system wide or per process statistics

[stardust496]

Greetings,

I have a simple question - I have a program on linux . This program will first pin itself to a particular core and then fork and the forked child will pin itself to a core on a different physical socket

Now, I launch this program from vtune's CLI amplxe-cli -collect nehalem_memory-access collector

Now, the stats that are gathered - are these per physical socket or systemwide for all physical sockets?

Is there a way to see the stats for only one physical socket (since I do not care about what the forked child on a different socket is doing)

Thank you very much!

[TimP]

There are system-wide statistics and per-thread statistics. According to what you said, you should be able to identify the thread you are interested in.

[Peter_W_Intel]

Hi,

If the user run the program which forks child, whatever it works on same core or on a different core -the user should use system wide data collection. Allperformance data are for all physical (and logical) cores, collecting data for special core(s)is not allowed.

For example, use - "amplxe-cl collectnehalem_memory-access -follow-child -analyze-system -- appname"

Sometime, if your mainprocess ran shortly - data collection will stop evenfolked progress still ran. It's better use this example, "amplxe-cl collectnehalem_memory-access -analyze-system -duration 360" (you have to launch your application manually). You may stop data collection before duration timeis out.

Regards, Peter