Re: perf-stat per thread results

From: Ingo Molnar
Date: Fri May 13 2011 - 11:44:34 EST



* David Ahern <dsahern@xxxxxxxxx> wrote:

> On 05/13/11 09:14, Wim Heirman wrote:
> > Hi all,
> >
> > I'm using perf-stat to look at hardware performance counters for a
> > parallel program. Is there a way to get counter values for each thread
> > individually, rather than aggregated for the whole process? I know I
> > can attach to a specific thread using --tid=, but due to the time
> > required to find the tid and attach/detach this isn't accurate for
> > short-running programs.
>
> perf-stat requires changes to dump counters per thread; it currently
> sums all threads into a single value.
>
> > Or, alternatively, can I use perf record --stat and get an exact count
> > for each performance counter?
>
> perf-record does not read values from hardware counters.

It's supposed to do that if --stat is specified, and it used to work - see this
commit:

8d51327090ac: perf report: Fix and improve the displaying of per-thread event counters

and the output there:

# PID TID cache-misses cache-references
4658 4659 495581 3238779
4658 4662 498246 3236823
4658 4663 499531 3243162

which appears to be roughly what Wim is asking for, AFAICT.

But this seems to have regressed meanwhile.

Thanks,

Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/