Remove access to the DataLayout in the TargetMachine

[mehdi_amini]

Summary

Replace getDataLayout() with a createDataLayout() method to make
explicit that it is intended to create a DataLayout only and not
accessing it for other purpose.

This change is the last of a series of commits dedicated to have a
single DataLayout during compilation by using always the one owned
by the module.

[mehdi_amini]

lib/Target/TargetMachineC.cpp

186

Please review this in particular

[mehdi_amini]

Update based on D11110 (ExecutionEngine owning a DataLayout now)

[mehdi_amini]

Cleanup, previous update was incomplete.

[echristo]

I think parts of this are going to conflict with another patch. I'll wait for an update before commenting further.

-eric

[mehdi_amini]

Rebase (minor conflict)

[• rafael]

lib/CodeGen/AsmPrinter/AsmPrinter.cpp

143

For another patch, but could the AsmPrinter constructor take the pointer size?

lib/ExecutionEngine/MCJIT/MCJIT.cpp

71

Can't it use the DL from the module?

lib/ExecutionEngine/Orc/OrcMCJITReplacement.h

143

same here.

160

whitespace change.

lib/LTO/LTOCodeGenerator.cpp

524

I think this is from the days when we supported the module not having a DL. We should probably try to remove it in another patch.

lib/LTO/LTOModule.cpp

235

same here.

[mehdi_amini]

Thanks for the review Raphael.

lib/CodeGen/AsmPrinter/AsmPrinter.cpp

143

Yes it is an option, I'll look into it.

lib/ExecutionEngine/MCJIT/MCJIT.cpp

71

The problem is that the JIT supports adding a module but also removing a module from its list. Keeping a reference to the DL of the module used for creation seems dangerous on this aspect.

lib/LTO/LTOCodeGenerator.cpp

524

I'm not very familiar with the JIT use case, but how is the client creating the module? And how can he makes sure that the DL on the module that he will give to the JIT is the one the TargetMachine expects?
Is he supposed to keep a reference to the TargetMachine when creating the JIT?

[• rafael]

rafael wrote:

I think this is from the days when we supported the module not having a DL. We should probably try to remove it in another patch.

I'm not very familiar with the JIT use case, but how is the client creating the module? And how can he makes sure that the DL on the module that he will give to the JIT is the one the TargetMachine expects?
Is he supposed to keep a reference to the TargetMachine when creating the JIT?

I am not too familiar with the JIT, but that is probably a reasonable
requirement for LTO.

Cheers,
Rafael

[lhames]

lib/LTO/LTOCodeGenerator.cpp

524

This is LTO code, and the LTO and JIT cases may be different.

Historically the JIT hasn't required a data layout. That seems like a reasonable restriction for us to add, but we'd have to give clients a heads-up first. For now, the JIT code should still set the data layout if there isn't one present.

[echristo]

lib/LTO/LTOCodeGenerator.cpp

524

Well, you have, you're just relied on the TargetMachine for this. Also, I imagine there's a DataLayout in the module since it's required now anyhow right? :)

[lhames]

lib/LTO/LTOCodeGenerator.cpp

524

Sorry: To clarify, what I meant was that we haven't historically required JIT clients to supply Modules with a DataLayout attached. In fact, until fairly recently MCJIT would happily consume & codegen IR without a DataLayout.

When I wrote Orc I added the data layout requirement to ensure that symbol naming was consistent between JIT'd and static code. To avoid changing the contract on clients, MCJIT and the in-tree Orc stacks (OrcMCJITReplacement, OrcLazy) take a TargetMachine and use it to attach a DataLayout if there isn't already one present.

I'd be open to changing the contract on MCJIT and Orc to require a DataLayout on the modules if that would be more consistent with LTO.

[echristo]

lib/LTO/LTOCodeGenerator.cpp

524

That makes sense. After Mehdi's earlier patches DataLayout is required on the Module, but we're probably not enforcing it. I'm working on how to help fix consumers of the TargetMachine DataLayout construction routines to construct a DataLayout ahead of time.

[mehdi_amini]

Fix the C API for backward compatibility

[dblaikie]

The C API back compat looks plausible to me - but it's not an area I dabble in, so I've no idea how legitimate it is as an idea. Totally seems like a fair way to do back-compat.

lib/Target/TargetMachineC.cpp

41

This could be a unique_ptr<TargetMachine>, perhaps?

[mehdi_amini]

Update the C API using a std::unique_ptr as suggested by D. Blaikie

[echristo]

Looks like a good intermediate way to go.

-eric

[mehdi_amini]

landed

examples/Kaleidoscope/Orc/fully_lazy/toy.cpp

 class IRGenContext {
 public:
 
   IRGenContext(SessionContext &S)
     : Session(S),
       M(new Module(GenerateUniqueName("jit_module_"),
                    Session.getLLVMContext())),
       Builder(Session.getLLVMContext()) {
-    M->setDataLayout(*Session.getTarget().getDataLayout());
+    M->setDataLayout(Session.getTarget().createDataLayout());
   }
 
   SessionContext& getSession() { return Session; }
   Module& getM() const { return *M; }
   std::unique_ptr<Module> takeM() { return std::move(M); }
   IRBuilder<>& getBuilder() { return Builder; }
   LLVMContext& getLLVMContext() { return Session.getLLVMContext(); }
   Function* getPrototype(const std::string &Name);
: Session(Session),
                        reinterpret_cast<uintptr_t>(EarthShatteringKaboom),
                        64) {}
 
   std::string mangle(const std::string &Name) {
     std::string MangledName;
     {
       raw_string_ostream MangledNameStream(MangledName);
       Mangler::getNameWithPrefix(MangledNameStream, Name,
-                                 *Session.getTarget().getDataLayout());
+                                 Session.getTarget().createDataLayout());
     }
     return MangledName;
   }
 
   void addFunctionAST(std::unique_ptr<FunctionAST> FnAST) {
     std::cerr << "Adding AST: " << FnAST->Proto->Name << "\n";
     FunctionDefs[mangle(FnAST->Proto->Name)] = std::move(FnAST);
   }

examples/Kaleidoscope/Orc/initial/toy.cpp

 class IRGenContext {
 public:
 
   IRGenContext(SessionContext &S)
     : Session(S),
       M(new Module(GenerateUniqueName("jit_module_"),
                    Session.getLLVMContext())),
       Builder(Session.getLLVMContext()) {
-    M->setDataLayout(*Session.getTarget().getDataLayout());
+    M->setDataLayout(Session.getTarget().createDataLayout());
   }
 
   SessionContext& getSession() { return Session; }
   Module& getM() const { return *M; }
   std::unique_ptr<Module> takeM() { return std::move(M); }
   IRBuilder<>& getBuilder() { return Builder; }
   LLVMContext& getLLVMContext() { return Session.getLLVMContext(); }
   Function* getPrototype(const std::string &Name);

 
 class KaleidoscopeJIT {
 public:
   typedef ObjectLinkingLayer<> ObjLayerT;
   typedef IRCompileLayer<ObjLayerT> CompileLayerT;
   typedef CompileLayerT::ModuleSetHandleT ModuleHandleT;
 
   KaleidoscopeJIT(SessionContext &Session)
-      : DL(*Session.getTarget().getDataLayout()),
+      : DL(Session.getTarget().createDataLayout()),
         CompileLayer(ObjectLayer, SimpleCompiler(Session.getTarget())) {}
 
   std::string mangle(const std::string &Name) {
     std::string MangledName;
     {
       raw_string_ostream MangledNameStream(MangledName);
       Mangler::getNameWithPrefix(MangledNameStream, Name, DL);
     }
JITSymbol findSymbol(const std::string &Name) {
     return CompileLayer.findSymbol(Name, true);
   }
 
   JITSymbol findUnmangledSymbol(const std::string Name) {
     return findSymbol(mangle(Name));
   }
 
 private:
-  const DataLayout &DL;
+  const DataLayout DL;
   ObjLayerT ObjectLayer;
   CompileLayerT CompileLayer;
 };
 
 static void HandleDefinition(SessionContext &S, KaleidoscopeJIT &J) {
   if (auto F = ParseDefinition()) {
     if (auto M = IRGen(S, *F)) {
       S.addPrototypeAST(llvm::make_unique<PrototypeAST>(*F->Proto));

examples/Kaleidoscope/Orc/lazy_codegen/toy.cpp

 class IRGenContext {
 public:
 
   IRGenContext(SessionContext &S)
     : Session(S),
       M(new Module(GenerateUniqueName("jit_module_"),
                    Session.getLLVMContext())),
       Builder(Session.getLLVMContext()) {
-    M->setDataLayout(*Session.getTarget().getDataLayout());
+    M->setDataLayout(Session.getTarget().createDataLayout());
   }
 
   SessionContext& getSession() { return Session; }
   Module& getM() const { return *M; }
   std::unique_ptr<Module> takeM() { return std::move(M); }
   IRBuilder<>& getBuilder() { return Builder; }
   LLVMContext& getLLVMContext() { return Session.getLLVMContext(); }
   Function* getPrototype(const std::string &Name);

 public:
   typedef ObjectLinkingLayer<> ObjLayerT;
   typedef IRCompileLayer<ObjLayerT> CompileLayerT;
   typedef LazyEmittingLayer<CompileLayerT> LazyEmitLayerT;
 
   typedef LazyEmitLayerT::ModuleSetHandleT ModuleHandleT;
 
   KaleidoscopeJIT(SessionContext &Session)
-      : DL(*Session.getTarget().getDataLayout()),
+      : DL(Session.getTarget().createDataLayout()),
         CompileLayer(ObjectLayer, SimpleCompiler(Session.getTarget())),
         LazyEmitLayer(CompileLayer) {}
 
   std::string mangle(const std::string &Name) {
     std::string MangledName;
     {
       raw_string_ostream MangledNameStream(MangledName);
       Mangler::getNameWithPrefix(MangledNameStream, Name, DL);
JITSymbol findSymbol(const std::string &Name) {
     return LazyEmitLayer.findSymbol(Name, true);
   }
 
   JITSymbol findUnmangledSymbol(const std::string Name) {
     return findSymbol(mangle(Name));
   }
 
 private:
-  const DataLayout &DL;
+  const DataLayout DL;
   ObjLayerT ObjectLayer;
   CompileLayerT CompileLayer;
   LazyEmitLayerT LazyEmitLayer;
 };
 
 static void HandleDefinition(SessionContext &S, KaleidoscopeJIT &J) {
   if (auto F = ParseDefinition()) {
     if (auto M = IRGen(S, *F)) {

examples/Kaleidoscope/Orc/lazy_irgen/toy.cpp

 class IRGenContext {
 public:
 
   IRGenContext(SessionContext &S)
     : Session(S),
       M(new Module(GenerateUniqueName("jit_module_"),
                    Session.getLLVMContext())),
       Builder(Session.getLLVMContext()) {
-    M->setDataLayout(*Session.getTarget().getDataLayout());
+    M->setDataLayout(Session.getTarget().createDataLayout());
   }
 
   SessionContext& getSession() { return Session; }
   Module& getM() const { return *M; }
   std::unique_ptr<Module> takeM() { return std::move(M); }
   IRBuilder<>& getBuilder() { return Builder; }
   LLVMContext& getLLVMContext() { return Session.getLLVMContext(); }
   Function* getPrototype(const std::string &Name);
: Session(Session),
       CompileLayer(ObjectLayer, SimpleCompiler(Session.getTarget())),
       LazyEmitLayer(CompileLayer) {}
 
   std::string mangle(const std::string &Name) {
     std::string MangledName;
     {
       raw_string_ostream MangledNameStream(MangledName);
       Mangler::getNameWithPrefix(MangledNameStream, Name,
-                                 *Session.getTarget().getDataLayout());
+                                 Session.getTarget().createDataLayout());
     }
     return MangledName;
   }
 
   void addFunctionAST(std::unique_ptr<FunctionAST> FnAST) {
     std::cerr << "Adding AST: " << FnAST->Proto->Name << "\n";
     FunctionDefs[mangle(FnAST->Proto->Name)] = std::move(FnAST);
   }

include/llvm/Target/TargetMachine.h

 protected: // Can only create subclasses.
   TargetMachine(const Target &T, StringRef DataLayoutString,
                 const Triple &TargetTriple, StringRef CPU, StringRef FS,
                 const TargetOptions &Options);
 
   /// The Target that this machine was created for.
   const Target &TheTarget;
 
-  /// For ABI type size and alignment.
+  /// DataLayout for the target: keep ABI type size and alignment.
+  ///
+  /// The DataLayout is created based on the string representation provided
+  /// during construction. It is kept here only to avoid reparsing the string
+  /// but should not really be used during compilation, because it has an
+  /// internal cache that is context specific.
   const DataLayout DL;
 
   /// Triple string, CPU name, and target feature strings the TargetMachine
   /// instance is created with.
   Triple TargetTriple;
   std::string TargetCPU;
   std::string TargetFS;
 
public:
 
   /// This method returns a pointer to the specified type of
   /// TargetSubtargetInfo.  In debug builds, it verifies that the object being
   /// returned is of the correct type.
   template <typename STC> const STC &getSubtarget(const Function &F) const {
     return *static_cast<const STC*>(getSubtargetImpl(F));
   }
 
-  /// This method returns a pointer to the DataLayout for the target. It should
-  /// be unchanging for every subtarget.
-  const DataLayout *getDataLayout() const { return &DL; }
+  /// Create a DataLayout.
+  const DataLayout createDataLayout() const { return DL; }
+
+  /// Get the pointer size for this target.
+  ///
+  /// This is the only time the DataLayout in the TargetMachine is used.
+  unsigned getPointerSize() const { return DL.getPointerSize(); }
 
   /// \brief Reset the target options based on the function's attributes.
   // FIXME: Remove TargetOptions that affect per-function code generation
   // from TargetMachine.
   void resetTargetOptions(const Function &F) const;
 
   /// Return target specific asm information.
   const MCAsmInfo *getMCAsmInfo() const { return AsmInfo; }

lib/CodeGen/AsmPrinter/AsmPrinter.cpp

 const TargetLoweringObjectFile &AsmPrinter::getObjFileLowering() const {
   return *TM.getObjFileLowering();
 }
 
 const DataLayout &AsmPrinter::getDataLayout() const {
   return MMI->getModule()->getDataLayout();
 }
 
-unsigned AsmPrinter::getPointerSize() const {
-  return TM.getDataLayout()->getPointerSize();
-}
+// Do not use the cached DataLayout because some client use it without a Module
+// (llmv-dsymutil, llvm-dwarfdump).
+unsigned AsmPrinter::getPointerSize() const { return TM.getPointerSize(); }
 
 const MCSubtargetInfo &AsmPrinter::getSubtargetInfo() const {
   assert(MF && "getSubtargetInfo requires a valid MachineFunction!");
   return MF->getSubtarget<MCSubtargetInfo>();
 }
 
 void AsmPrinter::EmitToStreamer(MCStreamer &S, const MCInst &Inst) {
   S.EmitInstruction(Inst, getSubtargetInfo());

lib/ExecutionEngine/MCJIT/MCJIT.cpp

MCJIT::createJIT(std::unique_ptr<Module> M,
 
   return new MCJIT(std::move(M), std::move(TM), std::move(MemMgr),
                    std::move(Resolver));
 }
 
 MCJIT::MCJIT(std::unique_ptr<Module> M, std::unique_ptr<TargetMachine> TM,
              std::shared_ptr<MCJITMemoryManager> MemMgr,
              std::shared_ptr<RuntimeDyld::SymbolResolver> Resolver)
-    : ExecutionEngine(*TM->getDataLayout(), std::move(M)), TM(std::move(TM)),
+    : ExecutionEngine(TM->createDataLayout(), std::move(M)), TM(std::move(TM)),
       Ctx(nullptr), MemMgr(std::move(MemMgr)),
       Resolver(*this, std::move(Resolver)), Dyld(*this->MemMgr, this->Resolver),
       ObjCache(nullptr) {
   // FIXME: We are managing our modules, so we do not want the base class
   // ExecutionEngine to manage them as well. To avoid double destruction
   // of the first (and only) module added in ExecutionEngine constructor
   // we remove it from EE and will destruct it ourselves.
   //

lib/ExecutionEngine/Orc/OrcMCJITReplacement.h

public:
   static void Register() {
     OrcMCJITReplacementCtor = createOrcMCJITReplacement;
   }
 
   OrcMCJITReplacement(
       std::shared_ptr<MCJITMemoryManager> MemMgr,
       std::shared_ptr<RuntimeDyld::SymbolResolver> ClientResolver,
       std::unique_ptr<TargetMachine> TM)
-      : ExecutionEngine(*TM->getDataLayout()), TM(std::move(TM)),
+      : ExecutionEngine(TM->createDataLayout()), TM(std::move(TM)),
         MemMgr(*this, std::move(MemMgr)), Resolver(*this),
         ClientResolver(std::move(ClientResolver)), NotifyObjectLoaded(*this),
         NotifyFinalized(*this),
         ObjectLayer(NotifyObjectLoaded, NotifyFinalized),
         CompileLayer(ObjectLayer, SimpleCompiler(*this->TM)),
         LazyEmitLayer(CompileLayer) {}
 
   void addModule(std::unique_ptr<Module> M) override {
 
     // If this module doesn't have a DataLayout attached then attach the
     // default.
     if (M->getDataLayout().isDefault()) {
       M->setDataLayout(getDataLayout());
     } else {
       assert(M->getDataLayout() == getDataLayout() && "DataLayout Mismatch");
     }
     Modules.push_back(std::move(M));
     std::vector<Module *> Ms;
     Ms.push_back(&*Modules.back());
     LazyEmitLayer.addModuleSet(std::move(Ms), &MemMgr, &Resolver);
   }
 
   void addObjectFile(std::unique_ptr<object::ObjectFile> O) override {
     std::vector<std::unique_ptr<object::ObjectFile>> Objs;
     Objs.push_back(std::move(O));

lib/LTO/LTOCodeGenerator.cpp

bool LTOCodeGenerator::optimize(bool DisableInline,
 
   // Mark which symbols can not be internalized
   this->applyScopeRestrictions();
 
   // Instantiate the pass manager to organize the passes.
   legacy::PassManager passes;
 
   // Add an appropriate DataLayout instance for this module...
-  mergedModule->setDataLayout(*TargetMach->getDataLayout());
+  mergedModule->setDataLayout(TargetMach->createDataLayout());
 
   passes.add(
       createTargetTransformInfoWrapperPass(TargetMach->getTargetIRAnalysis()));
 
   Triple TargetTriple(TargetMach->getTargetTriple());
   PassManagerBuilder PMB;
   PMB.DisableGVNLoadPRE = DisableGVNLoadPRE;
   PMB.LoopVectorize = !DisableVectorization;

lib/LTO/LTOModule.cpp

if (Triple.isOSDarwin()) {
     else if (Triple.getArch() == llvm::Triple::x86)
       CPU = "yonah";
     else if (Triple.getArch() == llvm::Triple::aarch64)
       CPU = "cyclone";
   }
 
   TargetMachine *target = march->createTargetMachine(TripleStr, CPU, FeatureStr,
                                                      options);
-  M->setDataLayout(*target->getDataLayout());
+  M->setDataLayout(target->createDataLayout());
 
   std::unique_ptr<object::IRObjectFile> IRObj(
       new object::IRObjectFile(Buffer, std::move(M)));
 
   LTOModule *Ret;
   if (OwnedContext)
     Ret = new LTOModule(std::move(IRObj), target, std::move(OwnedContext));
   else

lib/Target/Sparc/SparcISelLowering.cpp

static SPCC::CondCodes FPCondCCodeToFCC(ISD::CondCode CC) {
   case ISD::SETONE: return SPCC::FCC_LG;
   case ISD::SETUEQ: return SPCC::FCC_UE;
   }
 }
 
 SparcTargetLowering::SparcTargetLowering(TargetMachine &TM,
                                          const SparcSubtarget &STI)
     : TargetLowering(TM), Subtarget(&STI) {
-  auto &DL = *TM.getDataLayout();
+  MVT PtrVT = MVT::getIntegerVT(8 * TM.getPointerSize());
 
   // Set up the register classes.
   addRegisterClass(MVT::i32, &SP::IntRegsRegClass);
   addRegisterClass(MVT::f32, &SP::FPRegsRegClass);
   addRegisterClass(MVT::f64, &SP::DFPRegsRegClass);
   addRegisterClass(MVT::f128, &SP::QFPRegsRegClass);
   if (Subtarget->is64Bit())
     addRegisterClass(MVT::i64, &SP::I64RegsRegClass);
for (MVT VT : MVT::integer_valuetypes())
     setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
 
   // Turn FP truncstore into trunc + store.
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
   setTruncStoreAction(MVT::f128, MVT::f32, Expand);
   setTruncStoreAction(MVT::f128, MVT::f64, Expand);
 
   // Custom legalize GlobalAddress nodes into LO/HI parts.
-  setOperationAction(ISD::GlobalAddress, getPointerTy(DL), Custom);
-  setOperationAction(ISD::GlobalTLSAddress, getPointerTy(DL), Custom);
-  setOperationAction(ISD::ConstantPool, getPointerTy(DL), Custom);
-  setOperationAction(ISD::BlockAddress, getPointerTy(DL), Custom);
+  setOperationAction(ISD::GlobalAddress, PtrVT, Custom);
+  setOperationAction(ISD::GlobalTLSAddress, PtrVT, Custom);
+  setOperationAction(ISD::ConstantPool, PtrVT, Custom);
+  setOperationAction(ISD::BlockAddress, PtrVT, Custom);
 
   // Sparc doesn't have sext_inreg, replace them with shl/sra
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Expand);
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8 , Expand);
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1 , Expand);
 
   // Sparc has no REM or DIVREM operations.
   setOperationAction(ISD::UREM, MVT::i32, Expand);

lib/Target/SystemZ/SystemZISelLowering.cpp

static MachineOperand earlyUseOperand(MachineOperand Op) {
   if (Op.isReg())
     Op.setIsKill(false);
   return Op;
 }
 
 SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &TM,
                                              const SystemZSubtarget &STI)
     : TargetLowering(TM), Subtarget(STI) {
-  auto &DL = *TM.getDataLayout();
-  MVT PtrVT = getPointerTy(DL);
+  MVT PtrVT = MVT::getIntegerVT(8 * TM.getPointerSize());
 
   // Set up the register classes.
   if (Subtarget.hasHighWord())
     addRegisterClass(MVT::i32, &SystemZ::GRX32BitRegClass);
   else
     addRegisterClass(MVT::i32, &SystemZ::GR32BitRegClass);
   addRegisterClass(MVT::i64, &SystemZ::GR64BitRegClass);
   if (Subtarget.hasVector()) {

lib/Target/TargetMachineC.cpp

 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetSubtargetInfo.h"
 #include <cassert>
 #include <cstdlib>
 #include <cstring>
 
 using namespace llvm;
 
+
+// The TargetMachine uses to offer access to a DataLayout member. This is reflected
+// in the C API. For backward compatibility reason, this structure allows to keep
+// a DataLayout member accessible to C client that have a handle to a
+// LLVMTargetMachineRef.
+struct LLVMOpaqueTargetMachine {
+  TargetMachine *Machine;
+  DataLayout DL;
+};
+
+
 inline TargetMachine *unwrap(LLVMTargetMachineRef P) {
-  return reinterpret_cast<TargetMachine*>(P);
+  return P->Machine;
 }
 inline Target *unwrap(LLVMTargetRef P) {
   return reinterpret_cast<Target*>(P);
 }
 inline LLVMTargetMachineRef wrap(const TargetMachine *P) {
-  return
-    reinterpret_cast<LLVMTargetMachineRef>(const_cast<TargetMachine*>(P));
+  return new LLVMOpaqueTargetMachine{ const_cast<TargetMachine*>(P),  P->createDataLayout() };
 }
 inline LLVMTargetRef wrap(const Target * P) {
   return reinterpret_cast<LLVMTargetRef>(const_cast<Target*>(P));
 }
 
 LLVMTargetRef LLVMGetFirstTarget() {
   if (TargetRegistry::targets().begin() == TargetRegistry::targets().end()) {
     return nullptr;
LLVMTargetMachineRef LLVMCreateTargetMachine(LLVMTargetRef T,
   TargetOptions opt;
   return wrap(unwrap(T)->createTargetMachine(Triple, CPU, Features, opt, RM,
     CM, OL));
 }
 
 
 void LLVMDisposeTargetMachine(LLVMTargetMachineRef T) {
   delete unwrap(T);
+  delete T;
 }
 
 LLVMTargetRef LLVMGetTargetMachineTarget(LLVMTargetMachineRef T) {
   const Target* target = &(unwrap(T)->getTarget());
   return wrap(target);
 }
 
 char* LLVMGetTargetMachineTriple(LLVMTargetMachineRef T) {
   std::string StringRep = unwrap(T)->getTargetTriple().str();
   return strdup(StringRep.c_str());
 }
 
 char* LLVMGetTargetMachineCPU(LLVMTargetMachineRef T) {
   std::string StringRep = unwrap(T)->getTargetCPU();
   return strdup(StringRep.c_str());
 }
 
 char* LLVMGetTargetMachineFeatureString(LLVMTargetMachineRef T) {
   std::string StringRep = unwrap(T)->getTargetFeatureString();
   return strdup(StringRep.c_str());
 }
 
+/// @deprecated: see "struct LLVMOpaqueTargetMachine" description above
 LLVMTargetDataRef LLVMGetTargetMachineData(LLVMTargetMachineRef T) {
-  return wrap(unwrap(T)->getDataLayout());
+  return wrap(&T->DL);
 }
 
 void LLVMSetTargetMachineAsmVerbosity(LLVMTargetMachineRef T,
                                       LLVMBool VerboseAsm) {
   unwrap(T)->Options.MCOptions.AsmVerbose = VerboseAsm;
 }
 
 static LLVMBool LLVMTargetMachineEmit(LLVMTargetMachineRef T, LLVMModuleRef M,
                                       raw_pwrite_stream &OS,
                                       LLVMCodeGenFileType codegen,
                                       char **ErrorMessage) {
   TargetMachine* TM = unwrap(T);
   Module* Mod = unwrap(M);
 
   legacy::PassManager pass;
 
   std::string error;
 
-  const DataLayout *td = TM->getDataLayout();
-
-  if (!td) {
-    error = "No DataLayout in TargetMachine";
-    *ErrorMessage = strdup(error.c_str());
-    return true;
-  }
-  Mod->setDataLayout(*td);
+  Mod->setDataLayout(TM->createDataLayout());
 
   TargetMachine::CodeGenFileType ft;
   switch (codegen) {
     case LLVMAssemblyFile:
       ft = TargetMachine::CGFT_AssemblyFile;
       break;
     default:
       ft = TargetMachine::CGFT_ObjectFile;

lib/Target/X86/X86ISelLowering.cpp

 static SDValue getMOVL(SelectionDAG &DAG, SDLoc dl, EVT VT, SDValue V1,
                        SDValue V2);
 
 X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
                                      const X86Subtarget &STI)
     : TargetLowering(TM), Subtarget(&STI) {
   X86ScalarSSEf64 = Subtarget->hasSSE2();
   X86ScalarSSEf32 = Subtarget->hasSSE1();
-  TD = TM.getDataLayout();
+  MVT PtrVT = MVT::getIntegerVT(8 * TM.getPointerSize());
 
   // Set up the TargetLowering object.
   static const MVT IntVTs[] = { MVT::i8, MVT::i16, MVT::i32, MVT::i64 };
 
   // X86 is weird. It always uses i8 for shift amounts and setcc results.
   setBooleanContents(ZeroOrOneBooleanContent);
   // X86-SSE is even stranger. It uses -1 or 0 for vector masks.
   setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
if (Subtarget->is64Bit() && !Subtarget->isTargetWin64()) {
     // TargetInfo::CharPtrBuiltinVaList
     setOperationAction(ISD::VAARG           , MVT::Other, Expand);
     setOperationAction(ISD::VACOPY          , MVT::Other, Expand);
   }
 
   setOperationAction(ISD::STACKSAVE,          MVT::Other, Expand);
   setOperationAction(ISD::STACKRESTORE,       MVT::Other, Expand);
 
-  setOperationAction(ISD::DYNAMIC_STACKALLOC, getPointerTy(*TD), Custom);
+  setOperationAction(ISD::DYNAMIC_STACKALLOC, PtrVT, Custom);
 
   // GC_TRANSITION_START and GC_TRANSITION_END need custom lowering.
   setOperationAction(ISD::GC_TRANSITION_START, MVT::Other, Custom);
   setOperationAction(ISD::GC_TRANSITION_END, MVT::Other, Custom);
 
   if (!Subtarget->useSoftFloat() && X86ScalarSSEf64) {
     // f32 and f64 use SSE.
     // Set up the FP register classes.
case CallingConv::X86_StdCall: {
       const AttributeSet &Attrs = Func->getAttributes();
 
       if (!Attrs.isEmpty() && !Func->isVarArg()) {
         unsigned InRegCount = 0;
         unsigned Idx = 1;
 
         for (FunctionType::param_iterator I = FTy->param_begin(),
              E = FTy->param_end(); I != E; ++I, ++Idx)
-          if (Attrs.hasAttribute(Idx, Attribute::InReg))
+          if (Attrs.hasAttribute(Idx, Attribute::InReg)) {
+            auto &DL = DAG.getDataLayout();
             // FIXME: should only count parameters that are lowered to integers.
-            InRegCount += (TD->getTypeSizeInBits(*I) + 31) / 32;
+            InRegCount += (DL.getTypeSizeInBits(*I) + 31) / 32;
+          }
 
         if (InRegCount > 2) {
           report_fatal_error("Nest register in use - reduce number of inreg"
                              " parameters!");
         }
       }
       break;
     }

tools/llc/llc.cpp

static int compileModule(char **argv, LLVMContext &Context) {
   TargetLibraryInfoImpl TLII(Triple(M->getTargetTriple()));
 
   // The -disable-simplify-libcalls flag actually disables all builtin optzns.
   if (DisableSimplifyLibCalls)
     TLII.disableAllFunctions();
   PM.add(new TargetLibraryInfoWrapperPass(TLII));
 
   // Add the target data from the target machine, if it exists, or the module.
-  if (const DataLayout *DL = Target->getDataLayout())
-    M->setDataLayout(*DL);
+  M->setDataLayout(Target->createDataLayout());
 
   // Override function attributes based on CPUStr, FeaturesStr, and command line
   // flags.
   setFunctionAttributes(CPUStr, FeaturesStr, *M);
 
   if (RelaxAll.getNumOccurrences() > 0 &&
       FileType != TargetMachine::CGFT_ObjectFile)
     errs() << argv[0]

tools/lli/OrcLazyJIT.h

public:
 
   typedef std::function<
             std::unique_ptr<CompileCallbackMgr>(IRDumpLayerT&,
                                                 RuntimeDyld::MemoryManager&,
                                                 LLVMContext&)>
     CallbackManagerBuilder;
 
   static CallbackManagerBuilder createCallbackManagerBuilder(Triple T);
+  const DataLayout &DL;
 
-  OrcLazyJIT(std::unique_ptr<TargetMachine> TM, LLVMContext &Context,
-             CallbackManagerBuilder &BuildCallbackMgr)
-    : TM(std::move(TM)),
-      ObjectLayer(),
+  OrcLazyJIT(std::unique_ptr<TargetMachine> TM, const DataLayout &DL,
+             LLVMContext &Context, CallbackManagerBuilder &BuildCallbackMgr)
+      : DL(DL), TM(std::move(TM)), ObjectLayer(),
         CompileLayer(ObjectLayer, orc::SimpleCompiler(*this->TM)),
         IRDumpLayer(CompileLayer, createDebugDumper()),
         CCMgr(BuildCallbackMgr(IRDumpLayer, CCMgrMemMgr, Context)),
         CODLayer(IRDumpLayer, *CCMgr, false),
-      CXXRuntimeOverrides([this](const std::string &S) { return mangle(S); }) {}
+        CXXRuntimeOverrides(
+            [this](const std::string &S) { return mangle(S); }) {}
 
   ~OrcLazyJIT() {
     // Run any destructors registered with __cxa_atexit.
     CXXRuntimeOverrides.runDestructors();
     // Run any IR destructors.
     for (auto &DtorRunner : IRStaticDestructorRunners)
       DtorRunner.runViaLayer(CODLayer);
   }
 
   template <typename PtrTy>
   static PtrTy fromTargetAddress(orc::TargetAddress Addr) {
     return reinterpret_cast<PtrTy>(static_cast<uintptr_t>(Addr));
   }
 
   ModuleHandleT addModule(std::unique_ptr<Module> M) {
     // Attach a data-layout if one isn't already present.
     if (M->getDataLayout().isDefault())
-      M->setDataLayout(*TM->getDataLayout());
+      M->setDataLayout(DL);
 
     // Record the static constructors and destructors. We have to do this before
     // we hand over ownership of the module to the JIT.
     std::vector<std::string> CtorNames, DtorNames;
     for (auto Ctor : orc::getConstructors(*M))
       CtorNames.push_back(mangle(Ctor.Func->getName()));
     for (auto Dtor : orc::getDestructors(*M))
       DtorNames.push_back(mangle(Dtor.Func->getName()));
orc::JITSymbol findSymbol(const std::string &Name) {
     return CODLayer.findSymbol(mangle(Name), true);
   }
 
   orc::JITSymbol findSymbolIn(ModuleHandleT H, const std::string &Name) {
     return CODLayer.findSymbolIn(H, mangle(Name), true);
   }
 
 private:
-
   std::string mangle(const std::string &Name) {
     std::string MangledName;
     {
       raw_string_ostream MangledNameStream(MangledName);
-      Mangler::getNameWithPrefix(MangledNameStream, Name, *TM->getDataLayout());
+      Mangler::getNameWithPrefix(MangledNameStream, Name, DL);
     }
     return MangledName;
   }
 
   static TransformFtor createDebugDumper();
 
   std::unique_ptr<TargetMachine> TM;
   SectionMemoryManager CCMgrMemMgr;

tools/lli/OrcLazyJIT.cpp

int llvm::runOrcLazyJIT(std::unique_ptr<Module> M, int ArgC, char* ArgV[]) {
   // manager for this target. Bail out.
   if (!CallbackMgrBuilder) {
     errs() << "No callback manager available for target '"
            << TM->getTargetTriple().str() << "'.\n";
     return 1;
   }
 
   // Everything looks good. Build the JIT.
-  OrcLazyJIT J(std::move(TM), Context, CallbackMgrBuilder);
+  auto &DL = M->getDataLayout();
+  OrcLazyJIT J(std::move(TM), DL, Context, CallbackMgrBuilder);
 
   // Add the module, look up main and run it.
   auto MainHandle = J.addModule(std::move(M));
   auto MainSym = J.findSymbolIn(MainHandle, "main");
 
   if (!MainSym) {
     errs() << "Could not find main function.\n";
     return 1;
   }
 
   typedef int (*MainFnPtr)(int, char*[]);
   auto Main = OrcLazyJIT::fromTargetAddress<MainFnPtr>(MainSym.getAddress());
   return Main(ArgC, ArgV);
 }