1. MUL and DIV on SMIs.
2. When calling GenericBinaryOpStub from a virtual frame.
3. When generating code for a loop counter.
Overall performance gain is about 0.6%.
Review URL: http://codereview.chromium.org/555098

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3708 ce2b1a6d-e550-0410-aec6-3dcde31c8c00