[llvm] Reduce memory footprint of Debug metadata nodes (#71227)

Using a combination of reordering fields and using empty SubclassData32
/ SubclassData1, it's possible to improve the size of data structures
used to store debug info in the IR:

Before:

DILexicalBlock: 24
DILexicalBlockFile: 24
DIModule: 24
DITemplateParameter: 24
DICommonBlock: 24
DIMacro: 24
DICompileUnit: 56
DIType: 48
DINamespace: 24
DIVariable: 24
DIGlobalVariable: 32
DILocalVariable: 32
DILabel: 24

After:

DILexicalBlock: 24
DILexicalBlockFile: 16
DIModule: 16
DITemplateParameter: 16
DICommonBlock: 16
DIMacro: 16
DICompileUnit: 48
DIType: 40
DINamespace: 16
DIVariable: 24
DIGlobalVariable: 24
DILocalVariable: 32
DILabel: 16

2 files changed, 90 insertions, 58 deletions

diff --git a/llvm/include/llvm/IR/DebugInfoMetadata.h b/llvm/include/llvm/IR/DebugInfoMetadata.h
index 1fe0543..3d9ee1c 100644
--- a/llvm/include/llvm/IR/DebugInfoMetadata.h
+++ b/llvm/include/llvm/IR/DebugInfoMetadata.h
@@ -128,6 +128,8 @@ public:
 /// A metadata node with a DWARF tag (i.e., a constant named \c DW_TAG_*,
 /// defined in llvm/BinaryFormat/Dwarf.h).  Called \a DINode because it's
 /// potentially used for non-DWARF output.
+///
+/// Uses the SubclassData16 Metadata slot.
 class DINode : public MDNode {
   friend class LLVMContextImpl;
   friend class MDNode;
@@ -227,6 +229,8 @@ public:
 /// (possibly empty) null-separated \a MDString header that contains arbitrary
 /// fields.  The remaining operands are \a dwarf_operands(), and are pointers
 /// to other metadata.
+///
+/// Uses the SubclassData32 Metadata slot.
 class GenericDINode : public DINode {
   friend class LLVMContextImpl;
   friend class MDNode;
@@ -695,12 +699,13 @@ std::optional<StringRef> DIScope::getSource() const {
 /// TODO: Remove the hardcoded name and context, since many types don't use
 /// them.
 /// TODO: Split up flags.
+///
+/// Uses the SubclassData32 Metadata slot.
 class DIType : public DIScope {
   unsigned Line;
   DIFlags Flags;
   uint64_t SizeInBits;
   uint64_t OffsetInBits;
-  uint32_t AlignInBits;
 
 protected:
   DIType(LLVMContext &C, unsigned ID, StorageType Storage, unsigned Tag,
@@ -716,7 +721,7 @@ protected:
     this->Line = Line;
     this->Flags = Flags;
     this->SizeInBits = SizeInBits;
-    this->AlignInBits = AlignInBits;
+    this->SubclassData32 = AlignInBits;
     this->OffsetInBits = OffsetInBits;
   }
 
@@ -735,7 +740,7 @@ public:
 
   unsigned getLine() const { return Line; }
   uint64_t getSizeInBits() const { return SizeInBits; }
-  uint32_t getAlignInBits() const { return AlignInBits; }
+  uint32_t getAlignInBits() const { return SubclassData32; }
   uint32_t getAlignInBytes() const { return getAlignInBits() / CHAR_BIT; }
   uint64_t getOffsetInBits() const { return OffsetInBits; }
   DIFlags getFlags() const { return Flags; }
@@ -1389,13 +1394,13 @@ public:
 
 private:
   unsigned SourceLanguage;
-  bool IsOptimized;
   unsigned RuntimeVersion;
-  unsigned EmissionKind;
   uint64_t DWOId;
+  unsigned EmissionKind;
+  unsigned NameTableKind;
+  bool IsOptimized;
   bool SplitDebugInlining;
   bool DebugInfoForProfiling;
-  unsigned NameTableKind;
   bool RangesBaseAddress;
 
   DICompileUnit(LLVMContext &C, StorageType Storage, unsigned SourceLanguage,
@@ -1876,6 +1881,10 @@ public:
 /// Debug location.
 ///
 /// A debug location in source code, used for debug info and otherwise.
+///
+/// Uses the SubclassData1, SubclassData16 and SubclassData32
+/// Metadata slots.
+
 class DILocation : public MDNode {
   friend class LLVMContextImpl;
   friend class MDNode;
@@ -2161,17 +2170,20 @@ public:
   }
 };
 
+/// Debug lexical block.
+///
+/// Uses the SubclassData32 Metadata slot.
 class DILexicalBlock : public DILexicalBlockBase {
   friend class LLVMContextImpl;
   friend class MDNode;
 
-  unsigned Line;
   uint16_t Column;
 
   DILexicalBlock(LLVMContext &C, StorageType Storage, unsigned Line,
                  unsigned Column, ArrayRef<Metadata *> Ops)
-      : DILexicalBlockBase(C, DILexicalBlockKind, Storage, Ops), Line(Line),
+      : DILexicalBlockBase(C, DILexicalBlockKind, Storage, Ops),
         Column(Column) {
+    SubclassData32 = Line;
     assert(Column < (1u << 16) && "Expected 16-bit column");
   }
   ~DILexicalBlock() = default;
@@ -2206,7 +2218,7 @@ public:
 
   TempDILexicalBlock clone() const { return cloneImpl(); }
 
-  unsigned getLine() const { return Line; }
+  unsigned getLine() const { return SubclassData32; }
   unsigned getColumn() const { return Column; }
 
   static bool classof(const Metadata *MD) {
@@ -2218,12 +2230,11 @@ class DILexicalBlockFile : public DILexicalBlockBase {
   friend class LLVMContextImpl;
   friend class MDNode;
 
-  unsigned Discriminator;
-
   DILexicalBlockFile(LLVMContext &C, StorageType Storage,
                      unsigned Discriminator, ArrayRef<Metadata *> Ops)
-      : DILexicalBlockBase(C, DILexicalBlockFileKind, Storage, Ops),
-        Discriminator(Discriminator) {}
+      : DILexicalBlockBase(C, DILexicalBlockFileKind, Storage, Ops) {
+    SubclassData32 = Discriminator;
+  }
   ~DILexicalBlockFile() = default;
 
   static DILexicalBlockFile *getImpl(LLVMContext &Context, DILocalScope *Scope,
@@ -2255,7 +2266,7 @@ public:
                     (Scope, File, Discriminator))
 
   TempDILexicalBlockFile clone() const { return cloneImpl(); }
-  unsigned getDiscriminator() const { return Discriminator; }
+  unsigned getDiscriminator() const { return SubclassData32; }
 
   static bool classof(const Metadata *MD) {
     return MD->getMetadataID() == DILexicalBlockFileKind;
@@ -2338,12 +2349,13 @@ DILocation::cloneByMultiplyingDuplicationFactor(unsigned DF) const {
   return std::nullopt;
 }
 
+/// Debug lexical block.
+///
+/// Uses the SubclassData1 Metadata slot.
 class DINamespace : public DIScope {
   friend class LLVMContextImpl;
   friend class MDNode;
 
-  unsigned ExportSymbols : 1;
-
   DINamespace(LLVMContext &Context, StorageType Storage, bool ExportSymbols,
               ArrayRef<Metadata *> Ops);
   ~DINamespace() = default;
@@ -2373,7 +2385,7 @@ public:
 
   TempDINamespace clone() const { return cloneImpl(); }
 
-  bool getExportSymbols() const { return ExportSymbols; }
+  bool getExportSymbols() const { return SubclassData1; }
   DIScope *getScope() const { return cast_or_null<DIScope>(getRawScope()); }
   StringRef getName() const { return getStringOperand(2); }
 
@@ -2387,11 +2399,11 @@ public:
 
 /// Represents a module in the programming language, for example, a Clang
 /// module, or a Fortran module.
+///
+/// Uses the SubclassData1 and SubclassData32 Metadata slots.
 class DIModule : public DIScope {
   friend class LLVMContextImpl;
   friend class MDNode;
-  unsigned LineNo;
-  bool IsDecl;
 
   DIModule(LLVMContext &Context, StorageType Storage, unsigned LineNo,
            bool IsDecl, ArrayRef<Metadata *> Ops);
@@ -2443,8 +2455,8 @@ public:
   StringRef getConfigurationMacros() const { return getStringOperand(3); }
   StringRef getIncludePath() const { return getStringOperand(4); }
   StringRef getAPINotesFile() const { return getStringOperand(5); }
-  unsigned getLineNo() const { return LineNo; }
-  bool getIsDecl() const { return IsDecl; }
+  unsigned getLineNo() const { return SubclassData32; }
+  bool getIsDecl() const { return SubclassData1; }
 
   Metadata *getRawScope() const { return getOperand(1); }
   MDString *getRawName() const { return getOperandAs<MDString>(2); }
@@ -2460,13 +2472,15 @@ public:
 };
 
 /// Base class for template parameters.
+///
+/// Uses the SubclassData1 Metadata slot.
 class DITemplateParameter : public DINode {
 protected:
-  bool IsDefault;
-
   DITemplateParameter(LLVMContext &Context, unsigned ID, StorageType Storage,
                       unsigned Tag, bool IsDefault, ArrayRef<Metadata *> Ops)
-      : DINode(Context, ID, Storage, Tag, Ops), IsDefault(IsDefault) {}
+      : DINode(Context, ID, Storage, Tag, Ops) {
+    SubclassData1 = IsDefault;
+  }
   ~DITemplateParameter() = default;
 
 public:
@@ -2475,7 +2489,7 @@ public:
 
   MDString *getRawName() const { return getOperandAs<MDString>(0); }
   Metadata *getRawType() const { return getOperand(1); }
-  bool isDefault() const { return IsDefault; }
+  bool isDefault() const { return SubclassData1; }
 
   static bool classof(const Metadata *MD) {
     return MD->getMetadataID() == DITemplateTypeParameterKind ||
@@ -2572,9 +2586,10 @@ public:
 };
 
 /// Base class for variables.
+///
+/// Uses the SubclassData32 Metadata slot.
 class DIVariable : public DINode {
   unsigned Line;
-  uint32_t AlignInBits;
 
 protected:
   DIVariable(LLVMContext &C, unsigned ID, StorageType Storage, signed Line,
@@ -2587,7 +2602,7 @@ public:
   StringRef getName() const { return getStringOperand(1); }
   DIFile *getFile() const { return cast_or_null<DIFile>(getRawFile()); }
   DIType *getType() const { return cast_or_null<DIType>(getRawType()); }
-  uint32_t getAlignInBits() const { return AlignInBits; }
+  uint32_t getAlignInBits() const { return SubclassData32; }
   uint32_t getAlignInBytes() const { return getAlignInBits() / CHAR_BIT; }
   /// Determines the size of the variable's type.
   std::optional<uint64_t> getSizeInBits() const;
@@ -3161,9 +3176,10 @@ public:
   }
 };
 
+/// Debug common block.
+///
+/// Uses the SubclassData32 Metadata slot.
 class DICommonBlock : public DIScope {
-  unsigned LineNo;
-
   friend class LLVMContextImpl;
   friend class MDNode;
 
@@ -3205,7 +3221,7 @@ public:
   }
   StringRef getName() const { return getStringOperand(2); }
   DIFile *getFile() const { return cast_or_null<DIFile>(getRawFile()); }
-  unsigned getLineNo() const { return LineNo; }
+  unsigned getLineNo() const { return SubclassData32; }
 
   Metadata *getRawScope() const { return getOperand(0); }
   Metadata *getRawDecl() const { return getOperand(1); }
@@ -3310,12 +3326,11 @@ public:
 
 /// Label.
 ///
+/// Uses the SubclassData32 Metadata slot.
 class DILabel : public DINode {
   friend class LLVMContextImpl;
   friend class MDNode;
 
-  unsigned Line;
-
   DILabel(LLVMContext &C, StorageType Storage, unsigned Line,
           ArrayRef<Metadata *> Ops);
   ~DILabel() = default;
@@ -3353,7 +3368,7 @@ public:
   DILocalScope *getScope() const {
     return cast_or_null<DILocalScope>(getRawScope());
   }
-  unsigned getLine() const { return Line; }
+  unsigned getLine() const { return SubclassData32; }
   StringRef getName() const { return getStringOperand(1); }
   DIFile *getFile() const { return cast_or_null<DIFile>(getRawFile()); }
 
@@ -3455,15 +3470,17 @@ public:
 };
 
 /// An imported module (C++ using directive or similar).
+///
+/// Uses the SubclassData32 Metadata slot.
 class DIImportedEntity : public DINode {
   friend class LLVMContextImpl;
   friend class MDNode;
 
-  unsigned Line;
-
   DIImportedEntity(LLVMContext &C, StorageType Storage, unsigned Tag,
                    unsigned Line, ArrayRef<Metadata *> Ops)
-      : DINode(C, DIImportedEntityKind, Storage, Tag, Ops), Line(Line) {}
+      : DINode(C, DIImportedEntityKind, Storage, Tag, Ops) {
+    SubclassData32 = Line;
+  }
   ~DIImportedEntity() = default;
 
   static DIImportedEntity *getImpl(LLVMContext &Context, unsigned Tag,
@@ -3499,7 +3516,7 @@ public:
 
   TempDIImportedEntity clone() const { return cloneImpl(); }
 
-  unsigned getLine() const { return Line; }
+  unsigned getLine() const { return SubclassData32; }
   DIScope *getScope() const { return cast_or_null<DIScope>(getRawScope()); }
   DINode *getEntity() const { return cast_or_null<DINode>(getRawEntity()); }
   StringRef getName() const { return getStringOperand(2); }
@@ -3567,6 +3584,8 @@ public:
 /// \c DW_MACINFO_*, defined in llvm/BinaryFormat/Dwarf.h).  Called \a
 /// DIMacroNode
 /// because it's potentially used for non-DWARF output.
+///
+/// Uses the SubclassData16 Metadata slot.
 class DIMacroNode : public MDNode {
   friend class LLVMContextImpl;
   friend class MDNode;
@@ -3611,15 +3630,18 @@ public:
   }
 };
 
+/// Macro
+///
+/// Uses the SubclassData32 Metadata slot.
 class DIMacro : public DIMacroNode {
   friend class LLVMContextImpl;
   friend class MDNode;
 
-  unsigned Line;
-
   DIMacro(LLVMContext &C, StorageType Storage, unsigned MIType, unsigned Line,
           ArrayRef<Metadata *> Ops)
-      : DIMacroNode(C, DIMacroKind, Storage, MIType, Ops), Line(Line) {}
+      : DIMacroNode(C, DIMacroKind, Storage, MIType, Ops) {
+    SubclassData32 = Line;
+  }
   ~DIMacro() = default;
 
   static DIMacro *getImpl(LLVMContext &Context, unsigned MIType, unsigned Line,
@@ -3649,7 +3671,7 @@ public:
 
   TempDIMacro clone() const { return cloneImpl(); }
 
-  unsigned getLine() const { return Line; }
+  unsigned getLine() const { return SubclassData32; }
 
   StringRef getName() const { return getStringOperand(0); }
   StringRef getValue() const { return getStringOperand(1); }
@@ -3662,15 +3684,18 @@ public:
   }
 };
 
+/// Macro file
+///
+/// Uses the SubclassData32 Metadata slot.
 class DIMacroFile : public DIMacroNode {
   friend class LLVMContextImpl;
   friend class MDNode;
 
-  unsigned Line;
-
   DIMacroFile(LLVMContext &C, StorageType Storage, unsigned MIType,
               unsigned Line, ArrayRef<Metadata *> Ops)
-      : DIMacroNode(C, DIMacroFileKind, Storage, MIType, Ops), Line(Line) {}
+      : DIMacroNode(C, DIMacroFileKind, Storage, MIType, Ops) {
+    SubclassData32 = Line;
+  }
   ~DIMacroFile() = default;
 
   static DIMacroFile *getImpl(LLVMContext &Context, unsigned MIType,
@@ -3711,7 +3736,7 @@ public:
     replaceOperandWith(1, Elements.get());
   }
 
-  unsigned getLine() const { return Line; }
+  unsigned getLine() const { return SubclassData32; }
   DIFile *getFile() const { return cast_or_null<DIFile>(getRawFile()); }
 
   DIMacroNodeArray getElements() const {
diff --git a/llvm/lib/IR/DebugInfoMetadata.cpp b/llvm/lib/IR/DebugInfoMetadata.cpp
index 943826c..927aefb 100644
--- a/llvm/lib/IR/DebugInfoMetadata.cpp
+++ b/llvm/lib/IR/DebugInfoMetadata.cpp
@@ -915,11 +915,11 @@ DICompileUnit::DICompileUnit(LLVMContext &C, StorageType Storage,
                              bool DebugInfoForProfiling, unsigned NameTableKind,
                              bool RangesBaseAddress, ArrayRef<Metadata *> Ops)
     : DIScope(C, DICompileUnitKind, Storage, dwarf::DW_TAG_compile_unit, Ops),
-      SourceLanguage(SourceLanguage), IsOptimized(IsOptimized),
-      RuntimeVersion(RuntimeVersion), EmissionKind(EmissionKind), DWOId(DWOId),
-      SplitDebugInlining(SplitDebugInlining),
+      SourceLanguage(SourceLanguage), RuntimeVersion(RuntimeVersion),
+      DWOId(DWOId), EmissionKind(EmissionKind), NameTableKind(NameTableKind),
+      IsOptimized(IsOptimized), SplitDebugInlining(SplitDebugInlining),
       DebugInfoForProfiling(DebugInfoForProfiling),
-      NameTableKind(NameTableKind), RangesBaseAddress(RangesBaseAddress) {
+      RangesBaseAddress(RangesBaseAddress) {
   assert(Storage != Uniqued);
 }
 
@@ -1181,8 +1181,9 @@ DILexicalBlockFile *DILexicalBlockFile::getImpl(LLVMContext &Context,
 
 DINamespace::DINamespace(LLVMContext &Context, StorageType Storage,
                          bool ExportSymbols, ArrayRef<Metadata *> Ops)
-    : DIScope(Context, DINamespaceKind, Storage, dwarf::DW_TAG_namespace, Ops),
-      ExportSymbols(ExportSymbols) {}
+    : DIScope(Context, DINamespaceKind, Storage, dwarf::DW_TAG_namespace, Ops) {
+  SubclassData1 = ExportSymbols;
+}
 DINamespace *DINamespace::getImpl(LLVMContext &Context, Metadata *Scope,
                                   MDString *Name, bool ExportSymbols,
                                   StorageType Storage, bool ShouldCreate) {
@@ -1196,8 +1197,9 @@ DINamespace *DINamespace::getImpl(LLVMContext &Context, Metadata *Scope,
 DICommonBlock::DICommonBlock(LLVMContext &Context, StorageType Storage,
                              unsigned LineNo, ArrayRef<Metadata *> Ops)
     : DIScope(Context, DICommonBlockKind, Storage, dwarf::DW_TAG_common_block,
-              Ops),
-      LineNo(LineNo) {}
+              Ops) {
+  SubclassData32 = LineNo;
+}
 DICommonBlock *DICommonBlock::getImpl(LLVMContext &Context, Metadata *Scope,
                                       Metadata *Decl, MDString *Name,
                                       Metadata *File, unsigned LineNo,
@@ -1211,8 +1213,10 @@ DICommonBlock *DICommonBlock::getImpl(LLVMContext &Context, Metadata *Scope,
 
 DIModule::DIModule(LLVMContext &Context, StorageType Storage, unsigned LineNo,
                    bool IsDecl, ArrayRef<Metadata *> Ops)
-    : DIScope(Context, DIModuleKind, Storage, dwarf::DW_TAG_module, Ops),
-      LineNo(LineNo), IsDecl(IsDecl) {}
+    : DIScope(Context, DIModuleKind, Storage, dwarf::DW_TAG_module, Ops) {
+  SubclassData1 = IsDecl;
+  SubclassData32 = LineNo;
+}
 DIModule *DIModule::getImpl(LLVMContext &Context, Metadata *File,
                             Metadata *Scope, MDString *Name,
                             MDString *ConfigurationMacros,
@@ -1301,8 +1305,9 @@ DILocalVariable::getImpl(LLVMContext &Context, Metadata *Scope, MDString *Name,
 DIVariable::DIVariable(LLVMContext &C, unsigned ID, StorageType Storage,
                        signed Line, ArrayRef<Metadata *> Ops,
                        uint32_t AlignInBits)
-    : DINode(C, ID, Storage, dwarf::DW_TAG_variable, Ops), Line(Line),
-      AlignInBits(AlignInBits) {}
+    : DINode(C, ID, Storage, dwarf::DW_TAG_variable, Ops), Line(Line) {
+  SubclassData32 = AlignInBits;
+}
 std::optional<uint64_t> DIVariable::getSizeInBits() const {
   // This is used by the Verifier so be mindful of broken types.
   const Metadata *RawType = getRawType();
@@ -1328,7 +1333,9 @@ std::optional<uint64_t> DIVariable::getSizeInBits() const {
 
 DILabel::DILabel(LLVMContext &C, StorageType Storage, unsigned Line,
                  ArrayRef<Metadata *> Ops)
-    : DINode(C, DILabelKind, Storage, dwarf::DW_TAG_label, Ops), Line(Line) {}
+    : DINode(C, DILabelKind, Storage, dwarf::DW_TAG_label, Ops) {
+  SubclassData32 = Line;
+}
 DILabel *DILabel::getImpl(LLVMContext &Context, Metadata *Scope, MDString *Name,
                           Metadata *File, unsigned Line, StorageType Storage,
                           bool ShouldCreate) {