[clang] Implement gcc_struct attribute on Itanium targets

clang/docs/ReleaseNotes.rst

@@ -379,6 +379,12 @@ Attribute Changes in Clang
   multilibs for printf features. Requires cooperation with the libc
   implementation.
 
+- On targets with Itanium C++ ABI, Clang now supports ``[[gnu:gcc_struct]]``
+  with the behavior similar to one existing in GCC. In particular, whenever
+  ``-mms-bitfields`` command line option is provided (or if Microsoft-compatible
+  structure layout is default on the target), ``[[gnu::gcc_struct]]`` requests
+  the compiler to follow Itanium rules for the layout of an annotated structure.
+
 Improvements to Clang's diagnostics
 -----------------------------------
 - Diagnostics messages now refer to ``structured binding`` instead of ``decomposition``,

clang/include/clang/AST/ASTContext.h

@@ -2810,6 +2810,10 @@ class ASTContext : public RefCountedBase<ASTContext> {
   /// runtime, such as those using the Itanium C++ ABI.
   CharUnits getExnObjectAlignment() const;
 
+  /// Return whether unannotated records are treated as if they have
+  /// [[gnu::ms_struct]].
+  bool defaultsToMsStruct() const;
+
   /// Get or compute information about the layout of the specified
   /// record (struct/union/class) \p D, which indicates its size and field
   /// position information.

clang/include/clang/Basic/Attr.td

@@ -4387,8 +4387,13 @@ def CFGuard : InheritableAttr, TargetSpecificAttr<TargetWindows> {
 def MSStruct : InheritableAttr {
   let Spellings = [GCC<"ms_struct">];
   let Subjects = SubjectList<[Record]>;
-  let Documentation = [Undocumented];
-  let SimpleHandler = 1;
+  let Documentation = [MSStructDocs];
+}
+
+def GCCStruct : InheritableAttr {
+  let Spellings = [GCC<"gcc_struct">];
+  let Subjects = SubjectList<[Record]>;
+  let Documentation = [MSStructDocs]; // Covers this attribute too.
 }
 
 def DLLExport : InheritableAttr, TargetSpecificAttr<TargetHasDLLImportExport> {

clang/include/clang/Basic/AttrDocs.td

@@ -9682,3 +9682,24 @@ The following aspects are currently supported:
 - ``float``: The call has a floating point argument
   }];
 }
+
+def MSStructDocs : Documentation {
+  let Category = DocCatDecl;
+  let Content = [{
+The ``ms_struct`` and ``gcc_struct`` attributes request the compiler to enter a
+special record layout compatibility mode which mimics the layout of Microsoft or
+Itanium C++ ABI respectively. Obviously, if the current C++ ABI matches the
+requested ABI, the attribute does nothing. However, if it does not, annotated
+structure or class is laid out in a special compatibility mode, which slightly
+changes offsets for fields and bit-fields. The intention is to match the layout
+of the requested ABI for structures which only use C features.
+
+Note that the default behavior can be controlled by ``-mms-bitfields`` and
+``-mno-ms-bitfields`` switches and via ``#pragma ms_struct``.
+
+The primary difference is for bitfields, where the MS variant only packs
+adjacent fields into the same allocation unit if they have integral types
+of the same size, while the GCC/Itanium variant packs all fields in a bitfield
+tightly.
+  }];
+}

clang/include/clang/Basic/DiagnosticASTKinds.td

@@ -1040,6 +1040,9 @@ def warn_npot_ms_struct : Warning<
   "data types with sizes that aren't a power of two">,
   DefaultError, InGroup<IncompatibleMSStruct>;
 
+def err_itanium_layout_unimplemented : Error<
+  "Itanium-compatible layout for the Microsoft C++ ABI is not yet supported">;
+
 // -Wpadded-bitfield
 def warn_padded_struct_bitfield : Warning<
   "padding %select{struct|interface|class}0 %1 with %2 "

clang/include/clang/Basic/LangOptions.def

@@ -104,7 +104,9 @@ LANGOPT(AssumeNothrowExceptionDtor , 1, 0, NotCompatible, "Assume exception obje
 LANGOPT(TraditionalCPP    , 1, 0, NotCompatible, "traditional CPP emulation")
 LANGOPT(RTTI              , 1, 1, NotCompatible, "run-time type information")
 LANGOPT(RTTIData          , 1, 1, NotCompatible, "emit run-time type information data")
-LANGOPT(MSBitfields       , 1, 0, NotCompatible, "Microsoft-compatible structure layout")
+ENUM_LANGOPT(LayoutCompatibility, LayoutCompatibilityKind, 2,
+             LayoutCompatibilityKind::Default, NotCompatible,
+             "Microsoft-compatible structure layout")
 LANGOPT(MSVolatile        , 1, 0, NotCompatible, "Microsoft-compatible volatile loads and stores")
 LANGOPT(Freestanding      , 1, 0, NotCompatible, "freestanding implementation")
 LANGOPT(NoBuiltin         , 1, 0, NotCompatible, "disable builtin functions")

clang/include/clang/Basic/LangOptions.h

@@ -421,6 +421,16 @@ class LangOptionsBase {
     None,
   };
 
+  enum class LayoutCompatibilityKind {
+    /// Use default layout rules of the target.
+    Default = 0,
+    /// Use Itanium rules for bit-field layout and fundamental types alignment.
+    Itanium = 1,
+    /// Use Microsoft C++ ABI rules for bit-field layout and fundamental types
+    /// alignment.
+    Microsoft = 2,
+  };
+
   // Define simple language options (with no accessors).
 #define LANGOPT(Name, Bits, Default, Compatibility, Description)               \
   unsigned Name : Bits;

clang/include/clang/Options/Options.td

@@ -5219,9 +5219,7 @@ def : Joined<["-"], "mmacosx-version-min=">,
   Visibility<[ClangOption, CC1Option, FC1Option, FlangOption]>,
   Group<m_Group>, Alias<mmacos_version_min_EQ>;
 def mms_bitfields : Flag<["-"], "mms-bitfields">, Group<m_Group>,
-  Visibility<[ClangOption, CC1Option]>,
-  HelpText<"Set the default structure layout to be compatible with the Microsoft compiler standard">,
-  MarshallingInfoFlag<LangOpts<"MSBitfields">>;
+  HelpText<"Set the default structure layout to be compatible with the Microsoft compiler standard">;
 def moutline : Flag<["-"], "moutline">, Group<f_clang_Group>,
   Visibility<[ClangOption, CC1Option]>,
     HelpText<"Enable function outlining (AArch64 only)">;
@@ -5230,6 +5228,12 @@ def mno_outline : Flag<["-"], "mno-outline">, Group<f_clang_Group>,
     HelpText<"Disable function outlining (AArch64 only)">;
 def mno_ms_bitfields : Flag<["-"], "mno-ms-bitfields">, Group<m_Group>,
   HelpText<"Do not set the default structure layout to be compatible with the Microsoft compiler standard">;
+def fms_layout_compatibility_EQ : Joined<["-"], "fms-layout-compatibility=">,
+  Visibility<[CC1Option]>,
+  HelpText<"Structure layout compatibility with Microsoft C++ ABI">,
+  Values<"default,itanium,microsoft">,
+  NormalizedValues<["Default", "Itanium", "Microsoft"]>, NormalizedValuesScope<"LangOptions::LayoutCompatibilityKind">,
+  MarshallingInfoEnum<LangOpts<"LayoutCompatibility">, "Default">;
 def mskip_rax_setup : Flag<["-"], "mskip-rax-setup">, Group<m_Group>,
   Visibility<[ClangOption, CC1Option]>,
   HelpText<"Skip setting up RAX register when passing variable arguments (x86 only)">,

clang/lib/AST/Decl.cpp

@@ -5244,7 +5244,14 @@ void RecordDecl::completeDefinition() {
 /// This which can be turned on with an attribute, pragma, or the
 /// -mms-bitfields command-line option.
 bool RecordDecl::isMsStruct(const ASTContext &C) const {
-  return hasAttr<MSStructAttr>() || C.getLangOpts().MSBitfields == 1;
+  if (hasAttr<GCCStructAttr>())
+    return false;
+  if (hasAttr<MSStructAttr>())
+    return true;
+  auto LayoutCompatibility = C.getLangOpts().getLayoutCompatibility();
+  if (LayoutCompatibility == LangOptions::LayoutCompatibilityKind::Default)
+    return C.defaultsToMsStruct();
+  return LayoutCompatibility == LangOptions::LayoutCompatibilityKind::Microsoft;
 }
 
 void RecordDecl::reorderDecls(const SmallVectorImpl<Decl *> &Decls) {

clang/lib/AST/RecordLayoutBuilder.cpp

@@ -2794,6 +2794,13 @@ void MicrosoftRecordLayoutBuilder::initializeLayout(const RecordDecl *RD) {
     UseExternalLayout = Source->layoutRecordType(
         RD, External.Size, External.Align, External.FieldOffsets,
         External.BaseOffsets, External.VirtualBaseOffsets);
+
+  if (!RD->isMsStruct(Context)) {
+    auto Location = RD->getLocation();
+    if (Location.isValid())
+      Context.getDiagnostics().Report(Location,
+                                      diag::err_itanium_layout_unimplemented);
+  }
 }
 
 void
@@ -3358,6 +3365,11 @@ void MicrosoftRecordLayoutBuilder::computeVtorDispSet(
   }
 }
 
+bool ASTContext::defaultsToMsStruct() const {
+  return getTargetInfo().hasMicrosoftRecordLayout() ||
+         getTargetInfo().getTriple().isWindowsGNUEnvironment();
+}
+
 /// getASTRecordLayout - Get or compute information about the layout of the
 /// specified record (struct/union/class), which indicates its size and field
 /// position information.

clang/lib/Driver/ToolChains/Clang.cpp

@@ -5889,9 +5889,12 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   if (KernelOrKext && RawTriple.isOSDarwin())
     CmdArgs.push_back("-fforbid-guard-variables");
 
-  if (Args.hasFlag(options::OPT_mms_bitfields, options::OPT_mno_ms_bitfields,
-                   Triple.isWindowsGNUEnvironment())) {
-    CmdArgs.push_back("-mms-bitfields");
+  if (Arg *A = Args.getLastArg(options::OPT_mms_bitfields,
+                               options::OPT_mno_ms_bitfields)) {
+    if (A->getOption().matches(options::OPT_mms_bitfields))
+      CmdArgs.push_back("-fms-layout-compatibility=microsoft");
+    else
+      CmdArgs.push_back("-fms-layout-compatibility=itanium");
   }
 
   if (Triple.isOSCygMing()) {

clang/lib/Sema/SemaDecl.cpp

@@ -18992,9 +18992,7 @@ ExprResult Sema::VerifyBitField(SourceLocation FieldLoc,
     // ABI.
     bool CStdConstraintViolation =
         BitfieldIsOverwide && !getLangOpts().CPlusPlus;
-    bool MSBitfieldViolation =
-        Value.ugt(TypeStorageSize) &&
-        (IsMsStruct || Context.getTargetInfo().getCXXABI().isMicrosoft());
+    bool MSBitfieldViolation = Value.ugt(TypeStorageSize) && IsMsStruct;
     if (CStdConstraintViolation || MSBitfieldViolation) {
       unsigned DiagWidth =
           CStdConstraintViolation ? TypeWidth : TypeStorageSize;

clang/lib/Sema/SemaDeclAttr.cpp

@@ -6173,6 +6173,36 @@ static void handleMSConstexprAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
   D->addAttr(::new (S.Context) MSConstexprAttr(S.Context, AL));
 }
 
+static void handleMSStructAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
+  if (const auto *First = D->getAttr<GCCStructAttr>()) {
+    S.Diag(AL.getLoc(), diag::err_attributes_are_not_compatible)
+        << AL << First << 0;
+    S.Diag(First->getLocation(), diag::note_conflicting_attribute);
+    return;
+  }
+  if (const auto *Preexisting = D->getAttr<MSStructAttr>()) {
+    if (Preexisting->isImplicit())
+      D->dropAttr<MSStructAttr>();
+  }
+
+  D->addAttr(::new (S.Context) MSStructAttr(S.Context, AL));
+}
+
+static void handleGCCStructAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
+  if (const auto *First = D->getAttr<MSStructAttr>()) {
+    if (First->isImplicit()) {
+      D->dropAttr<MSStructAttr>();
+    } else {
+      S.Diag(AL.getLoc(), diag::err_attributes_are_not_compatible)
+          << AL << First << 0;
+      S.Diag(First->getLocation(), diag::note_conflicting_attribute);
+      return;
+    }
+  }
+
+  D->addAttr(::new (S.Context) GCCStructAttr(S.Context, AL));
+}
+
 static void handleAbiTagAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
   SmallVector<StringRef, 4> Tags;
   for (unsigned I = 0, E = AL.getNumArgs(); I != E; ++I) {
@@ -7983,6 +8013,14 @@ ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D, const ParsedAttr &AL,
   case ParsedAttr::AT_ModularFormat:
     handleModularFormat(S, D, AL);
     break;
+
+  case ParsedAttr::AT_MSStruct:
+    handleMSStructAttr(S, D, AL);
+    break;
+
+  case ParsedAttr::AT_GCCStruct:
+    handleGCCStructAttr(S, D, AL);
+    break;
   }
 }
 

clang/lib/Sema/SemaDeclCXX.cpp

@@ -7288,20 +7288,27 @@ void Sema::CheckCompletedCXXClass(Scope *S, CXXRecordDecl *Record) {
       CheckCompletedMemberFunction(MD);
   }
 
-  // ms_struct is a request to use the same ABI rules as MSVC.  Check
-  // whether this class uses any C++ features that are implemented
-  // completely differently in MSVC, and if so, emit a diagnostic.
-  // That diagnostic defaults to an error, but we allow projects to
-  // map it down to a warning (or ignore it).  It's a fairly common
-  // practice among users of the ms_struct pragma to mass-annotate
-  // headers, sweeping up a bunch of types that the project doesn't
-  // really rely on MSVC-compatible layout for.  We must therefore
-  // support "ms_struct except for C++ stuff" as a secondary ABI.
+  // {ms,gcc}_struct is a request to change ABI rules to either follow
+  // Microsoft or Itanium C++ ABI. However, even if these attributes are
+  // present, we do not layout classes following foreign ABI rules, but
+  // instead enter a special "compatibility mode", which only changes
+  // alignments of fundamental types and layout of bit fields.
+  // Check whether this class uses any C++ features that are implemented
+  // completely differently in the requested ABI, and if so, emit a
+  // diagnostic. That diagnostic defaults to an error, but we allow
+  // projects to map it down to a warning (or ignore it). It's a fairly
+  // common practice among users of the ms_struct pragma to
+  // mass-annotate headers, sweeping up a bunch of types that the
+  // project doesn't really rely on MSVC-compatible layout for. We must
+  // therefore support "ms_struct except for C++ stuff" as a secondary
+  // ABI.
   // Don't emit this diagnostic if the feature was enabled as a
   // language option (as opposed to via a pragma or attribute), as
   // the option -mms-bitfields otherwise essentially makes it impossible
   // to build C++ code, unless this diagnostic is turned off.
-  if (Record->isMsStruct(Context) && !Context.getLangOpts().MSBitfields &&
+  if (Context.getLangOpts().getLayoutCompatibility() ==
+          LangOptions::LayoutCompatibilityKind::Default &&
+      Record->isMsStruct(Context) != Context.defaultsToMsStruct() &&
       (Record->isPolymorphic() || Record->getNumBases())) {
     Diag(Record->getLocation(), diag::warn_cxx_ms_struct);
   }

clang/test/CodeGen/gcc_struct-msvc-todo-1.c

@@ -0,0 +1,6 @@
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-pc-windows-msvc -verify %s
+
+// expected-error@+1 {{Itanium-compatible layout for the Microsoft C++ ABI is not yet supported}}
+struct {
+    int a;
+} __attribute__((gcc_struct)) t1;

clang/test/CodeGen/gcc_struct-msvc-todo-2.c

@@ -0,0 +1,6 @@
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-pc-windows-msvc -fms-layout-compatibility=itanium -verify %s
+
+// expected-error@+1 {{Itanium-compatible layout for the Microsoft C++ ABI is not yet supported}}
+struct {
+    int a;
+} t1;

clang/test/CodeGen/gcc_struct.c

@@ -0,0 +1,18 @@
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-pc-linux-gnu %s
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-pc-linux-gnu -fms-layout-compatibility=microsoft %s
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-pc-windows-gnu %s
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-pc-windows-gnu -fms-layout-compatibility=itanium %s
+
+struct {
+    int a : 24;
+    char b : 8;
+} __attribute__((gcc_struct)) t1;
+_Static_assert(sizeof(t1) == 4, "");
+
+#pragma ms_struct on
+struct {
+    int a : 24;
+    char b : 8;
+} __attribute__((gcc_struct)) t2;
+_Static_assert(sizeof(t2) == 4, "");
+#pragma ms_struct off

clang/test/CodeGen/mingw-long-double.c

@@ -1,6 +1,6 @@
 // RUN: %clang_cc1 -triple i686-windows-gnu -emit-llvm -o - %s \
 // RUN:    | FileCheck %s --check-prefix=GNU32
-// RUN: %clang_cc1 -triple i686-windows-gnu -emit-llvm -o - %s -mms-bitfields \
+// RUN: %clang_cc1 -triple i686-windows-gnu -emit-llvm -o - %s -fms-layout-compatibility=microsoft \
 // RUN:    | FileCheck %s --check-prefix=GNU32
 // RUN: %clang_cc1 -triple x86_64-windows-gnu -emit-llvm -o - %s \
 // RUN:    | FileCheck %s --check-prefix=GNU64

clang/test/CodeGen/mms-bitfields.c

@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -triple i386-apple-darwin10 -mms-bitfields -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang_cc1 -triple i386-apple-darwin10 -fms-layout-compatibility=microsoft -emit-llvm %s -o - | FileCheck %s
 
 struct s1 {
   int       f32;

clang/test/Driver/ms-bitfields.c

@@ -1,8 +1,15 @@
-// RUN: %clang -### --target=x86_64-linux-gnu %s 2>&1 | FileCheck %s -check-prefix=NO-MSBITFIELDS
-// RUN: %clang -### --target=x86_64-windows-gnu %s 2>&1 | FileCheck %s -check-prefix=MSBITFIELDS
-// RUN: %clang -### -mno-ms-bitfields -mms-bitfields %s 2>&1 | FileCheck %s -check-prefix=MSBITFIELDS
-// RUN: %clang -### -mms-bitfields -mno-ms-bitfields %s 2>&1 | FileCheck %s -check-prefix=NO-MSBITFIELDS
+// RUN: %clang -### --target=x86_64-linux-gnu %s 2>&1 | FileCheck %s -check-prefix=DEFAULT-LAYOUT
+// RUN: %clang -### --target=x86_64-windows-gnu %s 2>&1 | FileCheck %s -check-prefix=DEFAULT-LAYOUT
+// RUN: %clang -### --target=x86_64-windows-msvc %s 2>&1 | FileCheck %s -check-prefix=DEFAULT-LAYOUT
+// RUN: %clang -### -mms-bitfields %s 2>&1 | FileCheck %s -check-prefix=MICROSOFT-LAYOUT
+// RUN: %clang -### -mno-ms-bitfields %s 2>&1 | FileCheck %s -check-prefix=ITANIUM-LAYOUT
+// RUN: %clang -### -mno-ms-bitfields -mms-bitfields %s 2>&1 | FileCheck %s -check-prefix=MICROSOFT-LAYOUT
+// RUN: %clang -### -mms-bitfields -mno-ms-bitfields %s 2>&1 | FileCheck %s -check-prefix=ITANIUM-LAYOUT
 
-// MSBITFIELDS: -mms-bitfields
-// NO-MSBITFIELDS-NOT: -mms-bitfields
+// DEFAULT-LAYOUT-NOT: -fms-layout-compatibility=itanium
+// DEFAULT-LAYOUT-NOT: -fms-layout-compatibility=microsoft
+// MICROSOFT-LAYOUT: -fms-layout-compatibility=microsoft
+// MICROSOFT-LAYOUT-NOT: -fms-layout-compatibility=itanium
+// ITANIUM-LAYOUT: -fms-layout-compatibility=itanium
+// ITANIUM-LAYOUT-NOT: -fms-layout-compatibility=microsoft
 

clang/test/Layout/itanium-union-bitfield.cpp

@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -emit-llvm-only -triple %itanium_abi_triple -fdump-record-layouts %s 2>/dev/null \
+// RUN: %clang_cc1 -emit-llvm-only -triple %itanium_abi_triple -fms-layout-compatibility=itanium -fdump-record-layouts %s 2>/dev/null \
 // RUN:            | FileCheck %s
 
 // On z/OS, a bit-field has single byte alignment.  Add aligned(4) on z/OS so the union has

clang/test/Misc/pragma-attribute-supported-attributes-list.test

@@ -87,6 +87,7 @@
 // CHECK-NEXT: FlagEnum (SubjectMatchRule_enum)
 // CHECK-NEXT: Flatten (SubjectMatchRule_function)
 // CHECK-NEXT: FunctionReturnThunks (SubjectMatchRule_function)
+// CHECK-NEXT: GCCStruct (SubjectMatchRule_record)
 // CHECK-NEXT: GNUInline (SubjectMatchRule_function)
 // CHECK-NEXT: HIPManaged (SubjectMatchRule_variable)
 // CHECK-NEXT: HLSLVkLocation (SubjectMatchRule_variable_is_parameter, SubjectMatchRule_field, SubjectMatchRule_function)