[Serialization] Remove delta encoding optimization (llvm#145670)

hokein · hokein · commit 6cd71d31a1af · 2025-07-08T09:03:45.000+02:00
See the discussion in llvm#145529. This will slightly increase the PCM size (~5%), some data (in-memory preamble size in clangd): - SemaExpr.cpp: 77MB -> 80MB - FindTarget.cpp: 71MB -> 75MB
diff --git a/clang/include/clang/Serialization/ASTReader.h b/clang/include/clang/Serialization/ASTReader.h
@@ -446,8 +446,6 @@ class ASTReader
   using ModuleReverseIterator = ModuleManager::ModuleReverseIterator;
 
 private:
-  using LocSeq = SourceLocationSequence;
-
   /// The receiver of some callbacks invoked by ASTReader.
   std::unique_ptr<ASTReaderListener> Listener;
 
@@ -2426,18 +2424,16 @@ class ASTReader
   /// Read a source location from raw form and return it in its
   /// originating module file's source location space.
   std::pair<SourceLocation, unsigned>
-  ReadUntranslatedSourceLocation(RawLocEncoding Raw,
-                                 LocSeq *Seq = nullptr) const {
-    return SourceLocationEncoding::decode(Raw, Seq);
+  ReadUntranslatedSourceLocation(RawLocEncoding Raw) const {
+    return SourceLocationEncoding::decode(Raw);
   }
 
   /// Read a source location from raw form.
-  SourceLocation ReadSourceLocation(ModuleFile &MF, RawLocEncoding Raw,
-                                    LocSeq *Seq = nullptr) const {
+  SourceLocation ReadSourceLocation(ModuleFile &MF, RawLocEncoding Raw) const {
     if (!MF.ModuleOffsetMap.empty())
       ReadModuleOffsetMap(MF);
 
-    auto [Loc, ModuleFileIndex] = ReadUntranslatedSourceLocation(Raw, Seq);
+    auto [Loc, ModuleFileIndex] = ReadUntranslatedSourceLocation(Raw);
     ModuleFile *OwningModuleFile =
         ModuleFileIndex == 0 ? &MF : MF.TransitiveImports[ModuleFileIndex - 1];
 
@@ -2465,9 +2461,9 @@ class ASTReader
 
   /// Read a source location.
   SourceLocation ReadSourceLocation(ModuleFile &ModuleFile,
-                                    const RecordDataImpl &Record, unsigned &Idx,
-                                    LocSeq *Seq = nullptr) {
-    return ReadSourceLocation(ModuleFile, Record[Idx++], Seq);
+                                    const RecordDataImpl &Record,
+                                    unsigned &Idx) {
+    return ReadSourceLocation(ModuleFile, Record[Idx++]);
   }
 
   /// Read a FileID.
@@ -2486,7 +2482,7 @@ class ASTReader
 
   /// Read a source range.
   SourceRange ReadSourceRange(ModuleFile &F, const RecordData &Record,
-                              unsigned &Idx, LocSeq *Seq = nullptr);
+                              unsigned &Idx);
 
   static llvm::BitVector ReadBitVector(const RecordData &Record,
                                        const StringRef Blob);
diff --git a/clang/include/clang/Serialization/ASTRecordReader.h b/clang/include/clang/Serialization/ASTRecordReader.h
@@ -32,7 +32,6 @@ class OMPChildren;
 class ASTRecordReader
     : public serialization::DataStreamBasicReader<ASTRecordReader> {
   using ModuleFile = serialization::ModuleFile;
-  using LocSeq = SourceLocationSequence;
 
   ASTReader *Reader;
   ModuleFile *F;
@@ -160,7 +159,7 @@ class ASTRecordReader
   TypeSourceInfo *readTypeSourceInfo();
 
   /// Reads the location information for a type.
-  void readTypeLoc(TypeLoc TL, LocSeq *Seq = nullptr);
+  void readTypeLoc(TypeLoc TL);
 
   /// Map a local type ID within a given AST file to a global type ID.
   serialization::TypeID getGlobalTypeID(serialization::TypeID LocalID) const {
@@ -287,13 +286,13 @@ class ASTRecordReader
   void readOpenACCRoutineDeclAttr(OpenACCRoutineDeclAttr *A);
 
   /// Read a source location, advancing Idx.
-  SourceLocation readSourceLocation(LocSeq *Seq = nullptr) {
-    return Reader->ReadSourceLocation(*F, Record, Idx, Seq);
+  SourceLocation readSourceLocation() {
+    return Reader->ReadSourceLocation(*F, Record, Idx);
   }
 
   /// Read a source range, advancing Idx.
-  SourceRange readSourceRange(LocSeq *Seq = nullptr) {
-    return Reader->ReadSourceRange(*F, Record, Idx, Seq);
+  SourceRange readSourceRange() {
+    return Reader->ReadSourceRange(*F, Record, Idx);
   }
 
   /// Read an arbitrary constant value, advancing Idx.
diff --git a/clang/include/clang/Serialization/ASTRecordWriter.h b/clang/include/clang/Serialization/ASTRecordWriter.h
@@ -28,7 +28,6 @@ class TypeLoc;
 /// An object for streaming information to a record.
 class ASTRecordWriter
     : public serialization::DataStreamBasicWriter<ASTRecordWriter> {
-  using LocSeq = SourceLocationSequence;
 
   ASTWriter *Writer;
   ASTWriter::RecordDataImpl *Record;
@@ -139,8 +138,8 @@ class ASTRecordWriter
   void AddFunctionDefinition(const FunctionDecl *FD);
 
   /// Emit a source location.
-  void AddSourceLocation(SourceLocation Loc, LocSeq *Seq = nullptr) {
-    return Writer->AddSourceLocation(Loc, *Record, Seq);
+  void AddSourceLocation(SourceLocation Loc) {
+    return Writer->AddSourceLocation(Loc, *Record);
   }
   void writeSourceLocation(SourceLocation Loc) {
     AddSourceLocation(Loc);
@@ -166,8 +165,8 @@ class ASTRecordWriter
   }
 
   /// Emit a source range.
-  void AddSourceRange(SourceRange Range, LocSeq *Seq = nullptr) {
-    return Writer->AddSourceRange(Range, *Record, Seq);
+  void AddSourceRange(SourceRange Range) {
+    return Writer->AddSourceRange(Range, *Record);
   }
 
   void writeBool(bool Value) {
@@ -237,7 +236,7 @@ class ASTRecordWriter
   void AddTypeSourceInfo(TypeSourceInfo *TInfo);
 
   /// Emits source location information for a type. Does not emit the type.
-  void AddTypeLoc(TypeLoc TL, LocSeq *Seq = nullptr);
+  void AddTypeLoc(TypeLoc TL);
 
   /// Emits a template argument location info.
   void AddTemplateArgumentLocInfo(TemplateArgument::ArgKind Kind,
diff --git a/clang/include/clang/Serialization/ASTWriter.h b/clang/include/clang/Serialization/ASTWriter.h
@@ -112,8 +112,6 @@ class ASTWriter : public ASTDeserializationListener,
   using TypeIdxMap = llvm::DenseMap<QualType, serialization::TypeIdx,
                                     serialization::UnsafeQualTypeDenseMapInfo>;
 
-  using LocSeq = SourceLocationSequence;
-
   /// The bitstream writer used to emit this precompiled header.
   llvm::BitstreamWriter &Stream;
 
@@ -732,16 +730,14 @@ class ASTWriter : public ASTDeserializationListener,
   void AddFileID(FileID FID, RecordDataImpl &Record);
 
   /// Emit a source location.
-  void AddSourceLocation(SourceLocation Loc, RecordDataImpl &Record,
-                         LocSeq *Seq = nullptr);
+  void AddSourceLocation(SourceLocation Loc, RecordDataImpl &Record);
 
   /// Return the raw encodings for source locations.
   SourceLocationEncoding::RawLocEncoding
-  getRawSourceLocationEncoding(SourceLocation Loc, LocSeq *Seq = nullptr);
+  getRawSourceLocationEncoding(SourceLocation Loc);
 
   /// Emit a source range.
-  void AddSourceRange(SourceRange Range, RecordDataImpl &Record,
-                      LocSeq *Seq = nullptr);
+  void AddSourceRange(SourceRange Range, RecordDataImpl &Record);
 
   /// Emit a reference to an identifier.
   void AddIdentifierRef(const IdentifierInfo *II, RecordDataImpl &Record);
diff --git a/clang/include/clang/Serialization/SourceLocationEncoding.h b/clang/include/clang/Serialization/SourceLocationEncoding.h
@@ -25,8 +25,6 @@
 //  * C: The macro bit. We rotate it to the lowest bit so that we can save some
 //  space in case the index of the module file is 0.
 //
-// Specially, if the index of the module file is 0, we allow to encode a
-// sequence of locations we store only differences between successive elements.
 //
 //===----------------------------------------------------------------------===//
 
@@ -38,7 +36,6 @@
 #define LLVM_CLANG_SERIALIZATION_SOURCELOCATIONENCODING_H
 
 namespace clang {
-class SourceLocationSequence;
 
 /// Serialized encoding of SourceLocations without context.
 /// Optimized to have small unsigned values (=> small after VBR encoding).
@@ -54,119 +51,22 @@ class SourceLocationEncoding {
   static UIntTy decodeRaw(UIntTy Raw) {
     return (Raw >> 1) | (Raw << (UIntBits - 1));
   }
-  friend SourceLocationSequence;
 
 public:
   using RawLocEncoding = uint64_t;
 
   static RawLocEncoding encode(SourceLocation Loc, UIntTy BaseOffset,
-                               unsigned BaseModuleFileIndex,
-                               SourceLocationSequence * = nullptr);
-  static std::pair<SourceLocation, unsigned>
-  decode(RawLocEncoding, SourceLocationSequence * = nullptr);
-};
-
-/// Serialized encoding of a sequence of SourceLocations.
-///
-/// Optimized to produce small values when locations with the sequence are
-/// similar. Each element can be delta-encoded against the last nonzero element.
-///
-/// Sequences should be started by creating a SourceLocationSequence::State,
-/// and then passed around as SourceLocationSequence*. Example:
-///
-///   // establishes a sequence
-///   void EmitTopLevelThing() {
-///     SourceLocationSequence::State Seq;
-///     EmitContainedThing(Seq);
-///     EmitRecursiveThing(Seq);
-///   }
-///
-///   // optionally part of a sequence
-///   void EmitContainedThing(SourceLocationSequence *Seq = nullptr) {
-///     Record.push_back(SourceLocationEncoding::encode(SomeLoc, Seq));
-///   }
-///
-///   // establishes a sequence if there isn't one already
-///   void EmitRecursiveThing(SourceLocationSequence *ParentSeq = nullptr) {
-///     SourceLocationSequence::State Seq(ParentSeq);
-///     Record.push_back(SourceLocationEncoding::encode(SomeLoc, Seq));
-///     EmitRecursiveThing(Seq);
-///   }
-///
-class SourceLocationSequence {
-  using UIntTy = SourceLocation::UIntTy;
-  using EncodedTy = uint64_t;
-  constexpr static auto UIntBits = SourceLocationEncoding::UIntBits;
-  static_assert(sizeof(EncodedTy) > sizeof(UIntTy), "Need one extra bit!");
-
-  // Prev stores the rotated last nonzero location.
-  UIntTy &Prev;
-
-  // Zig-zag encoding turns small signed integers into small unsigned integers.
-  // 0 => 0, -1 => 1, 1 => 2, -2 => 3, ...
-  static UIntTy zigZag(UIntTy V) {
-    UIntTy Sign = (V & (1 << (UIntBits - 1))) ? UIntTy(-1) : UIntTy(0);
-    return Sign ^ (V << 1);
-  }
-  static UIntTy zagZig(UIntTy V) { return (V >> 1) ^ -(V & 1); }
-
-  SourceLocationSequence(UIntTy &Prev) : Prev(Prev) {}
-
-  EncodedTy encodeRaw(UIntTy Raw) {
-    if (Raw == 0)
-      return 0;
-    UIntTy Rotated = SourceLocationEncoding::encodeRaw(Raw);
-    if (Prev == 0)
-      return Prev = Rotated;
-    UIntTy Delta = Rotated - Prev;
-    Prev = Rotated;
-    // Exactly one 33 bit value is possible! (1 << 32).
-    // This is because we have two representations of zero: trivial & relative.
-    return 1 + EncodedTy{zigZag(Delta)};
-  }
-  UIntTy decodeRaw(EncodedTy Encoded) {
-    if (Encoded == 0)
-      return 0;
-    if (Prev == 0)
-      return SourceLocationEncoding::decodeRaw(Prev = Encoded);
-    return SourceLocationEncoding::decodeRaw(Prev += zagZig(Encoded - 1));
-  }
-
-public:
-  SourceLocation decode(EncodedTy Encoded) {
-    return SourceLocation::getFromRawEncoding(decodeRaw(Encoded));
-  }
-  EncodedTy encode(SourceLocation Loc) {
-    return encodeRaw(Loc.getRawEncoding());
-  }
-
-  class State;
-};
-
-/// This object establishes a SourceLocationSequence.
-class SourceLocationSequence::State {
-  UIntTy Prev = 0;
-  SourceLocationSequence Seq;
-
-public:
-  // If Parent is provided and non-null, then this root becomes part of that
-  // enclosing sequence instead of establishing a new one.
-  State(SourceLocationSequence *Parent = nullptr)
-      : Seq(Parent ? Parent->Prev : Prev) {}
-
-  // Implicit conversion for uniform use of roots vs propagated sequences.
-  operator SourceLocationSequence *() { return &Seq; }
+                               unsigned BaseModuleFileIndex);
+  static std::pair<SourceLocation, unsigned> decode(RawLocEncoding);
 };
 
 inline SourceLocationEncoding::RawLocEncoding
 SourceLocationEncoding::encode(SourceLocation Loc, UIntTy BaseOffset,
-                               unsigned BaseModuleFileIndex,
-                               SourceLocationSequence *Seq) {
+                               unsigned BaseModuleFileIndex) {
   // If the source location is a local source location, we can try to optimize
   // the similar sequences to only record the differences.
   if (!BaseOffset)
-    return Seq ? Seq->encode(Loc) : encodeRaw(Loc.getRawEncoding());
-
+    return encodeRaw(Loc.getRawEncoding());
   if (Loc.isInvalid())
     return 0;
 
@@ -183,13 +83,11 @@ SourceLocationEncoding::encode(SourceLocation Loc, UIntTy BaseOffset,
   return Encoded;
 }
 inline std::pair<SourceLocation, unsigned>
-SourceLocationEncoding::decode(RawLocEncoding Encoded,
-                               SourceLocationSequence *Seq) {
+SourceLocationEncoding::decode(RawLocEncoding Encoded) {
   unsigned ModuleFileIndex = Encoded >> 32;
 
   if (!ModuleFileIndex)
-    return {Seq ? Seq->decode(Encoded)
-                : SourceLocation::getFromRawEncoding(decodeRaw(Encoded)),
+    return {SourceLocation::getFromRawEncoding(decodeRaw(Encoded)),
             ModuleFileIndex};
 
   Encoded &= llvm::maskTrailingOnes<RawLocEncoding>(32);
diff --git a/clang/lib/Serialization/ASTReader.cpp b/clang/lib/Serialization/ASTReader.cpp
@@ -1914,10 +1914,9 @@ bool ASTReader::ReadSLocEntry(int ID) {
   }
 
   case SM_SLOC_EXPANSION_ENTRY: {
-    LocSeq::State Seq;
-    SourceLocation SpellingLoc = ReadSourceLocation(*F, Record[1], Seq);
-    SourceLocation ExpansionBegin = ReadSourceLocation(*F, Record[2], Seq);
-    SourceLocation ExpansionEnd = ReadSourceLocation(*F, Record[3], Seq);
+    SourceLocation SpellingLoc = ReadSourceLocation(*F, Record[1]);
+    SourceLocation ExpansionBegin = ReadSourceLocation(*F, Record[2]);
+    SourceLocation ExpansionEnd = ReadSourceLocation(*F, Record[3]);
     SourceMgr.createExpansionLoc(SpellingLoc, ExpansionBegin, ExpansionEnd,
                                  Record[5], Record[4], ID,
                                  BaseOffset + Record[0]);
@@ -7072,13 +7071,10 @@ QualType ASTReader::readTypeRecord(TypeID ID) {
 namespace clang {
 
 class TypeLocReader : public TypeLocVisitor<TypeLocReader> {
-  using LocSeq = SourceLocationSequence;
-
   ASTRecordReader &Reader;
-  LocSeq *Seq;
 
-  SourceLocation readSourceLocation() { return Reader.readSourceLocation(Seq); }
-  SourceRange readSourceRange() { return Reader.readSourceRange(Seq); }
+  SourceLocation readSourceLocation() { return Reader.readSourceLocation(); }
+  SourceRange readSourceRange() { return Reader.readSourceRange(); }
 
   TypeSourceInfo *GetTypeSourceInfo() {
     return Reader.readTypeSourceInfo();
@@ -7093,8 +7089,7 @@ class TypeLocReader : public TypeLocVisitor<TypeLocReader> {
   }
 
 public:
-  TypeLocReader(ASTRecordReader &Reader, LocSeq *Seq)
-      : Reader(Reader), Seq(Seq) {}
+  TypeLocReader(ASTRecordReader &Reader) : Reader(Reader) {}
 
   // We want compile-time assurance that we've enumerated all of
   // these, so unfortunately we have to declare them first, then
@@ -7458,9 +7453,8 @@ void TypeLocReader::VisitDependentBitIntTypeLoc(
   TL.setNameLoc(readSourceLocation());
 }
 
-void ASTRecordReader::readTypeLoc(TypeLoc TL, LocSeq *ParentSeq) {
-  LocSeq::State Seq(ParentSeq);
-  TypeLocReader TLR(*this, Seq);
+void ASTRecordReader::readTypeLoc(TypeLoc TL) {
+  TypeLocReader TLR(*this);
   for (; !TL.isNull(); TL = TL.getNextTypeLoc())
     TLR.Visit(TL);
 }
@@ -10016,9 +10010,9 @@ ASTRecordReader::readNestedNameSpecifierLoc() {
 }
 
 SourceRange ASTReader::ReadSourceRange(ModuleFile &F, const RecordData &Record,
-                                       unsigned &Idx, LocSeq *Seq) {
-  SourceLocation beg = ReadSourceLocation(F, Record, Idx, Seq);
-  SourceLocation end = ReadSourceLocation(F, Record, Idx, Seq);
+                                       unsigned &Idx) {
+  SourceLocation beg = ReadSourceLocation(F, Record, Idx);
+  SourceLocation end = ReadSourceLocation(F, Record, Idx);
   return SourceRange(beg, end);
 }
 
diff --git a/clang/lib/Serialization/ASTWriter.cpp b/clang/lib/Serialization/ASTWriter.cpp
diff --git a/clang/unittests/Serialization/SourceLocationEncodingTest.cpp b/clang/unittests/Serialization/SourceLocationEncodingTest.cpp
