Commit b55add7e authored by lrn@chromium.org's avatar lrn@chromium.org

Optimized scanner to avoid virtual calls for every character read.

Review URL: http://codereview.chromium.org/5545006

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@5935 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
parent 670aea02
......@@ -1165,14 +1165,22 @@ void ObjectTemplate::SetInternalFieldCount(int value) {
ScriptData* ScriptData::PreCompile(const char* input, int length) {
unibrow::Utf8InputBuffer<> buf(input, length);
return i::ParserApi::PreParse(i::Handle<i::String>(), &buf, NULL);
i::Utf8ToUC16CharacterStream stream(
reinterpret_cast<const unsigned char*>(input), length);
return i::ParserApi::PreParse(&stream, NULL);
}
ScriptData* ScriptData::PreCompile(v8::Handle<String> source) {
i::Handle<i::String> str = Utils::OpenHandle(*source);
return i::ParserApi::PreParse(str, NULL, NULL);
if (str->IsExternalTwoByteString()) {
i::ExternalTwoByteStringUC16CharacterStream stream(
i::Handle<i::ExternalTwoByteString>::cast(str), 0, str->length());
return i::ParserApi::PreParse(&stream, NULL);
} else {
i::GenericStringUC16CharacterStream stream(str, 0, str->length());
return i::ParserApi::PreParse(&stream, NULL);
}
}
......
......@@ -231,6 +231,8 @@ static inline void CheckNonEqualsHelper(const char* file,
#define CHECK_GT(a, b) CHECK((a) > (b))
#define CHECK_GE(a, b) CHECK((a) >= (b))
#define CHECK_LT(a, b) CHECK((a) < (b))
#define CHECK_LE(a, b) CHECK((a) <= (b))
// This is inspired by the static assertion facility in boost. This
......
......@@ -461,7 +461,14 @@ Handle<SharedFunctionInfo> Compiler::Compile(Handle<String> source,
ScriptDataImpl* pre_data = input_pre_data;
if (pre_data == NULL
&& source_length >= FLAG_min_preparse_length) {
pre_data = ParserApi::PartialPreParse(source, NULL, extension);
if (source->IsExternalTwoByteString()) {
ExternalTwoByteStringUC16CharacterStream stream(
Handle<ExternalTwoByteString>::cast(source), 0, source->length());
pre_data = ParserApi::PartialPreParse(&stream, extension);
} else {
GenericStringUC16CharacterStream stream(source, 0, source->length());
pre_data = ParserApi::PartialPreParse(&stream, extension);
}
}
// Create a script object describing the script to be compiled.
......
This diff is collapsed.
......@@ -169,14 +169,12 @@ class ParserApi {
static bool Parse(CompilationInfo* info);
// Generic preparser generating full preparse data.
static ScriptDataImpl* PreParse(Handle<String> source,
unibrow::CharacterStream* stream,
static ScriptDataImpl* PreParse(UC16CharacterStream* source,
v8::Extension* extension);
// Preparser that only does preprocessing that makes sense if only used
// immediately after.
static ScriptDataImpl* PartialPreParse(Handle<String> source,
unibrow::CharacterStream* stream,
static ScriptDataImpl* PartialPreParse(UC16CharacterStream* source,
v8::Extension* extension);
};
......@@ -435,18 +433,26 @@ class Parser {
Vector<const char*> args);
protected:
FunctionLiteral* ParseLazy(Handle<SharedFunctionInfo> info,
UC16CharacterStream* source,
ZoneScope* zone_scope);
enum Mode {
PARSE_LAZILY,
PARSE_EAGERLY
};
// Called by ParseProgram after setting up the scanner.
FunctionLiteral* DoParseProgram(Handle<String> source,
bool in_global_context,
ZoneScope* zone_scope);
// Report syntax error
void ReportUnexpectedToken(Token::Value token);
void ReportInvalidPreparseData(Handle<String> name, bool* ok);
void ReportMessage(const char* message, Vector<const char*> args);
bool inside_with() const { return with_nesting_level_ > 0; }
Scanner& scanner() { return scanner_; }
V8JavaScriptScanner& scanner() { return scanner_; }
Mode mode() const { return mode_; }
ScriptDataImpl* pre_data() const { return pre_data_; }
......@@ -548,7 +554,7 @@ class Parser {
INLINE(Token::Value peek()) {
if (stack_overflow_) return Token::ILLEGAL;
return scanner_.peek();
return scanner().peek();
}
INLINE(Token::Value Next()) {
......@@ -560,9 +566,11 @@ class Parser {
}
if (StackLimitCheck().HasOverflowed()) {
// Any further calls to Next or peek will return the illegal token.
// The current call must return the next token, which might already
// have been peek'ed.
stack_overflow_ = true;
}
return scanner_.Next();
return scanner().Next();
}
INLINE(void Consume(Token::Value token));
......@@ -702,7 +710,14 @@ class JsonParser BASE_EMBEDDED {
// Parse JSON input as a single JSON value.
// Returns null handle and sets exception if parsing failed.
static Handle<Object> Parse(Handle<String> source) {
return JsonParser().ParseJson(source);
if (source->IsExternalTwoByteString()) {
ExternalTwoByteStringUC16CharacterStream stream(
Handle<ExternalTwoByteString>::cast(source), 0, source->length());
return JsonParser().ParseJson(source, &stream);
} else {
GenericStringUC16CharacterStream stream(source, 0, source->length());
return JsonParser().ParseJson(source, &stream);
}
}
private:
......@@ -710,7 +725,7 @@ class JsonParser BASE_EMBEDDED {
~JsonParser() { }
// Parse a string containing a single JSON value.
Handle<Object> ParseJson(Handle<String>);
Handle<Object> ParseJson(Handle<String> script, UC16CharacterStream* source);
// Parse a single JSON value from input (grammar production JSONValue).
// A JSON value is either a (double-quoted) string literal, a number literal,
// one of "true", "false", or "null", or an object or array literal.
......
......@@ -39,39 +39,82 @@ namespace v8 {
namespace internal {
// UTF16Buffer based on a v8::UnicodeInputStream.
class InputStreamUTF16Buffer : public UTF16Buffer {
class InputStreamUTF16Buffer : public UC16CharacterStream {
public:
explicit InputStreamUTF16Buffer(UnicodeInputStream* stream)
: UTF16Buffer(),
stream_(stream) { }
explicit InputStreamUTF16Buffer(v8::UnicodeInputStream* stream)
: UC16CharacterStream(),
stream_(stream),
pushback_active_(false) {
buffer_cursor_ = buffer_end_ = buffer_ + kPushBackSize;
}
virtual ~InputStreamUTF16Buffer() { }
virtual void PushBack(uc32 ch) {
virtual void PushBack(uc16 ch) {
ASSERT(pos_ > 0);
if (buffer_cursor_ > buffer_) {
// While we can stay within the buffer, just do so.
*--buffer_cursor_ = ch;
pos_--;
return;
}
if (!pushback_active_) {
// Push back the entire buffer to the stream and let the
// stream handle pushbacks from now.
// We leave buffer_cursor_ == buffer_end_, so the next read
// will fill the buffer from the current position.
// This should happen exceedingly rarely.
while (buffer_end_ > buffer_) {
stream_->PushBack(*--buffer_end_);
}
buffer_cursor_ = buffer_end_;
pushback_active_ = true;
}
stream_->PushBack(ch);
pos_--;
}
virtual uc32 Advance() {
uc32 result = stream_->Next();
if (result >= 0) pos_++;
return result;
protected:
virtual bool ReadBlock() {
// Copy the top of the buffer into the pushback area.
pushback_active_ = false;
int32_t value;
uc16* buffer_start = buffer_ + kPushBackSize;
buffer_cursor_ = buffer_end_ = buffer_start;
while ((value = stream_->Next()) >= 0) {
if (value > static_cast<int32_t>(unibrow::Utf8::kMaxThreeByteChar)) {
value = unibrow::Utf8::kBadChar;
}
// buffer_end_ is a const pointer, but buffer_ is writable.
buffer_start[buffer_end_++ - buffer_start] = static_cast<uc16>(value);
if (buffer_end_ == buffer_ + kPushBackSize + kBufferSize) break;
}
return buffer_end_ > buffer_start;
}
virtual void SeekForward(int pos) {
virtual unsigned SlowSeekForward(unsigned pos) {
// Seeking in the input is not used by preparsing.
// It's only used by the real parser based on preparser data.
UNIMPLEMENTED();
return 0;
}
private:
static const unsigned kBufferSize = 512;
static const unsigned kPushBackSize = 16;
v8::UnicodeInputStream* const stream_;
// Buffer holding first kPushBackSize characters of pushback buffer,
// then kBufferSize chars of read-ahead.
// The pushback buffer is only used if pushing back characters past
// the start of a block.
uc16 buffer_[kBufferSize + kPushBackSize];
bool pushback_active_;
};
class StandAloneJavaScriptScanner : public JavaScriptScanner {
public:
void Initialize(UTF16Buffer* source) {
void Initialize(UC16CharacterStream* source) {
source_ = source;
literal_flags_ = kLiteralString | kLiteralIdentifier;
Init();
......
......@@ -1078,6 +1078,7 @@ PreParser::Expression PreParser::ParseFunctionLiteral(bool* ok) {
Expect(i::Token::RBRACE, CHECK_OK);
// Position right after terminal '}'.
int end_pos = scanner_->location().end_pos;
log_->LogFunction(function_block_pos, end_pos,
function_scope.materialized_literal_count(),
......
......@@ -34,12 +34,6 @@
namespace v8 {
namespace internal {
// ----------------------------------------------------------------------------
// UTF16Buffer
UTF16Buffer::UTF16Buffer()
: pos_(0), end_(kNoEndPosition) { }
// ----------------------------------------------------------------------------
// LiteralCollector
......@@ -92,7 +86,7 @@ bool ScannerConstants::IsIdentifier(unibrow::CharacterStream* buffer) {
// ----------------------------------------------------------------------------
// Scanner
Scanner::Scanner() : source_(NULL) {}
Scanner::Scanner() { }
uc32 Scanner::ScanHexEscape(uc32 c, int length) {
......@@ -142,8 +136,7 @@ uc32 Scanner::ScanOctalEscape(uc32 c, int length) {
// ----------------------------------------------------------------------------
// JavaScriptScanner
JavaScriptScanner::JavaScriptScanner()
: has_line_terminator_before_next_(false) {}
JavaScriptScanner::JavaScriptScanner() : Scanner() {}
Token::Value JavaScriptScanner::Next() {
......@@ -503,13 +496,23 @@ void JavaScriptScanner::Scan() {
void JavaScriptScanner::SeekForward(int pos) {
source_->SeekForward(pos - 1);
Advance();
// This function is only called to seek to the location
// of the end of a function (at the "}" token). It doesn't matter
// whether there was a line terminator in the part we skip.
has_line_terminator_before_next_ = false;
// After this call, we will have the token at the given position as
// the "next" token. The "current" token will be invalid.
if (pos == next_.location.beg_pos) return;
int current_pos = source_pos();
ASSERT_EQ(next_.location.end_pos, current_pos);
// Positions inside the lookahead token aren't supported.
ASSERT(pos >= current_pos);
if (pos != current_pos) {
source_->SeekForward(pos - source_->pos());
Advance();
// This function is only called to seek to the location
// of the end of a function (at the "}" token). It doesn't matter
// whether there was a line terminator in the part we skip.
has_line_terminator_before_next_ = false;
}
Scan();
ASSERT_EQ(Token::RBRACE, next_.token);
}
......
......@@ -52,31 +52,75 @@ inline int HexValue(uc32 c) {
return -1;
}
// ----------------------------------------------------------------------------
// UTF16Buffer - scanner input source with pushback.
class UTF16Buffer {
// ---------------------------------------------------------------------
// Buffered stream of characters, using an internal UC16 buffer.
class UC16CharacterStream {
public:
UTF16Buffer();
virtual ~UTF16Buffer() {}
UC16CharacterStream() : pos_(0) { }
virtual ~UC16CharacterStream() { }
// Returns and advances past the next UC16 character in the input
// stream. If there are no more characters, it returns a negative
// value.
inline int32_t Advance() {
if (buffer_cursor_ < buffer_end_ || ReadBlock()) {
pos_++;
return *(buffer_cursor_++);
}
// Note: currently the following increment is necessary to avoid a
// parser problem! The scanner treats the final kEndOfInput as
// a character with a position, and does math relative to that
// position.
pos_++;
virtual void PushBack(uc32 ch) = 0;
// Returns a value < 0 when the buffer end is reached.
virtual uc32 Advance() = 0;
virtual void SeekForward(int pos) = 0;
return kEndOfInput;
}
int pos() const { return pos_; }
// Return the current position in the character stream.
// Starts at zero.
inline unsigned pos() const { return pos_; }
// Skips forward past the next character_count UC16 characters
// in the input, or until the end of input if that comes sooner.
// Returns the number of characters actually skipped. If less
// than character_count,
inline unsigned SeekForward(unsigned character_count) {
unsigned buffered_chars =
static_cast<unsigned>(buffer_end_ - buffer_cursor_);
if (character_count <= buffered_chars) {
buffer_cursor_ += character_count;
pos_ += character_count;
return character_count;
}
return SlowSeekForward(character_count);
}
static const int kNoEndPosition = 1;
// Pushes back the most recently read UC16 character, i.e.,
// the value returned by the most recent call to Advance.
// Must not be used right after calling SeekForward.
virtual void PushBack(uc16 character) = 0;
protected:
// Initial value of end_ before the input stream is initialized.
int pos_; // Current position in the buffer.
int end_; // Position where scanning should stop (EOF).
static const int32_t kEndOfInput = -1;
// Ensures that the buffer_cursor_ points to the character at
// position pos_ of the input, if possible. If the position
// is at or after the end of the input, return false. If there
// are more characters available, return true.
virtual bool ReadBlock() = 0;
virtual unsigned SlowSeekForward(unsigned character_count) = 0;
const uc16* buffer_cursor_;
const uc16* buffer_end_;
unsigned pos_;
};
// ---------------------------------------------------------------------
// Constants used by scanners.
class ScannerConstants : AllStatic {
public:
typedef unibrow::Utf8InputBuffer<1024> Utf8Decoder;
......@@ -277,7 +321,7 @@ class Scanner {
// Low-level scanning support.
void Advance() { c0_ = source_->Advance(); }
void PushBack(uc32 ch) {
source_->PushBack(ch);
source_->PushBack(c0_);
c0_ = ch;
}
......@@ -307,8 +351,8 @@ class Scanner {
TokenDesc current_; // desc for current token (as returned by Next())
TokenDesc next_; // desc for next token (one token look-ahead)
// Input stream. Must be initialized to an UTF16Buffer.
UTF16Buffer* source_;
// Input stream. Must be initialized to an UC16CharacterStream.
UC16CharacterStream* source_;
// Buffer to hold literal values (identifiers, strings, numbers)
// using '\x00'-terminated UTF-8 encoding. Handles allocation internally.
......
This diff is collapsed.
......@@ -35,67 +35,97 @@
namespace v8 {
namespace internal {
// UTF16 buffer to read characters from a character stream.
class CharacterStreamUTF16Buffer: public UTF16Buffer {
// A buffered character stream based on a random access character
// source (ReadBlock can be called with pos_ pointing to any position,
// even positions before the current).
class BufferedUC16CharacterStream: public UC16CharacterStream {
public:
CharacterStreamUTF16Buffer();
virtual ~CharacterStreamUTF16Buffer() {}
void Initialize(Handle<String> data,
unibrow::CharacterStream* stream,
int start_position,
int end_position);
virtual void PushBack(uc32 ch);
virtual uc32 Advance();
virtual void SeekForward(int pos);
private:
List<uc32> pushback_buffer_;
uc32 last_;
unibrow::CharacterStream* stream_;
List<uc32>* pushback_buffer() { return &pushback_buffer_; }
BufferedUC16CharacterStream();
virtual ~BufferedUC16CharacterStream();
virtual void PushBack(uc16 character);
protected:
static const unsigned kBufferSize = 512;
static const unsigned kPushBackStepSize = 16;
virtual unsigned SlowSeekForward(unsigned delta);
virtual bool ReadBlock();
virtual void SlowPushBack(uc16 character);
virtual unsigned BufferSeekForward(unsigned delta) = 0;
virtual unsigned FillBuffer(unsigned position, unsigned length) = 0;
const uc16* pushback_limit_;
uc16 buffer_[kBufferSize];
};
// UTF16 buffer to read characters from an external string.
template <typename StringType, typename CharType>
class ExternalStringUTF16Buffer: public UTF16Buffer {
// Generic string stream.
class GenericStringUC16CharacterStream: public BufferedUC16CharacterStream {
public:
ExternalStringUTF16Buffer();
virtual ~ExternalStringUTF16Buffer() {}
void Initialize(Handle<StringType> data,
int start_position,
int end_position);
virtual void PushBack(uc32 ch);
virtual uc32 Advance();
virtual void SeekForward(int pos);
private:
const CharType* raw_data_; // Pointer to the actual array of characters.
GenericStringUC16CharacterStream(Handle<String> data,
unsigned start_position,
unsigned end_position);
virtual ~GenericStringUC16CharacterStream();
protected:
virtual unsigned BufferSeekForward(unsigned delta);
virtual unsigned FillBuffer(unsigned position, unsigned length);
Handle<String> string_;
unsigned start_position_;
unsigned length_;
};
// Initializes a UTF16Buffer as input stream, using one of a number
// of strategies depending on the available character sources.
class StreamInitializer {
// UC16 stream based on a literal UTF-8 string.
class Utf8ToUC16CharacterStream: public BufferedUC16CharacterStream {
public:
UTF16Buffer* Init(Handle<String> source,
unibrow::CharacterStream* stream,
int start_position,
int end_position);
private:
// Different UTF16 buffers used to pull characters from. Based on input one of
// these will be initialized as the actual data source.
CharacterStreamUTF16Buffer char_stream_buffer_;
ExternalStringUTF16Buffer<ExternalTwoByteString, uint16_t>
two_byte_string_buffer_;
ExternalStringUTF16Buffer<ExternalAsciiString, char> ascii_string_buffer_;
// Used to convert the source string into a character stream when a stream
// is not passed to the scanner.
SafeStringInputBuffer safe_string_input_buffer_;
Utf8ToUC16CharacterStream(const byte* data, unsigned length);
virtual ~Utf8ToUC16CharacterStream();
protected:
virtual unsigned BufferSeekForward(unsigned delta);
virtual unsigned FillBuffer(unsigned char_position, unsigned length);
void SetRawPosition(unsigned char_position);
const byte* raw_data_;
unsigned raw_data_length_; // Measured in bytes, not characters.
unsigned raw_data_pos_;
// The character position of the character at raw_data[raw_data_pos_].
// Not necessarily the same as pos_.
unsigned raw_character_position_;
};
// UTF16 buffer to read characters from an external string.
class ExternalTwoByteStringUC16CharacterStream: public UC16CharacterStream {
public:
ExternalTwoByteStringUC16CharacterStream(Handle<ExternalTwoByteString> data,
int start_position,
int end_position);
virtual ~ExternalTwoByteStringUC16CharacterStream();
virtual void PushBack(uc16 character) {
ASSERT(buffer_cursor_ > raw_data_);
buffer_cursor_--;
pos_--;
}
protected:
virtual unsigned SlowSeekForward(unsigned delta) {
// Fast case always handles seeking.
return 0;
}
virtual bool ReadBlock() {
// Entire string is read at start.
return false;
}
Handle<ExternalTwoByteString> source_;
const uc16* raw_data_; // Pointer to the actual array of characters.
};
// ----------------------------------------------------------------------------
// V8JavaScriptScanner
// JavaScript scanner getting its input from either a V8 String or a unicode
......@@ -103,19 +133,9 @@ class StreamInitializer {
class V8JavaScriptScanner : public JavaScriptScanner {
public:
V8JavaScriptScanner() {}
// Initialize the Scanner to scan source.
void Initialize(Handle<String> source, int literal_flags = kAllLiterals);
void Initialize(Handle<String> source,
unibrow::CharacterStream* stream,
int literal_flags = kAllLiterals);
void Initialize(Handle<String> source,
int start_position, int end_position,
V8JavaScriptScanner();
void Initialize(UC16CharacterStream* source,
int literal_flags = kAllLiterals);
protected:
StreamInitializer stream_initializer_;
};
......@@ -123,8 +143,7 @@ class JsonScanner : public Scanner {
public:
JsonScanner();
// Initialize the Scanner to scan source.
void Initialize(Handle<String> source);
void Initialize(UC16CharacterStream* source);
// Returns the next token.
Token::Value Next();
......@@ -138,7 +157,7 @@ class JsonScanner : public Scanner {
// Recognizes all of the single-character tokens directly, or calls a function
// to scan a number, string or identifier literal.
// The only allowed whitespace characters between tokens are tab,
// carrige-return, newline and space.
// carriage-return, newline and space.
void ScanJson();
// A JSON number (production JSONNumber) is a subset of the valid JavaScript
......@@ -159,60 +178,8 @@ class JsonScanner : public Scanner {
// are the only valid JSON identifiers (productions JSONBooleanLiteral,
// JSONNullLiteral).
Token::Value ScanJsonIdentifier(const char* text, Token::Value token);
StreamInitializer stream_initializer_;
};
// ExternalStringUTF16Buffer
template <typename StringType, typename CharType>
ExternalStringUTF16Buffer<StringType, CharType>::ExternalStringUTF16Buffer()
: raw_data_(NULL) { }
template <typename StringType, typename CharType>
void ExternalStringUTF16Buffer<StringType, CharType>::Initialize(
Handle<StringType> data,
int start_position,
int end_position) {
ASSERT(!data.is_null());
raw_data_ = data->resource()->data();
ASSERT(end_position <= data->length());
if (start_position > 0) {
SeekForward(start_position);
}
end_ =
end_position != kNoEndPosition ? end_position : data->length();
}
template <typename StringType, typename CharType>
uc32 ExternalStringUTF16Buffer<StringType, CharType>::Advance() {
if (pos_ < end_) {
return raw_data_[pos_++];
} else {
// note: currently the following increment is necessary to avoid a
// test-parser problem!
pos_++;
return static_cast<uc32>(-1);
}
}
template <typename StringType, typename CharType>
void ExternalStringUTF16Buffer<StringType, CharType>::PushBack(uc32 ch) {
pos_--;
ASSERT(pos_ >= Scanner::kCharacterLookaheadBufferSize);
ASSERT(raw_data_[pos_ - Scanner::kCharacterLookaheadBufferSize] == ch);
}
template <typename StringType, typename CharType>
void ExternalStringUTF16Buffer<StringType, CharType>::SeekForward(int pos) {
pos_ = pos;
}
} } // namespace v8::internal
#endif // V8_SCANNER_H_
This diff is collapsed.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment