[RTTI] Align rtti types to prevent over-alignment

[dmgreen]

Summary

Previously the alignment on the newly created rtti/typeinfo data was largely
not set, meaning that DataLayout::getPreferredAlignment was free to overalign
it to 16 bytes. This causes unnecessary code bloat.

[efriedma]

Could we make CreateOrReplaceCXXRuntimeVariable take the alignment as an argument, so we can be sure we're consistently setting the alignment for these variables? This seems easy to mess up if we're scattering calls all over the place.

Sort of orthogonal, but CreateOrReplaceCXXRuntimeVariable should probably also call setUnnamedAddr, instead of making each caller call it.

[dmgreen]

I've become less sure about what the various alignments should be here. There seem to be multiple ways to calculate such things, I'm not sure which are best. For example, for want of a better option I've used getPointerAlign in ItaniumRTTIBuilder::BuildTypeInfo.

[efriedma]

getClassPointerAlignment is not really relevant here; that's the alignment of a pointer to an object with the type of the class.

For the LLVM IR structs which don't have a corresponding clang type, you can probably just use DataLayout::getABITypeAlignment(). I think that's just the alignment of a pointer in practice, but the intent is more clear.

lib/CodeGen/MicrosoftCXXABI.cpp

2027

This is an array of int.

[dmgreen]

Now using getABITypeAlignment, plus added a simple test

[efriedma]

LGTM with one change.

lib/CodeGen/ItaniumCXXABI.cpp

2730

Please use getTypeAlignInChars(getContext().CharTy), or something like that, so it's clear where the "1" is coming from.

[dmgreen]

That sounds like a good idea. Thanks for the help on this one.

lib/CodeGen/CGVTT.cpp

llvm::GlobalVariable *CodeGenVTables::GetAddrOfVTT(const CXXRecordDecl *RD) {
 
   // This will also defer the definition of the VTT.
   (void) CGM.getCXXABI().getAddrOfVTable(RD, CharUnits());
 
   VTTBuilder Builder(CGM.getContext(), RD, /*GenerateDefinition=*/false);
 
   llvm::ArrayType *ArrayType =
     llvm::ArrayType::get(CGM.Int8PtrTy, Builder.getVTTComponents().size());
+  unsigned Align = CGM.getClassPointerAlignment(RD).getQuantity();
 
-  llvm::GlobalVariable *GV =
-    CGM.CreateOrReplaceCXXRuntimeVariable(Name, ArrayType,
-                                          llvm::GlobalValue::ExternalLinkage);
+  llvm::GlobalVariable *GV = CGM.CreateOrReplaceCXXRuntimeVariable(
+      Name, ArrayType, llvm::GlobalValue::ExternalLinkage, Align);
   GV->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
   return GV;
 }
 
 uint64_t CodeGenVTables::getSubVTTIndex(const CXXRecordDecl *RD,
                                         BaseSubobject Base) {
   BaseSubobjectPairTy ClassSubobjectPair(RD, Base);
 

lib/CodeGen/CGVTables.cpp

CodeGenVTables::GenerateConstructionVTable(const CXXRecordDecl *RD,
   // Construction vtable symbols are not part of the Itanium ABI, so we cannot
   // guarantee that they actually will be available externally. Instead, when
   // emitting an available_externally VTT, we provide references to an internal
   // linkage construction vtable. The ABI only requires complete-object vtables
   // to be the same for all instances of a type, not construction vtables.
   if (Linkage == llvm::GlobalVariable::AvailableExternallyLinkage)
     Linkage = llvm::GlobalVariable::InternalLinkage;
 
+  unsigned Align = CGM.getClassPointerAlignment(RD).getQuantity();
+
   // Create the variable that will hold the construction vtable.
   llvm::GlobalVariable *VTable =
-    CGM.CreateOrReplaceCXXRuntimeVariable(Name, VTType, Linkage);
+      CGM.CreateOrReplaceCXXRuntimeVariable(Name, VTType, Linkage, Align);
   CGM.setGVProperties(VTable, RD);
 
   // V-tables are always unnamed_addr.
   VTable->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
 
   llvm::Constant *RTTI = CGM.GetAddrOfRTTIDescriptor(
       CGM.getContext().getTagDeclType(Base.getBase()));
 

lib/CodeGen/CodeGenModule.h

llvm::Constant *GetAddrOfGlobal(GlobalDecl GD,
                                     = NotForDefinition);
 
   /// Will return a global variable of the given type. If a variable with a
   /// different type already exists then a new  variable with the right type
   /// will be created and all uses of the old variable will be replaced with a
   /// bitcast to the new variable.
   llvm::GlobalVariable *
   CreateOrReplaceCXXRuntimeVariable(StringRef Name, llvm::Type *Ty,
-                                    llvm::GlobalValue::LinkageTypes Linkage);
+                                    llvm::GlobalValue::LinkageTypes Linkage,
+                                    unsigned Alignment);
 
   llvm::Function *
   CreateGlobalInitOrDestructFunction(llvm::FunctionType *ty, const Twine &name,
                                      const CGFunctionInfo &FI,
                                      SourceLocation Loc = SourceLocation(),
                                      bool TLS = false);
 
   /// Return the AST address space of the underlying global variable for D, as

lib/CodeGen/CodeGenModule.cpp

else if (isa<CXXMethodDecl>(D)) {
     llvm::FunctionType *Ty = getTypes().GetFunctionType(FI);
     return GetAddrOfFunction(GD, Ty, /*ForVTable=*/false, /*DontDefer=*/false,
                              IsForDefinition);
   } else
     return GetAddrOfGlobalVar(cast<VarDecl>(D), /*Ty=*/nullptr,
                               IsForDefinition);
 }
 
-llvm::GlobalVariable *
-CodeGenModule::CreateOrReplaceCXXRuntimeVariable(StringRef Name,
-                                      llvm::Type *Ty,
-                                      llvm::GlobalValue::LinkageTypes Linkage) {
+llvm::GlobalVariable *CodeGenModule::CreateOrReplaceCXXRuntimeVariable(
+    StringRef Name, llvm::Type *Ty, llvm::GlobalValue::LinkageTypes Linkage,
+    unsigned Alignment) {
   llvm::GlobalVariable *GV = getModule().getNamedGlobal(Name);
   llvm::GlobalVariable *OldGV = nullptr;
 
   if (GV) {
     // Check if the variable has the right type.
     if (GV->getType()->getElementType() == Ty)
       return GV;
 
if (OldGV) {
 
     OldGV->eraseFromParent();
   }
 
   if (supportsCOMDAT() && GV->isWeakForLinker() &&
       !GV->hasAvailableExternallyLinkage())
     GV->setComdat(TheModule.getOrInsertComdat(GV->getName()));
 
+  GV->setAlignment(Alignment);
+
   return GV;
 }
 
 /// GetAddrOfGlobalVar - Return the llvm::Constant for the address of the
 /// given global variable.  If Ty is non-null and if the global doesn't exist,
 /// then it will be created with the specified type instead of whatever the
 /// normal requested type would be. If IsForDefinition is true, it is guaranteed
 /// that an actual global with type Ty will be returned, not conversion of a

lib/CodeGen/ItaniumCXXABI.cpp

void ItaniumCXXABI::emitVTableDefinitions(CodeGenVTables &CGVT,
   VTable->setLinkage(Linkage);
 
   if (CGM.supportsCOMDAT() && VTable->isWeakForLinker())
     VTable->setComdat(CGM.getModule().getOrInsertComdat(VTable->getName()));
 
   // Set the right visibility.
   CGM.setGVProperties(VTable, RD);
 
-  // Use pointer alignment for the vtable. Otherwise we would align them based
-  // on the size of the initializer which doesn't make sense as only single
-  // values are read.
-  unsigned PAlign = CGM.getTarget().getPointerAlign(0);
-  VTable->setAlignment(getContext().toCharUnitsFromBits(PAlign).getQuantity());
-
   // If this is the magic class __cxxabiv1::__fundamental_type_info,
   // we will emit the typeinfo for the fundamental types. This is the
   // same behaviour as GCC.
   const DeclContext *DC = RD->getDeclContext();
   if (RD->getIdentifier() &&
       RD->getIdentifier()->isStr("__fundamental_type_info") &&
       isa<NamespaceDecl>(DC) && cast<NamespaceDecl>(DC)->getIdentifier() &&
       cast<NamespaceDecl>(DC)->getIdentifier()->isStr("__cxxabiv1") &&
llvm::GlobalVariable *ItaniumCXXABI::getAddrOfVTable(const CXXRecordDecl *RD,
   SmallString<256> Name;
   llvm::raw_svector_ostream Out(Name);
   getMangleContext().mangleCXXVTable(RD, Out);
 
   const VTableLayout &VTLayout =
       CGM.getItaniumVTableContext().getVTableLayout(RD);
   llvm::Type *VTableType = CGM.getVTables().getVTableType(VTLayout);
 
+  // Use pointer alignment for the vtable. Otherwise we would align them based
+  // on the size of the initializer which doesn't make sense as only single
+  // values are read.
+  unsigned PAlign = CGM.getTarget().getPointerAlign(0);
+
   VTable = CGM.CreateOrReplaceCXXRuntimeVariable(
-      Name, VTableType, llvm::GlobalValue::ExternalLinkage);
+      Name, VTableType, llvm::GlobalValue::ExternalLinkage,
+      getContext().toCharUnitsFromBits(PAlign).getQuantity());
   VTable->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
 
   CGM.setGVProperties(VTable, RD);
 
   return VTable;
 }
 
 CGCallee ItaniumCXXABI::getVirtualFunctionPointer(CodeGenFunction &CGF,
llvm::GlobalVariable *ItaniumRTTIBuilder::GetAddrOfTypeName(
 
   // We know that the mangled name of the type starts at index 4 of the
   // mangled name of the typename, so we can just index into it in order to
   // get the mangled name of the type.
   llvm::Constant *Init = llvm::ConstantDataArray::getString(VMContext,
                                                             Name.substr(4));
 
   llvm::GlobalVariable *GV =
-    CGM.CreateOrReplaceCXXRuntimeVariable(Name, Init->getType(), Linkage);
+      CGM.CreateOrReplaceCXXRuntimeVariable(Name, Init->getType(), Linkage, 1);
 
   GV->setInitializer(Init);
 
   return GV;
 }
 
 llvm::Constant *
 ItaniumRTTIBuilder::GetAddrOfExternalRTTIDescriptor(QualType Ty) {
llvm::Constant *NewPtr =
       llvm::ConstantExpr::getBitCast(GV, OldGV->getType());
     OldGV->replaceAllUsesWith(NewPtr);
     OldGV->eraseFromParent();
   }
 
   if (CGM.supportsCOMDAT() && GV->isWeakForLinker())
     GV->setComdat(M.getOrInsertComdat(GV->getName()));
 
+  CharUnits Align =
+      CGM.getContext().toCharUnitsFromBits(CGM.getTarget().getPointerAlign(0));
+  GV->setAlignment(Align.getQuantity());
+
   // The Itanium ABI specifies that type_info objects must be globally
   // unique, with one exception: if the type is an incomplete class
   // type or a (possibly indirect) pointer to one.  That exception
   // affects the general case of comparing type_info objects produced
   // by the typeid operator, which is why the comparison operators on
   // std::type_info generally use the type_info name pointers instead
   // of the object addresses.  However, the language's built-in uses
   // of RTTI generally require class types to be complete, even when

lib/CodeGen/MicrosoftCXXABI.cpp

MicrosoftCXXABI::getAddrOfVBTable(const VPtrInfo &VBT, const CXXRecordDecl *RD,
   getMangleContext().mangleCXXVBTable(RD, VBT.MangledPath, Out);
   StringRef Name = OutName.str();
 
   llvm::ArrayType *VBTableType =
       llvm::ArrayType::get(CGM.IntTy, 1 + VBT.ObjectWithVPtr->getNumVBases());
 
   assert(!CGM.getModule().getNamedGlobal(Name) &&
          "vbtable with this name already exists: mangling bug?");
+  unsigned Align = CGM.getClassPointerAlignment(RD).getQuantity();
   llvm::GlobalVariable *GV =
-      CGM.CreateOrReplaceCXXRuntimeVariable(Name, VBTableType, Linkage);
+      CGM.CreateOrReplaceCXXRuntimeVariable(Name, VBTableType, Linkage, Align);
   GV->setUnnamedAddr(llvm::GlobalValue::UnnamedAddr::Global);
 
   if (RD->hasAttr<DLLImportAttr>())
     GV->setDLLStorageClass(llvm::GlobalValue::DLLImportStorageClass);
   else if (RD->hasAttr<DLLExportAttr>())
     GV->setDLLStorageClass(llvm::GlobalValue::DLLExportStorageClass);
 
   if (!GV->hasExternalLinkage())

test/CodeGenCXX/vtable-align.cpp

 // RUN: %clang_cc1 %s -triple=i386-apple-darwin10 -emit-llvm -o - | FileCheck %s -check-prefix=CHECK-32
 // RUN: %clang_cc1 %s -triple=x86_64-apple-darwin10 -emit-llvm -o - | FileCheck %s -check-prefix=CHECK-64
 
 struct A {
   virtual void f();
   virtual void g();
   virtual void h();
 };
 
 void A::f() {}
 
 // CHECK-32: @_ZTV1A = unnamed_addr constant { [5 x i8*] } { [5 x i8*] [i8* null, i8* bitcast ({ i8*, i8* }* @_ZTI1A to i8*), i8* bitcast (void (%struct.A*)* @_ZN1A1fEv to i8*), i8* bitcast (void (%struct.A*)* @_ZN1A1gEv to i8*), i8* bitcast (void (%struct.A*)* @_ZN1A1hEv to i8*)] }, align 4
-
+// CHECK-32: @_ZTS1A = constant [3 x i8] c"1A\00", align 1
+// CHECK-32: @_ZTI1A = constant { i8*, i8* } { i8* bitcast (i8** getelementptr inbounds (i8*, i8** @_ZTVN10__cxxabiv117__class_type_infoE, i32 2) to i8*), i8* getelementptr inbounds ([3 x i8], [3 x i8]* @_ZTS1A, i32 0, i32 0) }, align 4
 // CHECK-64: @_ZTV1A = unnamed_addr constant { [5 x i8*] } { [5 x i8*] [i8* null, i8* bitcast ({ i8*, i8* }* @_ZTI1A to i8*), i8* bitcast (void (%struct.A*)* @_ZN1A1fEv to i8*), i8* bitcast (void (%struct.A*)* @_ZN1A1gEv to i8*), i8* bitcast (void (%struct.A*)* @_ZN1A1hEv to i8*)] }, align 8
+// CHECK-64: @_ZTS1A = constant [3 x i8] c"1A\00", align 1
+// CHECK-64: @_ZTI1A = constant { i8*, i8* } { i8* bitcast (i8** getelementptr inbounds (i8*, i8** @_ZTVN10__cxxabiv117__class_type_infoE, i64 2) to i8*), i8* getelementptr inbounds ([3 x i8], [3 x i8]* @_ZTS1A, i32 0, i32 0) }, align 8

test/CodeGenCXX/vtable-linkage.cpp

 }
 
 // B has a key function that is not defined in this translation unit so its vtable
 // has external linkage.
 // CHECK-DAG: @_ZTV1B = external unnamed_addr constant
 
 // C has no key function, so its vtable should have weak_odr linkage
 // and hidden visibility (rdar://problem/7523229).
-// CHECK-DAG: @_ZTV1C = linkonce_odr unnamed_addr constant {{.*}}, comdat,
-// CHECK-DAG: @_ZTS1C = linkonce_odr constant {{.*}}, comdat{{$}}
-// CHECK-DAG: @_ZTI1C = linkonce_odr constant {{.*}}, comdat{{$}}
-// CHECK-DAG: @_ZTT1C = linkonce_odr unnamed_addr constant {{.*}}, comdat{{$}}
+// CHECK-DAG: @_ZTV1C = linkonce_odr unnamed_addr constant {{.*}}, comdat, align 8{{$}}
+// CHECK-DAG: @_ZTS1C = linkonce_odr constant {{.*}}, comdat, align 1{{$}}
+// CHECK-DAG: @_ZTI1C = linkonce_odr constant {{.*}}, comdat, align 8{{$}}
+// CHECK-DAG: @_ZTT1C = linkonce_odr unnamed_addr constant {{.*}}, comdat, align 8{{$}}
 
 // D has a key function that is defined in this translation unit so its vtable is
 // defined in the translation unit.
 // CHECK-DAG: @_ZTV1D = unnamed_addr constant
 // CHECK-DAG: @_ZTS1D = constant
 // CHECK-DAG: @_ZTI1D = constant
 
 // E<char> is an explicit specialization with a key function defined
 // in this translation unit, so its vtable should have external
 // linkage.
 // CHECK-DAG: @_ZTV1EIcE = unnamed_addr constant
 // CHECK-DAG: @_ZTS1EIcE = constant
 // CHECK-DAG: @_ZTI1EIcE = constant
 
 // E<short> is an explicit template instantiation with a key function
 // defined in this translation unit, so its vtable should have
 // weak_odr linkage.
 // CHECK-DAG: @_ZTV1EIsE = weak_odr unnamed_addr constant {{.*}}, comdat,
-// CHECK-DAG: @_ZTS1EIsE = weak_odr constant {{.*}}, comdat{{$}}
-// CHECK-DAG: @_ZTI1EIsE = weak_odr constant {{.*}}, comdat{{$}}
+// CHECK-DAG: @_ZTS1EIsE = weak_odr constant {{.*}}, comdat, align 1{{$}}
+// CHECK-DAG: @_ZTI1EIsE = weak_odr constant {{.*}}, comdat, align 8{{$}}
 
 // F<short> is an explicit template instantiation without a key
 // function, so its vtable should have weak_odr linkage
 // CHECK-DAG: @_ZTV1FIsE = weak_odr unnamed_addr constant {{.*}}, comdat,
-// CHECK-DAG: @_ZTS1FIsE = weak_odr constant {{.*}}, comdat{{$}}
-// CHECK-DAG: @_ZTI1FIsE = weak_odr constant {{.*}}, comdat{{$}}
+// CHECK-DAG: @_ZTS1FIsE = weak_odr constant {{.*}}, comdat, align 1{{$}}
+// CHECK-DAG: @_ZTI1FIsE = weak_odr constant {{.*}}, comdat, align 8{{$}}
 
 // E<long> is an implicit template instantiation with a key function
 // defined in this translation unit, so its vtable should have
 // linkonce_odr linkage.
 // CHECK-DAG: @_ZTV1EIlE = linkonce_odr unnamed_addr constant {{.*}}, comdat,
-// CHECK-DAG: @_ZTS1EIlE = linkonce_odr constant {{.*}}, comdat{{$}}
-// CHECK-DAG: @_ZTI1EIlE = linkonce_odr constant {{.*}}, comdat{{$}}
+// CHECK-DAG: @_ZTS1EIlE = linkonce_odr constant {{.*}}, comdat, align 1{{$}}
+// CHECK-DAG: @_ZTI1EIlE = linkonce_odr constant {{.*}}, comdat, align 8{{$}}
 
 // F<long> is an implicit template instantiation with no key function,
 // so its vtable should have linkonce_odr linkage.
 // CHECK-DAG: @_ZTV1FIlE = linkonce_odr unnamed_addr constant {{.*}}, comdat,
-// CHECK-DAG: @_ZTS1FIlE = linkonce_odr constant {{.*}}, comdat{{$}}
-// CHECK-DAG: @_ZTI1FIlE = linkonce_odr constant {{.*}}, comdat{{$}}
+// CHECK-DAG: @_ZTS1FIlE = linkonce_odr constant {{.*}}, comdat, align 1{{$}}
+// CHECK-DAG: @_ZTI1FIlE = linkonce_odr constant {{.*}}, comdat, align 8{{$}}
 
 // F<int> is an explicit template instantiation declaration without a
 // key function, so its vtable should have external linkage.
 // CHECK-DAG: @_ZTV1FIiE = external unnamed_addr constant
 // CHECK-OPT-DAG: @_ZTV1FIiE = available_externally unnamed_addr constant
 
 // E<int> is an explicit template instantiation declaration. It has a
 // key function is not instantiated, so we know that vtable definition

 // namespace.
 // CHECK-DAG: @_ZTVN12_GLOBAL__N_11AE = internal unnamed_addr constant
 // CHECK-DAG: @_ZTSN12_GLOBAL__N_11AE = internal constant
 // CHECK-DAG: @_ZTIN12_GLOBAL__N_11AE = internal constant
 
 // F<char> is an explicit specialization without a key function, so
 // its vtable should have linkonce_odr linkage.
 // CHECK-DAG: @_ZTV1FIcE = linkonce_odr unnamed_addr constant {{.*}}, comdat,
-// CHECK-DAG: @_ZTS1FIcE = linkonce_odr constant {{.*}}, comdat{{$}}
-// CHECK-DAG: @_ZTI1FIcE = linkonce_odr constant {{.*}}, comdat{{$}}
+// CHECK-DAG: @_ZTS1FIcE = linkonce_odr constant {{.*}}, comdat, align 1{{$}}
+// CHECK-DAG: @_ZTI1FIcE = linkonce_odr constant {{.*}}, comdat, align 8{{$}}
 
 // CHECK-DAG: @_ZTV1GIiE = linkonce_odr unnamed_addr constant {{.*}}, comdat,
 template <typename T>
 class G {
 public:
   G() {}
   virtual void f0();
   virtual void f1();