Optimize StringImpl::lower() for 8-bit strings

Source/wtf/text/StringImpl.cpp

 /*
  * Copyright (C) 1999 Lars Knoll (knoll@kde.org)
  *           (C) 1999 Antti Koivisto (koivisto@kde.org)
  *           (C) 2001 Dirk Mueller ( mueller@kde.org )
  * Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2013 Apple Inc. All rights reserved.
  * Copyright (C) 2006 Andrew Wellington (proton@wiretapped.net)
  *
  * This library is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Library General Public
  * License as published by the Free Software Foundation; either
@@ skipping 484 matching lines @@
         return U16_GET_SUPPLEMENTARY(characters16()[i], characters16()[i + 1]);
     return 0;
 }
 
 PassRefPtr<StringImpl> StringImpl::lower()
 {
     // Note: This is a hot function in the Dromaeo benchmark, specifically the
     // no-op code path up through the first 'return' statement.
 
     // First scan the string for uppercase and non-ASCII characters:
-    bool noUpper = true;
-    UChar ored = 0;
     if (is8Bit()) {
-        const LChar* end = characters8() + m_length;
-        for (const LChar* chp = characters8(); chp != end; ++chp) {
-            if (UNLIKELY(isASCIIUpper(*chp)))
-                noUpper = false;
-            ored |= *chp;
+        unsigned needsLowering = m_length;

[tkent, 2015/06/15 01:12:56]

The name |needsLowering| sounds a boolean.  should be |firstIndexToBeLowered| or
something?

+        for (unsigned i = 0; i < m_length; ++i) {
+            LChar ch = characters8()[i];
+            if (UNLIKELY(isASCIIUpper(ch) || ch & ~0x7F)) {
+                needsLowering = i;
+                break;
+            }
         }
+
         // Nothing to do if the string is all ASCII with no uppercase.
-        if (noUpper && !(ored & ~0x7F))
+        if (needsLowering == m_length)
             return this;
 
-        RELEASE_ASSERT(m_length <= static_cast<unsigned>(numeric_limits<int32_t>::max()));
-        int32_t length = m_length;
+        LChar* data8;
+        RefPtr<StringImpl> newImpl = createUninitialized(m_length, data8);
+        memcpy(data8, characters8(), needsLowering);
 
-        LChar* data8;
-        RefPtr<StringImpl> newImpl = createUninitialized(length, data8);
-
-        if (!(ored & ~0x7F)) {
-            for (int32_t i = 0; i < length; ++i)
-                data8[i] = toASCIILower(characters8()[i]);
-
-            return newImpl.release();
+        for (unsigned i = needsLowering; i < m_length; ++i) {
+            LChar ch = characters8()[i];
+            data8[i] = UNLIKELY(ch & ~0x7F) ? static_cast<LChar>(Unicode::toLower(ch))
+                                            : toASCIILower(ch);
         }
 
-        // Do a slower implementation for cases that include non-ASCII Latin-1 characters.
-        for (int32_t i = 0; i < length; ++i)
-            data8[i] = static_cast<LChar>(Unicode::toLower(characters8()[i]));
-
         return newImpl.release();
     }
 
+    bool noUpper = true;
+    UChar ored = 0;
+
     const UChar* end = characters16() + m_length;
     for (const UChar* chp = characters16(); chp != end; ++chp) {
         if (UNLIKELY(isASCIIUpper(*chp)))
             noUpper = false;
         ored |= *chp;
     }
     // Nothing to do if the string is all ASCII with no uppercase.
     if (noUpper && !(ored & ~0x7F))
         return this;
 
@@ skipping 1575 matching lines @@
         } else if (localeIdMatchesLang(localeIdentifier, "lt")) {
             // TODO(rob.buis) implement upper-casing rules for lt
             // like in StringImpl::upper(locale).
         }
     }
 
     return toUpper(c);
 }
 
 } // namespace WTF