Some optimizations for packer.js.

src/heap.cc

Index: src/heap.cc
===================================================================
--- src/heap.cc	(revision 3335)
+++ src/heap.cc	(working copy)
@@ -1774,6 +1774,38 @@
   }
 
   int length = first_length + second_length;
+
+  // Optimization for packer.js.  This popular library uses two-character
+  // strings as keys in a hash.  Check whether we already have the string
+  // in the symbol table to prevent creation of many unneccesary strings.
+  if (length == 2) {
+    unsigned c1 = first->Get(0);
+    unsigned c2 = second->Get(0);
+    String* symbol;

[Søren Thygesen Gjesse, 2009/11/19 21:42:57]

Please add a comment on why digits only will not work here.

+    if ((c1 - '0' > '9' - '0' ||

[Søren Thygesen Gjesse, 2009/11/19 21:42:57]

Why c1 - '0' > '9' - '0' instead of c1 > '9'?

[Erik Corry, 2009/11/20 10:12:46]

I moved this to a different function so I could point out in a comment that the
variables are unsigned and thus what you suggest would be completely different.

+         c2 - '0' > '9' - '0') &&
+        symbol_table()->LookupTwoCharsIfExists(c1, c2, &symbol)) {
+      return symbol;
+    // Now we know the length is 2, we might as well make use of that fact
+    // when building the new string.
+    } else if ((c1 | c2) <= String::kMaxAsciiCharCodeU) {  // We can do this
+      ASSERT(IsPowerOf2(String::kMaxAsciiCharCodeU + 1));  // because of this.
+      Object* result = AllocateRawAsciiString(2);
+      if (result->IsFailure()) return result;
+      char* dest = SeqAsciiString::cast(result)->GetChars();
+      dest[0] = c1;
+      dest[1] = c2;
+      return result;
+    } else {
+      Object* result = AllocateRawTwoByteString(2);
+      if (result->IsFailure()) return result;
+      uc16* dest = SeqTwoByteString::cast(result)->GetChars();
+      dest[0] = c1;
+      dest[1] = c2;
+      return result;
+    }
+  }
+
   bool is_ascii = first->IsAsciiRepresentation()
       && second->IsAsciiRepresentation();
 
@@ -1843,6 +1875,35 @@
   if (length == 1) {
     return Heap::LookupSingleCharacterStringFromCode(
         buffer->Get(start));
+  } else if (length == 2) {

[Søren Thygesen Gjesse, 2009/11/19 21:42:57]

Refactor duplicated code.

[Erik Corry, 2009/11/20 10:12:46]

Done.

+    // Optimization for packer.js.  This popular library uses two-character
+    // strings as keys in a hash.  Check whether we already have the string
+    // in the symbol table to prevent creation of many unneccesary strings.
+    unsigned c1 = buffer->Get(start);
+    unsigned c2 = buffer->Get(start + 1);
+    String* symbol;
+    if ((c1 - '0' > '9' - '0' ||
+         c2 - '0' > '9' - '0') &&
+        symbol_table()->LookupTwoCharsIfExists(c1, c2, &symbol)) {
+      return symbol;
+    // Now we know the length is 2, we might as well make use of that fact
+    // when building the new string.
+    } else if ((c1 | c2) <= String::kMaxAsciiCharCodeU) {  // We can do this
+      ASSERT(IsPowerOf2(String::kMaxAsciiCharCodeU + 1));  // because of this.
+      Object* result = AllocateRawAsciiString(2);
+      if (result->IsFailure()) return result;
+      char* dest = SeqAsciiString::cast(result)->GetChars();
+      dest[0] = c1;
+      dest[1] = c2;
+      return result;
+    } else {
+      Object* result = AllocateRawTwoByteString(2);
+      if (result->IsFailure()) return result;
+      uc16* dest = SeqTwoByteString::cast(result)->GetChars();
+      dest[0] = c1;
+      dest[1] = c2;
+      return result;
+    }
   }
 
   // Make an attempt to flatten the buffer to reduce access time.