Another step on the way to context snapshots. We can now refer to

objects in the startup heap from a partial snapshot.  This happens
through the partial snapshot cache.  A startup snapshot and a
partial snapshot are created together so that the startup snapshot
contains the partial snapshot cache entries needed.
Review URL: http://codereview.chromium.org/548149

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3713 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/heap.cc

@@ -3344,6 +3344,11 @@ void Heap::IterateRSet(PagedSpace* space, ObjectSlotCallback copy_object_func) {
 
 void Heap::IterateRoots(ObjectVisitor* v, VisitMode mode) {
   IterateStrongRoots(v, mode);
+  IterateWeakRoots(v, mode);
+}
+
+
+void Heap::IterateWeakRoots(ObjectVisitor* v, VisitMode mode) {
   v->VisitPointer(reinterpret_cast<Object**>(&roots_[kSymbolTableRootIndex]));
   v->Synchronize("symbol_table");
   if (mode != VISIT_ALL_IN_SCAVENGE) {
@@ -3398,6 +3403,20 @@ void Heap::IterateStrongRoots(ObjectVisitor* v, VisitMode mode) {
   // Iterate over pointers being held by inactive threads.
   ThreadManager::Iterate(v);
   v->Synchronize("threadmanager");
+
+  // Iterate over the pointers the Serialization/Deserialization code is
+  // holding.
+  // During garbage collection this keeps the partial snapshot cache alive.
+  // During deserialization of the startup snapshot this creates the partial
+  // snapshot cache and deserializes the objects it refers to.  During
+  // serialization this does nothing, since the partial snapshot cache is
+  // empty.  However the next thing we do is create the partial snapshot,
+  // filling up the partial snapshot cache with objects it needs as we go.
+  SerializerDeserializer::Iterate(v);
+  // We don't do a v->Synchronize call here, because in debug mode that will
+  // output a flag to the snapshot.  However at this point the serializer and
+  // deserializer are deliberately a little unsynchronized (see above) so the
+  // checking of the sync flag in the snapshot would fail.
 }
 
 

src/heap.h

@@ -690,6 +690,8 @@ class Heap : public AllStatic {
   static void IterateRoots(ObjectVisitor* v, VisitMode mode);
   // Iterates over all strong roots in the heap.
   static void IterateStrongRoots(ObjectVisitor* v, VisitMode mode);
+  // Iterates over all the other roots in the heap.
+  static void IterateWeakRoots(ObjectVisitor* v, VisitMode mode);
 
   // Iterates remembered set of an old space.
   static void IterateRSet(PagedSpace* space, ObjectSlotCallback callback);

src/mksnapshot.cc

@@ -164,10 +164,10 @@ int main(int argc, char** argv) {
   }
   context.Dispose();
   CppByteSink sink(argv[1]);
-  i::Serializer ser(&sink);
   // This results in a somewhat smaller snapshot, probably because it gets rid
   // of some things that are cached between garbage collections.
   i::Heap::CollectAllGarbage(true);
+  i::StartupSerializer ser(&sink);
   ser.Serialize();
   return 0;
 }

src/serialize.h

@@ -185,9 +185,14 @@ class SnapshotByteSource {
   f(14, 32) \
   f(15, 36)
 
-// The SerDes class is a common superclass for Serializer and Deserializer
-// which is used to store common constants and methods used by both.
-class SerDes: public ObjectVisitor {
+// The Serializer/Deserializer class is a common superclass for Serializer and
+// Deserializer which is used to store common constants and methods used by
+// both.
+class SerializerDeserializer: public ObjectVisitor {
+ public:
+  static void Iterate(ObjectVisitor* visitor);
+  static void SetSnapshotCacheSize(int size);
+
  protected:
   enum DataType {
     RAW_DATA_SERIALIZATION = 0,
@@ -202,7 +207,8 @@ class SerDes: public ObjectVisitor {
     START_NEW_PAGE_SERIALIZATION = 37,
     NATIVES_STRING_RESOURCE = 38,
     ROOT_SERIALIZATION = 39,
-    // Free: 40-47.
+    PARTIAL_SNAPSHOT_CACHE_ENTRY = 40,
+    // Free: 41-47.
     BACKREF_SERIALIZATION = 48,
     // One per space, must be kSpaceMask aligned.
     // Free: 57-63.
@@ -227,17 +233,21 @@ class SerDes: public ObjectVisitor {
   static inline bool SpaceIsPaged(int space) {
     return space >= FIRST_PAGED_SPACE && space <= LAST_PAGED_SPACE;
   }
+
+  static int partial_snapshot_cache_length_;
+  static const int kPartialSnapshotCacheCapacity = 1024;
+  static Object* partial_snapshot_cache_[];
 };
 
 
 
 // A Deserializer reads a snapshot and reconstructs the Object graph it defines.
-class Deserializer: public SerDes {
+class Deserializer: public SerializerDeserializer {
  public:
   // Create a deserializer from a snapshot byte source.
   explicit Deserializer(SnapshotByteSource* source);
 
-  virtual ~Deserializer() { }
+  virtual ~Deserializer();
 
   // Deserialize the snapshot into an empty heap.
   void Deserialize();
@@ -249,8 +259,6 @@ class Deserializer: public SerDes {
   virtual void Synchronize(const char* tag);
 #endif
 
-  static void TearDown();
-
  private:
   virtual void VisitPointers(Object** start, Object** end);
 
@@ -272,7 +280,7 @@ class Deserializer: public SerDes {
   // (In large object space we are keeping track of individual objects
   // rather than pages.)  In new space we just need the address of the
   // first object and the others will flow from that.
-  List<Address> pages_[SerDes::kNumberOfSpaces];
+  List<Address> pages_[SerializerDeserializer::kNumberOfSpaces];
 
   SnapshotByteSource* source_;
   static ExternalReferenceDecoder* external_reference_decoder_;
@@ -300,13 +308,62 @@ class SnapshotByteSink {
 };
 
 
-class Serializer : public SerDes {
+// Mapping objects to their location after deserialization.
+// This is used during building, but not at runtime by V8.
+class SerializationAddressMapper {
+ public:
+  SerializationAddressMapper()
+      : serialization_map_(new HashMap(&SerializationMatchFun)),
+        no_allocation_(new AssertNoAllocation()) { }
+
+  ~SerializationAddressMapper() {
+    delete serialization_map_;
+    delete no_allocation_;
+  }
+
+  bool IsMapped(HeapObject* obj) {
+    return serialization_map_->Lookup(Key(obj), Hash(obj), false) != NULL;
+  }
+
+  int MappedTo(HeapObject* obj) {
+    ASSERT(IsMapped(obj));
+    return static_cast<int>(reinterpret_cast<intptr_t>(
+        serialization_map_->Lookup(Key(obj), Hash(obj), false)->value));
+  }
+
+  void AddMapping(HeapObject* obj, int to) {
+    ASSERT(!IsMapped(obj));
+    HashMap::Entry* entry =
+        serialization_map_->Lookup(Key(obj), Hash(obj), true);
+    entry->value = Value(to);
+  }
+
+ private:
+  static bool SerializationMatchFun(void* key1, void* key2) {
+    return key1 == key2;
+  }
+
+  static uint32_t Hash(HeapObject* obj) {
+    return static_cast<int32_t>(reinterpret_cast<intptr_t>(obj->address()));
+  }
+
+  static void* Key(HeapObject* obj) {
+    return reinterpret_cast<void*>(obj->address());
+  }
+
+  static void* Value(int v) {
+    return reinterpret_cast<void*>(v);
+  }
+
+  HashMap* serialization_map_;
+  AssertNoAllocation* no_allocation_;
+  DISALLOW_COPY_AND_ASSIGN(SerializationAddressMapper);
+};
+
+
+class Serializer : public SerializerDeserializer {
  public:
   explicit Serializer(SnapshotByteSink* sink);
-  // Serialize the current state of the heap.
-  void Serialize();
-  // Serialize a single object and the objects reachable from it.
-  void SerializePartial(Object** obj);
   void VisitPointers(Object** start, Object** end);
   // You can call this after serialization to find out how much space was used
   // in each space.
@@ -327,15 +384,20 @@ class Serializer : public SerDes {
   // going on.
   static void TooLateToEnableNow() { too_late_to_enable_now_ = true; }
   static bool enabled() { return serialization_enabled_; }
+  SerializationAddressMapper* address_mapper() { return &address_mapper_; }
 #ifdef DEBUG
   virtual void Synchronize(const char* tag);
 #endif
 
- private:
+ protected:
   enum ReferenceRepresentation {
     TAGGED_REPRESENTATION,      // A tagged object reference.
     CODE_TARGET_REPRESENTATION  // A reference to first instruction in target.
   };
+  static const int kInvalidRootIndex = -1;
+  virtual int RootIndex(HeapObject* heap_object) = 0;
+  virtual bool ShouldBeInThePartialSnapshotCache(HeapObject* o) = 0;
+
   class ObjectSerializer : public ObjectVisitor {
    public:
     ObjectSerializer(Serializer* serializer,
@@ -371,7 +433,12 @@ class Serializer : public SerDes {
     int bytes_processed_so_far_;
   };
 
-  void SerializeObject(Object* o, ReferenceRepresentation representation);
+  virtual void SerializeObject(Object* o,
+                               ReferenceRepresentation representation) = 0;
+  void SerializeReferenceToPreviousObject(
+      int space,
+      int address,
+      ReferenceRepresentation reference_representation);
   void InitializeAllocators();
   // This will return the space for an object.  If the object is in large
   // object space it may return kLargeCode or kLargeFixedArray in order
@@ -386,8 +453,6 @@ class Serializer : public SerDes {
   int EncodeExternalReference(Address addr) {
     return external_reference_encoder_->Encode(addr);
   }
-  int RootIndex(HeapObject* heap_object);
-  static const int kInvalidRootIndex = -1;
 
   // Keep track of the fullness of each space in order to generate
   // relative addresses for back references.  Large objects are
@@ -397,11 +462,11 @@ class Serializer : public SerDes {
   SnapshotByteSink* sink_;
   int current_root_index_;
   ExternalReferenceEncoder* external_reference_encoder_;
-  bool partial_;
   static bool serialization_enabled_;
   // Did we already make use of the fact that serialization was not enabled?
   static bool too_late_to_enable_now_;
   int large_object_total_;
+  SerializationAddressMapper address_mapper_;
 
   friend class ObjectSerializer;
   friend class Deserializer;
@@ -409,6 +474,62 @@ class Serializer : public SerDes {
   DISALLOW_COPY_AND_ASSIGN(Serializer);
 };
 
+
+class PartialSerializer : public Serializer {
+ public:
+  PartialSerializer(Serializer* startup_snapshot_serializer,
+                    SnapshotByteSink* sink)
+    : Serializer(sink),
+      startup_serializer_(startup_snapshot_serializer) {
+  }
+
+  // Serialize the objects reachable from a single object pointer.
+  virtual void Serialize(Object** o);
+  virtual void SerializeObject(Object* o,
+                               ReferenceRepresentation representation);
+
+ protected:
+  virtual int RootIndex(HeapObject* o);
+  virtual int PartialSnapshotCacheIndex(HeapObject* o);
+  virtual bool ShouldBeInThePartialSnapshotCache(HeapObject* o) {
+    return o->IsString() || o->IsSharedFunctionInfo();
+  }
+
+ private:
+  Serializer* startup_serializer_;
+  DISALLOW_COPY_AND_ASSIGN(PartialSerializer);
+};
+
+
+class StartupSerializer : public Serializer {
+ public:
+  explicit StartupSerializer(SnapshotByteSink* sink) : Serializer(sink) {
+    // Clear the cache of objects used by the partial snapshot.  After the
+    // strong roots have been serialized we can create a partial snapshot
+    // which will repopulate the cache with objects neede by that partial
+    // snapshot.
+    partial_snapshot_cache_length_ = 0;
+  }
+  // Serialize the current state of the heap.  The order is:
+  // 1) Strong references.
+  // 2) Partial snapshot cache.
+  // 3) Weak references (eg the symbol table).
+  virtual void SerializeStrongReferences();
+  virtual void SerializeObject(Object* o,
+                               ReferenceRepresentation representation);
+  void SerializeWeakReferences();
+  void Serialize() {
+    SerializeStrongReferences();
+    SerializeWeakReferences();
+  }
+
+ private:
+  virtual int RootIndex(HeapObject* o) { return kInvalidRootIndex; }
+  virtual bool ShouldBeInThePartialSnapshotCache(HeapObject* o) {
+    return false;
+  }
+};
+
 } }  // namespace v8::internal
 
 #endif  // V8_SERIALIZE_H_

src/snapshot-common.cc

@@ -59,42 +59,4 @@ bool Snapshot::Initialize(const char* snapshot_file) {
   return false;
 }
 
-
-class FileByteSink : public SnapshotByteSink {
- public:
-  explicit FileByteSink(const char* snapshot_file) {
-    fp_ = OS::FOpen(snapshot_file, "wb");
-    if (fp_ == NULL) {
-      PrintF("Unable to write to snapshot file \"%s\"\n", snapshot_file);
-      exit(1);
-    }
-  }
-  virtual ~FileByteSink() {
-    if (fp_ != NULL) {
-      fclose(fp_);
-    }
-  }
-  virtual void Put(int byte, const char* description) {
-    if (fp_ != NULL) {
-      fputc(byte, fp_);
-    }
-  }
-  virtual int Position() {
-    return ftell(fp_);
-  }
-
- private:
-  FILE* fp_;
-};
-
-
-bool Snapshot::WriteToFile(const char* snapshot_file) {
-  FileByteSink file(snapshot_file);
-  Serializer ser(&file);
-  ser.Serialize();
-  return true;
-}
-
-
-
 } }  // namespace v8::internal

src/v8.cc

@@ -146,7 +146,6 @@ void V8::TearDown() {
 
   Heap::TearDown();
   Logger::TearDown();
-  Deserializer::TearDown();
 
   is_running_ = false;
   has_been_disposed_ = true;

test/cctest/test-serialize.cc

@@ -39,6 +39,8 @@
 #include "cctest.h"
 #include "spaces.h"
 #include "objects.h"
+#include "natives.h"
+#include "bootstrapper.h"
 
 using namespace v8::internal;
 
@@ -169,6 +171,75 @@ TEST(ExternalReferenceDecoder) {
 }
 
 
+class FileByteSink : public SnapshotByteSink {
+ public:
+  explicit FileByteSink(const char* snapshot_file) {
+    fp_ = OS::FOpen(snapshot_file, "wb");
+    file_name_ = snapshot_file;
+    if (fp_ == NULL) {
+      PrintF("Unable to write to snapshot file \"%s\"\n", snapshot_file);
+      exit(1);
+    }
+  }
+  virtual ~FileByteSink() {
+    if (fp_ != NULL) {
+      fclose(fp_);
+    }
+  }
+  virtual void Put(int byte, const char* description) {
+    if (fp_ != NULL) {
+      fputc(byte, fp_);
+    }
+  }
+  virtual int Position() {
+    return ftell(fp_);
+  }
+  void WriteSpaceUsed(
+      int new_space_used,
+      int pointer_space_used,
+      int data_space_used,
+      int code_space_used,
+      int map_space_used,
+      int cell_space_used,
+      int large_space_used);
+
+ private:
+  FILE* fp_;
+  const char* file_name_;
+};
+
+
+void FileByteSink::WriteSpaceUsed(
+      int new_space_used,
+      int pointer_space_used,
+      int data_space_used,
+      int code_space_used,
+      int map_space_used,
+      int cell_space_used,
+      int large_space_used) {
+  int file_name_length = strlen(file_name_) + 10;
+  Vector<char> name = Vector<char>::New(file_name_length + 1);
+  OS::SNPrintF(name, "%s.size", file_name_);
+  FILE* fp = OS::FOpen(name.start(), "w");
+  fprintf(fp, "new %d\n", new_space_used);
+  fprintf(fp, "pointer %d\n", pointer_space_used);
+  fprintf(fp, "data %d\n", data_space_used);
+  fprintf(fp, "code %d\n", code_space_used);
+  fprintf(fp, "map %d\n", map_space_used);
+  fprintf(fp, "cell %d\n", cell_space_used);
+  fprintf(fp, "large %d\n", large_space_used);
+  fclose(fp);
+}
+
+
+static bool WriteToFile(const char* snapshot_file) {
+  FileByteSink file(snapshot_file);
+  StartupSerializer ser(&file);
+  ser.Serialize();
+  return true;
+}
+
+
 static void Serialize() {
   // We have to create one context.  One reason for this is so that the builtins
   // can be loaded from v8natives.js and their addresses can be processed.  This
@@ -176,7 +247,7 @@ static void Serialize() {
   // that would confuse the serialization/deserialization process.
   v8::Persistent<v8::Context> env = v8::Context::New();
   env.Dispose();
-  Snapshot::WriteToFile(FLAG_testing_serialization_file);
+  WriteToFile(FLAG_testing_serialization_file);
 }
 
 
@@ -273,95 +344,63 @@ DEPENDENT_TEST(DeserializeFromSecondSerializationAndRunScript2,
 }
 
 
-class FileByteSink : public SnapshotByteSink {
- public:
-  explicit FileByteSink(const char* snapshot_file) {
-    fp_ = OS::FOpen(snapshot_file, "wb");
-    file_name_ = snapshot_file;
-    if (fp_ == NULL) {
-      PrintF("Unable to write to snapshot file \"%s\"\n", snapshot_file);
-      exit(1);
-    }
-  }
-  virtual ~FileByteSink() {
-    if (fp_ != NULL) {
-      fclose(fp_);
-    }
-  }
-  virtual void Put(int byte, const char* description) {
-    if (fp_ != NULL) {
-      fputc(byte, fp_);
+TEST(PartialSerialization) {
+  Serializer::Enable();
+  v8::V8::Initialize();
+
+  v8::Persistent<v8::Context> env = v8::Context::New();
+  ASSERT(!env.IsEmpty());
+  env->Enter();
+  // Make sure all builtin scripts are cached.
+  { HandleScope scope;
+    for (int i = 0; i < Natives::GetBuiltinsCount(); i++) {
+      Bootstrapper::NativesSourceLookup(i);
     }
   }
-  virtual int Position() {
-    return ftell(fp_);
+  Heap::CollectAllGarbage(true);
+  Heap::CollectAllGarbage(true);
+
+  Object* raw_foo;
+  {
+    v8::HandleScope handle_scope;
+    v8::Local<v8::String> foo = v8::String::New("foo");
+    ASSERT(!foo.IsEmpty());
+    raw_foo = *(v8::Utils::OpenHandle(*foo));
   }
-  void WriteSpaceUsed(
-      int new_space_used,
-      int pointer_space_used,
-      int data_space_used,
-      int code_space_used,
-      int map_space_used,
-      int cell_space_used,
-      int large_space_used);
 
- private:
-  FILE* fp_;
-  const char* file_name_;
-};
+  int file_name_length = strlen(FLAG_testing_serialization_file) + 10;
+  Vector<char> startup_name = Vector<char>::New(file_name_length + 1);
+  OS::SNPrintF(startup_name, "%s.startup", FLAG_testing_serialization_file);
 
+  env->Exit();
+  env.Dispose();
 
-void FileByteSink::WriteSpaceUsed(
-      int new_space_used,
-      int pointer_space_used,
-      int data_space_used,
-      int code_space_used,
-      int map_space_used,
-      int cell_space_used,
-      int large_space_used) {
-  int file_name_length = StrLength(file_name_) + 10;
-  Vector<char> name = Vector<char>::New(file_name_length + 1);
-  OS::SNPrintF(name, "%s.size", file_name_);
-  FILE* fp = OS::FOpen(name.start(), "w");
-  fprintf(fp, "new %d\n", new_space_used);
-  fprintf(fp, "pointer %d\n", pointer_space_used);
-  fprintf(fp, "data %d\n", data_space_used);
-  fprintf(fp, "code %d\n", code_space_used);
-  fprintf(fp, "map %d\n", map_space_used);
-  fprintf(fp, "cell %d\n", cell_space_used);
-  fprintf(fp, "large %d\n", large_space_used);
-  fclose(fp);
+  FileByteSink startup_sink(startup_name.start());
+  StartupSerializer startup_serializer(&startup_sink);
+  startup_serializer.SerializeStrongReferences();
+
+  FileByteSink partial_sink(FLAG_testing_serialization_file);
+  PartialSerializer p_ser(&startup_serializer, &partial_sink);
+  p_ser.Serialize(&raw_foo);
+  startup_serializer.SerializeWeakReferences();
+  partial_sink.WriteSpaceUsed(p_ser.CurrentAllocationAddress(NEW_SPACE),
+                              p_ser.CurrentAllocationAddress(OLD_POINTER_SPACE),
+                              p_ser.CurrentAllocationAddress(OLD_DATA_SPACE),
+                              p_ser.CurrentAllocationAddress(CODE_SPACE),
+                              p_ser.CurrentAllocationAddress(MAP_SPACE),
+                              p_ser.CurrentAllocationAddress(CELL_SPACE),
+                              p_ser.CurrentAllocationAddress(LO_SPACE));
 }
 
 
-TEST(PartialSerialization) {
-  Serializer::Enable();
-  v8::V8::Initialize();
-  v8::Persistent<v8::Context> env = v8::Context::New();
-  env->Enter();
-
-  v8::HandleScope handle_scope;
-  v8::Local<v8::String> foo = v8::String::New("foo");
-
-  FileByteSink file(FLAG_testing_serialization_file);
-  Serializer ser(&file);
-  i::Handle<i::String> internal_foo = v8::Utils::OpenHandle(*foo);
-  Object* raw_foo = *internal_foo;
-  ser.SerializePartial(&raw_foo);
-  file.WriteSpaceUsed(ser.CurrentAllocationAddress(NEW_SPACE),
-                      ser.CurrentAllocationAddress(OLD_POINTER_SPACE),
-                      ser.CurrentAllocationAddress(OLD_DATA_SPACE),
-                      ser.CurrentAllocationAddress(CODE_SPACE),
-                      ser.CurrentAllocationAddress(MAP_SPACE),
-                      ser.CurrentAllocationAddress(CELL_SPACE),
-                      ser.CurrentAllocationAddress(LO_SPACE));
-}
+DEPENDENT_TEST(PartialDeserialization, PartialSerialization) {
+  int file_name_length = strlen(FLAG_testing_serialization_file) + 10;
+  Vector<char> startup_name = Vector<char>::New(file_name_length + 1);
+  OS::SNPrintF(startup_name, "%s.startup", FLAG_testing_serialization_file);
 
+  CHECK(Snapshot::Initialize(startup_name.start()));
 
-DEPENDENT_TEST(PartialDeserialization, PartialSerialization) {
-  v8::V8::Initialize();
   const char* file_name = FLAG_testing_serialization_file;
-  int file_name_length = StrLength(file_name) + 10;
   Vector<char> name = Vector<char>::New(file_name_length + 1);
   OS::SNPrintF(name, "%s.size", file_name);
   FILE* fp = OS::FOpen(name.start(), "r");
@@ -392,11 +431,25 @@ DEPENDENT_TEST(PartialDeserialization, PartialSerialization) {
                      large_size);
   int snapshot_size = 0;
   byte* snapshot = ReadBytes(file_name, &snapshot_size);
-  SnapshotByteSource source(snapshot, snapshot_size);
-  Deserializer deserializer(&source);
+
   Object* root;
-  deserializer.DeserializePartial(&root);
-  CHECK(root->IsString());
+  {
+    SnapshotByteSource source(snapshot, snapshot_size);
+    Deserializer deserializer(&source);
+    deserializer.DeserializePartial(&root);
+    CHECK(root->IsString());
+  }
+  v8::HandleScope handle_scope;
+  Handle<Object>root_handle(root);
+
+  Object* root2;
+  {
+    SnapshotByteSource source(snapshot, snapshot_size);
+    Deserializer deserializer(&source);
+    deserializer.DeserializePartial(&root2);
+    CHECK(root2->IsString());
+    CHECK(*root_handle == root2);
+  }
 }