Changeset 43144 in webkit for trunk/JavaScriptCore/parser/Lexer.cpp

Timestamp:

May 2, 2009, 4:33:56 AM (16 years ago)

Author:

[email protected]

Message:

2009-05-02 Maciej Stachowiak <​[email protected]>

Reviewed by Cameron Zwarich.

• speed up the lexer in various ways

~2% command-line SunSpider speedup

• parser/Lexer.cpp: (JSC::Lexer::setCode): Moved below shift() so it can inline. (JSC::Lexer::scanRegExp): Use resize(0) instead of clear() on Vectors, since the intent here is not to free the underlying buffer. (JSC::Lexer::lex): ditto; also, change the loop logic a bit for the main lexing loop to avoid branching on !m_done twice per iteration. Now we only check it once. (JSC::Lexer::shift): Make this ALWAYS_INLINE and tag an unusual branch as UNLIKELY
• parser/Lexer.h: (JSC::Lexer::makeIdentifier): force to be ALWAYS_INLINE
• wtf/Vector.h: (WTF::::append): force to be ALWAYS_INLINE (may have helped in ways other than parsing but it wasn't getting inlined in a hot code path in the lexer)

File:

• trunk/JavaScriptCore/parser/Lexer.cpp (modified) (7 diffs)

trunk/JavaScriptCore/parser/Lexer.cpp

@@ -89 +89 @@
 }
 
-void Lexer::setCode(const SourceCode& source)
-{
-    yylineno = source.firstLine();
-    m_restrKeyword = false;
-    m_delimited = false;
-    m_eatNextIdentifier = false;
-    m_stackToken = -1;
-    m_lastToken = -1;
-
-    m_position = source.startOffset();
-    m_source = &source;
-    m_code = source.provider()->data();
-    m_length = source.endOffset();
-    m_skipLF = false;
-    m_skipCR = false;
-    m_error = false;
-    m_atLineStart = true;
-
-    // read first characters
-    shift(4);
-}
-
-void Lexer::shift(unsigned p)
+ALWAYS_INLINE void Lexer::shift(unsigned p)
 {
     // ECMA-262 calls for stripping Cf characters here, but we only do this for BOM,
@@ -132 +110 @@
             m_nextOffset3 = m_position;
             m_next3 = m_code[m_position++];
-        } while (m_next3 == 0xFEFF);
-    }
+        } while (UNLIKELY(m_next3 == 0xFEFF));
+    }
+}
+
+void Lexer::setCode(const SourceCode& source)
+{
+    yylineno = source.firstLine();
+    m_restrKeyword = false;
+    m_delimited = false;
+    m_eatNextIdentifier = false;
+    m_stackToken = -1;
+    m_lastToken = -1;
+
+    m_position = source.startOffset();
+    m_source = &source;
+    m_code = source.provider()->data();
+    m_length = source.endOffset();
+    m_skipLF = false;
+    m_skipCR = false;
+    m_error = false;
+    m_atLineStart = true;
+
+    // read first characters
+    shift(4);
 }
 
@@ -156 +156 @@
     m_state = Start;
     unsigned short stringType = 0; // either single or double quotes
-    m_buffer8.clear();
-    m_buffer16.clear();
+    m_buffer8.resize(0);
+    m_buffer16.resize(0);
     m_done = false;
     m_terminator = false;
@@ -171 +171 @@
     }
     int startOffset = m_currentOffset;
-    while (!m_done) {
+    if (!m_done) {
+    while (true) {
         if (m_skipLF && m_current != '\n') // found \r but not \n afterwards
             m_skipLF = false;
@@ -452 +453 @@
         }
 
-        // move on to the next character
-        if (!m_done)
-            shift(1);
         if (m_state != Start && m_state != InSingleLineComment)
             m_atLineStart = false;
+        if (m_done)
+            break;
+
+        shift(1);
+    }
     }
 
@@ -843 +846 @@
 bool Lexer::scanRegExp()
 {
-    m_buffer16.clear();
+    m_buffer16.resize(0);
     bool lastWasEscape = false;
     bool inBrackets = false;
@@ -863 +866 @@
         } else { // end of regexp
             m_pattern = UString(m_buffer16);
-            m_buffer16.clear();
+            m_buffer16.resize(0);
             shift(1);
             break;