Page 1 of 1

Collecting performance counters of parallel programs by PAPI

PostPosted: Fri Jun 26, 2015 10:25 am
by khaleghzadeh
Hi all,

As you know, PAPI can be attached to a thread by PAPI_attach option. There is a question about this option. I would be thankful if someone answer me. Consider a PThread application consists of one main thread and two worker threads (Thread-0 and Thread-1) both run on core-0. If I define an Eventset collecting L1 misses of Thread-0 (the eventset has been attached to Thread-0), could you tell me the Eventset only collects L1 misses originate only form Thread-0 or it counts L1 misses from both Thread-0 and Thread-1?

Thanks

Re: Collecting performance counters of parallel programs by

PostPosted: Fri Jun 26, 2015 1:31 pm
by s_ragate
I think, when you attach papi eventset to a thread, the eventset counts should be specific to that thread IF AND ONLY IF it is a bound or kernel thread. More information on using PAPI for parallel program can be found in the following link,

http://icl.cs.utk.edu/projects/papi/wiki/Threads

Hope this answers your question.