[clang] Implement gcc_struct attribute on Itanium targets #71148

DanShaders · 2023-11-03T05:58:19Z

This commit implements gcc_struct attribute with the behavior similar to one in GCC. Current behavior is as follows:

When ItaniumRecordLayoutBuilder is used, [[gcc_struct]] will locally cancel the effect of -mms-bitfields on a record. If -mms-bitfields is not supplied and is not a default behavior on a target, [[gcc_struct]] will be a no-op. This should provide enough compatibility with GCC.

If C++ ABI is "Microsoft", [[gcc_struct]] will currently always produce a diagnostic, since support for it is not yet implemented in MicrosoftRecordLayoutBuilder. Note, however, that all the infrastructure is ready for the future implementation.

In particular, check for default value of -mms-bitfields is moved from a driver to ASTContext, since it now non-trivially depends on other supplied flags. This also, unfortunately, makes it impossible to use usual argument parsing for -m{no-,}ms-bitfields.

The patch doesn't introduce any backwards-incompatible changes, except for situations when cc1 is called directly with -mms-bitfields option.

Work towards #24757

llvmbot · 2023-11-03T05:58:50Z

@llvm/pr-subscribers-clang
@llvm/pr-subscribers-clang-driver

@llvm/pr-subscribers-clang-codegen

Author: Dan Klishch (DanShaders)

Changes

This implements gcc_struct attribute with the behavior similar to one in GCC.

In particular, when C++ ABI is not "Microsoft" (i. e. when ItaniumRecordBuilder is used), [[gcc_struct]] will locally cancel the effect of -mms-bitfields on a record. If -mms-bitfields is not supplied and is not a default behavior on a target, [[gcc_struct]] will be a no-op. This hopefully should provide compatibility with MinGW binaries.

On the other hand, if C++ ABI is "Microsoft" and a record is annotated with [[gcc_struct]], we do not have any compatibility concerns, so I decided not to complicate MicrosoftRecordLayoutBuilder and just to fully switch record layout to GenericItanium.

Note that -mno-ms-bitfields is no longer always a no-op. On x86_64-pc-windows-msvc, specifying only it should yield the same results as annotating all the records with [[gcc_struct]]. This makes it impossible to use usual argument parsing for -m{no-,}ms-bitfield in a driver since MSBitfields now has a default value, which non-trivially depends on other supplied flags.

Changes related to ms_struct should not hopefully cause any observable differences, thus the patch itself is fully backwards-compatible.

Fixes #24757

Full diff: https://github.com/llvm/llvm-project/pull/71148.diff

15 Files Affected:

(modified) clang/include/clang/Basic/Attr.td (+7)
(modified) clang/include/clang/Basic/LangOptions.def (-1)
(modified) clang/include/clang/Basic/LangOptions.h (+2)
(modified) clang/include/clang/Basic/TargetInfo.h (+4)
(modified) clang/include/clang/Driver/Options.td (+2-2)
(modified) clang/lib/AST/Decl.cpp (+7-2)
(modified) clang/lib/AST/RecordLayoutBuilder.cpp (+7-6)
(modified) clang/lib/CodeGen/CGRecordLayoutBuilder.cpp (+3-5)
(modified) clang/lib/Driver/ToolChains/Clang.cpp (+7-3)
(modified) clang/lib/Frontend/CompilerInvocation.cpp (+12)
(modified) clang/lib/Sema/SemaDecl.cpp (+1-3)
(modified) clang/lib/Sema/SemaDeclCXX.cpp (+8-1)
(added) clang/test/CodeGenCXX/ms-bitfield-class-layout.cpp (+41)
(added) clang/test/CodeGenCXX/ms_struct-gcc_struct.cpp (+29)
(modified) clang/test/Driver/ms-bitfields.c (+8-3)

diff --git a/clang/include/clang/Basic/Attr.td b/clang/include/clang/Basic/Attr.td
index 60b549999c155e5..34ec6c22481837e 100644
--- a/clang/include/clang/Basic/Attr.td
+++ b/clang/include/clang/Basic/Attr.td
@@ -3635,6 +3635,13 @@ def MSStruct : InheritableAttr {
   let SimpleHandler = 1;
 }
 
+def GCCStruct : InheritableAttr {
+  let Spellings = [GCC<"gcc_struct">];
+  let Subjects = SubjectList<[Record]>;
+  let Documentation = [Undocumented];
+  let SimpleHandler = 1;
+}
+
 def DLLExport : InheritableAttr, TargetSpecificAttr<TargetHasDLLImportExport> {
   let Spellings = [Declspec<"dllexport">, GCC<"dllexport">];
   let Subjects = SubjectList<[Function, Var, CXXRecord, ObjCInterface]>;
diff --git a/clang/include/clang/Basic/LangOptions.def b/clang/include/clang/Basic/LangOptions.def
index c0ea4ecb9806a5b..afbb3c724a8d193 100644
--- a/clang/include/clang/Basic/LangOptions.def
+++ b/clang/include/clang/Basic/LangOptions.def
@@ -148,7 +148,6 @@ LANGOPT(ExternCNoUnwind   , 1, 0, "Assume extern C functions don't unwind")
 LANGOPT(TraditionalCPP    , 1, 0, "traditional CPP emulation")
 LANGOPT(RTTI              , 1, 1, "run-time type information")
 LANGOPT(RTTIData          , 1, 1, "emit run-time type information data")
-LANGOPT(MSBitfields       , 1, 0, "Microsoft-compatible structure layout")
 LANGOPT(MSVolatile        , 1, 0, "Microsoft-compatible volatile loads and stores")
 LANGOPT(Freestanding, 1, 0, "freestanding implementation")
 LANGOPT(NoBuiltin         , 1, 0, "disable builtin functions")
diff --git a/clang/include/clang/Basic/LangOptions.h b/clang/include/clang/Basic/LangOptions.h
index 20a8ada60e0fe51..07eae8dad61813c 100644
--- a/clang/include/clang/Basic/LangOptions.h
+++ b/clang/include/clang/Basic/LangOptions.h
@@ -502,6 +502,8 @@ class LangOptions : public LangOptionsBase {
   // received as a result of a standard operator new (-fcheck-new)
   bool CheckNew = false;
 
+  std::optional<bool> MSBitfields;
+
   LangOptions();
 
   /// Set language defaults for the given input language and
diff --git a/clang/include/clang/Basic/TargetInfo.h b/clang/include/clang/Basic/TargetInfo.h
index b3c5cbfb319f01f..6865b6e23bc9628 100644
--- a/clang/include/clang/Basic/TargetInfo.h
+++ b/clang/include/clang/Basic/TargetInfo.h
@@ -1724,6 +1724,10 @@ class TargetInfo : public TransferrableTargetInfo,
   /// Whether to support HIP image/texture API's.
   virtual bool hasHIPImageSupport() const { return true; }
 
+  virtual bool defaultsToMsStruct() const {
+    return getCXXABI().isMicrosoft() || getTriple().isWindowsGNUEnvironment();
+  }
+
 protected:
   /// Copy type and layout related info.
   void copyAuxTarget(const TargetInfo *Aux);
diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index fcf6a4b2ccb2369..bde324a51c5f77f 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -4373,8 +4373,7 @@ def : Joined<["-"], "mmacosx-version-min=">,
   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)">;
@@ -4382,6 +4381,7 @@ def mno_outline : Flag<["-"], "mno-outline">, Group<f_clang_Group>,
   Visibility<[ClangOption, CC1Option]>,
     HelpText<"Disable function outlining (AArch64 only)">;
 def mno_ms_bitfields : Flag<["-"], "mno-ms-bitfields">, Group<m_Group>,
+  Visibility<[ClangOption, CC1Option]>,
   HelpText<"Do not set the default structure layout to be compatible with the Microsoft compiler standard">;
 def mskip_rax_setup : Flag<["-"], "mskip-rax-setup">, Group<m_Group>,
   Visibility<[ClangOption, CC1Option]>,
diff --git a/clang/lib/AST/Decl.cpp b/clang/lib/AST/Decl.cpp
index 6efc177d61c03ba..352af5899a0c02a 100644
--- a/clang/lib/AST/Decl.cpp
+++ b/clang/lib/AST/Decl.cpp
@@ -5009,7 +5009,7 @@ void RecordDecl::completeDefinition() {
   ASTContext &Ctx = getASTContext();
 
   // Layouts are dumped when computed, so if we are dumping for all complete
-  // types, we need to force usage to get types that wouldn't be used elsewhere.
+  // types, we need to force usage to pet types that wouldn't be used elsewhere.
   if (Ctx.getLangOpts().DumpRecordLayoutsComplete)
     (void)Ctx.getASTRecordLayout(this);
 }
@@ -5018,7 +5018,12 @@ 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<MSStructAttr>())
+    return true;
+  if (hasAttr<GCCStructAttr>())
+    return false;
+  return C.getLangOpts().MSBitfields.value_or(
+      C.getTargetInfo().defaultsToMsStruct());
 }
 
 void RecordDecl::reorderDecls(const SmallVectorImpl<Decl *> &Decls) {
diff --git a/clang/lib/AST/RecordLayoutBuilder.cpp b/clang/lib/AST/RecordLayoutBuilder.cpp
index 5d4f930fca50e2b..f38c6432a9992a6 100644
--- a/clang/lib/AST/RecordLayoutBuilder.cpp
+++ b/clang/lib/AST/RecordLayoutBuilder.cpp
@@ -2447,8 +2447,9 @@ static bool mustSkipTailPadding(TargetCXXABI ABI, const CXXRecordDecl *RD) {
   llvm_unreachable("bad tail-padding use kind");
 }
 
-static bool isMsLayout(const ASTContext &Context) {
-  return Context.getTargetInfo().getCXXABI().isMicrosoft();
+static bool isMsLayout(const ASTContext &Context, const RecordDecl *RD) {
+  return Context.getTargetInfo().getCXXABI().isMicrosoft() &&
+         RD->isMsStruct(Context);
 }
 
 // This section contains an implementation of struct layout that is, up to the
@@ -3341,7 +3342,7 @@ ASTContext::getASTRecordLayout(const RecordDecl *D) const {
 
   const ASTRecordLayout *NewEntry = nullptr;
 
-  if (isMsLayout(*this)) {
+  if (isMsLayout(*this, D)) {
     if (const auto *RD = dyn_cast<CXXRecordDecl>(D)) {
       EmptySubobjectMap EmptySubobjects(*this, RD);
       MicrosoftRecordLayoutBuilder Builder(*this, &EmptySubobjects);
@@ -3617,7 +3618,7 @@ static void DumpRecordLayout(raw_ostream &OS, const RecordDecl *RD,
     bool HasOwnVBPtr = Layout.hasOwnVBPtr();
 
     // Vtable pointer.
-    if (CXXRD->isDynamicClass() && !PrimaryBase && !isMsLayout(C)) {
+    if (CXXRD->isDynamicClass() && !PrimaryBase && !isMsLayout(C, RD)) {
       PrintOffset(OS, Offset, IndentLevel);
       OS << '(' << *RD << " vtable pointer)\n";
     } else if (HasOwnVFPtr) {
@@ -3717,7 +3718,7 @@ static void DumpRecordLayout(raw_ostream &OS, const RecordDecl *RD,
 
   PrintIndentNoOffset(OS, IndentLevel - 1);
   OS << "[sizeof=" << Layout.getSize().getQuantity();
-  if (CXXRD && !isMsLayout(C))
+  if (CXXRD && !isMsLayout(C, RD))
     OS << ", dsize=" << Layout.getDataSize().getQuantity();
   OS << ", align=" << Layout.getAlignment().getQuantity();
   if (C.getTargetInfo().defaultsToAIXPowerAlignment())
@@ -3756,7 +3757,7 @@ void ASTContext::DumpRecordLayout(const RecordDecl *RD, raw_ostream &OS,
   OS << "\nLayout: ";
   OS << "<ASTRecordLayout\n";
   OS << "  Size:" << toBits(Info.getSize()) << "\n";
-  if (!isMsLayout(*this))
+  if (!isMsLayout(*this, RD))
     OS << "  DataSize:" << toBits(Info.getDataSize()) << "\n";
   OS << "  Alignment:" << toBits(Info.getAlignment()) << "\n";
   if (Target->defaultsToAIXPowerAlignment())
diff --git a/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp b/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp
index cbfa79e10bfefcc..88a766adfe82a1c 100644
--- a/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp
+++ b/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp
@@ -104,10 +104,7 @@ struct CGRecordLowering {
   /// fields of the same formal type.  We want to emit a layout with
   /// these discrete storage units instead of combining them into a
   /// continuous run.
-  bool isDiscreteBitFieldABI() {
-    return Context.getTargetInfo().getCXXABI().isMicrosoft() ||
-           D->isMsStruct(Context);
-  }
+  bool isDiscreteBitFieldABI() { return D->isMsStruct(Context); }
 
   /// Helper function to check if we are targeting AAPCS.
   bool isAAPCS() const {
@@ -122,7 +119,8 @@ struct CGRecordLowering {
   ///
   /// Note specifically that the ms_struct attribute doesn't change this.
   bool isOverlappingVBaseABI() {
-    return !Context.getTargetInfo().getCXXABI().isMicrosoft();
+    return !Context.getTargetInfo().getCXXABI().isMicrosoft() ||
+           !D->isMsStruct(Context);
   }
 
   /// Wraps llvm::Type::getIntNTy with some implicit arguments.
diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index 79f7fba22570746..2602d1f6bedee9f 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -5645,9 +5645,13 @@ 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 (Args.hasArg(options::OPT_mms_bitfields) ||
+      Args.hasArg(options::OPT_mno_ms_bitfields)) {
+    if (Args.hasFlag(options::OPT_mms_bitfields, options::OPT_mno_ms_bitfields,
+                     false))
+      CmdArgs.push_back("-mms-bitfields");
+    else
+      CmdArgs.push_back("-mno-ms-bitfields");
   }
 
   if (Triple.isWindowsGNUEnvironment()) {
diff --git a/clang/lib/Frontend/CompilerInvocation.cpp b/clang/lib/Frontend/CompilerInvocation.cpp
index 637c6a35af6532b..a0576441c5d24a7 100644
--- a/clang/lib/Frontend/CompilerInvocation.cpp
+++ b/clang/lib/Frontend/CompilerInvocation.cpp
@@ -3619,6 +3619,13 @@ void CompilerInvocationBase::GenerateLangArgs(const LangOptions &Opts,
     GenerateArg(Consumer, OPT_fcxx_abi_EQ,
                 TargetCXXABI::getSpelling(*Opts.CXXABI));
 
+  if (Opts.MSBitfields.has_value()) {
+    if (Opts.MSBitfields.value())
+      GenerateArg(Consumer, OPT_mms_bitfields);
+    else
+      GenerateArg(Consumer, OPT_mno_ms_bitfields);
+  }
+
   if (Opts.RelativeCXXABIVTables)
     GenerateArg(Consumer, OPT_fexperimental_relative_cxx_abi_vtables);
   else
@@ -4153,6 +4160,11 @@ bool CompilerInvocation::ParseLangArgs(LangOptions &Opts, ArgList &Args,
                    options::OPT_fno_experimental_relative_cxx_abi_vtables,
                    TargetCXXABI::usesRelativeVTables(T));
 
+  if (Args.hasArg(options::OPT_mms_bitfields) ||
+      Args.hasArg(options::OPT_mno_ms_bitfields))
+    Opts.MSBitfields = Args.hasFlag(options::OPT_mms_bitfields,
+                                    options::OPT_mno_ms_bitfields, false);
+
   // RTTI is on by default.
   bool HasRTTI = Args.hasFlag(options::OPT_frtti, options::OPT_fno_rtti, true);
   Opts.OmitVTableRTTI =
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index a8bad12b670fc75..f925972e7d3d104 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -18192,9 +18192,7 @@ 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;
diff --git a/clang/lib/Sema/SemaDeclCXX.cpp b/clang/lib/Sema/SemaDeclCXX.cpp
index be79defbbfac6f1..9d42157534eda47 100644
--- a/clang/lib/Sema/SemaDeclCXX.cpp
+++ b/clang/lib/Sema/SemaDeclCXX.cpp
@@ -7265,7 +7265,14 @@ void Sema::CheckCompletedCXXClass(Scope *S, CXXRecordDecl *Record) {
   // 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 &&
+  // We do not support ABI compatibility mode in the other direction,
+  // however. Whenever we encounter type with gcc_struct attribute on a
+  // target with Microsoft C++ ABI, we switch layout completely to
+  // Itanium. Therefore, record ABI would be fully compatible and there
+  // is no point in emitting the diagnostic.
+  if (!Context.getLangOpts().MSBitfields.has_value() &&
+      Record->isMsStruct(Context) &&
+      !Context.getTargetInfo().defaultsToMsStruct() &&
       (Record->isPolymorphic() || Record->getNumBases())) {
     Diag(Record->getLocation(), diag::warn_cxx_ms_struct);
   }
diff --git a/clang/test/CodeGenCXX/ms-bitfield-class-layout.cpp b/clang/test/CodeGenCXX/ms-bitfield-class-layout.cpp
new file mode 100644
index 000000000000000..6fefd3f608bd051
--- /dev/null
+++ b/clang/test/CodeGenCXX/ms-bitfield-class-layout.cpp
@@ -0,0 +1,41 @@
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-none %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=ITANIUM %s < %t
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-windows-msvc %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=MSVC %s < %t
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-windows-gnu %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=ITANIUM %s < %t
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-none -mms-bitfields %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=ITANIUM %s < %t
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-windows-msvc -mms-bitfields %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=MSVC %s < %t
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-windows-gnu -mms-bitfields %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=ITANIUM %s < %t
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-none -mno-ms-bitfields %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=ITANIUM %s < %t
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-windows-msvc -mno-ms-bitfields %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=ITANIUM-EXCEPT-FOR-S3 %s < %t
+// RUN: %clang_cc1 -fdump-record-layouts -Wno-incompatible-ms-struct -emit-llvm-only -triple x86_64-none-windows-gnu -mno-ms-bitfields %s -o %t.ll > %t
+// RUN: FileCheck --check-prefix=ITANIUM %s < %t
+
+// ITANIUM:(S1 vtable pointer)
+// ITANIUM-EXCEPT-FOR-S3:(S1 vtable pointer)
+// MSVC:(S1 vftable pointer)
+struct S1 {
+    virtual void f() {}
+};
+
+// ITANIUM:(S2 vtable pointer)
+// ITANIUM-EXCEPT-FOR-S3:(S2 vtable pointer)
+// MSVC:(S2 vtable pointer)
+struct [[gnu::gcc_struct]] S2 {
+    virtual void f() {}
+};
+
+// ITANIUM:(S3 vtable pointer)
+// ITANIUM-EXCEPT-FOR-S3:(S3 vftable pointer)
+// MSVC:(S3 vftable pointer)
+struct [[gnu::ms_struct]] S3 {
+    virtual void f() {}
+};
+
+void use(S1 s1, S2 s2, S3 s3) { s1.f(); s2.f(); s3.f(); }
diff --git a/clang/test/CodeGenCXX/ms_struct-gcc_struct.cpp b/clang/test/CodeGenCXX/ms_struct-gcc_struct.cpp
new file mode 100644
index 000000000000000..803ca59ffca584b
--- /dev/null
+++ b/clang/test/CodeGenCXX/ms_struct-gcc_struct.cpp
@@ -0,0 +1,29 @@
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-none-none -verify=default-itanium %s
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-none-windows-msvc -verify=default-msvc %s
+// RUN: %clang_cc1 -emit-llvm-only -triple x86_64-none-windows-gnu -verify=default-msvc %s
+// RUN: %clang_cc1 -mms-bitfields -emit-llvm-only -triple x86_64-none-none -verify=default-msvc %s
+// RUN: %clang_cc1 -mms-bitfields -emit-llvm-only -triple x86_64-none-windows-msvc -verify=default-msvc %s
+// RUN: %clang_cc1 -mms-bitfields -emit-llvm-only -triple x86_64-none-windows-gnu -verify=default-msvc %s
+// RUN: %clang_cc1 -mno-ms-bitfields -emit-llvm-only -triple x86_64-none-none -verify=default-itanium %s
+// RUN: %clang_cc1 -mno-ms-bitfields -emit-llvm-only -triple x86_64-none-windows-msvc -verify=default-itanium %s
+// RUN: %clang_cc1 -mno-ms-bitfields -emit-llvm-only -triple x86_64-none-windows-gnu -verify=default-itanium %s
+
+struct [[gnu::gcc_struct]] S1 {
+    unsigned a : 1;
+    unsigned long long b : 1;
+};
+
+struct S2 {
+    unsigned a : 1;
+    unsigned long long b : 1;
+};
+
+struct [[gnu::ms_struct]] S3 {
+    unsigned a : 1;
+    unsigned long long b : 1;
+};
+
+int a1[sizeof(S1) == 8 ? 1 : -1];
+int a2[sizeof(S2) == 8 ? 1 : -1]; // default-msvc-error {{'a2' declared as an array with a negative size}}
+int a3[sizeof(S2) == 16 ? 1 : -1]; // default-itanium-error {{'a3' declared as an array with a negative size}}
+int a4[sizeof(S3) == 16 ? 1 : -1];
diff --git a/clang/test/Driver/ms-bitfields.c b/clang/test/Driver/ms-bitfields.c
index 031ed41e2aad678..4cd3f1380e41acf 100644
--- a/clang/test/Driver/ms-bitfields.c
+++ b/clang/test/Driver/ms-bitfields.c
@@ -1,8 +1,13 @@
-// 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 -### -target x86_64-linux-gnu %s 2>&1 | FileCheck %s -check-prefix=DEFAULT-MSBITFIELDS
+// RUN: %clang -### -target x86_64-windows-gnu %s 2>&1 | FileCheck %s -check-prefix=DEFAULT-MSBITFIELDS
+// RUN: %clang -### -target x86_64-windows-msvc %s 2>&1 | FileCheck %s -check-prefix=DEFAULT-MSBITFIELDS
+// RUN: %clang -### -mms-bitfields %s 2>&1 | FileCheck %s -check-prefix=MSBITFIELDS
+// RUN: %clang -### -mno-ms-bitfields %s 2>&1 | FileCheck %s -check-prefix=NO-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
 
 // MSBITFIELDS: -mms-bitfields
-// NO-MSBITFIELDS-NOT: -mms-bitfields
+// NO-MSBITFIELDS: -mno-ms-bitfields
+// DEFAULT-MSBITFIELDS-NOT: -mms-bitfields
+// DEFAULT-MSBITFIELDS-NOT: -mno-ms-bitfields

github-actions · 2023-11-03T06:06:54Z

✅ With the latest revision this PR passed the C/C++ code formatter.

DanShaders · 2023-11-03T06:11:39Z

And, I guess, CC @erichkeane

erichkeane

This needs attribute documentation and a release note.

Additionally, I think we should prohibit using this in Microsoft mode (that is, diagnose that) rather than having it be a no-op. We can enable it in the future, but if we permit it, folks will assume it works.

clang/include/clang/Basic/Attr.td

erichkeane

Docs need work, code is fine from what I can tell, but @efriedma-quic or @rjmccall need to take a look at the ABI components.

clang/docs/ReleaseNotes.rst

clang/include/clang/Basic/DiagnosticASTKinds.td

rjmccall · 2023-11-14T19:48:19Z

clang/lib/AST/Decl.cpp

Is there a reason we don't just set the lang opt correctly for the target? I'm pretty sure language option defaults depending on target settings is well-established.

But there is no TargetInfo object when I set MSBitfields in CompilerInvocation::ParseLangArgs.

And, for example, for C++ ABI we do the same:

llvm-project/clang/lib/AST/ASTContext.cpp

Lines 813 to 816 in d19616f

TargetCXXABI::Kind ASTContext::getCXXABIKind() const {

auto Kind = getTargetInfo().getCXXABI().getKind();

return getLangOpts().CXXABI.value_or(Kind);

}

You have a target triple, and you have the base CXXABI. What information are you looking to get from the TargetInfo?

I still don't quite understand what you want.

Yes, I can compute default value while parsing arguments, but I will duplicate logic of ASTContext::getCXXABIKind. And on top of that, I won't be able to remove TargetInfo::defaultsToMsStruct either, since I use it in SemaDeclCxx.

I always prefer to have less complicated logic in the options types, because I've seen it get out of hand. I don't feel too strongly about this, though, and I do see how you have uses of the components feeding into it.

clang/lib/Sema/SemaDeclCXX.cpp

DanShaders · 2023-11-15T18:42:58Z

@rjmccall Would you mind merging this then? (I don't have write access)

clang/include/clang/Basic/Attr.td

clang/include/clang/Basic/AttrDocs.td

AaronBallman

The attribute bits LGTM, but I added some more reviewers to double-check the driver changes.

rjmccall · 2023-11-15T21:51:33Z

I agree with the Sema/AST-level LGTM (but also don't feel comfortable approving the driver changes)

MaskRay · 2023-11-28T05:31:22Z

clang/lib/Frontend/CompilerInvocation.cpp

If we use BoolOption in Options.td, we won't need this custom code.

MaskRay · 2023-11-28T05:33:21Z

clang/include/clang/Driver/Options.td

The Driver and Frontend changes seem unneeded? -mno-ms-bitfields can remain a driver-only option that is not in CC1.

It's a convention that Driver supports both -mxxx and -mno-xxx while only one is supported by CC1.

Once again, yes, I can theoretically use BoolOption and leave everything as it is. However, this requires computing default value for -mms-bitfields in driver, which is possible but will result in code duplication and is in general less clean that the current solution. I provided some more details in the reply to rjmccall above.

In case you are strongly against breaking the convention, I can alteranatively make both -mms-bitfields and -mno-ms-bitfields driver-only options and invent a new enum option for cc1 along the lines of -flayout-compatiblity-type={default,microsoft,itanium}. I should note that this would only slightly change the way I pass arguments from the driver to cc1.

DanShaders · 2023-11-29T06:56:24Z

@MaskRay, I figured the alternative solution with the completely decoupled driver/frontend flags might actually be better than my initial approach. The only issue I see with it is that it requires s/-mms-bitfields/-fms-layout-compatibility=microsoft/g in quite a large number of tests which invoke frontend directly.

Would you mind taking a look at this again?

P.S.: Probably, -fbitfield-layout-compatibility={microsoft,itanium,default} or -fms-layout-compatibility={on,off,default} are more representative names for the frontend option.
I force-pushed the branch only to rebase on top of the latest main. The only changes are two last commits.

DanShaders · 2024-01-13T17:00:15Z

@MaskRay Bump

DanShaders · 2024-03-13T02:18:28Z

@rjmccall Re asking GCC developers about gcc_struct on non-x86: https://gcc.gnu.org/pipermail/gcc/2024-January/243154.html. Either GCC devs aren't really worried about this or I can't properly write emails (what's totally possible).

@MaskRay Any chance you can look at this PR again? Will you be more happy if instead of -fms-layout-compatibility there will be a copy of TargetInfo::defaultsToMsStruct in the driver?

rjmccall · 2024-03-13T02:22:50Z

@rjmccall Re asking GCC developers about gcc_struct on non-x86: https://gcc.gnu.org/pipermail/gcc/2024-January/243154.html. Either GCC devs aren't really worried about this or I can't properly write emails (what's totally possible).

Alright, well, we tried. I think rolling forward with what your proposal on non-x86 platforms is reasonable.

mstorsjo · 2024-03-13T08:23:47Z

@rjmccall Re asking GCC developers about gcc_struct on non-x86: https://gcc.gnu.org/pipermail/gcc/2024-January/243154.html. Either GCC devs aren't really worried about this or I can't properly write emails (what's totally possible).

Alright, well, we tried. I think rolling forward with what your proposal on non-x86 platforms is reasonable.

Yes, I don't foresee any issue with it. There's a proposed patchset to GCC for adding support for AArch64 on Windows, but it's all very early. So that shouldn't be taken as reference for how finer details should work in the end anyway. Even if they don't support these attributes from the start, the probably should/will in the end.

DanShaders · 2024-07-11T18:10:30Z

Ping @MaskRay

Changes in the force-push:

rebased on top of main;
moved defaultsToMsStruct to ASTContext since it also depends on auxiliary target.

mstorsjo · 2024-07-11T20:23:23Z

Ping @MaskRay

Changes in the force-push:

rebased on top of main;

moved defaultsToMsStruct to ASTContext since it also depends on auxiliary target.

Please update the subject of the PR - this doesn't stub out the attribute, it implements it, if I understand things correctly?

DanShaders · 2024-07-11T20:36:53Z

Fair enough

mstorsjo

I had a brief look over this, and it looks reasonable to me, although I'm not anywhere near competent enough to properly review it from the Clang perspective and approve it.

I don't see any added test that actually exercises the new gcc_struct attribute anywhere though - I'd like to see that, in a test that runs with -mms-bitfields or similar, and shows what effect the attribute has in that context.

pbo-linaro · 2024-11-21T18:53:34Z

Hi all, any update on this PR?
QEMU still actively block compiling with clang on windows for this reason, by failing at configure step, and it was decided to wait for official gcc_struct support in llvm.

DanShaders · 2024-11-22T20:06:02Z

I won't have any free time to revive this until after Christmas.

This commit implements gcc_struct attribute with the behavior similar to one in GCC. Current behavior is as follows: When ItaniumRecordLayoutBuilder is used, [[gcc_struct]] will locally cancel the effect of -mms-bitfields on a record. If -mms-bitfields is not supplied and is not a default behavior on a target, [[gcc_struct]] will be a no-op. This should provide enough compatibility with GCC. If C++ ABI is "Microsoft", [[gcc_struct]] will currently always produce a diagnostic, since support for it is not yet implemented in MicrosoftRecordLayoutBuilder. Note, however, that all the infrastructure is ready for the future implementation. In particular, check for default value of -mms-bitfields is moved from a driver to ASTContext, since it now non-trivially depends on other supplied flags. This also, unfortunately, makes it impossible to use usual argument parsing for -m{no-,}ms-bitfields. The patch doesn't introduce any backwards-incompatible changes, except for situations when cc1 is called directly with -mms-bitfields option. Work towards llvm#24757

DanShaders · 2025-01-13T21:11:57Z

@MaskRay Biannual ping

Rebased on top of main and actually added test for attribute itself which uncovered a bug :)

mstorsjo · 2025-01-13T21:18:38Z

Why the change from -mms-bitfields to -fms-layout-compatibility=microsoft in tests?

DanShaders · 2025-01-13T21:29:50Z

Short story: because otherwise -mno-ms-bitfields gets silently ignored for x86_64-pc-windows-msvc and I want clang to be very explicit about ignoring -mno-ms-bitfields. The user-facing flags remain the same (-m{no-,}ms-bitfields), the only time one ever encounters -fms-layout-compatibility= is when directly invoking frontend which we happen to do in tests.

More details here and in the following comment.

mstorsjo · 2025-01-13T21:35:34Z

Ah, I see, I missed that the changed tests were on the -cc1 level.

clang/test/CodeGen/gcc_struct.c

clang/include/clang/Basic/DiagnosticASTKinds.td

DanShaders · 2025-01-21T20:36:52Z

@MaskRay 7th ping :)

efriedma-quic

LGTM

MaskRay · 2025-01-25T22:11:47Z

clang/lib/Driver/ToolChains/Clang.cpp

-  if (Args.hasFlag(options::OPT_mms_bitfields, options::OPT_mno_ms_bitfields,
-                   Triple.isWindowsGNUEnvironment())) {
-    CmdArgs.push_back("-mms-bitfields");
+  if (Args.hasArg(options::OPT_mms_bitfields) ||


If the LangOption default varies across targets, and you want to forward the driver option to clang -cc1,

if (auto *A = Args.getLastArg(..., ...)) { if (A->matches(...)) CmdArgs.push_back... else CmdArgs.push_back... }

lygstate · 2025-08-10T15:56:46Z

mesa for clang64 also need this

./../src/compiler/glsl_types.c:749:8: warning: unknown attribute 'gcc_struct' ignored [-Wunknown-attributes]
  749 | struct PACKED explicit_matrix_key {
      |        ^~~~~~
../../src/util/macros.h:214:35: note: expanded from macro 'PACKED'
  214 | #    define PACKED __attribute__((gcc_struct,__packed__))

lygstate · 2025-08-10T17:24:29Z

@DanShaders do you still have time on this?

llvmbot added clang Clang issues not falling into any other category clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:codegen IR generation bugs: mangling, exceptions, etc. labels Nov 3, 2023

DanShaders force-pushed the gcc-struct branch from 1912ae4 to ea8a7ea Compare November 3, 2023 06:02

DanShaders force-pushed the gcc-struct branch 3 times, most recently from 7e268c2 to 0d6728f Compare November 4, 2023 01:33

DanShaders changed the title ~~[clang] Implement gcc_struct attribute~~ [clang] Stub out gcc_struct attribute Nov 4, 2023

DanShaders force-pushed the gcc-struct branch from 0d6728f to 1912422 Compare November 4, 2023 01:53

erichkeane reviewed Nov 13, 2023

View reviewed changes

clang/include/clang/Basic/Attr.td Outdated Show resolved Hide resolved

DanShaders force-pushed the gcc-struct branch from 1912422 to 1bfe264 Compare November 14, 2023 01:41

DanShaders requested a review from erichkeane November 14, 2023 07:13

erichkeane reviewed Nov 14, 2023

View reviewed changes

clang/docs/ReleaseNotes.rst Outdated Show resolved Hide resolved

clang/docs/ReleaseNotes.rst Outdated Show resolved Hide resolved

clang/include/clang/Basic/DiagnosticASTKinds.td Outdated Show resolved Hide resolved

rjmccall reviewed Nov 14, 2023

View reviewed changes

DanShaders requested a review from rjmccall November 15, 2023 16:37

AaronBallman reviewed Nov 15, 2023

View reviewed changes

clang/include/clang/Basic/Attr.td Outdated Show resolved Hide resolved

clang/include/clang/Basic/AttrDocs.td Outdated Show resolved Hide resolved

AaronBallman requested review from MaskRay and jansvoboda11 November 15, 2023 19:11

AaronBallman reviewed Nov 15, 2023

View reviewed changes

MaskRay reviewed Nov 28, 2023

View reviewed changes

DanShaders force-pushed the gcc-struct branch 2 times, most recently from 7baf4c5 to fc6a512 Compare November 29, 2023 02:43

DanShaders requested a review from MaskRay November 29, 2023 06:56

DanShaders force-pushed the gcc-struct branch from b26dd88 to d0484b0 Compare March 13, 2024 02:02

DanShaders force-pushed the gcc-struct branch from d0484b0 to 39f27a9 Compare July 11, 2024 18:08

DanShaders changed the title ~~[clang] Stub out gcc_struct attribute~~ [clang] Implement gcc_struct attribute on Itanium targets Jul 11, 2024

mstorsjo mentioned this pull request Aug 16, 2024

qemu: update to 9.1.0 msys2/MINGW-packages#21540

Merged

mstorsjo reviewed Aug 16, 2024

View reviewed changes

heljkon mentioned this pull request Sep 12, 2024

virglrenderer: Update to 1.1.0 msys2/MINGW-packages#21903

Merged

CAD97 mentioned this pull request Nov 1, 2024

[RFC] Allow packed types to transitively contain aligned types rust-lang/rfcs#3718

Open

DanShaders added 2 commits January 11, 2025 22:39

fixup! Add test for gcc_struct

dea03c8

DanShaders force-pushed the gcc-struct branch from 39f27a9 to dea03c8 Compare January 13, 2025 21:07

efriedma-quic reviewed Jan 13, 2025

View reviewed changes

clang/test/CodeGen/gcc_struct.c Outdated Show resolved Hide resolved

clang/include/clang/Basic/DiagnosticASTKinds.td Show resolved Hide resolved

tests for todo diagnostic & static_assert

f3254ce

efriedma-quic approved these changes Jan 21, 2025

View reviewed changes

MaskRay reviewed Jan 25, 2025

View reviewed changes

	TargetCXXABI::Kind ASTContext::getCXXABIKind() const {
	auto Kind = getTargetInfo().getCXXABI().getKind();
	return getLangOpts().CXXABI.value_or(Kind);
	}

[clang] Implement gcc_struct attribute on Itanium targets #71148

Are you sure you want to change the base?

[clang] Implement gcc_struct attribute on Itanium targets #71148

Uh oh!

Conversation

DanShaders commented Nov 3, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Nov 3, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Nov 3, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DanShaders commented Nov 3, 2023

Uh oh!

erichkeane left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

erichkeane left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

rjmccall Nov 14, 2023

Choose a reason for hiding this comment

Uh oh!

DanShaders Nov 14, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

DanShaders Nov 14, 2023

Choose a reason for hiding this comment

Uh oh!

rjmccall Nov 15, 2023

Choose a reason for hiding this comment

Uh oh!

DanShaders Nov 15, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rjmccall Nov 15, 2023

Choose a reason for hiding this comment

Uh oh!

Uh oh!

DanShaders commented Nov 15, 2023

Uh oh!

Uh oh!

Uh oh!

AaronBallman left a comment

Choose a reason for hiding this comment

Uh oh!

rjmccall commented Nov 15, 2023

Uh oh!

MaskRay Nov 28, 2023

Choose a reason for hiding this comment

Uh oh!

MaskRay Nov 28, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

DanShaders Nov 28, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

DanShaders commented Nov 29, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DanShaders commented Jan 13, 2024

Uh oh!

DanShaders commented Mar 13, 2024

Uh oh!

rjmccall commented Mar 13, 2024

Uh oh!

mstorsjo commented Mar 13, 2024

Uh oh!

DanShaders commented Jul 11, 2024

Uh oh!

DanShaders commented Nov 3, 2023 •

edited

Loading

llvmbot commented Nov 3, 2023 •

edited

Loading

github-actions bot commented Nov 3, 2023 •

edited

Loading

DanShaders Nov 14, 2023 •

edited

Loading

DanShaders Nov 15, 2023 •

edited

Loading

MaskRay Nov 28, 2023 •

edited

Loading

DanShaders Nov 28, 2023 •

edited

Loading

DanShaders commented Nov 29, 2023 •

edited

Loading

lygstate commented Aug 10, 2025 •

edited

Loading