Changeset 59061 in webkit for trunk/JavaScriptCore

Timestamp:

May 9, 2010, 2:16:16 PM (15 years ago)

Author:

[email protected]

Message:

Reserve a large-ish initial capacity for Lexer::m_buffer16.

Reviewed by Oliver Hunt.

SunSpider says 0.3% faster.

m_buffer16 is used when parsing complex strings -- for example, strings
with escape sequences in them. These kinds of strings can be really long,
and we want to avoid repeatedly copying as we grow m_buffer16.

The net memory cost is quite low, since it's proporitional to source
code we already have in memory, and we throw away m_buffer16 right when
we're done parsing.

• parser/Lexer.cpp:

(JSC::Lexer::Lexer): No need to reserve initial capacity in our constructor,
since setCode will be called before we're asked to lex anything.
(JSC::Lexer::setCode): Reserve enough space to lex half the source code
as a complex string without having to copy.
(JSC::Lexer::clear): No need to reserve initial capacity here either,
since setCode will be called before we're asked to lex anything.

Location:

trunk/JavaScriptCore

Files:

• ChangeLog (modified) (1 diff)
• parser/Lexer.cpp (modified) (3 diffs)

trunk/JavaScriptCore/ChangeLog

@@ +1 @@
+2010-05-09  Geoffrey Garen  <[email protected]>
+
+        Reviewed by Oliver Hunt.
+
+        Reserve a large-ish initial capacity for Lexer::m_buffer16.
+        
+        SunSpider says 0.3% faster.
+        
+        m_buffer16 is used when parsing complex strings -- for example, strings
+        with escape sequences in them. These kinds of strings can be really long,
+        and we want to avoid repeatedly copying as we grow m_buffer16.
+        
+        The net memory cost is quite low, since it's proporitional to source
+        code we already have in memory, and we throw away m_buffer16 right when
+        we're done parsing.
+
+        * parser/Lexer.cpp:
+        (JSC::Lexer::Lexer): No need to reserve initial capacity in our constructor,
+        since setCode will be called before we're asked to lex anything.
+        (JSC::Lexer::setCode): Reserve enough space to lex half the source code
+        as a complex string without having to copy.
+        (JSC::Lexer::clear): No need to reserve initial capacity here either,
+        since setCode will be called before we're asked to lex anything.
+
 2010-05-09  Laszlo Gombos  <[email protected]>
 

trunk/JavaScriptCore/parser/Lexer.cpp

@@ -53 +53 @@
     , m_keywordTable(JSC::mainTable)
 {
-    m_buffer8.reserveInitialCapacity(initialReadBufferCapacity);
-    m_buffer16.reserveInitialCapacity(initialReadBufferCapacity);
 }
 
@@ -149 +147 @@
     m_error = false;
     m_atLineStart = true;
+
+    m_buffer8.reserveInitialCapacity(initialReadBufferCapacity);
+    m_buffer16.reserveInitialCapacity((m_codeEnd - m_code) / 2);
 
     // ECMA-262 calls for stripping all Cf characters, but we only strip BOM characters.
@@ -1010 +1011 @@
 
     Vector<char> newBuffer8;
-    newBuffer8.reserveInitialCapacity(initialReadBufferCapacity);
     m_buffer8.swap(newBuffer8);
 
     Vector<UChar> newBuffer16;
-    newBuffer16.reserveInitialCapacity(initialReadBufferCapacity);
     m_buffer16.swap(newBuffer16);