[CIR] Add support for delegating constructors (#143932)

This change adds the necessary support for handling delegating
constructors in ClangIR. The implementation is kept as small as possible
by not handling any other sort of initialization (members, base classes,
etc.). That will be added in a future commit.

Author: andykaylor

clang/include/clang/CIR/MissingFeatures.h

@@ -172,6 +172,7 @@ struct MissingFeatures {
   static bool astVarDeclInterface() { return false; }
   static bool stackSaveOp() { return false; }
   static bool aggValueSlot() { return false; }
+  static bool aggValueSlotMayOverlap() { return false; }
   static bool generateDebugInfo() { return false; }
   static bool pointerOverflowSanitizer() { return false; }
   static bool fpConstraints() { return false; }
@@ -227,7 +228,6 @@ struct MissingFeatures {
   static bool implicitConstructorArgs() { return false; }
   static bool intrinsics() { return false; }
   static bool attributeNoBuiltin() { return false; }
-  static bool emitCtorPrologue() { return false; }
   static bool thunks() { return false; }
   static bool runCleanupsScope() { return false; }
 

clang/lib/CIR/CodeGen/CIRGenCall.cpp

@@ -189,8 +189,7 @@ void CIRGenFunction::emitDelegateCallArg(CallArgList &args,
     // For the most part, we just need to load the alloca, except that aggregate
     // r-values are actually pointers to temporaries.
   } else {
-    cgm.errorNYI(param->getSourceRange(),
-                 "emitDelegateCallArg: convertTempToRValue");
+    args.add(convertTempToRValue(local, type, loc), type);
   }
 
   // Deactivate the cleanup for the callee-destructed param that was pushed.

clang/lib/CIR/CodeGen/CIRGenClass.cpp

@@ -53,6 +53,21 @@ bool CIRGenFunction::isConstructorDelegationValid(
   return true;
 }
 
+/// This routine generates necessary code to initialize base classes and
+/// non-static data members belonging to this constructor.
+void CIRGenFunction::emitCtorPrologue(const CXXConstructorDecl *cd,
+                                      CXXCtorType ctorType,
+                                      FunctionArgList &args) {
+  if (cd->isDelegatingConstructor())
+    return emitDelegatingCXXConstructorCall(cd, args);
+
+  if (cd->getNumCtorInitializers() != 0) {
+    // There's much more to do here.
+    cgm.errorNYI(cd->getSourceRange(), "emitCtorPrologue: any initializer");
+    return;
+  }
+}
+
 Address CIRGenFunction::loadCXXThisAddress() {
   assert(curFuncDecl && "loading 'this' without a func declaration?");
   assert(isa<CXXMethodDecl>(curFuncDecl));
@@ -102,6 +117,29 @@ void CIRGenFunction::emitDelegateCXXConstructorCall(
                          /*Delegating=*/true, thisAddr, delegateArgs, loc);
 }
 
+void CIRGenFunction::emitDelegatingCXXConstructorCall(
+    const CXXConstructorDecl *ctor, const FunctionArgList &args) {
+  assert(ctor->isDelegatingConstructor());
+
+  Address thisPtr = loadCXXThisAddress();
+
+  assert(!cir::MissingFeatures::objCGC());
+  assert(!cir::MissingFeatures::sanitizers());
+  AggValueSlot aggSlot = AggValueSlot::forAddr(
+      thisPtr, Qualifiers(), AggValueSlot::IsDestructed,
+      AggValueSlot::IsNotAliased, AggValueSlot::MayOverlap,
+      AggValueSlot::IsNotZeroed);
+
+  emitAggExpr(ctor->init_begin()[0]->getInit(), aggSlot);
+
+  const CXXRecordDecl *classDecl = ctor->getParent();
+  if (cgm.getLangOpts().Exceptions && !classDecl->hasTrivialDestructor()) {
+    cgm.errorNYI(ctor->getSourceRange(),
+                 "emitDelegatingCXXConstructorCall: exception");
+    return;
+  }
+}
+
 Address CIRGenFunction::getAddressOfBaseClass(
     Address value, const CXXRecordDecl *derived,
     llvm::iterator_range<CastExpr::path_const_iterator> path,

clang/lib/CIR/CodeGen/CIRGenDecl.cpp

@@ -259,7 +259,12 @@ void CIRGenFunction::emitExprAsInit(const Expr *init, const ValueDecl *d,
     return;
   }
   case cir::TEK_Aggregate:
-    emitAggExpr(init, AggValueSlot::forLValue(lvalue));
+    // The overlap flag here should be calculated.
+    assert(!cir::MissingFeatures::aggValueSlotMayOverlap());
+    emitAggExpr(init,
+                AggValueSlot::forLValue(lvalue, AggValueSlot::IsDestructed,
+                                        AggValueSlot::IsNotAliased,
+                                        AggValueSlot::MayOverlap));
     return;
   }
   llvm_unreachable("bad evaluation kind");

clang/lib/CIR/CodeGen/CIRGenExpr.cpp

@@ -1261,6 +1261,23 @@ Address CIRGenFunction::emitArrayToPointerDecay(const Expr *e) {
   return Address(ptr, addr.getAlignment());
 }
 
+/// Given the address of a temporary variable, produce an r-value of its type.
+RValue CIRGenFunction::convertTempToRValue(Address addr, clang::QualType type,
+                                           clang::SourceLocation loc) {
+  LValue lvalue = makeAddrLValue(addr, type, AlignmentSource::Decl);
+  switch (getEvaluationKind(type)) {
+  case cir::TEK_Complex:
+    cgm.errorNYI(loc, "convertTempToRValue: complex type");
+    return RValue::get(nullptr);
+  case cir::TEK_Aggregate:
+    cgm.errorNYI(loc, "convertTempToRValue: aggregate type");
+    return RValue::get(nullptr);
+  case cir::TEK_Scalar:
+    return RValue::get(emitLoadOfScalar(lvalue, loc));
+  }
+  llvm_unreachable("bad evaluation kind");
+}
+
 /// Emit an `if` on a boolean condition, filling `then` and `else` into
 /// appropriated regions.
 mlir::LogicalResult CIRGenFunction::emitIfOnBoolExpr(const Expr *cond,
@@ -1473,6 +1490,10 @@ void CIRGenFunction::emitCXXConstructExpr(const CXXConstructExpr *e,
     type = Ctor_Complete;
     break;
   case CXXConstructionKind::Delegating:
+    // We should be emitting a constructor; GlobalDecl will assert this
+    type = curGD.getCtorType();
+    delegating = true;
+    break;
   case CXXConstructionKind::VirtualBase:
   case CXXConstructionKind::NonVirtualBase:
     cgm.errorNYI(e->getSourceRange(),

clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp

@@ -203,7 +203,11 @@ void AggExprEmitter::emitInitializationToLValue(Expr *e, LValue lv) {
     cgf.cgm.errorNYI("emitInitializationToLValue TEK_Complex");
     break;
   case cir::TEK_Aggregate:
-    cgf.emitAggExpr(e, AggValueSlot::forLValue(lv));
+    cgf.emitAggExpr(e, AggValueSlot::forLValue(lv, AggValueSlot::IsDestructed,
+                                               AggValueSlot::IsNotAliased,
+                                               AggValueSlot::MayOverlap,
+                                               dest.isZeroed()));
+
     return;
   case cir::TEK_Scalar:
     if (lv.isSimple())
@@ -284,6 +288,8 @@ LValue CIRGenFunction::emitAggExprToLValue(const Expr *e) {
   assert(hasAggregateEvaluationKind(e->getType()) && "Invalid argument!");
   Address temp = createMemTemp(e->getType(), getLoc(e->getSourceRange()));
   LValue lv = makeAddrLValue(temp, e->getType());
-  emitAggExpr(e, AggValueSlot::forLValue(lv));
+  emitAggExpr(e, AggValueSlot::forLValue(lv, AggValueSlot::IsNotDestructed,
+                                         AggValueSlot::IsNotAliased,
+                                         AggValueSlot::DoesNotOverlap));
   return lv;
 }

clang/lib/CIR/CodeGen/CIRGenFunction.cpp

@@ -526,14 +526,8 @@ void CIRGenFunction::emitConstructorBody(FunctionArgList &args) {
   // TODO: in restricted cases, we can emit the vbase initializers of a
   // complete ctor and then delegate to the base ctor.
 
-  assert(!cir::MissingFeatures::emitCtorPrologue());
-  if (ctor->isDelegatingConstructor()) {
-    // This will be handled in emitCtorPrologue, but we should emit a diagnostic
-    // rather than silently fail to delegate.
-    cgm.errorNYI(ctor->getSourceRange(),
-                 "emitConstructorBody: delegating ctor");
-    return;
-  }
+  // Emit the constructor prologue, i.e. the base and member initializers.
+  emitCtorPrologue(ctor, ctorType, args);
 
   // TODO(cir): propagate this result via mlir::logical result. Just unreachable
   // now just to have it handled.

clang/lib/CIR/CodeGen/CIRGenFunction.h

@@ -474,6 +474,9 @@ class CIRGenFunction : public CIRGenTypeCache {
 
   bool shouldNullCheckClassCastValue(const CastExpr *ce);
 
+  RValue convertTempToRValue(Address addr, clang::QualType type,
+                             clang::SourceLocation loc);
+
   static bool
   isConstructorDelegationValid(const clang::CXXConstructorDecl *ctor);
 
@@ -797,6 +800,16 @@ class CIRGenFunction : public CIRGenTypeCache {
                                        const CXXMethodDecl *md,
                                        ReturnValueSlot returnValue);
 
+  void emitCtorPrologue(const clang::CXXConstructorDecl *ctor,
+                        clang::CXXCtorType ctorType, FunctionArgList &args);
+
+  // It's important not to confuse this and emitDelegateCXXConstructorCall.
+  // Delegating constructors are the C++11 feature. The constructor delegate
+  // optimization is used to reduce duplication in the base and complete
+  // constructors where they are substantially the same.
+  void emitDelegatingCXXConstructorCall(const CXXConstructorDecl *ctor,
+                                        const FunctionArgList &args);
+
   mlir::LogicalResult emitDoStmt(const clang::DoStmt &s);
 
   /// Emit an expression as an initializer for an object (variable, field, etc.)

clang/lib/CIR/CodeGen/CIRGenValue.h

@@ -267,23 +267,64 @@ class AggValueSlot {
   Address addr;
   clang::Qualifiers quals;
 
+  /// This is set to true if some external code is responsible for setting up a
+  /// destructor for the slot.  Otherwise the code which constructs it should
+  /// push the appropriate cleanup.
+  LLVM_PREFERRED_TYPE(bool)
+  [[maybe_unused]] unsigned destructedFlag : 1;
+
   /// This is set to true if the memory in the slot is known to be zero before
   /// the assignment into it.  This means that zero fields don't need to be set.
-  bool zeroedFlag : 1;
+  LLVM_PREFERRED_TYPE(bool)
+  unsigned zeroedFlag : 1;
+
+  /// This is set to true if the slot might be aliased and it's not undefined
+  /// behavior to access it through such an alias.  Note that it's always
+  /// undefined behavior to access a C++ object that's under construction
+  /// through an alias derived from outside the construction process.
+  ///
+  /// This flag controls whether calls that produce the aggregate
+  /// value may be evaluated directly into the slot, or whether they
+  /// must be evaluated into an unaliased temporary and then memcpy'ed
+  /// over.  Since it's invalid in general to memcpy a non-POD C++
+  /// object, it's important that this flag never be set when
+  /// evaluating an expression which constructs such an object.
+  LLVM_PREFERRED_TYPE(bool)
+  [[maybe_unused]] unsigned aliasedFlag : 1;
+
+  /// This is set to true if the tail padding of this slot might overlap
+  /// another object that may have already been initialized (and whose
+  /// value must be preserved by this initialization). If so, we may only
+  /// store up to the dsize of the type. Otherwise we can widen stores to
+  /// the size of the type.
+  LLVM_PREFERRED_TYPE(bool)
+  [[maybe_unused]] unsigned overlapFlag : 1;
 
 public:
+  enum IsDestructed_t { IsNotDestructed, IsDestructed };
   enum IsZeroed_t { IsNotZeroed, IsZeroed };
+  enum IsAliased_t { IsNotAliased, IsAliased };
+  enum Overlap_t { MayOverlap, DoesNotOverlap };
 
-  AggValueSlot(Address addr, clang::Qualifiers quals, bool zeroedFlag)
-      : addr(addr), quals(quals), zeroedFlag(zeroedFlag) {}
+  AggValueSlot(Address addr, clang::Qualifiers quals, bool destructedFlag,
+               bool zeroedFlag, bool aliasedFlag, bool overlapFlag)
+      : addr(addr), quals(quals), destructedFlag(destructedFlag),
+        zeroedFlag(zeroedFlag), aliasedFlag(aliasedFlag),
+        overlapFlag(overlapFlag) {}
 
   static AggValueSlot forAddr(Address addr, clang::Qualifiers quals,
+                              IsDestructed_t isDestructed,
+                              IsAliased_t isAliased, Overlap_t mayOverlap,
                               IsZeroed_t isZeroed = IsNotZeroed) {
-    return AggValueSlot(addr, quals, isZeroed);
+    return AggValueSlot(addr, quals, isDestructed, isZeroed, isAliased,
+                        mayOverlap);
   }
 
-  static AggValueSlot forLValue(const LValue &lv) {
-    return forAddr(lv.getAddress(), lv.getQuals());
+  static AggValueSlot forLValue(const LValue &LV, IsDestructed_t isDestructed,
+                                IsAliased_t isAliased, Overlap_t mayOverlap,
+                                IsZeroed_t isZeroed = IsNotZeroed) {
+    return forAddr(LV.getAddress(), LV.getQuals(), isDestructed, isAliased,
+                   mayOverlap, isZeroed);
   }
 
   clang::Qualifiers getQualifiers() const { return quals; }

clang/test/CIR/CodeGen/ctor.cpp

@@ -67,3 +67,49 @@ void bar() {
 // CHECK-NEXT:    %[[THREE:.*]] = cir.const #cir.int<3> : !s32i
 // CHECK-NEXT:    cir.call @_ZN13VariadicStrukC1Eiz(%[[S_ADDR]], %[[ONE]], %[[TWO]], %[[THREE]])
 // CHECK-NEXT:    cir.return
+
+struct DelegatingStruk {
+  int a;
+  DelegatingStruk(int n) { a = n; }
+  DelegatingStruk() : DelegatingStruk(0) {}
+};
+
+void bam() {
+  DelegatingStruk s;
+}
+
+// CHECK:       cir.func @_ZN15DelegatingStrukC2Ei(%arg0: !cir.ptr<!rec_DelegatingStruk>
+// CHECK-SAME:                                     %arg1: !s32i
+// CHECK-NEXT:   %[[THIS_ADDR:.*]] = cir.alloca {{.*}} ["this", init]
+// CHECK-NEXT:   %[[N_ADDR:.*]] = cir.alloca {{.*}} ["n", init]
+// CHECK-NEXT:   cir.store %arg0, %[[THIS_ADDR]]
+// CHECK-NEXT:   cir.store %arg1, %[[N_ADDR]]
+// CHECK-NEXT:   %[[THIS:.*]] = cir.load{{.*}} %[[THIS_ADDR]]
+// CHECK-NEXT:   %[[N:.*]] = cir.load{{.*}} %[[N_ADDR]]
+// CHECK-NEXT:   %[[A_ADDR:.*]] = cir.get_member %[[THIS]][0] {name = "a"}
+// CHECK-NEXT:   cir.store{{.*}} %[[N]], %[[A_ADDR]]
+// CHECK-NEXT:   cir.return
+
+// CHECK:       cir.func @_ZN15DelegatingStrukC1Ei(%arg0: !cir.ptr<!rec_DelegatingStruk>
+// CHECK-SAME:                                     %arg1: !s32i
+// CHECK-NEXT:   %[[THIS_ADDR:.*]] = cir.alloca {{.*}} ["this", init]
+// CHECK-NEXT:   %[[N_ADDR:.*]] = cir.alloca {{.*}} ["n", init]
+// CHECK-NEXT:   cir.store %arg0, %[[THIS_ADDR]]
+// CHECK-NEXT:   cir.store %arg1, %[[N_ADDR]]
+// CHECK-NEXT:   %[[THIS:.*]] = cir.load{{.*}} %[[THIS_ADDR]]
+// CHECK-NEXT:   %[[N:.*]] = cir.load{{.*}} %[[N_ADDR]]
+// CHECK-NEXT:   cir.call @_ZN15DelegatingStrukC2Ei(%[[THIS]], %[[N]])
+// CHECK-NEXT:   cir.return
+
+// CHECK: cir.func @_ZN15DelegatingStrukC1Ev(%arg0: !cir.ptr<!rec_DelegatingStruk>
+// CHECK-NEXT:   %[[THIS_ADDR:.*]] = cir.alloca {{.*}} ["this", init]
+// CHECK-NEXT:   cir.store %arg0, %[[THIS_ADDR]]
+// CHECK-NEXT:   %[[THIS:.*]] = cir.load{{.*}} %[[THIS_ADDR]]
+// CHECK-NEXT:   %[[ZERO:.*]] = cir.const #cir.int<0> : !s32i
+// CHECK-NEXT:   cir.call @_ZN15DelegatingStrukC1Ei(%[[THIS]], %[[ZERO]])
+// CHECK-NEXT:   cir.return
+
+// CHECK: cir.func @_Z3bamv
+// CHECK-NEXT:    %[[S_ADDR:.*]] = cir.alloca {{.*}} ["s", init]
+// CHECK-NEXT:    cir.call @_ZN15DelegatingStrukC1Ev(%[[S_ADDR]])
+// CHECK-NEXT:    cir.return