Change TargetLoweringObjectFile ownership

[aditya_nandakumar]

Summary

This patch changes the ownership of TLOF from TargetLoweringBase to TargetMachine so that different subtargets could share the TLOF effectively.
I have made the change to X86 only currently and it passes make check. I could update the other targets after feedback.

Please review.

Thanks
Aditya

[echristo]

Hi Aditya!

I think this will work. You should return a unique_ptr out of the construction routine and use make_unique instead of new. Trying to come up with a way that doesn't require a unique_ptr here, but I'm not coming up with one.

Thanks!

-eric

[aditya_nandakumar]

Hi Eric

Apologies - I missed your review.
I have updated the Construction to return unique_ptr and uses make_unique as you suggested. I too am not sure how to eliminate the unique_ptr here.

Aditya

[echristo]

LGTM. Thanks!

Getting to all of the other targets shortly?

-eric

[aditya_nandakumar]

Thank you. I’ll update the remaining targets shortly.

[aditya_nandakumar]

Hi Eric,

Would it be preferable to add a virtual method - getTargetLoweringObjectFile() to TargetMachine ? All targets to could override this.

Thanks
Aditya

[echristo]

Agreed, have it just query the Subtarget for now to get it on targets that
haven't replaced.

That said, it'll make it easier if you just call it getObjFileLowering like
the current name (which I think is best).

Thanks!

-eric

[aditya_nandakumar]

Updated all target machines to own the TargetLoweringObjectFile.
Changed the method name to getObjFileLowering()

Make check passes . Please let me know if this is good to submit.

Aditya

[echristo]

You have extra whitespace cleanups in a couple of files, please remove those before committing. Are you going to fix the TargetLoweringBase constructor to no longer need both arguments since the TM can get everything in a follow up commit or would you like to do that as part of this?

[aditya_nandakumar]

Thanks Eric.
I’ll remove those whitespace cleanups (default in my editor) and commit soon.

I’d like to fix the TargetLoweringBase constructor in a followup commit.

Thanks
Aditya

[echristo]

Sounds good to me. Should just be a quick change. :)

-eric

include/llvm/Target/TargetLowering.h

     llvm_unreachable("Invalid content kind");
   }
 
-  /// NOTE: The constructor takes ownership of TLOF.
+  /// NOTE: The TargetMachine owns TLOF.
   explicit TargetLoweringBase(const TargetMachine &TM,
                               const TargetLoweringObjectFile *TLOF);
-  virtual ~TargetLoweringBase();
+  virtual ~TargetLoweringBase() {}
 
 protected:
   /// \brief Initialize all of the actions to default values.

include/llvm/Target/TargetMachine.h

 class VectorTargetTransformInfo;
 class formatted_raw_ostream;
 class raw_ostream;
+class TargetLoweringObjectFile;
 
 // The old pass manager infrastructure is hidden in a legacy namespace now.
 namespace legacy {

   virtual const TargetSubtargetInfo *getSubtargetImpl(const Function &) const {
     return getSubtargetImpl();
   }
+  virtual TargetLoweringObjectFile *getObjFileLowering() const {
+    return nullptr;
+  }
 
   /// getSubtarget - This method returns a pointer to the specified type of
   /// TargetSubtargetInfo.  In debug builds, it verifies that the object being

lib/CodeGen/TargetLoweringBase.cpp

   InitLibcallCallingConvs(LibcallCallingConvs);
 }
 
-TargetLoweringBase::~TargetLoweringBase() {
-  delete &TLOF;
-}
-
 void TargetLoweringBase::initActions() {
   // All operations default to being supported.
   memset(OpActions, 0, sizeof(OpActions));

lib/Target/AArch64/AArch64ISelLowering.cpp

                          cl::desc("Allow AArch64 SLI/SRI formation"),
                          cl::init(false));
 
-//===----------------------------------------------------------------------===//
-// AArch64 Lowering public interface.
-//===----------------------------------------------------------------------===//
-static TargetLoweringObjectFile *createTLOF(const Triple &TT) {
-  if (TT.isOSBinFormatMachO())
-    return new AArch64_MachoTargetObjectFile();
-
-  return new AArch64_ELFTargetObjectFile();
-}
-
 AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM)
-    : TargetLowering(TM, createTLOF(Triple(TM.getTargetTriple()))) {
+    : TargetLowering(TM, TM.getObjFileLowering()) {
   Subtarget = &TM.getSubtarget<AArch64Subtarget>();
 
   // AArch64 doesn't have comparisons which set GPRs or setcc instructions, so

 
       continue;
     }
-    
     if (VA.isRegLoc()) {
       // Arguments stored in registers.
       EVT RegVT = VA.getLocVT();

 
   // The index of an EXT is the first element if it is not UNDEF.
   // Watch out for the beginning UNDEFs. The EXT index should be the expected
-  // value of the first element.  E.g. 
+  // value of the first element.  E.g.
   // <-1, -1, 3, ...> is treated as <1, 2, 3, ...>.
   // <-1, -1, 0, 1, ...> is treated as <2*NumElts-2, 2*NumElts-1, 0, 1, ...>.
   // ExpectedElt is the last mask index plus 1.

lib/Target/AArch64/AArch64TargetMachine.h

 
 class AArch64TargetMachine : public LLVMTargetMachine {
 protected:
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   AArch64Subtarget Subtarget;
   mutable StringMap<std::unique_ptr<AArch64Subtarget>> SubtargetMap;
 

   /// \brief Register AArch64 analysis passes with a pass manager.
   void addAnalysisPasses(PassManagerBase &PM) override;
 
+  TargetLoweringObjectFile* getObjFileLowering() const override {
+    return TLOF.get();
+  }
+
 private:
   bool isLittle;
 };

lib/Target/AArch64/AArch64TargetMachine.cpp

 
 #include "AArch64.h"
 #include "AArch64TargetMachine.h"
+#include "AArch64TargetObjectFile.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/RegAllocRegistry.h"
 #include "llvm/IR/Function.h"

   RegisterTargetMachine<AArch64leTargetMachine> Z(TheARM64Target);
 }
 
+//===----------------------------------------------------------------------===//
+// AArch64 Lowering public interface.
+//===----------------------------------------------------------------------===//
+static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
+  if (TT.isOSBinFormatMachO())
+    return make_unique<AArch64_MachoTargetObjectFile>();
+
+  return make_unique<AArch64_ELFTargetObjectFile>();
+}
+
 /// TargetMachine ctor - Create an AArch64 architecture model.
 ///
 AArch64TargetMachine::AArch64TargetMachine(const Target &T, StringRef TT,

                                            CodeGenOpt::Level OL,
                                            bool LittleEndian)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      TLOF(createTLOF(Triple(getTargetTriple()))),
       Subtarget(TT, CPU, FS, *this, LittleEndian), isLittle(LittleEndian) {
   initAsmInfo();
 }

lib/Target/ARM/ARMISelLowering.cpp

   addTypeForNEON(VT, MVT::v2f64, MVT::v4i32);
 }
 
-static TargetLoweringObjectFile *createTLOF(const Triple &TT) {
-  if (TT.isOSBinFormatMachO())
-    return new TargetLoweringObjectFileMachO();
-  if (TT.isOSWindows())
-    return new TargetLoweringObjectFileCOFF();
-  return new ARMElfTargetObjectFile();
-}
-
 ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM)
-    : TargetLowering(TM, createTLOF(Triple(TM.getTargetTriple()))) {
+    : TargetLowering(TM, TM.getObjFileLowering()) {
   Subtarget = &TM.getSubtarget<ARMSubtarget>();
   RegInfo = TM.getSubtargetImpl()->getRegisterInfo();
   Itins = TM.getSubtargetImpl()->getInstrItineraryData();

   // Get widened type and narrowed type.
   MVT widenType;
   unsigned numElem = VT.getVectorNumElements();
-  
   EVT inputLaneType = Vec.getValueType().getVectorElementType();
   switch (inputLaneType.getSimpleVT().SimpleTy) {
     case MVT::i8: widenType = MVT::getVectorVT(MVT::i16, numElem); break;

lib/Target/ARM/ARMTargetMachine.h

 
 class ARMBaseTargetMachine : public LLVMTargetMachine {
 protected:
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   ARMSubtarget        Subtarget;
   bool isLittle;
   mutable StringMap<std::unique_ptr<ARMSubtarget>> SubtargetMap;

 
   // Pass Pipeline Configuration
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
+
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 };
 
 /// ARMTargetMachine - ARM target machine.

lib/Target/ARM/ARMTargetMachine.cpp

 #include "ARM.h"
 #include "ARMTargetMachine.h"
 #include "ARMFrameLowering.h"
+#include "ARMTargetObjectFile.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/MC/MCAsmInfo.h"

   RegisterTargetMachine<ThumbBETargetMachine> B(TheThumbBETarget);
 }
 
+static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
+  if (TT.isOSBinFormatMachO())
+    return make_unique<TargetLoweringObjectFileMachO>();
+  if (TT.isOSWindows())
+    return make_unique<TargetLoweringObjectFileCOFF>();
+  return make_unique<ARMElfTargetObjectFile>();
+}
+
 /// TargetMachine ctor - Create an ARM architecture model.
 ///
 ARMBaseTargetMachine::ARMBaseTargetMachine(const Target &T, StringRef TT,

                                            Reloc::Model RM, CodeModel::Model CM,
                                            CodeGenOpt::Level OL, bool isLittle)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      TLOF(createTLOF(Triple(getTargetTriple()))),
       Subtarget(TT, CPU, FS, *this, isLittle), isLittle(isLittle) {
 
   // Default to triple-appropriate float ABI

lib/Target/Hexagon/HexagonISelLowering.cpp

 //===----------------------------------------------------------------------===//
 
 HexagonTargetLowering::HexagonTargetLowering(const TargetMachine &targetmachine)
-    : TargetLowering(targetmachine, new HexagonTargetObjectFile()),
+    : TargetLowering(targetmachine, targetmachine.getObjFileLowering()),
       TM(targetmachine) {
 
   const HexagonSubtarget &Subtarget = TM.getSubtarget<HexagonSubtarget>();

lib/Target/Hexagon/HexagonTargetMachine.h

 class Module;
 
 class HexagonTargetMachine : public LLVMTargetMachine {
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   HexagonSubtarget Subtarget;
 
 public:

   static unsigned getModuleMatchQuality(const Module &M);
 
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
+
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 };
 
 extern bool flag_aligned_memcpy;

lib/Target/Hexagon/HexagonTargetMachine.cpp

                                            Reloc::Model RM, CodeModel::Model CM,
                                            CodeGenOpt::Level OL)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      TLOF(make_unique<HexagonTargetObjectFile>()),
       Subtarget(TT, CPU, FS, *this) {
     initAsmInfo();
 }

lib/Target/MSP430/MSP430ISelLowering.cpp

              clEnumValEnd));
 
 MSP430TargetLowering::MSP430TargetLowering(const TargetMachine &TM)
-    : TargetLowering(TM, new TargetLoweringObjectFileELF()) {
+    : TargetLowering(TM, TM.getObjFileLowering()) {
 
   // Set up the register classes.
   addRegisterClass(MVT::i8,  &MSP430::GR8RegClass);

lib/Target/MSP430/MSP430TargetMachine.h

 /// MSP430TargetMachine
 ///
 class MSP430TargetMachine : public LLVMTargetMachine {
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   MSP430Subtarget        Subtarget;
 
 public:

     return &Subtarget;
   }
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
+
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 }; // MSP430TargetMachine.
 
 } // end namespace llvm

lib/Target/MSP430/MSP430TargetMachine.cpp

 //===----------------------------------------------------------------------===//
 
 #include "MSP430TargetMachine.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
 #include "MSP430.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/MC/MCAsmInfo.h"

                                          Reloc::Model RM, CodeModel::Model CM,
                                          CodeGenOpt::Level OL)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      TLOF(make_unique<TargetLoweringObjectFileELF>()),
       Subtarget(TT, CPU, FS, *this) {
   initAsmInfo();
 }

lib/Target/Mips/MipsISelLowering.cpp

 
 MipsTargetLowering::MipsTargetLowering(const MipsTargetMachine &TM,
                                        const MipsSubtarget &STI)
-    : TargetLowering(TM, new MipsTargetObjectFile()), Subtarget(STI) {
+    : TargetLowering(TM, TM.getObjFileLowering()), Subtarget(STI) {
   // Mips does not have i1 type, so use i32 for
   // setcc operations results (slt, sgt, ...).
   setBooleanContents(ZeroOrOneBooleanContent);

lib/Target/Mips/MipsTargetMachine.h

 
 class MipsTargetMachine : public LLVMTargetMachine {
   bool isLittle;
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   MipsSubtarget *Subtarget;
   MipsSubtarget DefaultSubtarget;
   MipsSubtarget NoMips16Subtarget;

                     const TargetOptions &Options, Reloc::Model RM,
                     CodeModel::Model CM, CodeGenOpt::Level OL, bool isLittle);
 
-  virtual ~MipsTargetMachine() {}
-
   void addAnalysisPasses(PassManagerBase &PM) override;
 
   const MipsSubtarget *getSubtargetImpl() const override {

 
   // Pass Pipeline Configuration
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
+
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 };
 
 /// MipsebTargetMachine - Mips32/64 big endian target machine.

lib/Target/Mips/MipsTargetMachine.cpp

 #include "MipsSEISelDAGToDAG.h"
 #include "MipsSEISelLowering.h"
 #include "MipsSEInstrInfo.h"
+#include "MipsTargetObjectFile.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/PassManager.h"

                                      Reloc::Model RM, CodeModel::Model CM,
                                      CodeGenOpt::Level OL, bool isLittle)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
-      isLittle(isLittle), Subtarget(nullptr),
+      isLittle(isLittle),
+      TLOF(make_unique<MipsTargetObjectFile>()),
+      Subtarget(nullptr),
       DefaultSubtarget(TT, CPU, FS, isLittle, this),
       NoMips16Subtarget(TT, CPU, FS.empty() ? "-mips16" : FS.str() + ",-mips16",
                         isLittle, this),

lib/Target/NVPTX/NVPTXISelLowering.cpp

 
 // NVPTXTargetLowering Constructor.
 NVPTXTargetLowering::NVPTXTargetLowering(const NVPTXTargetMachine &TM)
-    : TargetLowering(TM, new NVPTXTargetObjectFile()), nvTM(&TM),
+    : TargetLowering(TM, TM.getObjFileLowering()), nvTM(&TM),
       nvptxSubtarget(TM.getSubtarget<NVPTXSubtarget>()) {
 
   // always lower memset, memcpy, and memmove intrinsics to load/store

                                       partVT.getTypeForEVT(F->getContext()));
           SDValue p;
           if (Ins[InsIdx].VT.getSizeInBits() > partVT.getSizeInBits()) {
-            ISD::LoadExtType ExtOp = Ins[InsIdx].Flags.isSExt() ? 
+            ISD::LoadExtType ExtOp = Ins[InsIdx].Flags.isSExt() ?
                                      ISD::SEXTLOAD : ISD::ZEXTLOAD;
             p = DAG.getExtLoad(ExtOp, dl, Ins[InsIdx].VT, Root, srcAddr,
                                MachinePointerInfo(srcValue), partVT, false,

           ObjectVT.getTypeForEVT(F->getContext()), llvm::ADDRESS_SPACE_PARAM));
       SDValue p;
        if (ObjectVT.getSizeInBits() < Ins[InsIdx].VT.getSizeInBits()) {
-        ISD::LoadExtType ExtOp = Ins[InsIdx].Flags.isSExt() ? 
+        ISD::LoadExtType ExtOp = Ins[InsIdx].Flags.isSExt() ?
                                        ISD::SEXTLOAD : ISD::ZEXTLOAD;
         p = DAG.getExtLoad(ExtOp, dl, Ins[InsIdx].VT, Root, Arg,
                            MachinePointerInfo(srcValue), ObjectVT, false, false,

lib/Target/NVPTX/NVPTXTargetMachine.h

 /// NVPTXTargetMachine
 ///
 class NVPTXTargetMachine : public LLVMTargetMachine {
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   NVPTXSubtarget Subtarget;
 
   // Hold Strings that can be free'd all together with NVPTXTargetMachine

                          bool = true) override {
     return true;
   }
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 
 }; // NVPTXTargetMachine.
 

lib/Target/NVPTX/NVPTXTargetMachine.cpp

 #include "NVPTX.h"
 #include "NVPTXAllocaHoisting.h"
 #include "NVPTXLowerAggrCopies.h"
+#include "NVPTXTargetObjectFile.h"
 #include "llvm/Analysis/Passes.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"

                                        Reloc::Model RM, CodeModel::Model CM,
                                        CodeGenOpt::Level OL, bool is64bit)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      TLOF(make_unique<NVPTXTargetObjectFile>()),
       Subtarget(TT, CPU, FS, *this, is64bit) {
   initAsmInfo();
 }

lib/Target/PowerPC/PPCISelLowering.cpp

 // FIXME: Remove this once the bug has been fixed!
 extern cl::opt<bool> ANDIGlueBug;
 
-static TargetLoweringObjectFile *createTLOF(const Triple &TT) {
-  // If it isn't a Mach-O file then it's going to be a linux ELF
-  // object file.
-  if (TT.isOSDarwin())
-    return new TargetLoweringObjectFileMachO();
-
-  return new PPC64LinuxTargetObjectFile();
-}
-
 PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM)
-    : TargetLowering(TM, createTLOF(Triple(TM.getTargetTriple()))),
+    : TargetLowering(TM, TM.getObjFileLowering()),
       Subtarget(*TM.getSubtargetImpl()) {
   setPow2SDivIsCheap();
 

       AddPromotedToType (ISD::SINT_TO_FP, MVT::i1,
                          isPPC64 ? MVT::i64 : MVT::i32);
       setOperationAction(ISD::UINT_TO_FP, MVT::i1, Promote);
-      AddPromotedToType (ISD::UINT_TO_FP, MVT::i1, 
+      AddPromotedToType (ISD::UINT_TO_FP, MVT::i1,
                          isPPC64 ? MVT::i64 : MVT::i32);
     } else {
       setOperationAction(ISD::SINT_TO_FP, MVT::i1, Custom);

 
 /// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
 /// a VMRGL* instruction with the specified unit size (1,2 or 4 bytes).
-/// The ShuffleKind distinguishes between big-endian merges with two 
+/// The ShuffleKind distinguishes between big-endian merges with two
 /// different inputs (0), either-endian merges with two identical inputs (1),
 /// and little-endian merges with two different inputs (2).  For the latter,
 /// the input operands are swapped (see PPCInstrAltivec.td).

 
 /// isVMRGHShuffleMask - Return true if this is a shuffle mask suitable for
 /// a VMRGH* instruction with the specified unit size (1,2 or 4 bytes).
-/// The ShuffleKind distinguishes between big-endian merges with two 
+/// The ShuffleKind distinguishes between big-endian merges with two
 /// different inputs (0), either-endian merges with two identical inputs (1),
 /// and little-endian merges with two different inputs (2).  For the latter,
 /// the input operands are swapped (see PPCInstrAltivec.td).

 
 /// isVSLDOIShuffleMask - If this is a vsldoi shuffle mask, return the shift
 /// amount, otherwise return -1.
-/// The ShuffleKind distinguishes between big-endian operations with two 
+/// The ShuffleKind distinguishes between big-endian operations with two
 /// different inputs (0), either-endian operations with two identical inputs
 /// (1), and little-endian operations with two different inputs (2).  For the
 /// latter, the input operands are swapped (see PPCInstrAltivec.td).

 
 #include "PPCGenCallingConv.inc"
 
-// Function whose sole purpose is to kill compiler warnings 
+// Function whose sole purpose is to kill compiler warnings
 // stemming from unused functions included from PPCGenCallingConv.inc.
 CCAssignFn *PPCTargetLowering::useFastISelCCs(unsigned Flag) const {
   return Flag ? CC_PPC64_ELF_FIS : RetCC_PPC64_ELF_FIS;

     EVT SVT = getSetCCResultType(*DAG.getContext(), N->getValueType(0));
     SDVTList VTs = DAG.getVTList(SVT, MVT::Other);
     SDValue NewInt = DAG.getNode(N->getOpcode(), dl, VTs, N->getOperand(0),
-                                 N->getOperand(1)); 
+                                 N->getOperand(1));
 
     Results.push_back(NewInt);
     Results.push_back(NewInt.getValue(1));

   for (SmallSet<SDNode *, 16>::iterator I = LoadRoots.begin(),
        IE = LoadRoots.end(); I != IE; ++I) {
     Queue.push_back(*I);
-       
     while (!Queue.empty()) {
       SDNode *LoadRoot = Queue.pop_back_val();
       if (!Visited.insert(LoadRoot))

   }
 
   // Visit all inputs, collect all binary operations (and, or, xor and
-  // select) that are all fed by extensions. 
+  // select) that are all fed by extensions.
   while (!BinOps.empty()) {
     SDValue BinOp = BinOps.back();
     BinOps.pop_back();

             BinOp.getOperand(i).getOpcode() == ISD::ANY_EXTEND) &&
            BinOp.getOperand(i).getOperand(0).getValueType() == MVT::i1) ||
           isa<ConstantSDNode>(BinOp.getOperand(i))) {
-        Inputs.push_back(BinOp.getOperand(i)); 
+        Inputs.push_back(BinOp.getOperand(i));
       } else if (BinOp.getOperand(i).getOpcode() == ISD::AND ||
                  BinOp.getOperand(i).getOpcode() == ISD::OR  ||
                  BinOp.getOperand(i).getOpcode() == ISD::XOR ||

     if (isa<ConstantSDNode>(Inputs[i]))
       continue;
     else
-      DAG.ReplaceAllUsesOfValueWith(Inputs[i], Inputs[i].getOperand(0)); 
+      DAG.ReplaceAllUsesOfValueWith(Inputs[i], Inputs[i].getOperand(0));
   }
 
   // Replace all operations (these are all the same, but have a different

   SmallPtrSet<SDNode *, 16> Visited;
 
   // Visit all inputs, collect all binary operations (and, or, xor and
-  // select) that are all fed by truncations. 
+  // select) that are all fed by truncations.
   while (!BinOps.empty()) {
     SDValue BinOp = BinOps.back();
     BinOps.pop_back();

 
       if (BinOp.getOperand(i).getOpcode() == ISD::TRUNCATE ||
           isa<ConstantSDNode>(BinOp.getOperand(i))) {
-        Inputs.push_back(BinOp.getOperand(i)); 
+        Inputs.push_back(BinOp.getOperand(i));
       } else if (BinOp.getOperand(i).getOpcode() == ISD::AND ||
                  BinOp.getOperand(i).getOpcode() == ISD::OR  ||
                  BinOp.getOperand(i).getOpcode() == ISD::XOR ||

   EVT ShiftAmountTy = getShiftAmountTy(N->getValueType(0));
   SDValue ShiftCst =
     DAG.getConstant(N->getValueSizeInBits(0)-PromBits, ShiftAmountTy);
-  return DAG.getNode(ISD::SRA, dl, N->getValueType(0), 
+  return DAG.getNode(ISD::SRA, dl, N->getValueType(0),
                      DAG.getNode(ISD::SHL, dl, N->getValueType(0),
                                  N->getOperand(0), ShiftCst), ShiftCst);
 }

     break;
   case ISD::SIGN_EXTEND:
   case ISD::ZERO_EXTEND:
-  case ISD::ANY_EXTEND: 
+  case ISD::ANY_EXTEND:
     return DAGCombineExtBoolTrunc(N, DCI);
   case ISD::TRUNCATE:
   case ISD::SETCC:

   case ISD::BRCOND: {
     SDValue Cond = N->getOperand(1);
     SDValue Target = N->getOperand(2);
- 
     if (Cond.getOpcode() == ISD::INTRINSIC_W_CHAIN &&
         cast<ConstantSDNode>(Cond.getOperand(1))->getZExtValue() ==
           Intrinsic::ppc_is_decremented_ctr_nonzero) {

lib/Target/PowerPC/PPCTargetMachine.h

 /// PPCTargetMachine - Common code between 32-bit and 64-bit PowerPC targets.
 ///
 class PPCTargetMachine : public LLVMTargetMachine {
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   PPCSubtarget Subtarget;
 
   mutable StringMap<std::unique_ptr<PPCSubtarget>> SubtargetMap;

 
   /// \brief Register PPC analysis passes with a pass manager.
   void addAnalysisPasses(PassManagerBase &PM) override;
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 };
 
 /// PPC32TargetMachine - PowerPC 32-bit target machine.

lib/Target/PowerPC/PPCTargetMachine.cpp

 //===----------------------------------------------------------------------===//
 
 #include "PPCTargetMachine.h"
+#include "PPCTargetObjectFile.h"
 #include "PPC.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/IR/Function.h"

   return FullFS;
 }
 
+static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
+  // If it isn't a Mach-O file then it's going to be a linux ELF
+  // object file.
+  if (TT.isOSDarwin())
+    return make_unique<TargetLoweringObjectFileMachO>();
+
+  return make_unique<PPC64LinuxTargetObjectFile>();
+}
+
 // The FeatureString here is a little subtle. We are modifying the feature string
 // with what are (currently) non-function specific overrides as it goes into the
 // LLVMTargetMachine constructor and then using the stored value in the

                                    CodeGenOpt::Level OL)
     : LLVMTargetMachine(T, TT, CPU, computeFSAdditions(FS, OL, TT), Options, RM,
                         CM, OL),
+      TLOF(createTLOF(Triple(getTargetTriple()))),
       Subtarget(TT, CPU, TargetFS, *this) {
   initAsmInfo();
 }

   PM.add(createBasicTargetTransformInfoPass(this));
   PM.add(createPPCTargetTransformInfoPass(this));
 }
-

lib/Target/R600/AMDGPUISelLowering.cpp

 }
 
 AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
-  TargetLowering(TM, new TargetLoweringObjectFileELF()) {
+  TargetLowering(TM, TM.getObjFileLowering()) {
 
   Subtarget = &TM.getSubtarget<AMDGPUSubtarget>();
 

lib/Target/R600/AMDGPUTargetMachine.h

 namespace llvm {
 
 class AMDGPUTargetMachine : public LLVMTargetMachine {
+  TargetLoweringObjectFile *TLOF;
   AMDGPUSubtarget Subtarget;
   AMDGPUIntrinsicInfo IntrinsicInfo;
 

 
   /// \brief Register R600 analysis passes with a pass manager.
   void addAnalysisPasses(PassManagerBase &PM) override;
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF;
+  }
 };
 
 } // End namespace llvm

lib/Target/R600/AMDGPUTargetMachine.cpp

 #include "SIInstrInfo.h"
 #include "llvm/Analysis/Passes.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/IR/Verifier.h"

                                          CodeModel::Model CM,
                                          CodeGenOpt::Level OptLevel)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OptLevel),
+      TLOF(new TargetLoweringObjectFileELF()),
       Subtarget(TT, CPU, FS, *this), IntrinsicInfo() {
   setRequiresStructuredCFG(true);
   initAsmInfo();
 }
 
 AMDGPUTargetMachine::~AMDGPUTargetMachine() {
+  delete TLOF;
 }
 
 namespace {

lib/Target/Sparc/SparcISelLowering.cpp

 }
 
 SparcTargetLowering::SparcTargetLowering(TargetMachine &TM)
-  : TargetLowering(TM, new SparcELFTargetObjectFile()) {
+  : TargetLowering(TM, TM.getObjFileLowering()) {
   Subtarget = &TM.getSubtarget<SparcSubtarget>();
 
   // Set up the register classes.

lib/Target/Sparc/SparcTargetMachine.h

 namespace llvm {
 
 class SparcTargetMachine : public LLVMTargetMachine {
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   SparcSubtarget Subtarget;
 public:
   SparcTargetMachine(const Target &T, StringRef TT,

 
   // Pass Pipeline Configuration
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 };
 
 /// SparcV8TargetMachine - Sparc 32-bit target machine

lib/Target/Sparc/SparcTargetMachine.cpp

 //===----------------------------------------------------------------------===//
 
 #include "SparcTargetMachine.h"
+#include "SparcTargetObjectFile.h"
 #include "Sparc.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/PassManager.h"

                                        CodeGenOpt::Level OL,
                                        bool is64bit)
   : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+    TLOF(make_unique<SparcELFTargetObjectFile>()),
     Subtarget(TT, CPU, FS, *this, is64bit) {
   initAsmInfo();
 }

lib/Target/SystemZ/SystemZISelLowering.cpp

 }
 
 SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &tm)
-    : TargetLowering(tm, new TargetLoweringObjectFileELF()),
+    : TargetLowering(tm, tm.getObjFileLowering()),
       Subtarget(tm.getSubtarget<SystemZSubtarget>()) {
   MVT PtrVT = getPointerTy();
 

lib/Target/SystemZ/SystemZTargetMachine.h

 class TargetFrameLowering;
 
 class SystemZTargetMachine : public LLVMTargetMachine {
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   SystemZSubtarget        Subtarget;
 
 public:

   }
   // Override LLVMTargetMachine
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 };
 
 } // end namespace llvm

lib/Target/SystemZ/SystemZTargetMachine.cpp

 #include "llvm/CodeGen/Passes.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Transforms/Scalar.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
 
 using namespace llvm;
 

                                            Reloc::Model RM, CodeModel::Model CM,
                                            CodeGenOpt::Level OL)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      TLOF(make_unique<TargetLoweringObjectFileELF>()),
       Subtarget(TT, CPU, FS, *this) {
   initAsmInfo();
 }

lib/Target/X86/X86ISelLowering.cpp

   return Insert256BitVector(V, V2, NumElems/2, DAG, dl);
 }
 
-static TargetLoweringObjectFile *createTLOF(const Triple &TT) {
-  if (TT.isOSBinFormatMachO()) {
-    if (TT.getArch() == Triple::x86_64)
-      return new X86_64MachoTargetObjectFile();
-    return new TargetLoweringObjectFileMachO();
-  }
 
-  if (TT.isOSLinux())
-    return new X86LinuxTargetObjectFile();
-  if (TT.isOSBinFormatELF())
-    return new TargetLoweringObjectFileELF();
-  if (TT.isKnownWindowsMSVCEnvironment())
-    return new X86WindowsTargetObjectFile();
-  if (TT.isOSBinFormatCOFF())
-    return new TargetLoweringObjectFileCOFF();
-  llvm_unreachable("unknown subtarget type");
-}
-
 // FIXME: This should stop caching the target machine as soon as
 // we can remove resetOperationActions et al.
 X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM)
-    : TargetLowering(TM, createTLOF(Triple(TM.getTargetTriple()))) {
+    : TargetLowering(TM, TM.getObjFileLowering()) {
   Subtarget = &TM.getSubtarget<X86Subtarget>();
   X86ScalarSSEf64 = Subtarget->hasSSE2();
   X86ScalarSSEf32 = Subtarget->hasSSE1();

       ValToCopy = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), ValToCopy);
 
     assert(VA.getLocInfo() != CCValAssign::FPExt &&
-           "Unexpected FP-extend for return value.");  
+           "Unexpected FP-extend for return value.");
 
     // If this is x86-64, and we disabled SSE, we can't return FP values,
     // or SSE or MMX vectors.

                                          MVT::getIntegerVT(VT.getSizeInBits()));
       DstVec = DAG.getNode(ISD::BITCAST, dl, VT, VecAsImm);
     }
-    else 
+    else
       DstVec = DAG.getUNDEF(VT);
     return DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, DstVec,
                        Op.getOperand(NonConstIdx),

 
 /// \brief Return true if \p N implements a horizontal binop and return the
 /// operands for the horizontal binop into V0 and V1.
-/// 
+///
 /// This is a helper function of PerformBUILD_VECTORCombine.
 /// This function checks that the build_vector \p N in input implements a
 /// horizontal operation. Parameter \p Opcode defines the kind of horizontal

   assert(BaseIdx * 2 <= LastIdx && "Invalid Indices in input!");
   assert(VT.isVector() && VT.getVectorNumElements() >= LastIdx &&
          "Invalid Vector in input!");
-  
   bool IsCommutable = (Opcode == ISD::ADD || Opcode == ISD::FADD);
   bool CanFold = true;
   unsigned ExpectedVExtractIdx = BaseIdx;

 }
 
 /// \brief Emit a sequence of two 128-bit horizontal add/sub followed by
-/// a concat_vector. 
+/// a concat_vector.
 ///
 /// This is a helper function of PerformBUILD_VECTORCombine.
 /// This function expects two 256-bit vectors called V0 and V1.
 /// At first, each vector is split into two separate 128-bit vectors.
 /// Then, the resulting 128-bit vectors are used to implement two
-/// horizontal binary operations. 
+/// horizontal binary operations.
 ///
 /// The kind of horizontal binary operation is defined by \p X86Opcode.
 ///

     // Try to match an SSE3 float HADD/HSUB.
     if (isHorizontalBinOp(BV, ISD::FADD, DAG, 0, NumElts, InVec0, InVec1))
       return DAG.getNode(X86ISD::FHADD, DL, VT, InVec0, InVec1);
-    
     if (isHorizontalBinOp(BV, ISD::FSUB, DAG, 0, NumElts, InVec0, InVec1))
       return DAG.getNode(X86ISD::FHSUB, DL, VT, InVec0, InVec1);
   } else if ((VT == MVT::v4i32 || VT == MVT::v8i16) && Subtarget->hasSSSE3()) {
     // Try to match an SSSE3 integer HADD/HSUB.
     if (isHorizontalBinOp(BV, ISD::ADD, DAG, 0, NumElts, InVec0, InVec1))
       return DAG.getNode(X86ISD::HADD, DL, VT, InVec0, InVec1);
-    
     if (isHorizontalBinOp(BV, ISD::SUB, DAG, 0, NumElts, InVec0, InVec1))
       return DAG.getNode(X86ISD::HSUB, DL, VT, InVec0, InVec1);
   }
-  
   if (!Subtarget->hasAVX())
     return SDValue();
 

       // Do this only if the target has AVX2.
       if (Subtarget->hasAVX2())
         return DAG.getNode(X86Opcode, DL, VT, InVec0, InVec1);
- 
       // Do not try to expand this build_vector into a pair of horizontal
       // add/sub if we can emit a pair of scalar add/sub.
       if (NumUndefsLO + 1 == Half || NumUndefsHI + 1 == Half)

 /// does not check for the availability of PALIGNR-based lowerings, only the
 /// applicability of this strategy to the given mask. This matches shuffle
 /// vectors that look like:
-/// 
+///
 ///   v8i16 [11, 12, 13, 14, 15, 0, 1, 2]
-/// 
+///
 /// Essentially it concatenates V1 and V2, shifts right by some number of
 /// elements, and takes the low elements as the result. Note that while this is
 /// specified as a *right shift* because x86 is little-endian, it is a *left

 
 /// Insert one bit to mask vector, like v16i1 or v8i1.
 /// AVX-512 feature.
-SDValue 
+SDValue
 X86TargetLowering::InsertBitToMaskVector(SDValue Op, SelectionDAG &DAG) const {
   SDLoc dl(Op);
   SDValue Vec = Op.getOperand(0);

     // insert element and then truncate the result.
     MVT ExtVecVT = (VecVT == MVT::v8i1 ?  MVT::v8i64 : MVT::v16i32);
     MVT ExtEltVT = (VecVT == MVT::v8i1 ?  MVT::i64 : MVT::i32);
-    SDValue ExtOp = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, ExtVecVT, 
+    SDValue ExtOp = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, ExtVecVT,
       DAG.getNode(ISD::ZERO_EXTEND, dl, ExtVecVT, Vec),
       DAG.getNode(ISD::ZERO_EXTEND, dl, ExtEltVT, Elt), Idx);
     return DAG.getNode(ISD::TRUNCATE, dl, VecVT, ExtOp);

       In = DAG.getNode(ISD::SIGN_EXTEND, DL, ExtVT, In);
       InVT = ExtVT;
     }
-    
     SDValue Cst = DAG.getTargetConstant(1, InVT.getVectorElementType());
     const Constant *C = (dyn_cast<ConstantSDNode>(Cst))->getConstantIntValue();
     SDValue CP = DAG.getConstantPool(C, getPointerTy());

     EltVT = VT.getVectorElementType();
     NumElts = VT.getVectorNumElements();
   }
-  
   unsigned EltBits = EltVT.getSizeInBits();
   LLVMContext *Context = DAG.getContext();
   // For FABS, mask is 0x7f...; for FNEG, mask is 0x80...

     return DAG.getNode(ISD::BITCAST, dl, VT,
                        DAG.getNode(BitOp, dl, VecVT, Operand, MaskCasted));
   }
-  
   // If not vector, then scalar.
   unsigned BitOp = IsFABS ? X86ISD::FAND : IsFNABS ? X86ISD::FOR : X86ISD::FXOR;
   SDValue Operand = IsFNABS ? Op0.getOperand(0) : Op0;

      if (Op0.getValueType() == MVT::i1)
        llvm_unreachable("Unexpected comparison operation for MVT::i1 operands");
   }
- 
   if ((Op0.getValueType() == MVT::i8 || Op0.getValueType() == MVT::i16 ||
        Op0.getValueType() == MVT::i32 || Op0.getValueType() == MVT::i64)) {
-    // Do the comparison at i32 if it's smaller, besides the Atom case. 
-    // This avoids subregister aliasing issues. Keep the smaller reference 
-    // if we're optimizing for size, however, as that'll allow better folding 
+    // Do the comparison at i32 if it's smaller, besides the Atom case.
+    // This avoids subregister aliasing issues. Keep the smaller reference
+    // if we're optimizing for size, however, as that'll allow better folding
     // of memory operations.
     if (Op0.getValueType() != MVT::i32 && Op0.getValueType() != MVT::i64 &&
         !DAG.getMachineFunction().getFunction()->getAttributes().hasAttribute(

     return SDValue();
 
   EVT VT = Op.getValueType();
-  
   // SSE1 has rsqrtss and rsqrtps.
   // TODO: Add support for AVX512 (v16f32).
   // It is likely not profitable to do this for f64 because a double-precision

 
        ((Subtarget->hasDQI() && Subtarget->hasVLX() &&
         VT.getSizeInBits() <= 256 && VTElt.getSizeInBits() >= 32)) ||
-    
        ((Subtarget->hasDQI() && VT.is512BitVector() &&
         VTElt.getSizeInBits() >= 32))))
     return DAG.getNode(X86ISD::VSEXT, dl, VT, In);
-    
   unsigned int NumElts = VT.getVectorNumElements();
 
   if (NumElts != 8 && NumElts != 16)

   switch(IntrData->Type) {
   default:
     llvm_unreachable("Unknown Intrinsic Type");
-    break;    
+    break;
   case RDSEED:
   case RDRAND: {
     // Emit the node with the right value type.

   // If possible, lower this packed shift into a vector multiply instead of
   // expanding it into a sequence of scalar shifts.
   // Do this only if the vector shift count is a constant build_vector.
-  if (Op.getOpcode() == ISD::SHL && 
+  if (Op.getOpcode() == ISD::SHL &&
       (VT == MVT::v8i16 || VT == MVT::v4i32 ||
        (Subtarget->hasInt256() && VT == MVT::v16i16)) &&
       ISD::isBuildVectorOfConstantSDNodes(Amt.getNode())) {

           CanBeSimplified = Amt2 == Amt->getOperand(j);
       }
     }
-    
     if (CanBeSimplified && isa<ConstantSDNode>(Amt1) &&
         isa<ConstantSDNode>(Amt2)) {
       // Replace this node with two shifts followed by a MOVSS/MOVSD.
       EVT CastVT = MVT::v4i32;
-      SDValue Splat1 = 
+      SDValue Splat1 =
         DAG.getConstant(cast<ConstantSDNode>(Amt1)->getAPIntValue(), VT);
       SDValue Shift1 = DAG.getNode(Op->getOpcode(), dl, VT, R, Splat1);
-      SDValue Splat2 = 
+      SDValue Splat2 =
         DAG.getConstant(cast<ConstantSDNode>(Amt2)->getAPIntValue(), VT);
       SDValue Shift2 = DAG.getNode(Op->getOpcode(), dl, VT, R, Splat2);
       if (TargetOpcode == X86ISD::MOVSD)

 
 // Replace 213-type (isel default) FMA3 instructions with 231-type for
 // accumulator loops. Writing back to the accumulator allows the coalescer
-// to remove extra copies in the loop.   
+// to remove extra copies in the loop.
 MachineBasicBlock *
 X86TargetLowering::emitFMA3Instr(MachineInstr *MI,
                                  MachineBasicBlock *MBB) const {

     EVT SVT = BC0.getValueType();
     unsigned Opcode = BC0.getOpcode();
     unsigned NumElts = VT.getVectorNumElements();
-    
     if (BC0.hasOneUse() && SVT.isVector() &&
         SVT.getVectorNumElements() * 2 == NumElts &&
         TLI.isOperationLegal(Opcode, VT)) {

       unsigned NumElems = Cond.getNumOperands();
       SDValue A = LHS;
       SDValue B = RHS;
-      
       if (isZero(Cond.getOperand(0))) {
         CanFold = true;
 

     // fold (blend A, B, allOnes) -> B
     if (ISD::isBuildVectorAllOnes(Mask.getNode()))
       return Op1;
-    
     // Simplify the case where the mask is a constant i32 value.
     if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Mask)) {
       if (C->isNullValue())

   // "load" ports instead of the dedicated "store" port.
   // E.g., on Haswell:
   // vmovaps %ymm1, (%r8, %rdi) can use port 2 or 3.
-  // vmovaps %ymm1, (%r8) can use port 2, 3, or 7.   
+  // vmovaps %ymm1, (%r8) can use port 2, 3, or 7.
   if (isLegalAddressingMode(AM, Ty))
     // Scale represents reg2 * scale, thus account for 1
     // as soon as we use a second register.

lib/Target/X86/X86TargetMachine.h

 
 class X86TargetMachine final : public LLVMTargetMachine {
   virtual void anchor();
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   X86Subtarget       Subtarget;
 
   mutable StringMap<std::unique_ptr<X86Subtarget>> SubtargetMap;

 
   // Set up the pass pipeline.
   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 };
 
 } // End llvm namespace

lib/Target/X86/X86TargetMachine.cpp

 
 #include "X86TargetMachine.h"
 #include "X86.h"
+#include "X86TargetObjectFile.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/IR/Function.h"
 #include "llvm/PassManager.h"

 
 void X86TargetMachine::anchor() { }
 
+static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
+  if (TT.isOSBinFormatMachO()) {
+    if (TT.getArch() == Triple::x86_64)
+      return make_unique<X86_64MachoTargetObjectFile>();
+    return make_unique<TargetLoweringObjectFileMachO>();
+  }
+
+  if (TT.isOSLinux())
+    return make_unique<X86LinuxTargetObjectFile>();
+  if (TT.isOSBinFormatELF())
+    return make_unique<TargetLoweringObjectFileELF>();
+  if (TT.isKnownWindowsMSVCEnvironment())
+    return make_unique<X86WindowsTargetObjectFile>();
+  if (TT.isOSBinFormatCOFF())
+    return make_unique<TargetLoweringObjectFileCOFF>();
+  llvm_unreachable("unknown subtarget type");
+}
+
 /// X86TargetMachine ctor - Create an X86 target.
 ///
 X86TargetMachine::X86TargetMachine(const Target &T, StringRef TT, StringRef CPU,

                                    Reloc::Model RM, CodeModel::Model CM,
                                    CodeGenOpt::Level OL)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      TLOF(createTLOF(Triple(getTargetTriple()))),
       Subtarget(TT, CPU, FS, *this, Options.StackAlignmentOverride) {
   // default to hard float ABI
   if (Options.FloatABIType == FloatABI::Default)

lib/Target/XCore/XCoreISelLowering.cpp

 }
 
 XCoreTargetLowering::XCoreTargetLowering(const TargetMachine &TM)
-    : TargetLowering(TM, new XCoreTargetObjectFile()), TM(TM),
+    : TargetLowering(TM, TM.getObjFileLowering()), TM(TM),
       Subtarget(TM.getSubtarget<XCoreSubtarget>()) {
 
   // Set up the register classes.

lib/Target/XCore/XCoreTargetMachine.h

 namespace llvm {
 
 class XCoreTargetMachine : public LLVMTargetMachine {
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   XCoreSubtarget Subtarget;
 public:
   XCoreTargetMachine(const Target &T, StringRef TT,

   TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
 
   void addAnalysisPasses(PassManagerBase &PM) override;
+  TargetLoweringObjectFile *getObjFileLowering() const override {
+    return TLOF.get();
+  }
 };
 
 } // end namespace llvm

lib/Target/XCore/XCoreTargetMachine.cpp

 //===----------------------------------------------------------------------===//
 
 #include "XCoreTargetMachine.h"
+#include "XCoreTargetObjectFile.h"
 #include "XCore.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/IR/Module.h"

                                        Reloc::Model RM, CodeModel::Model CM,
                                        CodeGenOpt::Level OL)
     : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OL),
+      TLOF(make_unique<XCoreTargetObjectFile>()),
       Subtarget(TT, CPU, FS, *this) {
   initAsmInfo();
 }