I have run stream. But it take effect on R2PCIe events. I can't understand this. If it take through on IMC, it clearly to easy understand.
I posted my response in the wrong place -- please see https://software.intel.com/en-us/forums/topic/569942#comment-1835300 for an explanation of why you are getting non-trivial measurements on the R2PCIe unit when running STREAM.
For more complete information about compiler optimizations, see our Optimization Notice.