See http://codereview.chromium.org/173470 for description. The x64 implementation follows the ia32 except it uses some additional registers to avoid spilling to the stack. Also tweaked the ia32 version a bit.

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@2805 ce2b1a6d-e550-0410-aec6-3dcde31c8c00