[CIR] Add support for polymorphic typeid with vtable lookup

This commit implements runtime type identification for polymorphic types,
completing typeid operator support in ClangIR.

Previously, only non-polymorphic (static) typeid was supported. This adds:

1. **Polymorphic typeid emission**: When typeid is applied to an expression
   of polymorphic class type, we now emit vtable lookup to get the runtime
   type_info pointer.

2. **Null pointer checking**: For pointer operands, we emit null checks and
   call emitBadTypeidCall (currently unreachable until exception support).
   Matches CodeGen behavior of always checking, even for references.

3. **Type safety checks**: emitTypeCheck ensures typeid is not used unsafely
   during object construction or destruction.

4. **Vtable layouts**: Supports both traditional (vtable[-1]) and relative
   (load.relative(vtable, -4)) layouts for Itanium ABI.

Implementation follows CodeGen closely:
- emitTypeidFromVTable handles the main logic with null checking
- CIRGenItaniumCXXABI::emitTypeid handles ABI-specific vtable access
- Uses existing getVTablePtr infrastructure
- shouldTypeidBeNullChecked matches CodeGen (always returns true)

The vtable layout for type_info access:
- Absolute: type_info* at vtable[-1] (8 bytes before vptr)
- Relative: type_info offset at vtable[-4] (4 bytes before vptr)

Test Plan:
- Added typeid-polymorphic.cpp with comprehensive tests
- Tests verify CIR output, LLVM lowering, and CodeGen comparison (OGCG)
- All 4 test cases have CIR + LLVM + OGCG checks
- Covers: basic usage, references, derived classes, const pointers
- Note: CIR lowering of PtrStrideOp does not emit 'inbounds' flag
  (general CIR lowering issue, not specific to this feature)
- All CIR tests pass


ghstack-source-id: 57a74ba
Pull-Request: #2005

Author: lanza

clang/lib/CIR/CodeGen/CIRGenCXXABI.h

@@ -366,6 +366,13 @@ class CIRGenCXXABI {
 
   virtual void emitBadCastCall(CIRGenFunction &CGF, mlir::Location loc) = 0;
 
+  virtual void emitBadTypeidCall(CIRGenFunction &CGF) = 0;
+  virtual bool shouldTypeidBeNullChecked(QualType SrcRecordTy) = 0;
+
+  virtual mlir::Value emitTypeid(CIRGenFunction &CGF, mlir::Location loc,
+                                 QualType SrcRecordTy, Address ThisPtr,
+                                 mlir::Type StdTypeInfoPtrTy) = 0;
+
   virtual mlir::Value
   getVirtualBaseClassOffset(mlir::Location loc, CIRGenFunction &CGF,
                             Address This, const CXXRecordDecl *ClassDecl,

clang/lib/CIR/CodeGen/CIRGenExprCXX.cpp

@@ -1810,6 +1810,58 @@ mlir::Value CIRGenFunction::emitDynamicCast(Address ThisAddr,
                                          destCirTy, isRefCast, ThisAddr);
 }
 
+static mlir::Value emitTypeidFromVTable(CIRGenFunction &CGF, mlir::Location loc,
+                                        const Expr *E,
+                                        mlir::Type stdTypeInfoPtrTy,
+                                        bool hasNullCheck) {
+  // Get the vtable pointer.
+  Address thisPtr = CGF.emitLValue(E).getAddress();
+
+  QualType srcRecordTy = E->getType();
+
+  // C++ [class.cdtor]p4:
+  //   If the operand of typeid refers to the object under construction or
+  //   destruction and the static type of the operand is neither the constructor
+  //   or destructor's class nor one of its bases, the behavior is undefined.
+  CGF.emitTypeCheck(CIRGenFunction::TCK_DynamicOperation, E->getExprLoc(),
+                    thisPtr.getPointer(), srcRecordTy);
+
+  // Whether we need an explicit null pointer check. For example, with the
+  // Microsoft ABI, if this is a call to __RTtypeid, the null pointer check and
+  // exception throw is inside the __RTtypeid(nullptr) call
+  if (hasNullCheck &&
+      CGF.CGM.getCXXABI().shouldTypeidBeNullChecked(srcRecordTy)) {
+    auto &builder = CGF.getBuilder();
+
+    // Create blocks for null and non-null paths
+    auto *currBlock = builder.getInsertionBlock();
+    auto *parentOp = currBlock->getParent()->getParentOp();
+    auto &region = parentOp->getRegion(0);
+
+    mlir::Block *badTypeidBlock = builder.createBlock(&region);
+    mlir::Block *endBlock = builder.createBlock(&region);
+
+    // Check if pointer is null
+    builder.setInsertionPointToEnd(currBlock);
+    mlir::Value nullPtr =
+        builder.getNullPtr(thisPtr.getPointer().getType(), loc);
+    mlir::Value isNull = builder.createCompare(loc, cir::CmpOpKind::eq,
+                                               thisPtr.getPointer(), nullPtr);
+
+    builder.create<cir::BrCondOp>(loc, isNull, badTypeidBlock, endBlock);
+
+    // Emit bad typeid path
+    builder.setInsertionPointToEnd(badTypeidBlock);
+    CGF.CGM.getCXXABI().emitBadTypeidCall(CGF);
+
+    // Continue on non-null path
+    builder.setInsertionPointToEnd(endBlock);
+  }
+
+  return CGF.CGM.getCXXABI().emitTypeid(CGF, loc, srcRecordTy, thisPtr,
+                                        stdTypeInfoPtrTy);
+}
+
 mlir::Value CIRGenFunction::emitCXXTypeidExpr(const CXXTypeidExpr *E) {
   auto loc = getLoc(E->getSourceRange());
 
@@ -1843,10 +1895,10 @@ mlir::Value CIRGenFunction::emitCXXTypeidExpr(const CXXTypeidExpr *E) {
   // If the operand is already the most derived object, no need to look up
   // vtable.
   if (E->isPotentiallyEvaluated() && !E->isMostDerived(getContext())) {
-    // This requires emitting code similar to dynamic_cast that looks up the
-    // type_info pointer from the vtable. Note that this path also needs to
-    // handle null checking when E->hasNullCheck() is true.
-    llvm_unreachable("NYI: typeid with polymorphic types (vtable lookup)");
+    // Polymorphic case: need runtime vtable lookup
+    mlir::Type typeInfoPtrTy = builder.getUInt8PtrTy();
+    return emitTypeidFromVTable(*this, loc, E->getExprOperand(), typeInfoPtrTy,
+                                E->hasNullCheck());
   }
 
   // For non-polymorphic types, just return the static RTTI descriptor.

clang/lib/CIR/CodeGen/CIRGenItaniumCXXABI.cpp

@@ -333,6 +333,12 @@ class CIRGenItaniumCXXABI : public CIRGenCXXABI {
 
   void emitBadCastCall(CIRGenFunction &CGF, mlir::Location loc) override;
 
+  void emitBadTypeidCall(CIRGenFunction &CGF) override;
+  bool shouldTypeidBeNullChecked(QualType SrcRecordTy) override;
+  mlir::Value emitTypeid(CIRGenFunction &CGF, mlir::Location loc,
+                         QualType SrcRecordTy, Address ThisPtr,
+                         mlir::Type StdTypeInfoPtrTy) override;
+
   mlir::Value
   getVirtualBaseClassOffset(mlir::Location loc, CIRGenFunction &CGF,
                             Address This, const CXXRecordDecl *ClassDecl,
@@ -2485,6 +2491,88 @@ void CIRGenItaniumCXXABI::emitBadCastCall(CIRGenFunction &CGF,
   emitCallToBadCast(CGF, loc);
 }
 
+void CIRGenItaniumCXXABI::emitBadTypeidCall(CIRGenFunction &CGF) {
+  auto loc = CGF.getLoc(SourceLocation());
+  // TODO: When exception support is complete, emit throw std::bad_typeid
+  // For now, emit unreachable since calling typeid on null is UB
+  cir::UnreachableOp::create(CGF.getBuilder(), loc);
+  CGF.getBuilder().clearInsertionPoint();
+}
+
+bool CIRGenItaniumCXXABI::shouldTypeidBeNullChecked(QualType srcRecordTy) {
+  // Match CodeGen behavior: always check for null.
+  // TODO(cir): The Itanium ABI 2.9.5p3 states only pointer operands need
+  // null checking (references cannot be null), but CodeGen conservatively
+  // always returns true. We match this behavior for consistency.
+  // Consider optimizing to: return srcRecordTy->isPointerType();
+  return true;
+}
+
+mlir::Value CIRGenItaniumCXXABI::emitTypeid(CIRGenFunction &CGF,
+                                            mlir::Location loc,
+                                            QualType srcRecordTy,
+                                            Address thisPtr,
+                                            mlir::Type stdTypeInfoPtrTy) {
+  auto &builder = CGF.getBuilder();
+  auto *classDecl = srcRecordTy->castAsCXXRecordDecl();
+
+  // Get the vtable pointer from the object
+  mlir::Value vTable = CGF.getVTablePtr(loc, thisPtr, classDecl);
+
+  mlir::Value typeInfoPtr;
+
+  if (CGM.getItaniumVTableContext().isRelativeLayout()) {
+    // Relative layout: type_info offset is at vptr[-4] (4 bytes before vptr)
+    // Load the offset and add it to the vtable pointer
+    auto int32Ty = builder.getSInt32Ty();
+    auto int8PtrTy = builder.getUInt8PtrTy();
+
+    // Cast vtable to i8* for byte arithmetic
+    auto vTableBytes = builder.createBitcast(loc, vTable, int8PtrTy);
+
+    // Get address of the offset: vtable - 4
+    auto offsetAddrPtr = cir::PtrStrideOp::create(
+        builder, loc, builder.getPointerTo(int32Ty), vTableBytes,
+        builder.getConstInt(loc, builder.getSInt64Ty(), -4));
+
+    // Load the 32-bit offset
+    auto offsetAddr =
+        Address(offsetAddrPtr, int32Ty, CharUnits::fromQuantity(4));
+    auto offset = builder.createLoad(loc, offsetAddr);
+
+    // Sign-extend offset to pointer width
+    auto offset64 = cir::CastOp::create(builder, loc, builder.getSInt64Ty(),
+                                        cir::CastKind::integral, offset);
+
+    // Add offset to vtable pointer: vtable + offset
+    auto typeInfoBytes = cir::PtrStrideOp::create(builder, loc, int8PtrTy,
+                                                  vTableBytes, offset64);
+
+    // Cast result to type_info pointer type
+    typeInfoPtr = builder.createBitcast(loc, typeInfoBytes, stdTypeInfoPtrTy);
+
+  } else {
+    // Absolute layout: type_info* is at vtable[-1]
+    // GEP to get address, then load
+
+    // Cast vtable pointer to a regular pointer type for ptr_stride
+    auto vTablePtr = builder.createBitcast(
+        loc, vTable, builder.getPointerTo(stdTypeInfoPtrTy));
+
+    // Get vtable[-1]
+    auto typeInfoAddrPtr = cir::PtrStrideOp::create(
+        builder, loc, builder.getPointerTo(stdTypeInfoPtrTy), vTablePtr,
+        builder.getConstInt(loc, builder.getSInt64Ty(), -1));
+
+    // Load the type_info pointer
+    auto typeInfoAddr =
+        Address(typeInfoAddrPtr, stdTypeInfoPtrTy, CGF.getPointerAlign());
+    typeInfoPtr = builder.createLoad(loc, typeInfoAddr);
+  }
+
+  return typeInfoPtr;
+}
+
 static CharUnits computeOffsetHint(ASTContext &astContext,
                                    const CXXRecordDecl *Src,
                                    const CXXRecordDecl *Dst) {

clang/test/CIR/CodeGen/typeid-polymorphic.cpp

@@ -0,0 +1,91 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-cir %s -o %t.cir
+// RUN: FileCheck --input-file=%t.cir %s --check-prefix=CIR
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ll
+// RUN: FileCheck --input-file=%t.ll %s --check-prefix=LLVM
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm %s -o %t.og.ll
+// RUN: FileCheck --input-file=%t.og.ll %s --check-prefix=OGCG
+
+namespace std {
+  class type_info {
+  public:
+    virtual ~type_info();
+  };
+}
+
+// Basic polymorphic class hierarchy
+struct Base {
+  virtual ~Base() {}
+  int x;
+};
+
+struct Derived : Base {
+  int y;
+};
+
+extern void use_typeinfo(const std::type_info*);
+
+// Test 1: Basic polymorphic typeid - pointer
+// CIR-LABEL: cir.func dso_local @_Z22test_polymorphic_basicP4Base
+void test_polymorphic_basic(Base* ptr) {
+  // CIR: cir.vtable.get_vptr
+  // CIR: cir.load{{.*}}!cir.vptr
+  // CIR: cir.cast bitcast
+  // CIR: cir.ptr_stride
+  // CIR: cir.load
+
+  // LLVM-LABEL: @_Z22test_polymorphic_basicP4Base
+  // LLVM: load ptr, ptr %
+  // LLVM: getelementptr ptr, ptr %{{.*}}, i64 -1
+  // LLVM: load ptr, ptr %
+
+  // OGCG-LABEL: @_Z22test_polymorphic_basicP4Base
+  // OGCG: load ptr, ptr %
+  // OGCG: getelementptr inbounds ptr, ptr %{{.*}}, i64 -1
+  // OGCG: load ptr, ptr %
+  use_typeinfo(&typeid(*ptr));
+}
+
+// Test 2: Polymorphic typeid - reference (no null check)
+// CIR-LABEL: cir.func dso_local @_Z14test_referenceR4Base
+void test_reference(Base& ref) {
+  // CIR-NOT: cir.cmp(eq
+  // CIR: cir.vtable.get_vptr
+  // CIR: cir.ptr_stride
+
+  // LLVM-LABEL: @_Z14test_referenceR4Base
+  // LLVM: load ptr, ptr %
+  // LLVM: getelementptr ptr, ptr %{{.*}}, i64 -1
+
+  // OGCG-LABEL: @_Z14test_referenceR4Base
+  // OGCG: load ptr, ptr %
+  // OGCG: getelementptr inbounds ptr, ptr %{{.*}}, i64 -1
+  use_typeinfo(&typeid(ref));
+}
+
+// Test 3: Derived class pointer
+// CIR-LABEL: cir.func dso_local @_Z20test_derived_pointerP7Derived
+void test_derived_pointer(Derived* ptr) {
+  // CIR: cir.vtable.get_vptr
+  // CIR: cir.ptr_stride
+
+  // LLVM-LABEL: @_Z20test_derived_pointerP7Derived
+  // LLVM: getelementptr ptr, ptr %{{.*}}, i64 -1
+
+  // OGCG-LABEL: @_Z20test_derived_pointerP7Derived
+  // OGCG: getelementptr inbounds ptr, ptr %{{.*}}, i64 -1
+  use_typeinfo(&typeid(*ptr));
+}
+
+// Test 4: Const qualified pointer
+// CIR-LABEL: cir.func dso_local @_Z14test_const_ptrPK4Base
+void test_const_ptr(const Base* ptr) {
+  // CIR: cir.vtable.get_vptr
+  // CIR: cir.ptr_stride
+
+  // LLVM-LABEL: @_Z14test_const_ptrPK4Base
+  // LLVM: getelementptr ptr, ptr %{{.*}}, i64 -1
+
+  // OGCG-LABEL: @_Z14test_const_ptrPK4Base
+  // OGCG: getelementptr inbounds ptr, ptr %{{.*}}, i64 -1
+  use_typeinfo(&typeid(*ptr));
+}