* Do a GC in mksnapshot to get rid of some extraneous junk.

* Make snapshot more compact by coding the tag and the space
in one byte.  Contract some common sequences to one byte.
* Use back references only within one page.  Index from the
start of the space otherwise.
* Serialize Smis as raw data rather than int-encoding them.
This takes a little more space but is faster.
Review URL: http://codereview.chromium.org/341079

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3208 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

Author: erik.corry@gmail.com

src/mksnapshot.cc

@@ -194,6 +194,9 @@ int main2(int argc, char** argv) {
   context.Dispose();
   CppByteSink sink(argv[1]);
   i::Serializer2 ser(&sink);
+  // This results in a somewhat smaller snapshot, probably because it gets rid
+  // of some things that are cached between garbage collections.
+  i::Heap::CollectAllGarbage(true);
   ser.Serialize();
   return 0;
 }

src/objects.h

@@ -2910,7 +2910,8 @@ class Code: public HeapObject {
   void CodeVerify();
 #endif
   // Code entry points are aligned to 32 bytes.
-  static const int kCodeAlignment = 32;
+  static const int kCodeAlignmentBits = 5;
+  static const int kCodeAlignment = 1 << kCodeAlignmentBits;
   static const int kCodeAlignmentMask = kCodeAlignment - 1;
 
   // Layout description.