http://software.intel.com/en-us/articles/high-clocks-per-instruction-retired-when-vectorizing-the-loop/feed/ en-us