Thanks for your reply. I had heard of PAPI before, but I never realized that
it is at a higher level than libpfm4. Looking at the docs, it does sound
quite a bit simpler than what I thought up. I guess I misunderstood the role
of libpfm4.
Oh well, at least some new IPC stuff so it wasn't for naught.
Thanks.
On Fri, Oct 29, 2010 at 7:53 PM, Corey Ashford
<cjash...@linux.vnet.ibm.com>wrote:
> On 10/29/2010 04:20 PM, DRAM Ninjas wrote:
> > Hello all,
> >
> > Up to this point I've been using task_smpl with libpfm4 to get
> > process-wide performance counters, but I would like to be able to get
> > counts for specific regions of interest. I was trying to search the
> > archives and found a relevant feature for perfmon2, but seemingly
> > nothing for libpfm4.
> >
> > Most of the things that I want to run these types of region of interest
> > tests on are my own little toy programs, so I can easily augment them
> > however I please. What I was thinking was just to setup a semaphore that
> > will start out locked. The parent process in task_smpl will just
> > sem_wait() right before the core loop, and the child will unlock the
> > semaphore when it gets to the region of interest.
> >
> > I'm new to writing IPC code. I've written a small example that behaves
> > as described above (where the parent doesn't continue executing until
> > the child gets to a certain point), but I'm not sure if there are any
> > pitfalls that I could be missing here. Does this sound like a feasible
> > strategy? Is there some easier way to achieve what I am looking to do?
> >
> > Thanks,
> > Paul
>
> Hello Paul,
>
> Have you thought about using a library like PAPI, called directly from
> your code to do what you want? Using PAPI would be far simpler than
> what you are attempting to accomplish using semaphores and other IPC
> mechanisms, I think. PAPI provides a layer on top of libpfm +
> perf_events (as well as other OS kernel APIs where available) which is
> designed primarily for measuring events from the context of the
> instrumented program rather than a remote monitoring program, such as
> "task_smpl".
>
> Another alternative would be to use libpfm+perf_events directly, instead
> of using PAPI. However, PAPI's abstraction is easier to use.
>
> PAPI info here:
>
> http://icl.cs.utk.edu/papi/
>
>
> - Corey
>
>
> ------------------------------------------------------------------------------
> Nokia and AT&T present the 2010 Calling All Innovators-North America
> contest
> Create new apps & games for the Nokia N8 for consumers in U.S. and Canada
> $10 million total in prizes - $4M cash, 500 devices, nearly $6M in
> marketing
> Develop with Nokia Qt SDK, Web Runtime, or Java and Publish to Ovi Store
> http://p.sf.net/sfu/nokia-dev2dev
> _______________________________________________
> perfmon2-devel mailing list
> perfmon2-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
>
------------------------------------------------------------------------------
Nokia and AT&T present the 2010 Calling All Innovators-North America contest
Create new apps & games for the Nokia N8 for consumers in U.S. and Canada
$10 million total in prizes - $4M cash, 500 devices, nearly $6M in marketing
Develop with Nokia Qt SDK, Web Runtime, or Java and Publish to Ovi Store
http://p.sf.net/sfu/nokia-dev2dev
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
