I am evaluating vtune on linux:
There is an option to delay the collection of samples by time. I think for more accurate testing, it would help if there was an option to trigger the collection with a symbol in the program. E.g. as soon as the function 'foo' is called, collection starts.
This would enable precise starts of profile runs that skip the initialization phases of your code.
There is also ability to use itt_resume and itt_pause API in the source of your application to more precisely pause/resume collection than start by time. Can this help in your situation?
Thanks & Regards, Dmitry