Analyzers
Talk to fellow users of Intel Analyzer tools (Intel VTune™ Profiler, Intel Advisor)

vTune and IPP

Derek_Woodman
Beginner
342 Views

I am using vTune 9.1 on Fedora 10.

The vTuneDemo works just fine and I can dive down into the actual code and everything. Under the "modules" of a thread of the vtunedemo process I can see the vtunedemo module, vmlinux-2.6.27.41-170.2.117.fc10.x86_64, and Other32. I can drill inside the vtunedemo, but not any of the others.


When I try to run a sampling activity on my application, it starts up and immediately thinks its finished. However, I have the application running in a loop several thousand times so it doesn't finish instantly like vtune says it does. Then when I look at the my process, it doesn't present a module for my application like the vtunde demo does. It just has the same vmlinux listed as above as well as Other64. If I change the vtune preference to run for 20 seconds no matter what, it still only gets 3 clock cycles on the my process which I know is wrong.

I can also run a call graph on the vtunedemo, but cannot run a call graph on my application.

I looked at the makefile for the vtunedemo program. I am using the same flags and everything, except I have to link with libippsem64t.so and libippiem64t.so because i am using ippi and ipps functions. Could this be an issue?
I have sped up an algorithm implementation by using the IPP and I want to get statistics on how IPP is doing this. I pretty much want to know the IPC and CPU utilization. I have been struggiling with vTune for awhile now. Any help will be greatly appreciated.
Thanks!
0 Kudos
1 Reply
Peter_W_Intel
Employee
342 Views

Hi Derek,

There are many possibilities on your problem.

1) Did you fork a child process, and main process terminates quickly? I saw "3 clock cycles on your process", so I guess. Please refer to http://software.intel.com/en-us/articles/sampling-data-collection-dont-know-child-processes-termination-on-linux/

2) You may not sanctify any target applicationbefore doing sampling data collection - I mean to manually launch your application, so performance data for all active applications in the system will be collected.

3) I think there is less samples in your code, but there should be more samples in IPP libraries.

4)For samples in IPP libraries, Iguess that you will seesome export functions from libippem64t.so and libippiem64t.so.

Hope ithelps!

Regards, Peter

0 Kudos
Reply