Community
cancel
Showing results for 
Search instead for 
Did you mean: 
mfcking
Beginner
110 Views

How many performance counters does VTune use on Xeon? 2 or 4 or 8 or 18?

Hello,
My first question is how many performance counters Xeon has? Some manual says 18, some says 4 for Non-HT Xeon and 8 for HT Xeon.
My second question is how many performance counters VTune uses on Xeon? It seems VTune only uses 2 of them, why not use all of them? The more counters, the less iterations, right?
Please shed some light,
Thanks,
L.Y.

Message Edited by mfcking@yahoo.com on 11-03-2005 05:00 PM

0 Kudos
3 Replies
David_A_Intel1
Employee
110 Views

Hi L.Y.:

You are very right! One run is optimal!! However, if you study the processor manual, you will see that selecting some event in one event control register limits the events available in the next register, and etc. So, the processor may very well have 18 registers, but based on the events you have selected, there may be no way to collect them all in one run.

The VTune analyzer uses an algorithm to attempt to collect as many events at a time as possible. It is probably not perfect and I have heard some discussion of a new algorithm that improves the use of the event registers. We'll have to wait an see if that shows up in a future release.

Regards,

mfcking
Beginner
110 Views

Hi Dave,

Thanks for your reply. I also noticed sometimes VTune got the problem when I combined two events together. If I sampled these two events in separate sampling activities, then VTune has no problem.

L.Y.

David_A_Intel1
Employee
110 Views

Hi L.Y.:

Can you give me an example of two events where this happens? If I can duplicate it, I can submit a defect to engineering.

Thanks,

Reply