeakLinkage(L))
+    Flags |= JITSymbolFlags::Exported;
+
+  if (isa<FunctionSummary>(S))
+    Flags |= JITSymbolFlags::Callable;
+
   return Flags;
 }
 
 Expected<JITSymbolFlags>
 llvm::JITSymbolFlags::fromObjectSymbol(const object::SymbolRef &Symbol) {
+  Expected<uint32_t> SymbolFlagsOrErr = Symbol.getFlags();
+  if (!SymbolFlagsOrErr)
+    // TODO: Test this error.
+    return SymbolFlagsOrErr.takeError();
+
   JITSymbolFlags Flags = JITSymbolFlags::None;
-  if (Symbol.getFlags() & object::BasicSymbolRef::SF_Weak)
+  if (*SymbolFlagsOrErr & object::BasicSymbolRef::SF_Weak)
     Flags |= JITSymbolFlags::Weak;
-  if (Symbol.getFlags() & object::BasicSymbolRef::SF_Common)
+  if (*SymbolFlagsOrErr & object::BasicSymbolRef::SF_Common)
     Flags |= JITSymbolFlags::Common;
-  if (Symbol.getFlags() & object::BasicSymbolRef::SF_Exported)
+  if (*SymbolFlagsOrErr & object::BasicSymbolRef::SF_Exported)
     Flags |= JITSymbolFlags::Exported;
 
   auto SymbolType = Symbol.getType();
@@ -58,8 +92,12 @@ llvm::JITSymbolFlags::fromObjectSymbol(const object::SymbolRef &Symbol) {
 
 ARMJITSymbolFlags
 llvm::ARMJITSymbolFlags::fromObjectSymbol(const object::SymbolRef &Symbol) {
+  Expected<uint32_t> SymbolFlagsOrErr = Symbol.getFlags();
+  if (!SymbolFlagsOrErr)
+    // TODO: Actually report errors helpfully.
+    report_fatal_error(SymbolFlagsOrErr.takeError());
   ARMJITSymbolFlags Flags;
-  if (Symbol.getFlags() & object::BasicSymbolRef::SF_Thumb)
+  if (*SymbolFlagsOrErr & object::BasicSymbolRef::SF_Thumb)
     Flags |= ARMJITSymbolFlags::Thumb;
   return Flags;
 }
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
index 2df71a5e5e74..7e9b0690ccea 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
@@ -214,8 +214,12 @@ RuntimeDyldImpl::loadObjectImpl(const object::ObjectFile &Obj) {
   {
     JITSymbolResolver::LookupSet Symbols;
     for (auto &Sym : Obj.symbols()) {
-      uint32_t Flags = Sym.getFlags();
-      if ((Flags & SymbolRef::SF_Common) || (Flags & SymbolRef::SF_Weak)) {
+      Expected<uint32_t> FlagsOrErr = Sym.getFlags();
+      if (!FlagsOrErr)
+        // TODO: Test this error.
+        return FlagsOrErr.takeError();
+      if ((*FlagsOrErr & SymbolRef::SF_Common) ||
+          (*FlagsOrErr & SymbolRef::SF_Weak)) {
         // Get symbol name.
         if (auto NameOrErr = Sym.getName())
           Symbols.insert(*NameOrErr);
@@ -234,10 +238,13 @@ RuntimeDyldImpl::loadObjectImpl(const object::ObjectFile &Obj) {
   LLVM_DEBUG(dbgs() << "Parse symbols:\n");
   for (symbol_iterator I = Obj.symbol_begin(), E = Obj.symbol_end(); I != E;
        ++I) {
-    uint32_t Flags = I->getFlags();
+    Expected<uint32_t> FlagsOrErr = I->getFlags();
+    if (!FlagsOrErr)
+      // TODO: Test this error.
+      return FlagsOrErr.takeError();
 
     // Skip undefined symbols.
-    if (Flags & SymbolRef::SF_Undefined)
+    if (*FlagsOrErr & SymbolRef::SF_Undefined)
       continue;
 
     // Get the symbol type.
@@ -287,7 +294,7 @@ RuntimeDyldImpl::loadObjectImpl(const object::ObjectFile &Obj) {
       }
     }
 
-    if (Flags & SymbolRef::SF_Absolute &&
+    if (*FlagsOrErr & SymbolRef::SF_Absolute &&
         SymType != object::SymbolRef::ST_File) {
       uint64_t Addr = 0;
       if (auto AddrOrErr = I->getAddress())
@@ -300,7 +307,7 @@ RuntimeDyldImpl::loadObjectImpl(const object::ObjectFile &Obj) {
       LLVM_DEBUG(dbgs() << "\tType: " << SymType << " (absolute) Name: " << Name
                         << " SID: " << SectionID
                         << " Offset: " << format("%p", (uintptr_t)Addr)
-                        << " flags: " << Flags << "\n");
+                        << " flags: " << *FlagsOrErr << "\n");
       GlobalSymbolTable[Name] = SymbolTableEntry(SectionID, Addr, *JITSymFlags);
     } else if (SymType == object::SymbolRef::ST_Function ||
                SymType == object::SymbolRef::ST_Data ||
@@ -332,7 +339,7 @@ RuntimeDyldImpl::loadObjectImpl(const object::ObjectFile &Obj) {
       LLVM_DEBUG(dbgs() << "\tType: " << SymType << " Name: " << Name
                         << " SID: " << SectionID
                         << " Offset: " << format("%p", (uintptr_t)SectOffset)
-                        << " flags: " << Flags << "\n");
+                        << " flags: " << *FlagsOrErr << "\n");
       GlobalSymbolTable[Name] =
           SymbolTableEntry(SectionID, SectOffset, *JITSymFlags);
     }
@@ -592,8 +599,11 @@ Error RuntimeDyldImpl::computeTotalAllocSize(const ObjectFile &Obj,
   uint32_t CommonAlign = 1;
   for (symbol_iterator I = Obj.symbol_begin(), E = Obj.symbol_end(); I != E;
        ++I) {
-    uint32_t Flags = I->getFlags();
-    if (Flags & SymbolRef::SF_Common) {
+    Expected<uint32_t> FlagsOrErr = I->getFlags();
+    if (!FlagsOrErr)
+      // TODO: Test this error.
+      return FlagsOrErr.takeError();
+    if (*FlagsOrErr & SymbolRef::SF_Common) {
       // Add the common symbols to a list.  We'll allocate them all below.
       uint64_t Size = I->getCommonSize();
       uint32_t Align = I->getAlignment();
@@ -1190,16 +1200,16 @@ Error RuntimeDyldImpl::resolveExternalSymbols() {
 
 void RuntimeDyldImpl::finalizeAsync(
     std::unique_ptr<RuntimeDyldImpl> This,
-    unique_function<void(Error)> OnEmitted,
-    std::unique_ptr<MemoryBuffer> UnderlyingBuffer) {
+    unique_function<void(object::OwningBinary<object::ObjectFile>, Error)>
+        OnEmitted,
+    object::OwningBinary<object::ObjectFile> O) {
 
   auto SharedThis = std::shared_ptr<RuntimeDyldImpl>(std::move(This));
   auto PostResolveContinuation =
-      [SharedThis, OnEmitted = std::move(OnEmitted),
-       UnderlyingBuffer = std::move(UnderlyingBuffer)](
+      [SharedThis, OnEmitted = std::move(OnEmitted), O = std::move(O)](
           Expected<JITSymbolResolver::LookupResult> Result) mutable {
         if (!Result) {
-          OnEmitted(Result.takeError());
+          OnEmitted(std::move(O), Result.takeError());
           return;
         }
 
@@ -1213,10 +1223,11 @@ void RuntimeDyldImpl::finalizeAsync(
         SharedThis->registerEHFrames();
         std::string ErrMsg;
         if (SharedThis->MemMgr.finalizeMemory(&ErrMsg))
-          OnEmitted(make_error<StringError>(std::move(ErrMsg),
+          OnEmitted(std::move(O),
+                    make_error<StringError>(std::move(ErrMsg),
                                             inconvertibleErrorCode()));
         else
-          OnEmitted(Error::success());
+          OnEmitted(std::move(O), Error::success());
       };
 
   JITSymbolResolver::LookupSet Symbols;
@@ -1403,32 +1414,35 @@ void RuntimeDyld::deregisterEHFrames() {
 // FIXME: Kill this with fire once we have a new JIT linker: this is only here
 // so that we can re-use RuntimeDyld's implementation without twisting the
 // interface any further for ORC's purposes.
-void jitLinkForORC(object::ObjectFile &Obj,
-                   std::unique_ptr<MemoryBuffer> UnderlyingBuffer,
-                   RuntimeDyld::MemoryManager &MemMgr,
-                   JITSymbolResolver &Resolver, bool ProcessAllSections,
-                   unique_function<Error(
-                       std::unique_ptr<RuntimeDyld::LoadedObjectInfo> LoadedObj,
-                       std::map<StringRef, JITEvaluatedSymbol>)>
-                       OnLoaded,
-                   unique_function<void(Error)> OnEmitted) {
+void jitLinkForORC(
+    object::OwningBinary<object::ObjectFile> O,
+    RuntimeDyld::MemoryManager &MemMgr, JITSymbolResolver &Resolver,
+    bool ProcessAllSections,
+    unique_function<
+        Error(const object::ObjectFile &Obj,
+              std::unique_ptr<RuntimeDyld::LoadedObjectInfo> LoadedObj,
+              std::map<StringRef, JITEvaluatedSymbol>)>
+        OnLoaded,
+    unique_function<void(object::OwningBinary<object::ObjectFile>, Error)>
+        OnEmitted) {
 
   RuntimeDyld RTDyld(MemMgr, Resolver);
   RTDyld.setProcessAllSections(ProcessAllSections);
 
-  auto Info = RTDyld.loadObject(Obj);
+  auto Info = RTDyld.loadObject(*O.getBinary());
 
   if (RTDyld.hasError()) {
-    OnEmitted(make_error<StringError>(RTDyld.getErrorString(),
-                                      inconvertibleErrorCode()));
+    OnEmitted(std::move(O), make_error<StringError>(RTDyld.getErrorString(),
+                                                    inconvertibleErrorCode()));
     return;
   }
 
-  if (auto Err = OnLoaded(std::move(Info), RTDyld.getSymbolTable()))
-    OnEmitted(std::move(Err));
+  if (auto Err =
+          OnLoaded(*O.getBinary(), std::move(Info), RTDyld.getSymbolTable()))
+    OnEmitted(std::move(O), std::move(Err));
 
   RuntimeDyldImpl::finalizeAsync(std::move(RTDyld.Dyld), std::move(OnEmitted),
-                                 std::move(UnderlyingBuffer));
+                                 std::move(O));
 }
 
 } // end namespace llvm
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldCOFF.cpp b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldCOFF.cpp
index 6e3cd7cd2cfc..1d8f1ac8ac8a 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldCOFF.cpp
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldCOFF.cpp
@@ -18,6 +18,7 @@
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/Triple.h"
 #include "llvm/Object/ObjectFile.h"
+#include "llvm/Support/FormatVariadic.h"
 
 using namespace llvm;
 using namespace llvm::object;
@@ -75,7 +76,42 @@ RuntimeDyldCOFF::loadObject(const object::ObjectFile &O) {
 
 uint64_t RuntimeDyldCOFF::getSymbolOffset(const SymbolRef &Sym) {
   // The value in a relocatable COFF object is the offset.
-  return Sym.getValue();
+  return cantFail(Sym.getValue());
+}
+
+uint64_t RuntimeDyldCOFF::getDLLImportOffset(unsigned SectionID, StubMap &Stubs,
+                                             StringRef Name,
+                                             bool SetSectionIDMinus1) {
+  LLVM_DEBUG(dbgs() << "Getting DLLImport entry for " << Name << "... ");
+  assert(Name.startswith(getImportSymbolPrefix()) && "Not a DLLImport symbol?");
+  RelocationValueRef Reloc;
+  Reloc.SymbolName = Name.data();
+  auto I = Stubs.find(Reloc);
+  if (I != Stubs.end()) {
+    LLVM_DEBUG(dbgs() << format("{0:x8}", I->second) << "\n");
+    return I->second;
+  }
+
+  assert(SectionID < Sections.size() && "SectionID out of range");
+  auto &Sec = Sections[SectionID];
+  auto EntryOffset = alignTo(Sec.getStubOffset(), PointerSize);
+  Sec.advanceStubOffset(EntryOffset + PointerSize - Sec.getStubOffset());
+  Stubs[Reloc] = EntryOffset;
+
+  RelocationEntry RE(SectionID, EntryOffset, PointerReloc, 0, false,
+                     Log2_64(PointerSize));
+  // Hack to tell I386/Thumb resolveRelocation that this isn't section relative.
+  if (SetSectionIDMinus1)
+    RE.Sections.SectionA = -1;
+  addRelocationForSymbol(RE, Name.drop_front(getImportSymbolPrefix().size()));
+
+  LLVM_DEBUG({
+    dbgs() << "Creating entry at "
+           << formatv("{0:x16} + {1:x8} ( {2:x16} )", Sec.getLoadAddress(),
+                      EntryOffset, Sec.getLoadAddress() + EntryOffset)
+           << "\n";
+  });
+  return EntryOffset;
 }
 
 bool RuntimeDyldCOFF::isCompatibleFile(const object::ObjectFile &Obj) const {
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldCOFF.h b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldCOFF.h
index 4efd18a2e6c5..41ee06c15448 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldCOFF.h
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldCOFF.h
@@ -36,9 +36,22 @@ public:
 
 protected:
   RuntimeDyldCOFF(RuntimeDyld::MemoryManager &MemMgr,
-                  JITSymbolResolver &Resolver)
-    : RuntimeDyldImpl(MemMgr, Resolver) {}
+                  JITSymbolResolver &Resolver, unsigned PointerSize,
+                  uint32_t PointerReloc)
+      : RuntimeDyldImpl(MemMgr, Resolver), PointerSize(PointerSize),
+        PointerReloc(PointerReloc) {
+    assert((PointerSize == 4 || PointerSize == 8) && "Unexpected pointer size");
+  }
+
   uint64_t getSymbolOffset(const SymbolRef &Sym);
+  uint64_t getDLLImportOffset(unsigned SectionID, StubMap &Stubs,
+                              StringRef Name, bool SetSectionIDMinus1 = false);
+
+  static constexpr StringRef getImportSymbolPrefix() { return "__imp_"; }
+
+private:
+  unsigned PointerSize;
+  uint32_t PointerReloc;
 };
 
 } // end namespace llvm
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldChecker.cpp b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldChecker.cpp
index 2ac0586ff324..e5e512672daa 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldChecker.cpp
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldChecker.cpp
@@ -9,6 +9,7 @@
 #include "llvm/ExecutionEngine/RuntimeDyldChecker.h"
 #include "RuntimeDyldCheckerImpl.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCDisassembler/MCDisassembler.h"
 #include "llvm/MC/MCInst.h"
@@ -704,10 +705,11 @@ bool RuntimeDyldCheckerImpl::checkAllRulesInBuffer(StringRef RulePrefix,
   bool DidAllTestsPass = true;
   unsigned NumRules = 0;
 
+  std::string CheckExpr;
   const char *LineStart = MemBuf->getBufferStart();
 
   // Eat whitespace.
-  while (LineStart != MemBuf->getBufferEnd() && std::isspace(*LineStart))
+  while (LineStart != MemBuf->getBufferEnd() && isSpace(*LineStart))
     ++LineStart;
 
   while (LineStart != MemBuf->getBufferEnd() && *LineStart != '\0') {
@@ -717,14 +719,23 @@ bool RuntimeDyldCheckerImpl::checkAllRulesInBuffer(StringRef RulePrefix,
       ++LineEnd;
 
     StringRef Line(LineStart, LineEnd - LineStart);
-    if (Line.startswith(RulePrefix)) {
-      DidAllTestsPass &= check(Line.substr(RulePrefix.size()));
-      ++NumRules;
+    if (Line.startswith(RulePrefix))
+      CheckExpr += Line.substr(RulePrefix.size()).str();
+
+    // If there's a check expr string...
+    if (!CheckExpr.empty()) {
+      // ... and it's complete then run it, otherwise remove the trailer '\'.
+      if (CheckExpr.back() != '\\') {
+        DidAllTestsPass &= check(CheckExpr);
+        CheckExpr.clear();
+        ++NumRules;
+      } else
+        CheckExpr.pop_back();
     }
 
     // Eat whitespace.
     LineStart = LineEnd;
-    while (LineStart != MemBuf->getBufferEnd() && std::isspace(*LineStart))
+    while (LineStart != MemBuf->getBufferEnd() && isSpace(*LineStart))
       ++LineStart;
   }
   return DidAllTestsPass && (NumRules != 0);
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
index 440ab4174a56..7c39ddc8b1da 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
@@ -399,6 +399,13 @@ void RuntimeDyldELF::resolveAArch64Relocation(const SectionEntry &Section,
   case ELF::R_AARCH64_ABS64:
     write(isBE, TargetPtr, Value + Addend);
     break;
+  case ELF::R_AARCH64_PLT32: {
+    uint64_t Result = Value + Addend - FinalAddress;
+    assert(static_cast<int64_t>(Result) >= INT32_MIN &&
+           static_cast<int64_t>(Result) <= INT32_MAX);
+    write(isBE, TargetPtr, static_cast<uint32_t>(Result));
+    break;
+  }
   case ELF::R_AARCH64_PREL32: {
     uint64_t Result = Value + Addend - FinalAddress;
     assert(static_cast<int64_t>(Result) >= INT32_MIN &&
@@ -554,7 +561,7 @@ void RuntimeDyldELF::setMipsABI(const ObjectFile &Obj) {
     IsMipsO32ABI = AbiVariant & ELF::EF_MIPS_ABI_O32;
     IsMipsN32ABI = AbiVariant & ELF::EF_MIPS_ABI2;
   }
-  IsMipsN64ABI = Obj.getFileFormatName().equals("ELF64-mips");
+  IsMipsN64ABI = Obj.getFileFormatName().equals("elf64-mips");
 }
 
 // Return the .TOC. section and offset.
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h
index ef0784e2273b..31892b7466e6 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h
@@ -88,12 +88,13 @@ class RuntimeDyldELF : public RuntimeDyldImpl {
 
   void setMipsABI(const ObjectFile &Obj) override;
 
-  Error findPPC64TOCSection(const ELFObjectFileBase &Obj,
+  Error findPPC64TOCSection(const object::ELFObjectFileBase &Obj,
                             ObjSectionToIDMap &LocalSections,
                             RelocationValueRef &Rel);
-  Error findOPDEntrySection(const ELFObjectFileBase &Obj,
+  Error findOPDEntrySection(const object::ELFObjectFileBase &Obj,
                             ObjSectionToIDMap &LocalSections,
                             RelocationValueRef &Rel);
+
 protected:
   size_t getGOTEntrySize() override;
 
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h
index cec7b92b8c48..a9346536fd09 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h
@@ -26,6 +26,7 @@
 #include "llvm/Support/Host.h"
 #include "llvm/Support/Mutex.h"
 #include "llvm/Support/SwapByteOrder.h"
+#include <deque>
 #include <map>
 #include <system_error>
 #include <unordered_map>
@@ -35,8 +36,6 @@ using namespace llvm::object;
 
 namespace llvm {
 
-class Twine;
-
 #define UNIMPLEMENTED_RELOC(RelType) \
   case RelType: \
     return make_error<RuntimeDyldError>("Unimplemented relocation: " #RelType)
@@ -74,7 +73,7 @@ class SectionEntry {
 public:
   SectionEntry(StringRef name, uint8_t *address, size_t size,
                size_t allocationSize, uintptr_t objAddress)
-      : Name(name), Address(address), Size(size),
+      : Name(std::string(name)), Address(address), Size(size),
         LoadAddress(reinterpret_cast<uintptr_t>(address)), StubOffset(size),
         AllocationSize(allocationSize), ObjAddress(objAddress) {
     // AllocationSize is used only in asserts, prevent an "unused private field"
@@ -190,13 +189,11 @@ public:
 
 class RelocationValueRef {
 public:
-  unsigned SectionID;
-  uint64_t Offset;
-  int64_t Addend;
-  const char *SymbolName;
+  unsigned SectionID = 0;
+  uint64_t Offset = 0;
+  int64_t Addend = 0;
+  const char *SymbolName = nullptr;
   bool IsStubThumb = false;
-  RelocationValueRef() : SectionID(0), Offset(0), Addend(0),
-                         SymbolName(nullptr) {}
 
   inline bool operator==(const RelocationValueRef &Other) const {
     return SectionID == Other.SectionID && Offset == Other.Offset &&
@@ -251,7 +248,9 @@ protected:
 
   // A list of all sections emitted by the dynamic linker.  These sections are
   // referenced in the code by means of their index in this list - SectionID.
-  typedef SmallVector<SectionEntry, 64> SectionList;
+  // Because references may be kept while the list grows, use a container that
+  // guarantees reference stability.
+  typedef std::deque<SectionEntry> SectionList;
   SectionList Sections;
 
   typedef unsigned SID; // Type for SectionIDs
@@ -319,32 +318,18 @@ protected:
   std::string ErrorStr;
 
   void writeInt16BE(uint8_t *Addr, uint16_t Value) {
-    if (IsTargetLittleEndian)
-      sys::swapByteOrder(Value);
-    *Addr       = (Value >> 8) & 0xFF;
-    *(Addr + 1) = Value & 0xFF;
+    llvm::support::endian::write<uint16_t, llvm::support::unaligned>(
+        Addr, Value, IsTargetLittleEndian ? support::little : support::big);
   }
 
   void writeInt32BE(uint8_t *Addr, uint32_t Value) {
-    if (IsTargetLittleEndian)
-      sys::swapByteOrder(Value);
-    *Addr       = (Value >> 24) & 0xFF;
-    *(Addr + 1) = (Value >> 16) & 0xFF;
-    *(Addr + 2) = (Value >> 8) & 0xFF;
-    *(Addr + 3) = Value & 0xFF;
+    llvm::support::endian::write<uint32_t, llvm::support::unaligned>(
+        Addr, Value, IsTargetLittleEndian ? support::little : support::big);
   }
 
   void writeInt64BE(uint8_t *Addr, uint64_t Value) {
-    if (IsTargetLittleEndian)
-      sys::swapByteOrder(Value);
-    *Addr       = (Value >> 56) & 0xFF;
-    *(Addr + 1) = (Value >> 48) & 0xFF;
-    *(Addr + 2) = (Value >> 40) & 0xFF;
-    *(Addr + 3) = (Value >> 32) & 0xFF;
-    *(Addr + 4) = (Value >> 24) & 0xFF;
-    *(Addr + 5) = (Value >> 16) & 0xFF;
-    *(Addr + 6) = (Value >> 8) & 0xFF;
-    *(Addr + 7) = Value & 0xFF;
+    llvm::support::endian::write<uint64_t, llvm::support::unaligned>(
+        Addr, Value, IsTargetLittleEndian ? support::little : support::big);
   }
 
   virtual void setMipsABI(const ObjectFile &Obj) {
@@ -548,9 +533,11 @@ public:
 
   void resolveLocalRelocations();
 
-  static void finalizeAsync(std::unique_ptr<RuntimeDyldImpl> This,
-                            unique_function<void(Error)> OnEmitted,
-                            std::unique_ptr<MemoryBuffer> UnderlyingBuffer);
+  static void finalizeAsync(
+      std::unique_ptr<RuntimeDyldImpl> This,
+      unique_function<void(object::OwningBinary<object::ObjectFile>, Error)>
+          OnEmitted,
+      object::OwningBinary<object::ObjectFile> O);
 
   void reassignSectionAddress(unsigned SectionID, uint64_t Addr);
 
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFAArch64.h b/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFAArch64.h
index a94f54f50ac4..14510e56b35a 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFAArch64.h
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFAArch64.h
@@ -89,7 +89,8 @@ private:
 public:
   RuntimeDyldCOFFAArch64(RuntimeDyld::MemoryManager &MM,
                          JITSymbolResolver &Resolver)
-      : RuntimeDyldCOFF(MM, Resolver), ImageBase(0) {}
+      : RuntimeDyldCOFF(MM, Resolver, 8, COFF::IMAGE_REL_ARM64_ADDR64),
+        ImageBase(0) {}
 
   unsigned getStubAlignment() override { return 8; }
 
@@ -161,7 +162,7 @@ public:
     uint64_t Offset = RelI->getOffset();
 
     // If there is no section, this must be an external reference.
-    const bool IsExtern = Section == Obj.section_end();
+    bool IsExtern = Section == Obj.section_end();
 
     // Determine the Addend used to adjust the relocation value.
     uint64_t Addend = 0;
@@ -169,6 +170,24 @@ public:
     uintptr_t ObjTarget = AddendSection.getObjAddress() + Offset;
     uint8_t *Displacement = (uint8_t *)ObjTarget;
 
+    unsigned TargetSectionID = -1;
+    uint64_t TargetOffset = -1;
+
+    if (TargetName.startswith(getImportSymbolPrefix())) {
+      TargetSectionID = SectionID;
+      TargetOffset = getDLLImportOffset(SectionID, Stubs, TargetName);
+      TargetName = StringRef();
+      IsExtern = false;
+    } else if (!IsExtern) {
+      if (auto TargetSectionIDOrErr = findOrEmitSection(
+              Obj, *Section, Section->isText(), ObjSectionToID))
+        TargetSectionID = *TargetSectionIDOrErr;
+      else
+        return TargetSectionIDOrErr.takeError();
+
+      TargetOffset = getSymbolOffset(*Symbol);
+    }
+
     switch (RelType) {
     case COFF::IMAGE_REL_ARM64_ADDR32:
     case COFF::IMAGE_REL_ARM64_ADDR32NB:
@@ -224,18 +243,10 @@ public:
                       << TargetName << " Addend " << Addend << "\n");
 #endif
 
-    unsigned TargetSectionID = -1;
     if (IsExtern) {
       RelocationEntry RE(SectionID, Offset, RelType, Addend);
       addRelocationForSymbol(RE, TargetName);
     } else {
-      if (auto TargetSectionIDOrErr = findOrEmitSection(
-              Obj, *Section, Section->isText(), ObjSectionToID)) {
-        TargetSectionID = *TargetSectionIDOrErr;
-      } else
-        return TargetSectionIDOrErr.takeError();
-
-      uint64_t TargetOffset = getSymbolOffset(*Symbol);
       RelocationEntry RE(SectionID, Offset, RelType, TargetOffset + Addend);
       addRelocationForSection(RE, TargetSectionID);
     }
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFI386.h b/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFI386.h
index 40910bea0c36..03c38260bece 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFI386.h
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFI386.h
@@ -25,7 +25,7 @@ class RuntimeDyldCOFFI386 : public RuntimeDyldCOFF {
 public:
   RuntimeDyldCOFFI386(RuntimeDyld::MemoryManager &MM,
                       JITSymbolResolver &Resolver)
-      : RuntimeDyldCOFF(MM, Resolver) {}
+      : RuntimeDyldCOFF(MM, Resolver, 4, COFF::IMAGE_REL_I386_DIR32) {}
 
   unsigned getMaxStubSize() const override {
     return 8; // 2-byte jmp instruction + 32-bit relative address + 2 byte pad
@@ -53,10 +53,28 @@ public:
     if (!SectionOrErr)
       return SectionOrErr.takeError();
     auto Section = *SectionOrErr;
+    bool IsExtern = Section == Obj.section_end();
 
     uint64_t RelType = RelI->getType();
     uint64_t Offset = RelI->getOffset();
 
+    unsigned TargetSectionID = -1;
+    uint64_t TargetOffset = -1;
+    if (TargetName.startswith(getImportSymbolPrefix())) {
+      TargetSectionID = SectionID;
+      TargetOffset = getDLLImportOffset(SectionID, Stubs, TargetName, true);
+      TargetName = StringRef();
+      IsExtern = false;
+    } else if (!IsExtern) {
+      if (auto TargetSectionIDOrErr = findOrEmitSection(
+              Obj, *Section, Section->isText(), ObjSectionToID))
+        TargetSectionID = *TargetSectionIDOrErr;
+      else
+        return TargetSectionIDOrErr.takeError();
+      if (RelType != COFF::IMAGE_REL_I386_SECTION)
+        TargetOffset = getSymbolOffset(*Symbol);
+    }
+
     // Determine the Addend used to adjust the relocation value.
     uint64_t Addend = 0;
     SectionEntry &AddendSection = Sections[SectionID];
@@ -83,16 +101,10 @@ public:
                       << " RelType: " << RelTypeName << " TargetName: "
                       << TargetName << " Addend " << Addend << "\n");
 
-    unsigned TargetSectionID = -1;
-    if (Section == Obj.section_end()) {
+    if (IsExtern) {
       RelocationEntry RE(SectionID, Offset, RelType, 0, -1, 0, 0, 0, false, 0);
       addRelocationForSymbol(RE, TargetName);
     } else {
-      if (auto TargetSectionIDOrErr =
-          findOrEmitSection(Obj, *Section, Section->isText(), ObjSectionToID))
-        TargetSectionID = *TargetSectionIDOrErr;
-      else
-        return TargetSectionIDOrErr.takeError();
 
       switch (RelType) {
       case COFF::IMAGE_REL_I386_ABSOLUTE:
@@ -103,7 +115,7 @@ public:
       case COFF::IMAGE_REL_I386_REL32: {
         RelocationEntry RE =
             RelocationEntry(SectionID, Offset, RelType, Addend, TargetSectionID,
-                            getSymbolOffset(*Symbol), 0, 0, false, 0);
+                            TargetOffset, 0, 0, false, 0);
         addRelocationForSection(RE, TargetSectionID);
         break;
       }
@@ -114,15 +126,14 @@ public:
         break;
       }
       case COFF::IMAGE_REL_I386_SECREL: {
-        RelocationEntry RE = RelocationEntry(SectionID, Offset, RelType,
-                                             getSymbolOffset(*Symbol) + Addend);
+        RelocationEntry RE =
+            RelocationEntry(SectionID, Offset, RelType, TargetOffset + Addend);
         addRelocationForSection(RE, TargetSectionID);
         break;
       }
       default:
         llvm_unreachable("unsupported relocation type");
       }
-
     }
 
     return ++RelI;
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFThumb.h b/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFThumb.h
index bb2e9626e0b0..721f2b14829a 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFThumb.h
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFThumb.h
@@ -48,7 +48,7 @@ class RuntimeDyldCOFFThumb : public RuntimeDyldCOFF {
 public:
   RuntimeDyldCOFFThumb(RuntimeDyld::MemoryManager &MM,
                        JITSymbolResolver &Resolver)
-      : RuntimeDyldCOFF(MM, Resolver) {}
+      : RuntimeDyldCOFF(MM, Resolver, 4, COFF::IMAGE_REL_ARM_ADDR32) {}
 
   unsigned getMaxStubSize() const override {
     return 16; // 8-byte load instructions, 4-byte jump, 4-byte padding
@@ -103,16 +103,29 @@ public:
                       << " RelType: " << RelTypeName << " TargetName: "
                       << TargetName << " Addend " << Addend << "\n");
 
+    bool IsExtern = Section == Obj.section_end();
     unsigned TargetSectionID = -1;
-    if (Section == Obj.section_end()) {
-      RelocationEntry RE(SectionID, Offset, RelType, 0, -1, 0, 0, 0, false, 0);
-      addRelocationForSymbol(RE, TargetName);
-    } else {
+    uint64_t TargetOffset = -1;
+
+    if (TargetName.startswith(getImportSymbolPrefix())) {
+      TargetSectionID = SectionID;
+      TargetOffset = getDLLImportOffset(SectionID, Stubs, TargetName, true);
+      TargetName = StringRef();
+      IsExtern = false;
+    } else if (!IsExtern) {
       if (auto TargetSectionIDOrErr =
           findOrEmitSection(Obj, *Section, Section->isText(), ObjSectionToID))
         TargetSectionID = *TargetSectionIDOrErr;
       else
         return TargetSectionIDOrErr.takeError();
+      if (RelType != COFF::IMAGE_REL_ARM_SECTION)
+        TargetOffset = getSymbolOffset(*Symbol);
+    }
+
+    if (IsExtern) {
+      RelocationEntry RE(SectionID, Offset, RelType, 0, -1, 0, 0, 0, false, 0);
+      addRelocationForSymbol(RE, TargetName);
+    } else {
 
       // We need to find out if the relocation is relative to a thumb function
       // so that we include the ISA selection bit when resolve the relocation
@@ -124,16 +137,16 @@ public:
         // This relocation is ignored.
         break;
       case COFF::IMAGE_REL_ARM_ADDR32: {
-        RelocationEntry RE = RelocationEntry(
-            SectionID, Offset, RelType, Addend, TargetSectionID,
-            getSymbolOffset(*Symbol), 0, 0, false, 0, IsTargetThumbFunc);
+        RelocationEntry RE =
+            RelocationEntry(SectionID, Offset, RelType, Addend, TargetSectionID,
+                            TargetOffset, 0, 0, false, 0, IsTargetThumbFunc);
         addRelocationForSection(RE, TargetSectionID);
         break;
       }
       case COFF::IMAGE_REL_ARM_ADDR32NB: {
         RelocationEntry RE =
             RelocationEntry(SectionID, Offset, RelType, Addend, TargetSectionID,
-                            getSymbolOffset(*Symbol), 0, 0, false, 0);
+                            TargetOffset, 0, 0, false, 0);
         addRelocationForSection(RE, TargetSectionID);
         break;
       }
@@ -144,24 +157,23 @@ public:
         break;
       }
       case COFF::IMAGE_REL_ARM_SECREL: {
-        RelocationEntry RE = RelocationEntry(SectionID, Offset, RelType,
-                                             getSymbolOffset(*Symbol) + Addend);
+        RelocationEntry RE =
+            RelocationEntry(SectionID, Offset, RelType, TargetOffset + Addend);
         addRelocationForSection(RE, TargetSectionID);
         break;
       }
       case COFF::IMAGE_REL_ARM_MOV32T: {
-        RelocationEntry RE = RelocationEntry(
-            SectionID, Offset, RelType, Addend, TargetSectionID,
-            getSymbolOffset(*Symbol), 0, 0, false, 0, IsTargetThumbFunc);
+        RelocationEntry RE =
+            RelocationEntry(SectionID, Offset, RelType, Addend, TargetSectionID,
+                            TargetOffset, 0, 0, false, 0, IsTargetThumbFunc);
         addRelocationForSection(RE, TargetSectionID);
         break;
       }
       case COFF::IMAGE_REL_ARM_BRANCH20T:
       case COFF::IMAGE_REL_ARM_BRANCH24T:
       case COFF::IMAGE_REL_ARM_BLX23T: {
-        RelocationEntry RE =
-            RelocationEntry(SectionID, Offset, RelType,
-                            getSymbolOffset(*Symbol) + Addend, true, 0);
+        RelocationEntry RE = RelocationEntry(SectionID, Offset, RelType,
+                                             TargetOffset + Addend, true, 0);
         addRelocationForSection(RE, TargetSectionID);
         break;
       }
@@ -256,7 +268,6 @@ public:
       EncodeImmediate(&Target[0],
                       (static_cast<uint32_t>(Result) >> 00) | ISASelectionBit);
       EncodeImmediate(&Target[4], static_cast<uint32_t>(Result) >> 16);
-
       break;
     }
     case COFF::IMAGE_REL_ARM_BRANCH20T: {
diff --git a/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFX86_64.h b/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFX86_64.h
index dc4af08583de..ebe3ca33d308 100644
--- a/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFX86_64.h
+++ b/llvm/lib/ExecutionEngine/RuntimeDyld/Targets/RuntimeDyldCOFFX86_64.h
@@ -56,7 +56,8 @@ private:
 public:
   RuntimeDyldCOFFX86_64(RuntimeDyld::MemoryManager &MM,
                         JITSymbolResolver &Resolver)
-    : RuntimeDyldCOFF(MM, Resolver), ImageBase(0) {}
+      : RuntimeDyldCOFF(MM, Resolver, 8, COFF::IMAGE_REL_AMD64_ADDR64),
+        ImageBase(0) {}
 
   unsigned getStubAlignment() override { return 1; }
 
@@ -202,7 +203,7 @@ public:
       return SectionOrError.takeError();
     object::section_iterator SecI = *SectionOrError;
     // If there is no section, this must be an external reference.
-    const bool IsExtern = SecI == Obj.section_end();
+    bool IsExtern = SecI == Obj.section_end();
 
     // Determine the Addend used to adjust the relocation value.
     uint64_t RelType = RelI->getType();
@@ -214,7 +215,25 @@ public:
     Expected<StringRef> TargetNameOrErr = Symbol->getName();
     if (!TargetNameOrErr)
       return TargetNameOrErr.takeError();
+
     StringRef TargetName = *TargetNameOrErr;
+    unsigned TargetSectionID = 0;
+    uint64_t TargetOffset = 0;
+
+    if (TargetName.startswith(getImportSymbolPrefix())) {
+      assert(IsExtern && "DLLImport not marked extern?");
+      TargetSectionID = SectionID;
+      TargetOffset = getDLLImportOffset(SectionID, Stubs, TargetName);
+      TargetName = StringRef();
+      IsExtern = false;
+    } else if (!IsExtern) {
+      if (auto TargetSectionIDOrErr =
+              findOrEmitSection(Obj, *SecI, SecI->isText(), ObjSectionToID))
+        TargetSectionID = *TargetSectionIDOrErr;
+      else
+        return TargetSectionIDOrErr.takeError();
+      TargetOffset = getSymbolOffset(*Symbol);
+    }
 
     switch (RelType) {
 
@@ -253,14 +272,6 @@ public:
       RelocationEntry RE(SectionID, Offset, RelType, Addend);
       addRelocationForSymbol(RE, TargetName);
     } else {
-      bool IsCode = SecI->isText();
-      unsigned TargetSectionID;
-      if (auto TargetSectionIDOrErr =
-          findOrEmitSection(Obj, *SecI, IsCode, ObjSectionToID))
-        TargetSectionID = *TargetSectionIDOrErr;
-      else
-        return TargetSectionIDOrErr.takeError();
-      uint64_t TargetOffset = getSymbolOffset(*Symbol);
       RelocationEntry RE(SectionID, Offset, RelType, TargetOffset + Addend);
       addRelocationForSection(RE, TargetSectionID);
     }