Improve speed of Utf8Write by always flattening the string first and

detecting the ASCII case.  Also rewrite Utf8Length to work on an
unflattened string.  Bug: http://code.google.com/p/v8/issues/detail?id=1665
Review URL: http://codereview.chromium.org/8304021

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@9661 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/api.cc

@@ -3634,13 +3634,25 @@ int String::WriteUtf8(char* buffer,
   if (IsDeadCheck(isolate, "v8::String::WriteUtf8()")) return 0;
   LOG_API(isolate, "String::WriteUtf8");
   ENTER_V8(isolate);
-  i::StringInputBuffer& write_input_buffer = *isolate->write_input_buffer();
   i::Handle<i::String> str = Utils::OpenHandle(this);
+  if (str->IsAsciiRepresentation()) {
+    if (capacity == -1) capacity = str->length() + 1;
+    int len = i::Min(capacity, str->length());
+    i::String::WriteToFlat(*str, buffer, 0, len);
+    if (nchars_ref != NULL) *nchars_ref = len;
+    if (!(options & NO_NULL_TERMINATION) && capacity > len) {
+      buffer[len] = '\0';
+      return len + 1;
+    }
+    return len;
+  }
+
+  i::StringInputBuffer& write_input_buffer = *isolate->write_input_buffer();
   isolate->string_tracker()->RecordWrite(str);
   if (options & HINT_MANY_WRITES_EXPECTED) {
     // Flatten the string for efficiency.  This applies whether we are
     // using StringInputBuffer or Get(i) to access the characters.
-    str->TryFlatten();
+    FlattenString(str);
   }
   write_input_buffer.Reset(0, *str);
   int len = str->length();

src/objects.cc

@@ -5519,24 +5519,6 @@ bool String::LooksValid() {
 }
 
 
-int String::Utf8Length() {
-  if (IsAsciiRepresentation()) return length();
-  // Attempt to flatten before accessing the string.  It probably
-  // doesn't make Utf8Length faster, but it is very likely that
-  // the string will be accessed later (for example by WriteUtf8)
-  // so it's still a good idea.
-  Heap* heap = GetHeap();
-  TryFlatten();
-  Access<StringInputBuffer> buffer(
-      heap->isolate()->objects_string_input_buffer());
-  buffer->Reset(0, this);
-  int result = 0;
-  while (buffer->has_more())
-    result += unibrow::Utf8::Length(buffer->GetNext());
-  return result;
-}
-
-
 String::FlatContent String::GetFlatContent() {
   int length = this->length();
   StringShape shape(this);
@@ -5960,6 +5942,73 @@ const unibrow::byte* String::ReadBlock(String* input,
 }
 
 
+// This method determines the type of string involved and then gets the UTF8
+// length of the string.  It doesn't flatten the string and has log(n) recursion
+// for a string of length n.
+int String::Utf8Length(String* input, int from, int to) {
+  if (from == to) return 0;
+  int total = 0;
+  while (true) {
+    if (input->IsAsciiRepresentation()) return total + to - from;
+    switch (StringShape(input).representation_tag()) {
+      case kConsStringTag: {
+        ConsString* str = ConsString::cast(input);
+        String* first = str->first();
+        String* second = str->second();
+        int first_length = first->length();
+        if (first_length - from < to - first_length) {
+          if (first_length > from) {
+            // Left hand side is shorter.
+            total += Utf8Length(first, from, first_length);
+            input = second;
+            from = 0;
+            to -= first_length;
+          } else {
+            // We only need the right hand side.
+            input = second;
+            from -= first_length;
+            to -= first_length;
+          }
+        } else {
+          if (first_length <= to) {
+            // Right hand side is shorter.
+            total += Utf8Length(second, 0, to - first_length);
+            input = first;
+            to = first_length;
+          } else {
+            // We only need the left hand side.
+            input = first;
+          }
+        }
+        continue;
+      }
+      case kExternalStringTag:
+      case kSeqStringTag: {
+        Vector<const uc16> vector = input->GetFlatContent().ToUC16Vector();
+        const uc16* p = vector.start();
+        for (int i = from; i < to; i++) {
+          total += unibrow::Utf8::Length(p[i]);
+        }
+        return total;
+      }
+      case kSlicedStringTag: {
+        SlicedString* str = SlicedString::cast(input);
+        int offset = str->offset();
+        input = str->parent();
+        from += offset;
+        to += offset;
+        continue;
+      }
+      default:
+        break;
+    }
+    UNREACHABLE();
+    return 0;
+  }
+  return 0;
+}
+
+
 void Relocatable::PostGarbageCollectionProcessing() {
   Isolate* isolate = Isolate::Current();
   Relocatable* current = isolate->relocatable_top();

src/objects.h

@@ -6073,7 +6073,8 @@ class String: public HeapObject {
       RobustnessFlag robustness_flag = FAST_STRING_TRAVERSAL,
       int* length_output = 0);
 
-  int Utf8Length();
+  inline int Utf8Length() { return Utf8Length(this, 0, length()); }
+  static int Utf8Length(String* input, int from, int to);
 
   // Return a 16 bit Unicode representation of the string.
   // The string should be nearly flat, otherwise the performance of

test/cctest/test-api.cc

@@ -5438,67 +5438,109 @@ static int StrNCmp16(uint16_t* a, uint16_t* b, int n) {
 
 
 THREADED_TEST(StringWrite) {
+  LocalContext context;
   v8::HandleScope scope;
   v8::Handle<String> str = v8_str("abcde");
   // abc<Icelandic eth><Unicode snowman>.
   v8::Handle<String> str2 = v8_str("abc\303\260\342\230\203");
+  const int kStride = 4;  // Must match stride in for loops in JS below.
+  CompileRun(
+      "var left = '';"
+      "for (var i = 0; i < 0xd800; i += 4) {"
+      "  left = left + String.fromCharCode(i);"
+      "}");
+  CompileRun(
+      "var right = '';"
+      "for (var i = 0; i < 0xd800; i += 4) {"
+      "  right = String.fromCharCode(i) + right;"
+      "}");
+  v8::Handle<v8::Object> global = Context::GetCurrent()->Global();
+  Handle<String> left_tree = global->Get(v8_str("left")).As<String>();
+  Handle<String> right_tree = global->Get(v8_str("right")).As<String>();
 
   CHECK_EQ(5, str2->Length());
+  CHECK_EQ(0xd800 / kStride, left_tree->Length());
+  CHECK_EQ(0xd800 / kStride, right_tree->Length());
 
   char buf[100];
-  char utf8buf[100];
+  char utf8buf[0xd800 * 3];
   uint16_t wbuf[100];
   int len;
   int charlen;
 
-  memset(utf8buf, 0x1, sizeof(utf8buf));
+  memset(utf8buf, 0x1, 1000);
   len = str2->WriteUtf8(utf8buf, sizeof(utf8buf), &charlen);
   CHECK_EQ(9, len);
   CHECK_EQ(5, charlen);
   CHECK_EQ(0, strcmp(utf8buf, "abc\303\260\342\230\203"));
 
-  memset(utf8buf, 0x1, sizeof(utf8buf));
+  memset(utf8buf, 0x1, 1000);
   len = str2->WriteUtf8(utf8buf, 8, &charlen);
   CHECK_EQ(8, len);
   CHECK_EQ(5, charlen);
   CHECK_EQ(0, strncmp(utf8buf, "abc\303\260\342\230\203\1", 9));
 
-  memset(utf8buf, 0x1, sizeof(utf8buf));
+  memset(utf8buf, 0x1, 1000);
   len = str2->WriteUtf8(utf8buf, 7, &charlen);
   CHECK_EQ(5, len);
   CHECK_EQ(4, charlen);
   CHECK_EQ(0, strncmp(utf8buf, "abc\303\260\1", 5));
 
-  memset(utf8buf, 0x1, sizeof(utf8buf));
+  memset(utf8buf, 0x1, 1000);
   len = str2->WriteUtf8(utf8buf, 6, &charlen);
   CHECK_EQ(5, len);
   CHECK_EQ(4, charlen);
   CHECK_EQ(0, strncmp(utf8buf, "abc\303\260\1", 5));
 
-  memset(utf8buf, 0x1, sizeof(utf8buf));
+  memset(utf8buf, 0x1, 1000);
   len = str2->WriteUtf8(utf8buf, 5, &charlen);
   CHECK_EQ(5, len);
   CHECK_EQ(4, charlen);
   CHECK_EQ(0, strncmp(utf8buf, "abc\303\260\1", 5));
 
-  memset(utf8buf, 0x1, sizeof(utf8buf));
+  memset(utf8buf, 0x1, 1000);
   len = str2->WriteUtf8(utf8buf, 4, &charlen);
   CHECK_EQ(3, len);
   CHECK_EQ(3, charlen);
   CHECK_EQ(0, strncmp(utf8buf, "abc\1", 4));
 
-  memset(utf8buf, 0x1, sizeof(utf8buf));
+  memset(utf8buf, 0x1, 1000);
   len = str2->WriteUtf8(utf8buf, 3, &charlen);
   CHECK_EQ(3, len);
   CHECK_EQ(3, charlen);
   CHECK_EQ(0, strncmp(utf8buf, "abc\1", 4));
 
-  memset(utf8buf, 0x1, sizeof(utf8buf));
+  memset(utf8buf, 0x1, 1000);
   len = str2->WriteUtf8(utf8buf, 2, &charlen);
   CHECK_EQ(2, len);
   CHECK_EQ(2, charlen);
   CHECK_EQ(0, strncmp(utf8buf, "ab\1", 3));
 
+  memset(utf8buf, 0x1, sizeof(utf8buf));
+  len = left_tree->Utf8Length();
+  int utf8_expected =
+      (0x80 + (0x800 - 0x80) * 2 + (0xd800 - 0x800) * 3) / kStride;
+  CHECK_EQ(utf8_expected, len);
+  len = left_tree->WriteUtf8(utf8buf, utf8_expected, &charlen);
+  CHECK_EQ(utf8_expected, len);
+  CHECK_EQ(0xd800 / kStride, charlen);
+  CHECK_EQ(0xed, static_cast<unsigned char>(utf8buf[utf8_expected - 3]));
+  CHECK_EQ(0x9f, static_cast<unsigned char>(utf8buf[utf8_expected - 2]));
+  CHECK_EQ(0xc0 - kStride,
+           static_cast<unsigned char>(utf8buf[utf8_expected - 1]));
+  CHECK_EQ(1, utf8buf[utf8_expected]);
+
+  memset(utf8buf, 0x1, sizeof(utf8buf));
+  len = right_tree->Utf8Length();
+  CHECK_EQ(utf8_expected, len);
+  len = right_tree->WriteUtf8(utf8buf, utf8_expected, &charlen);
+  CHECK_EQ(utf8_expected, len);
+  CHECK_EQ(0xd800 / kStride, charlen);
+  CHECK_EQ(0xed, static_cast<unsigned char>(utf8buf[0]));
+  CHECK_EQ(0x9f, static_cast<unsigned char>(utf8buf[1]));
+  CHECK_EQ(0xc0 - kStride, static_cast<unsigned char>(utf8buf[2]));
+  CHECK_EQ(1, utf8buf[utf8_expected]);
+
   memset(buf, 0x1, sizeof(buf));
   memset(wbuf, 0x1, sizeof(wbuf));
   len = str->WriteAscii(buf);
@@ -11440,6 +11482,7 @@ static void MorphAString(i::String* string,
 // Test that we can still flatten a string if the components it is built up
 // from have been turned into 16 bit strings in the mean time.
 THREADED_TEST(MorphCompositeStringTest) {
+  char utf_buffer[129];
   const char* c_string = "Now is the time for all good men"
                          " to come to the aid of the party";
   uint16_t* two_byte_string = AsciiToTwoByteString(c_string);
@@ -11468,6 +11511,17 @@ THREADED_TEST(MorphCompositeStringTest) {
     MorphAString(*v8::Utils::OpenHandle(*lhs), &ascii_resource, &uc16_resource);
     MorphAString(*v8::Utils::OpenHandle(*rhs), &ascii_resource, &uc16_resource);
 
+    // This should UTF-8 without flattening, since everything is ASCII.
+    Handle<String> cons = v8_compile("cons")->Run().As<String>();
+    CHECK_EQ(128, cons->Utf8Length());
+    int nchars = -1;
+    CHECK_EQ(129, cons->WriteUtf8(utf_buffer, -1, &nchars));
+    CHECK_EQ(128, nchars);
+    CHECK_EQ(0, strcmp(
+        utf_buffer,
+        "Now is the time for all good men to come to the aid of the party"
+        "Now is the time for all good men to come to the aid of the party"));
+
     // Now do some stuff to make sure the strings are flattened, etc.
     CompileRun(
         "/[^a-z]/.test(cons);"