Serendipitously arrange the tags so that String.length() becomes a branch-fre...

src/objects.h

Index: src/objects.h
===================================================================
--- src/objects.h	(revision 595)
+++ src/objects.h	(working copy)
@@ -395,29 +395,28 @@
 const uint32_t kStringTag = 0x0;
 const uint32_t kNotStringTag = 0x80;
 
-// If bit 7 is clear, bits 5 and 6 are the string's size (short, medium, or
-// long).
-const uint32_t kStringSizeMask = 0x60;
-const uint32_t kShortStringTag = 0x0;
-const uint32_t kMediumStringTag = 0x20;
-const uint32_t kLongStringTag = 0x40;
+// If bit 7 is clear, bits 5 and 6 are the string's size (short, medium or long).

[Mads Ager (chromium), 2008/10/27 09:37:00]

This is now bits 4 and 5?

Also, you should add a comment explaining that you use this encoding so that you
can perform the length shifts based on the tags in the instance type.

+const uint32_t kStringSizeMask = 0x18;
+const uint32_t kShortStringTag = 0x18;
+const uint32_t kMediumStringTag = 0x10;
+const uint32_t kLongStringTag = 0x00;
 
 // If bit 7 is clear, bit 4 indicates that the string is a symbol (if set) or

[Mads Ager (chromium), 2008/10/27 09:37:00]

And this is now bit 5?

 // not (if cleared).
-const uint32_t kIsSymbolMask = 0x10;
+const uint32_t kIsSymbolMask = 0x20;
 const uint32_t kNotSymbolTag = 0x0;
-const uint32_t kSymbolTag = 0x10;
+const uint32_t kSymbolTag = 0x20;
 
 // If bit 7 is clear, and the string representation is a sequential string,
 // then bit 3 indicates whether the string consists of two-byte characters or
 // one-byte characters.
-const uint32_t kStringEncodingMask = 0x8;
+const uint32_t kStringEncodingMask = 0x4;
 const uint32_t kTwoByteStringTag = 0x0;
-const uint32_t kAsciiStringTag = 0x8;
+const uint32_t kAsciiStringTag = 0x4;
 
 // If bit 7 is clear, the low-order 3 bits indicate the representation

[Mads Ager (chromium), 2008/10/27 09:37:00]

3 -> 2 bits.

 // of the string.
-const uint32_t kStringRepresentationMask = 0x07;
+const uint32_t kStringRepresentationMask = 0x03;
 enum StringRepresentationTag {
   kSeqStringTag = 0x0,
   kConsStringTag = 0x1,
@@ -3154,8 +3153,8 @@
   static const int kSize = kLengthOffset + kIntSize;
 
   // Limits on sizes of different types of strings.
-  static const int kMaxShortStringSize = 255;
-  static const int kMaxMediumStringSize = 65535;
+  static const int kMaxShortStringSize = 63;
+  static const int kMaxMediumStringSize = 16383;

[Mads Ager (chromium), 2008/10/27 09:37:00]

These reductions in the limits is the only thing that worries me.  It means that
a lot of strings could potentially have worse hash-codes.

 
   static const int kMaxArrayIndexSize = 10;
 
@@ -3176,13 +3175,13 @@
 
   // Array index strings this short can keep their index in the hash
   // field.
-  static const int kMaxCachedArrayIndexLength = 6;
+  static const int kMaxCachedArrayIndexLength = 7;
 
   // Shift constants for retriving length and hash code from
   // length/hash field.
   static const int kHashShift = kNofLengthBitFields;
-  static const int kShortLengthShift = 3 * kBitsPerByte;
-  static const int kMediumLengthShift = 2 * kBitsPerByte;
+  static const int kShortLengthShift = 26;

[Mads Ager (chromium), 2008/10/27 09:37:00]

Could we write these as kShortStringTag + kRepresentationTagSize and
kMediumStringTag + kRepresentationTagSize.  That would make it clearer to me
what the encoding is.

+  static const int kMediumLengthShift = 18;
   static const int kLongLengthShift = kHashShift;
 
   // Limit for truncation in short printing.