[CodeGen] Collect information about sizes of accesses and access types for TBAA

The information about access and type sizes is necessary for
producing TBAA metadata in the new size-aware format. With this
patch, D39955 and D39956 in place we should be able to change
CodeGenTBAA::createScalarTypeNode() and
CodeGenTBAA::getBaseTypeInfo() to generate metadata in the new
format under the -new-struct-path-tbaa command-line option. For
now, this new information remains unused.

Differential Revision: https://reviews.llvm.org/D40176


git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@319012 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/CGClass.cpp

@@ -2423,7 +2423,8 @@ void CodeGenFunction::InitializeVTablePointer(const VPtr &Vptr) {
   VTableAddressPoint = Builder.CreateBitCast(VTableAddressPoint, VTablePtrTy);
 
   llvm::StoreInst *Store = Builder.CreateStore(VTableAddressPoint, VTableField);
-  CGM.DecorateInstructionWithTBAA(Store, CGM.getTBAAVTablePtrAccessInfo());
+  TBAAAccessInfo TBAAInfo = CGM.getTBAAVTablePtrAccessInfo(VTablePtrTy);
+  CGM.DecorateInstructionWithTBAA(Store, TBAAInfo);
   if (CGM.getCodeGenOpts().OptimizationLevel > 0 &&
       CGM.getCodeGenOpts().StrictVTablePointers)
     CGM.DecorateInstructionWithInvariantGroup(Store, Vptr.VTableClass);
@@ -2517,7 +2518,8 @@ llvm::Value *CodeGenFunction::GetVTablePtr(Address This,
                                            const CXXRecordDecl *RD) {
   Address VTablePtrSrc = Builder.CreateElementBitCast(This, VTableTy);
   llvm::Instruction *VTable = Builder.CreateLoad(VTablePtrSrc, "vtable");
-  CGM.DecorateInstructionWithTBAA(VTable, CGM.getTBAAVTablePtrAccessInfo());
+  TBAAAccessInfo TBAAInfo = CGM.getTBAAVTablePtrAccessInfo(VTableTy);
+  CGM.DecorateInstructionWithTBAA(VTable, TBAAInfo);
 
   if (CGM.getCodeGenOpts().OptimizationLevel > 0 &&
       CGM.getCodeGenOpts().StrictVTablePointers)

lib/CodeGen/CodeGenModule.cpp

@@ -136,7 +136,7 @@ CodeGenModule::CodeGenModule(ASTContext &C, const HeaderSearchOptions &HSO,
   // Enable TBAA unless it's suppressed. ThreadSanitizer needs TBAA even at O0.
   if (LangOpts.Sanitize.has(SanitizerKind::Thread) ||
       (!CodeGenOpts.RelaxedAliasing && CodeGenOpts.OptimizationLevel > 0))
-    TBAA.reset(new CodeGenTBAA(Context, VMContext, CodeGenOpts, getLangOpts(),
+    TBAA.reset(new CodeGenTBAA(Context, TheModule, CodeGenOpts, getLangOpts(),
                                getCXXABI().getMangleContext()));
 
   // If debug info or coverage generation is enabled, create the CGDebugInfo
@@ -579,13 +579,20 @@ llvm::MDNode *CodeGenModule::getTBAATypeInfo(QualType QTy) {
 }
 
 TBAAAccessInfo CodeGenModule::getTBAAAccessInfo(QualType AccessType) {
-  return TBAAAccessInfo(getTBAATypeInfo(AccessType));
+  // Pointee values may have incomplete types, but they shall never be
+  // dereferenced.
+  if (AccessType->isIncompleteType())
+    return TBAAAccessInfo::getIncompleteInfo();
+
+  uint64_t Size = Context.getTypeSizeInChars(AccessType).getQuantity();
+  return TBAAAccessInfo(getTBAATypeInfo(AccessType), Size);
 }
 
-TBAAAccessInfo CodeGenModule::getTBAAVTablePtrAccessInfo() {
+TBAAAccessInfo
+CodeGenModule::getTBAAVTablePtrAccessInfo(llvm::Type *VTablePtrType) {
   if (!TBAA)
     return TBAAAccessInfo();
-  return TBAA->getVTablePtrAccessInfo();
+  return TBAA->getVTablePtrAccessInfo(VTablePtrType);
 }
 
 llvm::MDNode *CodeGenModule::getTBAAStructInfo(QualType QTy) {

lib/CodeGen/CodeGenModule.h

@@ -664,7 +664,7 @@ public:
 
   /// getTBAAVTablePtrAccessInfo - Get the TBAA information that describes an
   /// access to a virtual table pointer.
-  TBAAAccessInfo getTBAAVTablePtrAccessInfo();
+  TBAAAccessInfo getTBAAVTablePtrAccessInfo(llvm::Type *VTablePtrType);
 
   llvm::MDNode *getTBAAStructInfo(QualType QTy);
 

lib/CodeGen/CodeGenTBAA.cpp

@@ -25,16 +25,18 @@
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/Metadata.h"
+#include "llvm/IR/Module.h"
 #include "llvm/IR/Type.h"
 using namespace clang;
 using namespace CodeGen;
 
-CodeGenTBAA::CodeGenTBAA(ASTContext &Ctx, llvm::LLVMContext& VMContext,
+CodeGenTBAA::CodeGenTBAA(ASTContext &Ctx, llvm::Module &M,
                          const CodeGenOptions &CGO,
                          const LangOptions &Features, MangleContext &MContext)
-  : Context(Ctx), CodeGenOpts(CGO), Features(Features), MContext(MContext),
-    MDHelper(VMContext), Root(nullptr), Char(nullptr) {
-}
+  : Context(Ctx), Module(M), CodeGenOpts(CGO),
+    Features(Features), MContext(MContext), MDHelper(M.getContext()),
+    Root(nullptr), Char(nullptr)
+{}
 
 CodeGenTBAA::~CodeGenTBAA() {
 }
@@ -54,10 +56,10 @@ llvm::MDNode *CodeGenTBAA::getRoot() {
   return Root;
 }
 
-// For both scalar TBAA and struct-path aware TBAA, the scalar type has the
-// same format: name, parent node, and offset.
-llvm::MDNode *CodeGenTBAA::createTBAAScalarType(StringRef Name,
-                                                llvm::MDNode *Parent) {
+llvm::MDNode *CodeGenTBAA::createScalarTypeNode(StringRef Name,
+                                                llvm::MDNode *Parent,
+                                                uint64_t Size) {
+  (void)Size; // TODO: Support generation of size-aware type nodes.
   return MDHelper.createTBAAScalarTypeNode(Name, Parent);
 }
 
@@ -67,7 +69,7 @@ llvm::MDNode *CodeGenTBAA::getChar() {
   // these special powers only cover user-accessible memory, and doesn't
   // include things like vtables.
   if (!Char)
-    Char = createTBAAScalarType("omnipotent char", getRoot());
+    Char = createScalarTypeNode("omnipotent char", getRoot(), /* Size= */ 1);
 
   return Char;
 }
@@ -108,6 +110,8 @@ static bool isValidBaseType(QualType QTy) {
 }
 
 llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
+  uint64_t Size = Context.getTypeSizeInChars(Ty).getQuantity();
+
   // Handle builtin types.
   if (const BuiltinType *BTy = dyn_cast<BuiltinType>(Ty)) {
     switch (BTy->getKind()) {
@@ -138,7 +142,7 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
     // treating wchar_t, char16_t, and char32_t as distinct from their
     // "underlying types".
     default:
-      return createTBAAScalarType(BTy->getName(Features), getChar());
+      return createScalarTypeNode(BTy->getName(Features), getChar(), Size);
     }
   }
 
@@ -152,7 +156,7 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
   // TODO: Implement C++'s type "similarity" and consider dis-"similar"
   // pointers distinct.
   if (Ty->isPointerType() || Ty->isReferenceType())
-    return createTBAAScalarType("any pointer", getChar());
+    return createScalarTypeNode("any pointer", getChar(), Size);
 
   // Enum types are distinct types. In C++ they have "underlying types",
   // however they aren't related for TBAA.
@@ -167,7 +171,7 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type *Ty) {
     SmallString<256> OutName;
     llvm::raw_svector_ostream Out(OutName);
     MContext.mangleTypeName(QualType(ETy, 0), Out);
-    return createTBAAScalarType(OutName, getChar());
+    return createScalarTypeNode(OutName, getChar(), Size);
   }
 
   // For now, handle any other kind of type conservatively.
@@ -204,8 +208,11 @@ llvm::MDNode *CodeGenTBAA::getTypeInfo(QualType QTy) {
   return MetadataCache[Ty] = TypeNode;
 }
 
-TBAAAccessInfo CodeGenTBAA::getVTablePtrAccessInfo() {
-  return TBAAAccessInfo(createTBAAScalarType("vtable pointer", getRoot()));
+TBAAAccessInfo CodeGenTBAA::getVTablePtrAccessInfo(llvm::Type *VTablePtrType) {
+  llvm::DataLayout DL(&Module);
+  unsigned Size = DL.getPointerTypeSize(VTablePtrType);
+  return TBAAAccessInfo(createScalarTypeNode("vtable pointer", getRoot(), Size),
+                        Size);
 }
 
 bool
@@ -245,7 +252,7 @@ CodeGenTBAA::CollectFields(uint64_t BaseOffset,
   uint64_t Offset = BaseOffset;
   uint64_t Size = Context.getTypeSizeInChars(QTy).getQuantity();
   llvm::MDNode *TBAAType = MayAlias ? getChar() : getTypeInfo(QTy);
-  llvm::MDNode *TBAATag = getAccessTagInfo(TBAAAccessInfo(TBAAType));
+  llvm::MDNode *TBAATag = getAccessTagInfo(TBAAAccessInfo(TBAAType, Size));
   Fields.push_back(llvm::MDBuilder::TBAAStructField(Offset, Size, TBAATag));
   return true;
 }
@@ -268,19 +275,20 @@ CodeGenTBAA::getTBAAStructInfo(QualType QTy) {
 llvm::MDNode *CodeGenTBAA::getBaseTypeInfoHelper(const Type *Ty) {
   if (auto *TTy = dyn_cast<RecordType>(Ty)) {
     const RecordDecl *RD = TTy->getDecl()->getDefinition();
-
     const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
-    SmallVector <std::pair<llvm::MDNode*, uint64_t>, 4> Fields;
-    unsigned idx = 0;
-    for (RecordDecl::field_iterator i = RD->field_begin(),
-         e = RD->field_end(); i != e; ++i, ++idx) {
-      QualType FieldQTy = i->getType();
-      llvm::MDNode *FieldNode = isValidBaseType(FieldQTy) ?
+    SmallVector<llvm::MDBuilder::TBAAStructField, 4> Fields;
+    for (FieldDecl *Field : RD->fields()) {
+      QualType FieldQTy = Field->getType();
+      llvm::MDNode *TypeNode = isValidBaseType(FieldQTy) ?
           getBaseTypeInfo(FieldQTy) : getTypeInfo(FieldQTy);
-      if (!FieldNode)
+      if (!TypeNode)
         return BaseTypeMetadataCache[Ty] = nullptr;
-      Fields.push_back(std::make_pair(
-          FieldNode, Layout.getFieldOffset(idx) / Context.getCharWidth()));
+
+      uint64_t BitOffset = Layout.getFieldOffset(Field->getFieldIndex());
+      uint64_t Offset = Context.toCharUnitsFromBits(BitOffset).getQuantity();
+      uint64_t Size = Context.getTypeSizeInChars(FieldQTy).getQuantity();
+      Fields.push_back(llvm::MDBuilder::TBAAStructField(Offset, Size,
+                                                        TypeNode));
     }
 
     SmallString<256> OutName;
@@ -291,8 +299,15 @@ llvm::MDNode *CodeGenTBAA::getBaseTypeInfoHelper(const Type *Ty) {
     } else {
       OutName = RD->getName();
     }
+
+    // TODO: Support size-aware type nodes and create one here for the
+    // given aggregate type.
+
     // Create the struct type node with a vector of pairs (offset, type).
-    return MDHelper.createTBAAStructTypeNode(OutName, Fields);
+    SmallVector<std::pair<llvm::MDNode*, uint64_t>, 4> OffsetsAndTypes;
+    for (const auto &Field : Fields)
+        OffsetsAndTypes.push_back(std::make_pair(Field.TBAA, Field.Offset));
+    return MDHelper.createTBAAStructTypeNode(OutName, OffsetsAndTypes);
   }
 
   return nullptr;
@@ -314,14 +329,16 @@ llvm::MDNode *CodeGenTBAA::getBaseTypeInfo(QualType QTy) {
 }
 
 llvm::MDNode *CodeGenTBAA::getAccessTagInfo(TBAAAccessInfo Info) {
+  assert(!Info.isIncomplete() && "Access to an object of an incomplete type!");
+
   if (Info.isMayAlias())
-    Info = TBAAAccessInfo(getChar());
+    Info = TBAAAccessInfo(getChar(), Info.Size);
 
   if (!Info.AccessType)
     return nullptr;
 
   if (!CodeGenOpts.StructPathTBAA)
-    Info = TBAAAccessInfo(Info.AccessType);
+    Info = TBAAAccessInfo(Info.AccessType, Info.Size);
 
   llvm::MDNode *&N = AccessTagMetadataCache[Info];
   if (N)

lib/CodeGen/CodeGenTBAA.h

@@ -36,40 +36,53 @@ class CGRecordLayout;
 enum class TBAAAccessKind : unsigned {
   Ordinary,
   MayAlias,
+  Incomplete,
 };
 
 // TBAAAccessInfo - Describes a memory access in terms of TBAA.
 struct TBAAAccessInfo {
   TBAAAccessInfo(TBAAAccessKind Kind, llvm::MDNode *BaseType,
-                 llvm::MDNode *AccessType, uint64_t Offset)
-    : Kind(Kind), BaseType(BaseType), AccessType(AccessType), Offset(Offset)
+                 llvm::MDNode *AccessType, uint64_t Offset, uint64_t Size)
+    : Kind(Kind), BaseType(BaseType), AccessType(AccessType),
+      Offset(Offset), Size(Size)
   {}
 
   TBAAAccessInfo(llvm::MDNode *BaseType, llvm::MDNode *AccessType,
-                 uint64_t Offset)
-    : TBAAAccessInfo(TBAAAccessKind::Ordinary, BaseType, AccessType, Offset)
+                 uint64_t Offset, uint64_t Size)
+    : TBAAAccessInfo(TBAAAccessKind::Ordinary, BaseType, AccessType,
+                     Offset, Size)
   {}
 
-  explicit TBAAAccessInfo(llvm::MDNode *AccessType)
-    : TBAAAccessInfo(/* BaseType= */ nullptr, AccessType, /* Offset= */ 0)
+  explicit TBAAAccessInfo(llvm::MDNode *AccessType, uint64_t Size)
+    : TBAAAccessInfo(/* BaseType= */ nullptr, AccessType, /* Offset= */ 0, Size)
   {}
 
   TBAAAccessInfo()
-    : TBAAAccessInfo(/* AccessType= */ nullptr)
+    : TBAAAccessInfo(/* AccessType= */ nullptr, /* Size= */ 0)
   {}
 
   static TBAAAccessInfo getMayAliasInfo() {
-    return TBAAAccessInfo(TBAAAccessKind::MayAlias, /* BaseType= */ nullptr,
-                          /* AccessType= */ nullptr, /* Offset= */ 0);
+    return TBAAAccessInfo(TBAAAccessKind::MayAlias,
+                          /* BaseType= */ nullptr, /* AccessType= */ nullptr,
+                          /* Offset= */ 0, /* Size= */ 0);
   }
 
   bool isMayAlias() const { return Kind == TBAAAccessKind::MayAlias; }
 
+  static TBAAAccessInfo getIncompleteInfo() {
+    return TBAAAccessInfo(TBAAAccessKind::Incomplete,
+                          /* BaseType= */ nullptr, /* AccessType= */ nullptr,
+                          /* Offset= */ 0, /* Size= */ 0);
+  }
+
+  bool isIncomplete() const { return Kind == TBAAAccessKind::Incomplete; }
+
   bool operator==(const TBAAAccessInfo &Other) const {
     return Kind == Other.Kind &&
            BaseType == Other.BaseType &&
            AccessType == Other.AccessType &&
-           Offset == Other.Offset;
+           Offset == Other.Offset &&
+           Size == Other.Size;
   }
 
   bool operator!=(const TBAAAccessInfo &Other) const {
@@ -95,12 +108,16 @@ struct TBAAAccessInfo {
   /// Offset - The byte offset of the final access within the base one. Must be
   /// zero if the base access type is not specified.
   uint64_t Offset;
+
+  /// Size - The size of access, in bytes.
+  uint64_t Size;
 };
 
 /// CodeGenTBAA - This class organizes the cross-module state that is used
 /// while lowering AST types to LLVM types.
 class CodeGenTBAA {
   ASTContext &Context;
+  llvm::Module &Module;
   const CodeGenOptions &CodeGenOpts;
   const LangOptions &Features;
   MangleContext &MContext;
@@ -138,10 +155,10 @@ class CodeGenTBAA {
                      SmallVectorImpl<llvm::MDBuilder::TBAAStructField> &Fields,
                      bool MayAlias);
 
-  /// A wrapper function to create a scalar type. For struct-path aware TBAA,
-  /// the scalar type has the same format as the struct type: name, offset,
-  /// pointer to another node in the type DAG.
-  llvm::MDNode *createTBAAScalarType(StringRef Name, llvm::MDNode *Parent);
+  /// createScalarTypeNode - A wrapper function to create a metadata node
+  /// describing a scalar type.
+  llvm::MDNode *createScalarTypeNode(StringRef Name, llvm::MDNode *Parent,
+                                     uint64_t Size);
 
   /// getTypeInfoHelper - An internal helper function to generate metadata used
   /// to describe accesses to objects of the given type.
@@ -152,10 +169,8 @@ class CodeGenTBAA {
   llvm::MDNode *getBaseTypeInfoHelper(const Type *Ty);
 
 public:
-  CodeGenTBAA(ASTContext &Ctx, llvm::LLVMContext &VMContext,
-              const CodeGenOptions &CGO,
-              const LangOptions &Features,
-              MangleContext &MContext);
+  CodeGenTBAA(ASTContext &Ctx, llvm::Module &M, const CodeGenOptions &CGO,
+              const LangOptions &Features, MangleContext &MContext);
   ~CodeGenTBAA();
 
   /// getTypeInfo - Get metadata used to describe accesses to objects of the
@@ -164,7 +179,7 @@ public:
 
   /// getVTablePtrAccessInfo - Get the TBAA information that describes an
   /// access to a virtual table pointer.
-  TBAAAccessInfo getVTablePtrAccessInfo();
+  TBAAAccessInfo getVTablePtrAccessInfo(llvm::Type *VTablePtrType);
 
   /// getTBAAStructInfo - Get the TBAAStruct MDNode to be used for a memcpy of
   /// the given type.
@@ -200,6 +215,7 @@ template<> struct DenseMapInfo<clang::CodeGen::TBAAAccessInfo> {
       static_cast<clang::CodeGen::TBAAAccessKind>(UnsignedKey),
       DenseMapInfo<MDNode *>::getEmptyKey(),
       DenseMapInfo<MDNode *>::getEmptyKey(),
+      DenseMapInfo<uint64_t>::getEmptyKey(),
       DenseMapInfo<uint64_t>::getEmptyKey());
   }
 
@@ -209,6 +225,7 @@ template<> struct DenseMapInfo<clang::CodeGen::TBAAAccessInfo> {
       static_cast<clang::CodeGen::TBAAAccessKind>(UnsignedKey),
       DenseMapInfo<MDNode *>::getTombstoneKey(),
       DenseMapInfo<MDNode *>::getTombstoneKey(),
+      DenseMapInfo<uint64_t>::getTombstoneKey(),
       DenseMapInfo<uint64_t>::getTombstoneKey());
   }
 
@@ -217,7 +234,8 @@ template<> struct DenseMapInfo<clang::CodeGen::TBAAAccessInfo> {
     return DenseMapInfo<unsigned>::getHashValue(KindValue) ^
            DenseMapInfo<MDNode *>::getHashValue(Val.BaseType) ^
            DenseMapInfo<MDNode *>::getHashValue(Val.AccessType) ^
-           DenseMapInfo<uint64_t>::getHashValue(Val.Offset);
+           DenseMapInfo<uint64_t>::getHashValue(Val.Offset) ^
+           DenseMapInfo<uint64_t>::getHashValue(Val.Size);
   }
 
   static bool isEqual(const clang::CodeGen::TBAAAccessInfo &LHS,