I was wondering if there is a way from the command line to group the results by synchronization event. By synchronization event I mean each time a synchronization was called. For example, if the same barrier wait is called multiple times, I would like to see stats for each of these times individually.
Thank you very much in advance
By grouping results you mean that you want to see wait time on a barrier for each barrier instances?
What is the semantic - are the barrier instances called in the loop or they are used to sync different algorithm pieces (and as a result called from different functions for example)?
And what threading model do you use? OpenMP, TBB or row OS threads?
Thank you, Regards, Dmitry
Thank you for your reply.
Wait time at each barrier can be one thing. Other stats I am interested are similar to those you can get from "General Exploration" such as instructions retired, cache miss rates, etc ...
I am trying different applications that have different semantics.
The applications I am studying are using pthreads.