[ObjC] Allow declaring __weak pointer fields in C structs in ARC.

This patch uses the infrastructure added in r326307 for enabling
non-trivial fields to be declared in C structs to allow __weak fields in
C structs in ARC.

This recommits r327206, which was reverted because it caused
module-enabled builders to fail. I discovered that the
CXXRecordDecl::CanPassInRegisters flag wasn't being set correctly in
some cases after I moved it to RecordDecl.

Thanks to Eric Liu for helping me investigate the bug.

rdar://problem/33599681

https://reviews.llvm.org/D44095

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@327870 91177308-0d34-0410-b5e6-96231b3b80d8

22 files changed, 421 insertions, 51 deletions

diff --git a/include/clang/AST/Decl.h b/include/clang/AST/Decl.h
index e14187013a..41b47eacee 100644
--- a/include/clang/AST/Decl.h
+++ b/include/clang/AST/Decl.h
@@ -3553,6 +3553,12 @@ class RecordDecl : public TagDecl {
   bool NonTrivialToPrimitiveCopy : 1;
   bool NonTrivialToPrimitiveDestroy : 1;
 
+  /// True if this class can be passed in a non-address-preserving fashion
+  /// (such as in registers).
+  /// This does not imply anything about how the ABI in use will actually
+  /// pass an object of this class.
+  bool CanPassInRegisters : 1;
+
 protected:
   RecordDecl(Kind DK, TagKind TK, const ASTContext &C, DeclContext *DC,
              SourceLocation StartLoc, SourceLocation IdLoc,
@@ -3636,6 +3642,18 @@ public:
     NonTrivialToPrimitiveDestroy = V;
   }
 
+  /// Determine whether this class can be passed in registers. In C++ mode,
+  /// it must have at least one trivial, non-deleted copy or move constructor.
+  /// FIXME: This should be set as part of completeDefinition.
+  bool canPassInRegisters() const {
+    return CanPassInRegisters;
+  }
+
+  /// Set that we can pass this RecordDecl in registers.
+  void setCanPassInRegisters(bool CanPass) {
+    CanPassInRegisters = CanPass;
+  }
+
   /// \brief Determines whether this declaration represents the
   /// injected class name.
   ///
diff --git a/include/clang/AST/DeclCXX.h b/include/clang/AST/DeclCXX.h
index faa86e3781..e9f6f9ef6e 100644
--- a/include/clang/AST/DeclCXX.h
+++ b/include/clang/AST/DeclCXX.h
@@ -467,12 +467,6 @@ class CXXRecordDecl : public RecordDecl {
     /// constructor.
     unsigned HasDefaultedDefaultConstructor : 1;
 
-    /// \brief True if this class can be passed in a non-address-preserving
-    /// fashion (such as in registers) according to the C++ language rules.
-    /// This does not imply anything about how the ABI in use will actually
-    /// pass an object of this class.
-    unsigned CanPassInRegisters : 1;
-
     /// \brief True if a defaulted default constructor for this class would
     /// be constexpr.
     unsigned DefaultedDefaultConstructorIsConstexpr : 1;
@@ -1474,18 +1468,6 @@ public:
     return data().HasIrrelevantDestructor;
   }
 
-  /// \brief Determine whether this class has at least one trivial, non-deleted
-  /// copy or move constructor.
-  bool canPassInRegisters() const {
-    return data().CanPassInRegisters;
-  }
-
-  /// \brief Set that we can pass this RecordDecl in registers.
-  // FIXME: This should be set as part of completeDefinition.
-  void setCanPassInRegisters(bool CanPass) {
-    data().CanPassInRegisters = CanPass;
-  }
-
   /// Determine whether the triviality for the purpose of calls for this class
   /// is overridden to be trivial because this class or the type of one of its
   /// subobjects has attribute "trivial_abi".
diff --git a/include/clang/AST/Type.h b/include/clang/AST/Type.h
index 0a5c086f38..bb8944aa6e 100644
--- a/include/clang/AST/Type.h
+++ b/include/clang/AST/Type.h
@@ -1097,6 +1097,10 @@ public:
     /// with the ARC __strong qualifier.
     PDIK_ARCStrong,
 
+    /// The type is an Objective-C retainable pointer type that is qualified
+    /// with the ARC __weak qualifier.
+    PDIK_ARCWeak,
+
     /// The type is a struct containing a field whose type is not PCK_Trivial.
     PDIK_Struct
   };
@@ -1124,6 +1128,10 @@ public:
     /// with the ARC __strong qualifier.
     PCK_ARCStrong,
 
+    /// The type is an Objective-C retainable pointer type that is qualified
+    /// with the ARC __weak qualifier.
+    PCK_ARCWeak,
+
     /// The type is a struct containing a field whose type is neither
     /// PCK_Trivial nor PCK_VolatileTrivial.
     /// Note that a C++ struct type does not necessarily match this; C++ copying
@@ -1146,6 +1154,8 @@ public:
   /// source object is placed in an uninitialized state.
   PrimitiveCopyKind isNonTrivialToPrimitiveDestructiveMove() const;
 
+  bool canPassInRegisters() const;
+
   enum DestructionKind {
     DK_none,
     DK_cxx_destructor,
diff --git a/lib/AST/ASTImporter.cpp b/lib/AST/ASTImporter.cpp
index ac262e7062..6b9260d2d8 100644
--- a/lib/AST/ASTImporter.cpp
+++ b/lib/AST/ASTImporter.cpp
@@ -1105,7 +1105,6 @@ bool ASTNodeImporter::ImportDefinition(RecordDecl *From, RecordDecl *To,
       = FromData.HasConstexprNonCopyMoveConstructor;
     ToData.HasDefaultedDefaultConstructor
       = FromData.HasDefaultedDefaultConstructor;
-    ToData.CanPassInRegisters = FromData.CanPassInRegisters;
     ToData.DefaultedDefaultConstructorIsConstexpr
       = FromData.DefaultedDefaultConstructorIsConstexpr;
     ToData.HasConstexprDefaultConstructor
diff --git a/lib/AST/Decl.cpp b/lib/AST/Decl.cpp
index b640dccbd8..35364d6abc 100644
--- a/lib/AST/Decl.cpp
+++ b/lib/AST/Decl.cpp
@@ -3931,7 +3931,8 @@ RecordDecl::RecordDecl(Kind DK, TagKind TK, const ASTContext &C,
       HasObjectMember(false), HasVolatileMember(false),
       LoadedFieldsFromExternalStorage(false),
       NonTrivialToPrimitiveDefaultInitialize(false),
-      NonTrivialToPrimitiveCopy(false), NonTrivialToPrimitiveDestroy(false) {
+      NonTrivialToPrimitiveCopy(false), NonTrivialToPrimitiveDestroy(false),
+      CanPassInRegisters(true) {
   assert(classof(static_cast<Decl*>(this)) && "Invalid Kind!");
 }
 
diff --git a/lib/AST/DeclCXX.cpp b/lib/AST/DeclCXX.cpp
index 487386fdb5..271872352c 100644
--- a/lib/AST/DeclCXX.cpp
+++ b/lib/AST/DeclCXX.cpp
@@ -93,7 +93,6 @@ CXXRecordDecl::DefinitionData::DefinitionData(CXXRecordDecl *D)
       DeclaredNonTrivialSpecialMembersForCall(0), HasIrrelevantDestructor(true),
       HasConstexprNonCopyMoveConstructor(false),
       HasDefaultedDefaultConstructor(false),
-      CanPassInRegisters(true),
       DefaultedDefaultConstructorIsConstexpr(true),
       HasConstexprDefaultConstructor(false),
       HasNonLiteralTypeFieldsOrBases(false), ComputedVisibleConversions(false),
diff --git a/lib/AST/Type.cpp b/lib/AST/Type.cpp
index 99f5b1384a..2985aac66b 100644
--- a/lib/AST/Type.cpp
+++ b/lib/AST/Type.cpp
@@ -2214,11 +2214,14 @@ QualType::isNonTrivialToPrimitiveDefaultInitialize() const {
     if (RT->getDecl()->isNonTrivialToPrimitiveDefaultInitialize())
       return PDIK_Struct;
 
-  Qualifiers::ObjCLifetime Lifetime = getQualifiers().getObjCLifetime();
-  if (Lifetime == Qualifiers::OCL_Strong)
+  switch (getQualifiers().getObjCLifetime()) {
+  case Qualifiers::OCL_Strong:
     return PDIK_ARCStrong;
-
-  return PDIK_Trivial;
+  case Qualifiers::OCL_Weak:
+    return PDIK_ARCWeak;
+  default:
+    return PDIK_Trivial;
+  }
 }
 
 QualType::PrimitiveCopyKind QualType::isNonTrivialToPrimitiveCopy() const {
@@ -2228,10 +2231,14 @@ QualType::PrimitiveCopyKind QualType::isNonTrivialToPrimitiveCopy() const {
       return PCK_Struct;
 
   Qualifiers Qs = getQualifiers();
-  if (Qs.getObjCLifetime() == Qualifiers::OCL_Strong)
+  switch (Qs.getObjCLifetime()) {
+  case Qualifiers::OCL_Strong:
     return PCK_ARCStrong;
-
-  return Qs.hasVolatile() ? PCK_VolatileTrivial : PCK_Trivial;
+  case Qualifiers::OCL_Weak:
+    return PCK_ARCWeak;
+  default:
+    return Qs.hasVolatile() ? PCK_VolatileTrivial : PCK_Trivial;
+  }
 }
 
 QualType::PrimitiveCopyKind
@@ -2239,6 +2246,17 @@ QualType::isNonTrivialToPrimitiveDestructiveMove() const {
   return isNonTrivialToPrimitiveCopy();
 }
 
+bool QualType::canPassInRegisters() const {
+  if (const auto *RT =
+          getTypePtr()->getBaseElementTypeUnsafe()->getAs<RecordType>())
+    return RT->getDecl()->canPassInRegisters();
+
+  if (getQualifiers().getObjCLifetime() == Qualifiers::OCL_Weak)
+    return false;
+
+  return true;
+}
+
 bool Type::isLiteralType(const ASTContext &Ctx) const {
   if (isDependentType())
     return false;
diff --git a/lib/CodeGen/CGBlocks.cpp b/lib/CodeGen/CGBlocks.cpp
index 663f1d03fd..1a6904976f 100644
--- a/lib/CodeGen/CGBlocks.cpp
+++ b/lib/CodeGen/CGBlocks.cpp
@@ -1525,6 +1525,9 @@ computeCopyInfoForBlockCapture(const BlockDecl::Capture &CI, QualType T,
   case QualType::PCK_Struct:
     return std::make_pair(BlockCaptureEntityKind::NonTrivialCStruct,
                           BlockFieldFlags());
+  case QualType::PCK_ARCWeak:
+    // We need to register __weak direct captures with the runtime.
+    return std::make_pair(BlockCaptureEntityKind::ARCWeak, Flags);
   case QualType::PCK_ARCStrong:
     // We need to retain the copied value for __strong direct captures.
     // If it's a block pointer, we have to copy the block and assign that to
@@ -1542,10 +1545,6 @@ computeCopyInfoForBlockCapture(const BlockDecl::Capture &CI, QualType T,
     // Special rules for ARC captures:
     Qualifiers QS = T.getQualifiers();
 
-    // We need to register __weak direct captures with the runtime.
-    if (QS.getObjCLifetime() == Qualifiers::OCL_Weak)
-      return std::make_pair(BlockCaptureEntityKind::ARCWeak, Flags);
-
     // Non-ARC captures of retainable pointers are strong and
     // therefore require a call to _Block_object_assign.
     if (!QS.getObjCLifetime() && !LangOpts.ObjCAutoRefCount)
diff --git a/lib/CodeGen/CGNonTrivialStruct.cpp b/lib/CodeGen/CGNonTrivialStruct.cpp
index 3b0b173198..731938af86 100644
--- a/lib/CodeGen/CGNonTrivialStruct.cpp
+++ b/lib/CodeGen/CGNonTrivialStruct.cpp
@@ -77,6 +77,8 @@ struct DefaultInitializedTypeVisitor {
     switch (PDIK) {
     case QualType::PDIK_ARCStrong:
       return asDerived().visitARCStrong(FT, std::forward<Ts>(Args)...);
+    case QualType::PDIK_ARCWeak:
+      return asDerived().visitARCWeak(FT, std::forward<Ts>(Args)...);
     case QualType::PDIK_Struct:
       return asDerived().visitStruct(FT, std::forward<Ts>(Args)...);
     case QualType::PDIK_Trivial:
@@ -108,6 +110,8 @@ struct CopiedTypeVisitor {
     switch (PCK) {
     case QualType::PCK_ARCStrong:
       return asDerived().visitARCStrong(FT, std::forward<Ts>(Args)...);
+    case QualType::PCK_ARCWeak:
+      return asDerived().visitARCWeak(FT, std::forward<Ts>(Args)...);
     case QualType::PCK_Struct:
       return asDerived().visitStruct(FT, std::forward<Ts>(Args)...);
     case QualType::PCK_Trivial:
@@ -141,11 +145,6 @@ template <class Derived> struct StructVisitor {
 
   template <class... Ts> void visitTrivial(Ts... Args) {}
 
-  template <class... Ts> void visitARCWeak(Ts... Args) {
-    // FIXME: remove this when visitARCWeak is implemented in the subclasses.
-    llvm_unreachable("weak field is not expected");
-  }
-
   template <class... Ts> void visitCXXDestructor(Ts... Args) {
     llvm_unreachable("field of a C++ struct type is not expected");
   }
@@ -245,6 +244,13 @@ template <class Derived> struct GenFuncNameBase {
     appendStr(getVolatileOffsetStr(FT.isVolatileQualified(), FieldOffset));
   }
 
+  void visitARCWeak(QualType FT, const FieldDecl *FD,
+                    CharUnits CurStructOffset) {
+    appendStr("_w");
+    CharUnits FieldOffset = CurStructOffset + asDerived().getFieldOffset(FD);
+    appendStr(getVolatileOffsetStr(FT.isVolatileQualified(), FieldOffset));
+  }
+
   void visitStruct(QualType QT, const FieldDecl *FD,
                    CharUnits CurStructOffset) {
     CharUnits FieldOffset = CurStructOffset + asDerived().getFieldOffset(FD);
@@ -615,6 +621,12 @@ struct GenDestructor : StructVisitor<GenDestructor>,
         *CGF, getAddrWithOffset(Addrs[DstIdx], CurStackOffset, FD), QT);
   }
 
+  void visitARCWeak(QualType QT, const FieldDecl *FD, CharUnits CurStackOffset,
+                    std::array<Address, 1> Addrs) {
+    CGF->destroyARCWeak(
+        *CGF, getAddrWithOffset(Addrs[DstIdx], CurStackOffset, FD), QT);
+  }
+
   void callSpecialFunction(QualType FT, CharUnits Offset,
                            std::array<Address, 1> Addrs) {
     CGF->callCStructDestructor(
@@ -636,6 +648,12 @@ struct GenDefaultInitialize
         getAddrWithOffset(Addrs[DstIdx], CurStackOffset, FD), QT);
   }
 
+  void visitARCWeak(QualType QT, const FieldDecl *FD, CharUnits CurStackOffset,
+                    std::array<Address, 1> Addrs) {
+    CGF->EmitNullInitialization(
+        getAddrWithOffset(Addrs[DstIdx], CurStackOffset, FD), QT);
+  }
+
   template <class FieldKind, size_t... Is>
   void visitArray(FieldKind FK, QualType QT, const FieldDecl *FD,
                   CharUnits CurStackOffset, std::array<Address, 1> Addrs) {
@@ -678,6 +696,14 @@ struct GenCopyConstructor : GenBinaryFunc<GenCopyConstructor, false> {
     llvm::Value *Val = CGF->EmitARCRetain(QT, SrcVal);
     CGF->EmitStoreOfScalar(Val, CGF->MakeAddrLValue(Addrs[DstIdx], QT), true);
   }
+
+  void visitARCWeak(QualType QT, const FieldDecl *FD, CharUnits CurStackOffset,
+                    std::array<Address, 2> Addrs) {
+    Addrs[DstIdx] = getAddrWithOffset(Addrs[DstIdx], CurStackOffset, FD);
+    Addrs[SrcIdx] = getAddrWithOffset(Addrs[SrcIdx], CurStackOffset, FD);
+    CGF->EmitARCCopyWeak(Addrs[DstIdx], Addrs[SrcIdx]);
+  }
+
   void callSpecialFunction(QualType FT, CharUnits Offset,
                            std::array<Address, 2> Addrs) {
     CGF->callCStructCopyConstructor(CGF->MakeAddrLValue(Addrs[DstIdx], FT),
@@ -700,6 +726,14 @@ struct GenMoveConstructor : GenBinaryFunc<GenMoveConstructor, true> {
     CGF->EmitStoreOfScalar(SrcVal, CGF->MakeAddrLValue(Addrs[DstIdx], QT),
                            /* isInitialization */ true);
   }
+
+  void visitARCWeak(QualType QT, const FieldDecl *FD, CharUnits CurStackOffset,
+                    std::array<Address, 2> Addrs) {
+    Addrs[DstIdx] = getAddrWithOffset(Addrs[DstIdx], CurStackOffset, FD);
+    Addrs[SrcIdx] = getAddrWithOffset(Addrs[SrcIdx], CurStackOffset, FD);
+    CGF->EmitARCMoveWeak(Addrs[DstIdx], Addrs[SrcIdx]);
+  }
+
   void callSpecialFunction(QualType FT, CharUnits Offset,
                            std::array<Address, 2> Addrs) {
     CGF->callCStructMoveConstructor(CGF->MakeAddrLValue(Addrs[DstIdx], FT),
@@ -720,6 +754,14 @@ struct GenCopyAssignment : GenBinaryFunc<GenCopyAssignment, false> {
     CGF->EmitARCStoreStrong(CGF->MakeAddrLValue(Addrs[DstIdx], QT), SrcVal,
                             false);
   }
+
+  void visitARCWeak(QualType QT, const FieldDecl *FD, CharUnits CurStackOffset,
+                    std::array<Address, 2> Addrs) {
+    Addrs[DstIdx] = getAddrWithOffset(Addrs[DstIdx], CurStackOffset, FD);
+    Addrs[SrcIdx] = getAddrWithOffset(Addrs[SrcIdx], CurStackOffset, FD);
+    CGF->emitARCCopyAssignWeak(QT, Addrs[DstIdx], Addrs[SrcIdx]);
+  }
+
   void callSpecialFunction(QualType FT, CharUnits Offset,
                            std::array<Address, 2> Addrs) {
     CGF->callCStructCopyAssignmentOperator(
@@ -747,6 +789,13 @@ struct GenMoveAssignment : GenBinaryFunc<GenMoveAssignment, true> {
     CGF->EmitARCRelease(DstVal, ARCImpreciseLifetime);
   }
 
+  void visitARCWeak(QualType QT, const FieldDecl *FD, CharUnits CurStackOffset,
+                    std::array<Address, 2> Addrs) {
+    Addrs[DstIdx] = getAddrWithOffset(Addrs[DstIdx], CurStackOffset, FD);
+    Addrs[SrcIdx] = getAddrWithOffset(Addrs[SrcIdx], CurStackOffset, FD);
+    CGF->emitARCMoveAssignWeak(QT, Addrs[DstIdx], Addrs[SrcIdx]);
+  }
+
   void callSpecialFunction(QualType FT, CharUnits Offset,
                            std::array<Address, 2> Addrs) {
     CGF->callCStructMoveAssignmentOperator(
diff --git a/lib/CodeGen/CGObjC.cpp b/lib/CodeGen/CGObjC.cpp
index e812ef399e..6ece1d3bd7 100644
--- a/lib/CodeGen/CGObjC.cpp
+++ b/lib/CodeGen/CGObjC.cpp
@@ -2307,6 +2307,21 @@ void CodeGenFunction::EmitARCCopyWeak(Address dst, Address src) {
                        "objc_copyWeak");
 }
 
+void CodeGenFunction::emitARCCopyAssignWeak(QualType Ty, Address DstAddr,
+                                            Address SrcAddr) {
+  llvm::Value *Object = EmitARCLoadWeakRetained(SrcAddr);
+  Object = EmitObjCConsumeObject(Ty, Object);
+  EmitARCStoreWeak(DstAddr, Object, false);
+}
+
+void CodeGenFunction::emitARCMoveAssignWeak(QualType Ty, Address DstAddr,
+                                            Address SrcAddr) {
+  llvm::Value *Object = EmitARCLoadWeakRetained(SrcAddr);
+  Object = EmitObjCConsumeObject(Ty, Object);
+  EmitARCStoreWeak(DstAddr, Object, false);
+  EmitARCDestroyWeak(SrcAddr);
+}
+
 /// Produce the code to do a objc_autoreleasepool_push.
 ///   call i8* \@objc_autoreleasePoolPush(void)
 llvm::Value *CodeGenFunction::EmitObjCAutoreleasePoolPush() {
diff --git a/lib/CodeGen/CodeGenFunction.h b/lib/CodeGen/CodeGenFunction.h
index fa65b1d699..5852788e29 100644
--- a/lib/CodeGen/CodeGenFunction.h
+++ b/lib/CodeGen/CodeGenFunction.h
@@ -3551,6 +3551,8 @@ public:
   llvm::Value *EmitARCLoadWeak(Address addr);
   llvm::Value *EmitARCLoadWeakRetained(Address addr);
   llvm::Value *EmitARCStoreWeak(Address addr, llvm::Value *value, bool ignored);
+  void emitARCCopyAssignWeak(QualType Ty, Address DstAddr, Address SrcAddr);
+  void emitARCMoveAssignWeak(QualType Ty, Address DstAddr, Address SrcAddr);
   void EmitARCCopyWeak(Address dst, Address src);
   void EmitARCMoveWeak(Address dst, Address src);
   llvm::Value *EmitARCRetainAutorelease(QualType type, llvm::Value *value);
diff --git a/lib/CodeGen/TargetInfo.cpp b/lib/CodeGen/TargetInfo.cpp
index aadbd3016b..939679e794 100644
--- a/lib/CodeGen/TargetInfo.cpp
+++ b/lib/CodeGen/TargetInfo.cpp
@@ -140,8 +140,11 @@ bool SwiftABIInfo::isLegalVectorTypeForSwift(CharUnits vectorSize,
 static CGCXXABI::RecordArgABI getRecordArgABI(const RecordType *RT,
                                               CGCXXABI &CXXABI) {
   const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl());
-  if (!RD)
+  if (!RD) {
+    if (!RT->getDecl()->canPassInRegisters())
+      return CGCXXABI::RAA_Indirect;
     return CGCXXABI::RAA_Default;
+  }
   return CXXABI.getRecordArgABI(RD);
 }
 
@@ -153,6 +156,20 @@ static CGCXXABI::RecordArgABI getRecordArgABI(QualType T,
   return getRecordArgABI(RT, CXXABI);
 }
 
+static bool classifyReturnType(const CGCXXABI &CXXABI, CGFunctionInfo &FI,
+                               const ABIInfo &Info) {
+  QualType Ty = FI.getReturnType();
+
+  if (const auto *RT = Ty->getAs<RecordType>())
+    if (!isa<CXXRecordDecl>(RT->getDecl()) &&
+        !RT->getDecl()->canPassInRegisters()) {
+      FI.getReturnInfo() = Info.getNaturalAlignIndirect(Ty);
+      return true;
+    }
+
+  return CXXABI.classifyReturnType(FI);
+}
+
 /// Pass transparent unions as if they were the type of the first element. Sema
 /// should ensure that all elements of the union have the same "machine type".
 static QualType useFirstFieldIfTransparentUnion(QualType Ty) {
@@ -1749,7 +1766,7 @@ void X86_32ABIInfo::computeInfo(CGFunctionInfo &FI) const {
   } else
     State.FreeRegs = DefaultNumRegisterParameters;
 
-  if (!getCXXABI().classifyReturnType(FI)) {
+  if (!::classifyReturnType(getCXXABI(), FI, *this)) {
     FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), State);
   } else if (FI.getReturnInfo().isIndirect()) {
     // The C++ ABI is not aware of register usage, so we have to check if the
@@ -3545,7 +3562,7 @@ void X86_64ABIInfo::computeInfo(CGFunctionInfo &FI) const {
   unsigned FreeSSERegs = IsRegCall ? 16 : 8;
   unsigned NeededInt, NeededSSE;
 
-  if (!getCXXABI().classifyReturnType(FI)) {
+  if (!::classifyReturnType(getCXXABI(), FI, *this)) {
     if (IsRegCall && FI.getReturnType()->getTypePtr()->isRecordType() &&
         !FI.getReturnType()->getTypePtr()->isUnionType()) {
       FI.getReturnInfo() =
@@ -4895,7 +4912,7 @@ private:
   bool isIllegalVectorType(QualType Ty) const;
 
   void computeInfo(CGFunctionInfo &FI) const override {
-    if (!getCXXABI().classifyReturnType(FI))
+    if (!::classifyReturnType(getCXXABI(), FI, *this))
       FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
 
     for (auto &it : FI.arguments())
@@ -5626,7 +5643,7 @@ void WindowsARMTargetCodeGenInfo::setTargetAttributes(
 }
 
 void ARMABIInfo::computeInfo(CGFunctionInfo &FI) const {
-  if (!getCXXABI().classifyReturnType(FI))
+  if (!::classifyReturnType(getCXXABI(), FI, *this))
     FI.getReturnInfo() =
         classifyReturnType(FI.getReturnType(), FI.isVariadic());
 
diff --git a/lib/Sema/SemaDecl.cpp b/lib/Sema/SemaDecl.cpp
index ea97135303..6319e4b34c 100644
--- a/lib/Sema/SemaDecl.cpp
+++ b/lib/Sema/SemaDecl.cpp
@@ -15266,7 +15266,6 @@ void Sema::ActOnFields(Scope *S, SourceLocation RecLoc, Decl *EnclosingDecl,
 
     // Get the type for the field.
     const Type *FDTy = FD->getType().getTypePtr();
-    Qualifiers QS = FD->getType().getQualifiers();
 
     if (!FD->isAnonymousStructOrUnion()) {
       // Remember all fields written by the user.
@@ -15407,10 +15406,7 @@ void Sema::ActOnFields(Scope *S, SourceLocation RecLoc, Decl *EnclosingDecl,
       QualType T = Context.getObjCObjectPointerType(FD->getType());
       FD->setType(T);
     } else if (getLangOpts().allowsNonTrivialObjCLifetimeQualifiers() &&
-               Record && !ObjCFieldLifetimeErrReported &&
-               ((!getLangOpts().CPlusPlus &&
-                 QS.getObjCLifetime() == Qualifiers::OCL_Weak) ||
-                Record->isUnion())) {
+               Record && !ObjCFieldLifetimeErrReported && Record->isUnion()) {
       // It's an error in ARC or Weak if a field has lifetime.
       // We don't want to report this in a system header, though,
       // so we just make the field unavailable.
@@ -15456,6 +15452,8 @@ void Sema::ActOnFields(Scope *S, SourceLocation RecLoc, Decl *EnclosingDecl,
         Record->setNonTrivialToPrimitiveCopy(true);
       if (FT.isDestructedType())
         Record->setNonTrivialToPrimitiveDestroy(true);
+      if (!FT.canPassInRegisters())
+        Record->setCanPassInRegisters(false);
     }
 
     if (Record && FD->getType().isVolatileQualified())
diff --git a/lib/Serialization/ASTReaderDecl.cpp b/lib/Serialization/ASTReaderDecl.cpp
index ab99a17079..733748c293 100644
--- a/lib/Serialization/ASTReaderDecl.cpp
+++ b/lib/Serialization/ASTReaderDecl.cpp
@@ -742,6 +742,7 @@ ASTDeclReader::VisitRecordDeclImpl(RecordDecl *RD) {
   RD->setNonTrivialToPrimitiveDefaultInitialize(Record.readInt());
   RD->setNonTrivialToPrimitiveCopy(Record.readInt());
   RD->setNonTrivialToPrimitiveDestroy(Record.readInt());
+  RD->setCanPassInRegisters(Record.readInt());
   return Redecl;
 }
 
@@ -1587,7 +1588,6 @@ void ASTDeclReader::ReadCXXDefinitionData(
   Data.HasIrrelevantDestructor = Record.readInt();
   Data.HasConstexprNonCopyMoveConstructor = Record.readInt();
   Data.HasDefaultedDefaultConstructor = Record.readInt();
-  Data.CanPassInRegisters = Record.readInt();
   Data.DefaultedDefaultConstructorIsConstexpr = Record.readInt();
   Data.HasConstexprDefaultConstructor = Record.readInt();
   Data.HasNonLiteralTypeFieldsOrBases = Record.readInt();
@@ -1727,7 +1727,6 @@ void ASTDeclReader::MergeDefinitionData(
   MATCH_FIELD(HasIrrelevantDestructor)
   OR_FIELD(HasConstexprNonCopyMoveConstructor)
   OR_FIELD(HasDefaultedDefaultConstructor)
-  MATCH_FIELD(CanPassInRegisters)
   MATCH_FIELD(DefaultedDefaultConstructorIsConstexpr)
   OR_FIELD(HasConstexprDefaultConstructor)
   MATCH_FIELD(HasNonLiteralTypeFieldsOrBases)
@@ -4107,6 +4106,7 @@ void ASTDeclReader::UpdateDecl(Decl *D,
       bool HadRealDefinition =
           OldDD && (OldDD->Definition != RD ||
                     !Reader.PendingFakeDefinitionData.count(OldDD));
+      RD->setCanPassInRegisters(Record.readInt());
       ReadCXXRecordDefinition(RD, /*Update*/true);
 
       // Visible update is handled separately.
diff --git a/lib/Serialization/ASTWriter.cpp b/lib/Serialization/ASTWriter.cpp
index e63000c669..a6412ed470 100644
--- a/lib/Serialization/ASTWriter.cpp
+++ b/lib/Serialization/ASTWriter.cpp
@@ -5193,6 +5193,7 @@ void ASTWriter::WriteDeclUpdatesBlocks(RecordDataImpl &OffsetsRecord) {
       case UPD_CXX_INSTANTIATED_CLASS_DEFINITION: {
         auto *RD = cast<CXXRecordDecl>(D);
         UpdatedDeclContexts.insert(RD->getPrimaryContext());
+        Record.push_back(RD->canPassInRegisters());
         Record.AddCXXDefinitionData(RD);
         Record.AddOffset(WriteDeclContextLexicalBlock(
             *Context, const_cast<CXXRecordDecl *>(RD)));
@@ -6020,7 +6021,6 @@ void ASTRecordWriter::AddCXXDefinitionData(const CXXRecordDecl *D) {
   Record->push_back(Data.HasIrrelevantDestructor);
   Record->push_back(Data.HasConstexprNonCopyMoveConstructor);
   Record->push_back(Data.HasDefaultedDefaultConstructor);
-  Record->push_back(Data.CanPassInRegisters);
   Record->push_back(Data.DefaultedDefaultConstructorIsConstexpr);
   Record->push_back(Data.HasConstexprDefaultConstructor);
   Record->push_back(Data.HasNonLiteralTypeFieldsOrBases);
diff --git a/lib/Serialization/ASTWriterDecl.cpp b/lib/Serialization/ASTWriterDecl.cpp
index b3886b6555..6a426036a2 100644
--- a/lib/Serialization/ASTWriterDecl.cpp
+++ b/lib/Serialization/ASTWriterDecl.cpp
@@ -468,6 +468,7 @@ void ASTDeclWriter::VisitRecordDecl(RecordDecl *D) {
   Record.push_back(D->isNonTrivialToPrimitiveDefaultInitialize());
   Record.push_back(D->isNonTrivialToPrimitiveCopy());
   Record.push_back(D->isNonTrivialToPrimitiveDestroy());
+  Record.push_back(D->canPassInRegisters());
 
   if (D->getDeclContext() == D->getLexicalDeclContext() &&
       !D->hasAttrs() &&
@@ -1909,6 +1910,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1));
   // isNonTrivialToPrimitiveDestroy
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1));
+  Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 1)); // canPassInRegisters
 
   // DC
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6));   // LexicalOffset
diff --git a/test/CodeGenObjC/nontrivial-c-struct-exception.m b/test/CodeGenObjC/nontrivial-c-struct-exception.m
index a926a6d493..7db53bb742 100644
--- a/test/CodeGenObjC/nontrivial-c-struct-exception.m
+++ b/test/CodeGenObjC/nontrivial-c-struct-exception.m
@@ -1,12 +1,18 @@
-// RUN: %clang_cc1 -triple arm64-apple-ios11 -fobjc-arc -fblocks -fobjc-runtime=ios-11.0 -fobjc-exceptions -fexceptions -fobjc-arc-exceptions -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple arm64-apple-ios11 -fobjc-arc -fblocks -fobjc-runtime=ios-11.0 -fobjc-exceptions -fexceptions -emit-llvm -o - %s | FileCheck %s
 
 // CHECK: %[[STRUCT_STRONG:.*]] = type { i32, i8* }
+// CHECK: %[[STRUCT_WEAK:.*]] = type { i32, i8* }
 
 typedef struct {
   int i;
   id f1;
 } Strong;
 
+typedef struct {
+  int i;
+  __weak id f1;
+} Weak;
+
 // CHECK: define void @testStrongException()
 // CHECK: %[[AGG_TMP:.*]] = alloca %[[STRUCT_STRONG]], align 8
 // CHECK: %[[AGG_TMP1:.*]] = alloca %[[STRUCT_STRONG]], align 8
@@ -31,3 +37,26 @@ void calleeStrong(Strong, Strong);
 void testStrongException(void) {
   calleeStrong(genStrong(), genStrong());
 }
+
+// CHECK: define void @testWeakException()
+// CHECK: %[[AGG_TMP:.*]] = alloca %[[STRUCT_WEAK]], align 8
+// CHECK: %[[AGG_TMP1:.*]] = alloca %[[STRUCT_WEAK]], align 8
+// CHECK: call void @genWeak(%[[STRUCT_WEAK]]* sret %[[AGG_TMP]])
+// CHECK: invoke void @genWeak(%[[STRUCT_WEAK]]* sret %[[AGG_TMP1]])
+
+// CHECK: call void @calleeWeak(%[[STRUCT_WEAK]]* %[[AGG_TMP]], %[[STRUCT_WEAK]]* %[[AGG_TMP1]])
+// CHECK: ret void
+
+// CHECK: landingpad { i8*, i32 }
+// CHECK: %[[V3:.*]] = bitcast %[[STRUCT_WEAK]]* %[[AGG_TMP]] to i8**
+// CHECK: call void @__destructor_8_w8(i8** %[[V3]])
+// CHECK: br label
+
+// CHECK: resume
+
+Weak genWeak(void);
+void calleeWeak(Weak, Weak);
+
+void testWeakException(void) {
+  calleeWeak(genWeak(), genWeak());
+}
diff --git a/test/CodeGenObjC/weak-in-c-struct.m b/test/CodeGenObjC/weak-in-c-struct.m
new file mode 100644
index 0000000000..b0f4c08510
--- /dev/null
+++ b/test/CodeGenObjC/weak-in-c-struct.m
@@ -0,0 +1,193 @@
+// RUN: %clang_cc1 -triple arm64-apple-ios11 -fobjc-arc -fblocks -fobjc-runtime=ios-11.0 -emit-llvm -o - %s | FileCheck -check-prefix=ARM64 -check-prefix=COMMON %s
+// RUN: %clang_cc1 -triple thumbv7-apple-ios10 -fobjc-arc -fblocks -fobjc-runtime=ios-10.0 -emit-llvm -o - %s | FileCheck -check-prefix=COMMON %s
+// RUN: %clang_cc1 -triple x86_64-apple-macosx10.13 -fobjc-arc -fblocks -fobjc-runtime=macosx-10.13.0 -emit-llvm -o - %s | FileCheck -check-prefix=COMMON %s
+// RUN: %clang_cc1 -triple i386-apple-macosx10.13.0 -fobjc-arc -fblocks -fobjc-runtime=macosx-fragile-10.13.0 -emit-llvm -o - %s | FileCheck -check-prefix=COMMON %s
+
+typedef void (^BlockTy)(void);
+
+// COMMON: %[[STRUCT_WEAK:.*]] = type { i32, i8* }
+
+typedef struct {
+  int f0;
+  __weak id f1;
+} Weak;
+
+Weak getWeak(void);
+void calleeWeak(Weak);
+
+// ARM64: define void @test_constructor_destructor_Weak()
+// ARM64: %[[T:.*]] = alloca %[[STRUCT_WEAK]], align 8
+// ARM64: %[[V0:.*]] = bitcast %[[STRUCT_WEAK]]* %[[T]] to i8**
+// ARM64: call void @__default_constructor_8_w8(i8** %[[V0]])
+// ARM64: %[[V1:.*]] = bitcast %[[STRUCT_WEAK]]* %[[T]] to i8**
+// ARM64: call void @__destructor_8_w8(i8** %[[V1]])
+// ARM64: ret void
+
+// ARM64: define linkonce_odr hidden void @__default_constructor_8_w8(i8** %[[DST:.*]])
+// ARM64: %[[DST_ADDR:.*]] = alloca i8**, align 8
+// ARM64: store i8** %[[DST]], i8*** %[[DST_ADDR]], align 8
+// ARM64: %[[V0:.*]] = load i8**, i8*** %[[DST_ADDR]], align 8
+// ARM64: %[[V1]] = bitcast i8** %[[V0]] to i8*
+// ARM64: %[[V2:.*]] = getelementptr inbounds i8, i8* %[[V1]], i64 8
+// ARM64: %[[V3:.*]] = bitcast i8* %[[V2]] to i8**
+// ARM64: %[[V4:.*]] = bitcast i8** %[[V3]] to i8*
+// ARM64: call void @llvm.memset.p0i8.i64(i8* align 8 %[[V4]], i8 0, i64 8, i1 false)
+
+// ARM64: define linkonce_odr hidden void @__destructor_8_w8(i8** %[[DST:.*]])
+// ARM64: %[[DST_ADDR:.*]] = alloca i8**, align 8
+// ARM64: store i8** %[[DST]], i8*** %[[DST_ADDR]], align 8
+// ARM64: %[[V0:.*]] = load i8**, i8*** %[[DST_ADDR]], align 8
+// ARM64: %[[V1:.*]] = bitcast i8** %[[V0]] to i8*
+// ARM64: %[[V2:.*]] = getelementptr inbounds i8, i8* %[[V1]], i64 8
+// ARM64: %[[V3:.*]] = bitcast i8* %[[V2]] to i8**
+// ARM64: call void @objc_destroyWeak(i8** %[[V3]])
+
+void test_constructor_destructor_Weak(void) {
+  Weak t;
+}
+
+// ARM64: define void @test_copy_constructor_Weak(%[[STRUCT_WEAK]]* %{{.*}})
+// ARM64: call void @__copy_constructor_8_8_t0w4_w8(i8** %{{.*}}, i8** %{{.*}})
+// ARM64: call void @__destructor_8_w8(i8** %{{.*}})
+
+// ARM64: define linkonce_odr hidden void @__copy_constructor_8_8_t0w4_w8(i8** %[[DST:.*]], i8** %[[SRC:.*]])
+// ARM64: %[[DST_ADDR:.*]] = alloca i8**, align 8
+// ARM64: %[[SRC_ADDR:.*]] = alloca i8**, align 8
+// ARM64: store i8** %[[DST]], i8*** %[[DST_ADDR]], align 8
+// ARM64: store i8** %[[SRC]], i8*** %[[SRC_ADDR]], align 8
+// ARM64: %[[V0:.*]] = load i8**, i8*** %[[DST_ADDR]], align 8
+// ARM64: %[[V1:.*]] = load i8**, i8*** %[[SRC_ADDR]], align 8
+// ARM64: %[[V2:.*]] = bitcast i8** %[[V0]] to i32*
+// ARM64: %[[V3:.*]] = bitcast i8** %[[V1]] to i32*
+// ARM64: %[[V4:.*]] = load i32, i32* %[[V3]], align 8
+// ARM64: store i32 %[[V4]], i32* %[[V2]], align 8
+// ARM64: %[[V5:.*]] = bitcast i8** %[[V0]] to i8*
+// ARM64: %[[V6:.*]] = getelementptr inbounds i8, i8* %[[V5]], i64 8
+// ARM64: %[[V7:.*]] = bitcast i8* %[[V6]] to i8**
+// ARM64: %[[V8:.*]] = bitcast i8** %[[V1]] to i8*
+// ARM64: %[[V9:.*]] = getelementptr inbounds i8, i8* %[[V8]], i64 8
+// ARM64: %[[V10:.*]] = bitcast i8* %[[V9]] to i8**
+// ARM64: call void @objc_copyWeak(i8** %[[V7]], i8** %[[V10]])
+
+void test_copy_constructor_Weak(Weak *s) {
+  Weak t = *s;
+}
+
+// ARM64: define void @test_copy_assignment_Weak(%[[STRUCT_WEAK]]* %{{.*}}, %[[STRUCT_WEAK]]* %{{.*}})
+// ARM64: call void @__copy_assignment_8_8_t0w4_w8(i8** %{{.*}}, i8** %{{.*}})
+
+// ARM64: define linkonce_odr hidden void @__copy_assignment_8_8_t0w4_w8(i8** %[[DST:.*]], i8** %[[SRC:.*]])
+// ARM64: %[[DST_ADDR:.*]] = alloca i8**, align 8
+// ARM64: %[[SRC_ADDR:.*]] = alloca i8**, align 8
+// ARM64: store i8** %[[DST]], i8*** %[[DST_ADDR]], align 8
+// ARM64: store i8** %[[SRC]], i8*** %[[SRC_ADDR]], align 8
+// ARM64: %[[V0:.*]] = load i8**, i8*** %[[DST_ADDR]], align 8
+// ARM64: %[[V1:.*]] = load i8**, i8*** %[[SRC_ADDR]], align 8
+// ARM64: %[[V2:.*]] = bitcast i8** %[[V0]] to i32*
+// ARM64: %[[V3:.*]] = bitcast i8** %[[V1]] to i32*
+// ARM64: %[[V4:.*]] = load i32, i32* %[[V3]], align 8
+// ARM64: store i32 %[[V4]], i32* %[[V2]], align 8
+// ARM64: %[[V5:.*]] = bitcast i8** %[[V0]] to i8*
+// ARM64: %[[V6:.*]] = getelementptr inbounds i8, i8* %[[V5]], i64 8
+// ARM64: %[[V7:.*]] = bitcast i8* %[[V6]] to i8**
+// ARM64: %[[V8:.*]] = bitcast i8** %[[V1]] to i8*
+// ARM64: %[[V9:.*]] = getelementptr inbounds i8, i8* %[[V8]], i64 8
+// ARM64: %[[V10:.*]] = bitcast i8* %[[V9]] to i8**
+// ARM64: %[[V11:.*]] = call i8* @objc_loadWeakRetained(i8** %[[V10]])
+// ARM64: %[[V12:.*]] = call i8* @objc_storeWeak(i8** %[[V7]], i8* %[[V11]])
+// ARM64: call void @objc_release(i8* %[[V11]])
+
+void test_copy_assignment_Weak(Weak *d, Weak *s) {
+  *d = *s;
+}
+
+// ARM64: define internal void @__Block_byref_object_copy_(i8*, i8*)
+// ARM64: call void @__move_constructor_8_8_t0w4_w8(i8** %{{.*}}, i8** %{{.*}})
+
+// ARM64: define linkonce_odr hidden void @__move_constructor_8_8_t0w4_w8(i8** %[[DST:.*]], i8** %[[SRC:.*]])
+// ARM64: %[[DST_ADDR:.*]] = alloca i8**, align 8
+// ARM64: %[[SRC_ADDR:.*]] = alloca i8**, align 8
+// ARM64: store i8** %[[DST]], i8*** %[[DST_ADDR]], align 8
+// ARM64: store i8** %[[SRC]], i8*** %[[SRC_ADDR]], align 8
+// ARM64: %[[V0:.*]] = load i8**, i8*** %[[DST_ADDR]], align 8
+// ARM64: %[[V1:.*]] = load i8**, i8*** %[[SRC_ADDR]], align 8
+// ARM64: %[[V2:.*]] = bitcast i8** %[[V0]] to i32*
+// ARM64: %[[V3:.*]] = bitcast i8** %[[V1]] to i32*
+// ARM64: %[[V4:.*]] = load i32, i32* %[[V3]], align 8
+// ARM64: store i32 %[[V4]], i32* %[[V2]], align 8
+// ARM64: %[[V5:.*]] = bitcast i8** %[[V0]] to i8*
+// ARM64: %[[V6:.*]] = getelementptr inbounds i8, i8* %[[V5]], i64 8
+// ARM64: %[[V7:.*]] = bitcast i8* %[[V6]] to i8**
+// ARM64: %[[V8:.*]] = bitcast i8** %[[V1]] to i8*
+// ARM64: %[[V9:.*]] = getelementptr inbounds i8, i8* %[[V8]], i64 8
+// ARM64: %[[V10:.*]] = bitcast i8* %[[V9]] to i8**
+// ARM64: call void @objc_moveWeak(i8** %[[V7]], i8** %[[V10]])
+
+void test_move_constructor_Weak(void) {
+  __block Weak t;
+  BlockTy b = ^{ (void)t; };
+}
+
+// ARM64: define void @test_move_assignment_Weak(%[[STRUCT_WEAK]]* %{{.*}})
+// ARM64: call void @__move_assignment_8_8_t0w4_w8(i8** %{{.*}}, i8** %{{.*}})
+
+// ARM64: define linkonce_odr hidden void @__move_assignment_8_8_t0w4_w8(i8** %[[DST:.*]], i8** %[[SRC:.*]])
+// ARM64: %[[DST_ADDR:.*]] = alloca i8**, align 8
+// ARM64: %[[SRC_ADDR:.*]] = alloca i8**, align 8
+// ARM64: store i8** %[[DST]], i8*** %[[DST_ADDR]], align 8
+// ARM64: store i8** %[[SRC]], i8*** %[[SRC_ADDR]], align 8
+// ARM64: %[[V0:.*]] = load i8**, i8*** %[[DST_ADDR]], align 8
+// ARM64: %[[V1:.*]] = load i8**, i8*** %[[SRC_ADDR]], align 8
+// ARM64: %[[V2:.*]] = bitcast i8** %[[V0]] to i32*
+// ARM64: %[[V3:.*]] = bitcast i8** %[[V1]] to i32*
+// ARM64: %[[V4:.*]] = load i32, i32* %[[V3]], align 8
+// ARM64: store i32 %[[V4]], i32* %[[V2]], align 8
+// ARM64: %[[V5:.*]] = bitcast i8** %[[V0]] to i8*
+// ARM64: %[[V6:.*]] = getelementptr inbounds i8, i8* %[[V5]], i64 8
+// ARM64: %[[V7:.*]] = bitcast i8* %[[V6]] to i8**
+// ARM64: %[[V8:.*]] = bitcast i8** %[[V1]] to i8*
+// ARM64: %[[V9:.*]] = getelementptr inbounds i8, i8* %[[V8]], i64 8
+// ARM64: %[[V10:.*]] = bitcast i8* %[[V9]] to i8**
+// ARM64: %[[V11:.*]] = call i8* @objc_loadWeakRetained(i8** %[[V10]])
+// ARM64: %[[V12:.*]] = call i8* @objc_storeWeak(i8** %[[V7]], i8* %[[V11]])
+// ARM64: call void @objc_destroyWeak(i8** %[[V10]])
+// ARM64: call void @objc_release(i8* %[[V11]])
+
+void test_move_assignment_Weak(Weak *p) {
+  *p = getWeak();
+}
+
+// COMMON: define void @test_parameter_Weak(%[[STRUCT_WEAK]]* %[[A:.*]])
+// COMMON: %[[V0:.*]] = bitcast %[[STRUCT_WEAK]]* %[[A]] to i8**
+// COMMON: call void @__destructor_{{.*}}(i8** %[[V0]])
+
+void test_parameter_Weak(Weak a) {
+}
+
+// COMMON: define void @test_argument_Weak(%[[STRUCT_WEAK]]* %[[A:.*]])
+// COMMON: %[[A_ADDR:.*]] = alloca %[[STRUCT_WEAK]]*
+// COMMON: %[[AGG_TMP:.*]] = alloca %[[STRUCT_WEAK]]
+// COMMON: store %[[STRUCT_WEAK]]* %[[A]], %[[STRUCT_WEAK]]** %[[A_ADDR]]
+// COMMON: %[[V0:.*]] = load %[[STRUCT_WEAK]]*, %[[STRUCT_WEAK]]** %[[A_ADDR]]
+// COMMON: %[[V1:.*]] = bitcast %[[STRUCT_WEAK]]* %[[AGG_TMP]] to i8**
+// COMMON: %[[V2:.*]] = bitcast %[[STRUCT_WEAK]]* %[[V0]] to i8**
+// COMMON: call void @__copy_constructor_{{.*}}(i8** %[[V1]], i8** %[[V2]])
+// COMMON: call void @calleeWeak(%[[STRUCT_WEAK]]* %[[AGG_TMP]])
+// COMMON-NEXT: ret
+
+void test_argument_Weak(Weak *a) {
+  calleeWeak(*a);
+}
+
+// COMMON: define void @test_return_Weak(%[[STRUCT_WEAK]]* noalias sret %[[AGG_RESULT:.*]], %[[STRUCT_WEAK]]* %[[A:.*]])
+// COMMON: %[[A_ADDR:.*]] = alloca %[[STRUCT_WEAK]]*
+// COMMON: store %[[STRUCT_WEAK]]* %[[A]], %[[STRUCT_WEAK]]** %[[A_ADDR]]
+// COMMON: %[[V0:.*]] = load %[[STRUCT_WEAK]]*, %[[STRUCT_WEAK]]** %[[A_ADDR]]
+// COMMON: %[[V1:.*]] = bitcast %[[STRUCT_WEAK]]* %[[AGG_RESULT]] to i8**
+// COMMON: %[[V2:.*]] = bitcast %[[STRUCT_WEAK]]* %[[V0]] to i8**
+// COMMON: call void @__copy_constructor_{{.*}}(i8** %[[V1]], i8** %[[V2]])
+// COMMON: ret void
+
+Weak test_return_Weak(Weak *a) {
+  return *a;
+}
diff --git a/test/Modules/Inputs/module.map b/test/Modules/Inputs/module.map
index 4788daa431..93fe468873 100644
--- a/test/Modules/Inputs/module.map
+++ b/test/Modules/Inputs/module.map
@@ -460,3 +460,13 @@ module innerstructredef {
    header "innerstructredef.h"
   }
 }
+
+module template_nontrivial0 {
+  header "template-nontrivial0.h"
+  export *
+}
+
+module template_nontrivial1 {
+  header "template-nontrivial1.h"
+  export *
+}
diff --git a/test/Modules/Inputs/template-nontrivial0.h b/test/Modules/Inputs/template-nontrivial0.h
new file mode 100644
index 0000000000..cff080ee5c
--- /dev/null
+++ b/test/Modules/Inputs/template-nontrivial0.h
@@ -0,0 +1,13 @@
+template <class T>
+struct Class0 {
+  Class0();
+  Class0(const Class0<T> &);
+  ~Class0();
+  T *p;
+};
+
+struct S0 {
+  id x;
+};
+
+Class0<S0> returnNonTrivial();
diff --git a/test/Modules/Inputs/template-nontrivial1.h b/test/Modules/Inputs/template-nontrivial1.h
new file mode 100644
index 0000000000..24136f0cae
--- /dev/null
+++ b/test/Modules/Inputs/template-nontrivial1.h
@@ -0,0 +1,6 @@
+@import template_nontrivial0;
+
+struct S1 {
+  S1();
+  Class0<S0> a;
+};
diff --git a/test/Modules/templates.mm b/test/Modules/templates.mm
index 9c3f327acb..95e7a9caea 100644
--- a/test/Modules/templates.mm
+++ b/test/Modules/templates.mm
@@ -122,3 +122,13 @@ void testWithAttributes() {
   static_assert(alignof(decltype(a)) == 2, "");
   static_assert(alignof(decltype(b)) == 2, "");
 }
+
+// Check that returnNonTrivial doesn't return Class0<S0> directly in registers.
+
+// CHECK: declare void @_Z16returnNonTrivialv(%struct.Class0* sret)
+
+@import template_nontrivial0;
+@import template_nontrivial1;
+
+S1::S1() : a(returnNonTrivial()) {
+}