[CIR][CIRGen][TBAA] Add support for scalar types (#1220)

clang/include/clang/CIR/Dialect/Builder/CIRBaseBuilder.h

@@ -170,7 +170,7 @@ class CIRBaseBuilderTy : public mlir::OpBuilder {
                                /*alignment=*/intAttr,
                                /*mem_order=*/
                                cir::MemOrderAttr{},
-                               /*tbaa=*/mlir::ArrayAttr{});
+                               /*tbaa=*/cir::TBAAAttr{});
   }
 
   mlir::Value createAlignedLoad(mlir::Location loc, mlir::Value ptr,
@@ -357,7 +357,7 @@ class CIRBaseBuilderTy : public mlir::OpBuilder {
         val.getType())
       dst = createPtrBitcast(dst, val.getType());
     return create<cir::StoreOp>(loc, val, dst, _volatile, align, order,
-                                /*tbaa=*/mlir::ArrayAttr{});
+                                /*tbaa=*/cir::TBAAAttr{});
   }
 
   mlir::Value createAlloca(mlir::Location loc, cir::PointerType addrType,
@@ -405,7 +405,7 @@ class CIRBaseBuilderTy : public mlir::OpBuilder {
   cir::CopyOp createCopy(mlir::Value dst, mlir::Value src,
                          bool isVolatile = false) {
     return create<cir::CopyOp>(dst.getLoc(), dst, src, isVolatile,
-                               /*tbaa=*/mlir::ArrayAttr{});
+                               /*tbaa=*/cir::TBAAAttr{});
   }
 
   cir::MemCpyOp createMemCpy(mlir::Location loc, mlir::Value dst,

clang/include/clang/CIR/Dialect/IR/CIRAttrs.td

@@ -24,8 +24,9 @@ include "clang/CIR/Interfaces/ASTAttrInterfaces.td"
 // CIR Attrs
 //===----------------------------------------------------------------------===//
 
-class CIR_Attr<string name, string attrMnemonic, list<Trait> traits = []>
-    : AttrDef<CIR_Dialect, name, traits> {
+class CIR_Attr<string name, string attrMnemonic, list<Trait> traits = [],
+               string baseCppClass = "::mlir::Attribute">
+    : AttrDef<CIR_Dialect, name, traits, baseCppClass> {
   let mnemonic = attrMnemonic;
 }
 
@@ -1294,8 +1295,7 @@ def GlobalAnnotationValuesAttr : CIR_Attr<"GlobalAnnotationValues",
   let genVerifyDecl = 1;
 }
 
-def CIR_TBAAAttr : CIR_Attr<"TBAA", "tbaa", []> {
-}
+include "clang/CIR/Dialect/IR/CIRTBAAAttrs.td"
 
 include "clang/CIR/Dialect/IR/CIROpenCLAttrs.td"
 

clang/include/clang/CIR/Dialect/IR/CIROps.td

@@ -588,7 +588,7 @@ def LoadOp : CIR_Op<"load", [
                        UnitAttr:$is_volatile,
                        OptionalAttr<I64Attr>:$alignment,
                        OptionalAttr<MemOrder>:$mem_order,
-                       OptionalAttr<ArrayAttr>:$tbaa
+                       OptionalAttr<CIR_AnyTBAAAttr>:$tbaa
                        );
   let results = (outs CIR_AnyType:$result);
 
@@ -657,7 +657,7 @@ def StoreOp : CIR_Op<"store", [
                        UnitAttr:$is_volatile,
                        OptionalAttr<I64Attr>:$alignment,
                        OptionalAttr<MemOrder>:$mem_order,
-                       OptionalAttr<ArrayAttr>:$tbaa);
+                       OptionalAttr<CIR_AnyTBAAAttr>:$tbaa);
 
   let assemblyFormat = [{
     (`volatile` $is_volatile^)?
@@ -4068,7 +4068,7 @@ def CopyOp : CIR_Op<"copy",
   let arguments = (ins Arg<CIR_PointerType, "", [MemWrite]>:$dst,
                        Arg<CIR_PointerType, "", [MemRead]>:$src,
                        UnitAttr:$is_volatile,
-                       OptionalAttr<ArrayAttr>:$tbaa);
+                       OptionalAttr<CIR_TBAAAttr>:$tbaa);
   let summary = "Copies contents from a CIR pointer to another";
   let description = [{
     Given two CIR pointers, `src` and `dst`, `cir.copy` will copy the memory

clang/include/clang/CIR/Dialect/IR/CIRTBAAAttrs.td

@@ -0,0 +1,38 @@
+//===----------------------------------------------------------------------===//
+// TBAAAttr
+//===----------------------------------------------------------------------===//
+
+def CIR_TBAAAttr : CIR_Attr<"TBAA", "tbaa", []> {
+  let summary = "CIR dialect TBAA base attribute";
+}
+
+//===----------------------------------------------------------------------===//
+// TBAAScalarAttr
+//===----------------------------------------------------------------------===//
+
+def CIR_TBAAScalarAttr : CIR_Attr<"TBAAScalar", "tbaa_scalar", [], "TBAAAttr"> {
+  let summary = "Describes a scalar type in TBAA with an identifier.";
+
+  let parameters = (ins CIR_AnyScalarType : $type);
+
+  let description = [{
+    Define a TBAA attribute.
+
+    Example:
+    ```mlir
+    // CIR_TBAAScalarAttr
+    #tbaa_scalar = #cir.tbaa_scalar<type = !s32i>
+    #tbaa_scalar1 = #cir.tbaa_scalar<type = !u32i>
+    ```
+
+    See the following link for more details:
+    https://llvm.org/docs/LangRef.html#tbaa-metadata
+  }];
+
+  let assemblyFormat = "`<` struct(params) `>`";
+}
+
+def CIR_AnyTBAAAttr : AnyAttrOf<[
+  CIR_TBAAAttr,
+  CIR_TBAAScalarAttr
+]>;

clang/include/clang/CIR/Dialect/IR/CIRTypes.td

@@ -74,6 +74,28 @@ def CIR_IntType : CIR_Type<"Int", "int",
     static bool isValidPrimitiveIntBitwidth(unsigned width) {
       return width == 8 || width == 16 || width == 32 || width == 64;
     }
+
+    llvm::StringRef getTBAATypeName() const {
+      switch (getWidth()) {
+        case 1:
+        case 8: {
+          return "omnipotent char";
+        }
+        case 16: {
+          return "short";
+        }
+        case 32: {
+          return "int";
+        }
+        case 64: {
+          return "long";
+        }
+        default: {
+          llvm::errs() << "unknown type: " << *this << "\n";
+          return "unknown";
+        }
+      }
+    }
   }];
   let genVerifyDecl = 1;
 }
@@ -609,4 +631,11 @@ def CIR_AnyType : AnyTypeOf<[
   CIR_ComplexType
 ]>;
 
+def CIR_AnyScalarType : AnyTypeOf<[
+  CIR_IntType, CIR_PointerType, CIR_DataMemberType, CIR_MethodType,
+  CIR_BoolType, CIR_ArrayType, CIR_VectorType, CIR_FuncType, CIR_VoidType,
+  CIR_ExceptionType, CIR_AnyFloat, CIR_FP16, CIR_BFloat16,
+  CIR_ComplexType
+]>;
+
 #endif // MLIR_CIR_DIALECT_CIR_TYPES

clang/include/clang/CIR/MissingFeatures.h

@@ -58,7 +58,12 @@ struct MissingFeatures {
   // sanitizer related type check features
   static bool emitTypeCheck() { return false; }
   static bool tbaa() { return false; }
-  static bool tbaa_struct() { return false; }
+  static bool tbaaStruct() { return false; }
+  static bool tbaaTagForStruct() { return false; }
+  static bool tbaaVTablePtr() { return false; }
+  static bool tbaaIncompleteType() { return false; }
+  static bool tbaaMergeTBAAInfo() { return false; }
+  static bool tbaaMayAlias() { return false; }
   static bool cleanups() { return false; }
   static bool emitNullabilityCheck() { return false; }
   static bool ptrAuth() { return false; }

clang/lib/CIR/CodeGen/CIRGenBuilder.h

@@ -839,7 +839,7 @@ class CIRGenBuilderTy : public cir::CIRBaseBuilderTy {
     return create<cir::LoadOp>(
         loc, addr.getElementType(), addr.getPointer(), /*isDeref=*/false,
         /*is_volatile=*/isVolatile, /*alignment=*/mlir::IntegerAttr{},
-        /*mem_order=*/cir::MemOrderAttr{}, /*tbaa=*/mlir::ArrayAttr{});
+        /*mem_order=*/cir::MemOrderAttr{}, /*tbaa=*/cir::TBAAAttr{});
   }
 
   mlir::Value createAlignedLoad(mlir::Location loc, mlir::Type ty,

clang/lib/CIR/CodeGen/CIRGenExprAgg.cpp

@@ -1716,7 +1716,7 @@ void CIRGenFunction::emitAggregateCopy(LValue Dest, LValue Src, QualType Ty,
   // Determine the metadata to describe the position of any padding in this
   // memcpy, as well as the TBAA tags for the members of the struct, in case
   // the optimizer wishes to expand it in to scalar memory operations.
-  assert(!cir::MissingFeatures::tbaa_struct() && "tbaa.struct NYI");
+  assert(!cir::MissingFeatures::tbaaStruct() && "tbaa.struct NYI");
   if (CGM.getCodeGenOpts().NewStructPathTBAA) {
     TBAAAccessInfo TBAAInfo = CGM.mergeTBAAInfoForMemoryTransfer(
         Dest.getTBAAInfo(), Src.getTBAAInfo());

clang/lib/CIR/CodeGen/CIRGenModule.cpp

@@ -3988,7 +3988,7 @@ cir::TBAAAttr CIRGenModule::getTBAABaseTypeInfo(QualType QTy) {
   return tbaa->getBaseTypeInfo(QTy);
 }
 
-mlir::ArrayAttr CIRGenModule::getTBAAAccessTagInfo(TBAAAccessInfo tbaaInfo) {
+cir::TBAAAttr CIRGenModule::getTBAAAccessTagInfo(TBAAAccessInfo tbaaInfo) {
   if (!tbaa) {
     return nullptr;
   }

clang/lib/CIR/CodeGen/CIRGenModule.h

@@ -525,7 +525,7 @@ class CIRGenModule : public CIRGenTypeCache {
   /// type is not suitable for use in TBAA access tags.
   cir::TBAAAttr getTBAABaseTypeInfo(QualType QTy);
 
-  mlir::ArrayAttr getTBAAAccessTagInfo(TBAAAccessInfo tbaaInfo);
+  cir::TBAAAttr getTBAAAccessTagInfo(TBAAAccessInfo tbaaInfo);
 
   /// Get merged TBAA information for the purposes of type casts.
   TBAAAccessInfo mergeTBAAInfoForCast(TBAAAccessInfo SourceInfo,

clang/lib/CIR/CodeGen/CIRGenTBAA.cpp

@@ -1,11 +1,12 @@
 #include "CIRGenTBAA.h"
-#include "CIRGenCXXABI.h"
 #include "CIRGenTypes.h"
 #include "mlir/IR/BuiltinAttributes.h"
 #include "mlir/IR/MLIRContext.h"
 #include "mlir/Interfaces/DataLayoutInterfaces.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/RecordLayout.h"
+#include "clang/CIR/Dialect/IR/CIRTypes.h"
+#include "clang/CIR/MissingFeatures.h"
 #include "llvm/Support/ErrorHandling.h"
 namespace clang::CIRGen {
 
@@ -21,44 +22,159 @@ CIRGenTBAA::CIRGenTBAA(mlir::MLIRContext *mlirContext,
     : mlirContext(mlirContext), astContext(astContext), types(types),
       moduleOp(moduleOp), codeGenOpts(codeGenOpts), features(features) {}
 
+cir::TBAAAttr CIRGenTBAA::getChar() {
+  return cir::TBAAScalarAttr::get(mlirContext,
+                                  cir::IntType::get(mlirContext, 1, true));
+}
+
+static bool typeHasMayAlias(clang::QualType qty) {
+  // Tagged types have declarations, and therefore may have attributes.
+  if (auto *td = qty->getAsTagDecl())
+    if (td->hasAttr<MayAliasAttr>())
+      return true;
+
+  // Also look for may_alias as a declaration attribute on a typedef.
+  // FIXME: We should follow GCC and model may_alias as a type attribute
+  // rather than as a declaration attribute.
+  while (auto *tt = qty->getAs<TypedefType>()) {
+    if (tt->getDecl()->hasAttr<MayAliasAttr>())
+      return true;
+    qty = tt->desugar();
+  }
+  return false;
+}
+
+/// Check if the given type is a valid base type to be used in access tags.
+static bool isValidBaseType(clang::QualType qty) {
+  if (const clang::RecordType *tty = qty->getAs<clang::RecordType>()) {
+    const clang::RecordDecl *rd = tty->getDecl()->getDefinition();
+    // Incomplete types are not valid base access types.
+    if (!rd)
+      return false;
+    if (rd->hasFlexibleArrayMember())
+      return false;
+    // rd can be struct, union, class, interface or enum.
+    // For now, we only handle struct and class.
+    if (rd->isStruct() || rd->isClass())
+      return true;
+  }
+  return false;
+}
+
 cir::TBAAAttr CIRGenTBAA::getTypeInfo(clang::QualType qty) {
-  return tbaa_NYI(mlirContext);
+  // At -O0 or relaxed aliasing, TBAA is not emitted for regular types.
+  if (codeGenOpts.OptimizationLevel == 0 || codeGenOpts.RelaxedAliasing) {
+    return nullptr;
+  }
+
+  // If the type has the may_alias attribute (even on a typedef), it is
+  // effectively in the general char alias class.
+  if (typeHasMayAlias(qty)) {
+    assert(!cir::MissingFeatures::tbaaMayAlias());
+    return getChar();
+  }
+  // We need this function to not fall back to returning the "omnipotent char"
+  // type node for aggregate and union types. Otherwise, any dereference of an
+  // aggregate will result into the may-alias access descriptor, meaning all
+  // subsequent accesses to direct and indirect members of that aggregate will
+  // be considered may-alias too.
+  // function.
+  if (isValidBaseType(qty)) {
+    // TODO(cir): support TBAA with struct
+    return tbaa_NYI(mlirContext);
+  }
+
+  const clang::Type *ty = astContext.getCanonicalType(qty).getTypePtr();
+  if (metadataCache.contains(ty)) {
+    return metadataCache[ty];
+  }
+
+  // Note that the following helper call is allowed to add new nodes to the
+  // cache, which invalidates all its previously obtained iterators. So we
+  // first generate the node for the type and then add that node to the
+  // cache.
+  auto typeNode = cir::TBAAScalarAttr::get(mlirContext, types.ConvertType(qty));
+  return metadataCache[ty] = typeNode;
 }
 
 TBAAAccessInfo CIRGenTBAA::getAccessInfo(clang::QualType accessType) {
-  return TBAAAccessInfo();
+  // Pointee values may have incomplete types, but they shall never be
+  // dereferenced.
+  if (accessType->isIncompleteType()) {
+    assert(!cir::MissingFeatures::tbaaIncompleteType());
+    return TBAAAccessInfo::getIncompleteInfo();
+  }
+
+  if (typeHasMayAlias(accessType)) {
+    assert(!cir::MissingFeatures::tbaaMayAlias());
+    return TBAAAccessInfo::getMayAliasInfo();
+  }
+
+  uint64_t size = astContext.getTypeSizeInChars(accessType).getQuantity();
+  return TBAAAccessInfo(getTypeInfo(accessType), size);
 }
 
 TBAAAccessInfo CIRGenTBAA::getVTablePtrAccessInfo(mlir::Type vtablePtrType) {
+  // TODO(cir): support vtable ptr
+  assert(!cir::MissingFeatures::tbaaVTablePtr());
   return TBAAAccessInfo();
 }
 
 mlir::ArrayAttr CIRGenTBAA::getTBAAStructInfo(clang::QualType qty) {
-  return mlir::ArrayAttr::get(mlirContext, {});
+  assert(!cir::MissingFeatures::tbaaStruct() && "tbaa.struct NYI");
+  return mlir::ArrayAttr();
 }
 
 cir::TBAAAttr CIRGenTBAA::getBaseTypeInfo(clang::QualType qty) {
   return tbaa_NYI(mlirContext);
 }
 
-mlir::ArrayAttr CIRGenTBAA::getAccessTagInfo(TBAAAccessInfo tbaaInfo) {
-  return mlir::ArrayAttr::get(mlirContext, {tbaa_NYI(mlirContext)});
+cir::TBAAAttr CIRGenTBAA::getAccessTagInfo(TBAAAccessInfo tbaaInfo) {
+  assert(!tbaaInfo.isIncomplete() &&
+         "Access to an object of an incomplete type!");
+
+  if (tbaaInfo.isMayAlias()) {
+    assert(!cir::MissingFeatures::tbaaMayAlias());
+    tbaaInfo = TBAAAccessInfo(getChar(), tbaaInfo.size);
+  }
+  if (!tbaaInfo.accessType) {
+    return nullptr;
+  }
+
+  if (!codeGenOpts.StructPathTBAA)
+    tbaaInfo = TBAAAccessInfo(tbaaInfo.accessType, tbaaInfo.size);
+
+  if (!tbaaInfo.baseType) {
+    tbaaInfo.baseType = tbaaInfo.accessType;
+    assert(!tbaaInfo.offset &&
+           "Nonzero offset for an access with no base type!");
+  }
+  if (codeGenOpts.NewStructPathTBAA) {
+    llvm_unreachable("NYI");
+  }
+  if (tbaaInfo.baseType == tbaaInfo.accessType) {
+    return tbaaInfo.accessType;
+  }
+  return tbaa_NYI(mlirContext);
 }
 
 TBAAAccessInfo CIRGenTBAA::mergeTBAAInfoForCast(TBAAAccessInfo sourceInfo,
                                                 TBAAAccessInfo targetInfo) {
+  assert(!cir::MissingFeatures::tbaaMergeTBAAInfo());
   return TBAAAccessInfo();
 }
 
 TBAAAccessInfo
 CIRGenTBAA::mergeTBAAInfoForConditionalOperator(TBAAAccessInfo infoA,
                                                 TBAAAccessInfo infoB) {
+  assert(!cir::MissingFeatures::tbaaMergeTBAAInfo());
   return TBAAAccessInfo();
 }
 
 TBAAAccessInfo
 CIRGenTBAA::mergeTBAAInfoForMemoryTransfer(TBAAAccessInfo destInfo,
                                            TBAAAccessInfo srcInfo) {
+  assert(!cir::MissingFeatures::tbaaMergeTBAAInfo());
   return TBAAAccessInfo();
 }