linux-perf-users.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* duration_time regression
@ 2023-08-01 20:32 Andi Kleen
  2023-08-01 21:47 ` Ian Rogers
  0 siblings, 1 reply; 2+ messages in thread
From: Andi Kleen @ 2023-08-01 20:32 UTC (permalink / raw)
  To: linux-perf-users; +Cc: irogers, acme, namhyung


With recent perf duration_time output regressed:

e.g. with an older version I get duration_time only once:

% perf stat -e cycles,duration_time true

CPU0                 190,925      cycles
... all the other CPUs printing cycles
CPU0                5,047,008 ns   duration_time

       0.005047008 seconds time elapsed

     
with the current perf it is printed on all CPUs, but not counted on most
of them:

% perf stat -e cycles,duration_time true

# started on Tue Aug  1 13:29:26 2023

 Performance counter stats for 'system wide':

CPU0               61,301      cycles                                                                
... print cycles for all CPUS ...
CPU0            5,343,556 ns   duration_time                                                         
CPU1        <not counted> ns   duration_time                                                         
CPU2        <not counted> ns   duration_time                                                         
CPU3        <not counted> ns   duration_time                                                         
CPU4        <not counted> ns   duration_time                                                         
... repeat the same for all CPUs ...

which is fairly useless.

-Andi

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: duration_time regression
  2023-08-01 20:32 duration_time regression Andi Kleen
@ 2023-08-01 21:47 ` Ian Rogers
  0 siblings, 0 replies; 2+ messages in thread
From: Ian Rogers @ 2023-08-01 21:47 UTC (permalink / raw)
  To: Andi Kleen; +Cc: linux-perf-users, acme, namhyung

On Tue, Aug 1, 2023 at 1:32 PM Andi Kleen <ak@linux.intel.com> wrote:
>
>
> With recent perf duration_time output regressed:
>
> e.g. with an older version I get duration_time only once:
>
> % perf stat -e cycles,duration_time true
>
> CPU0                 190,925      cycles
> ... all the other CPUs printing cycles
> CPU0                5,047,008 ns   duration_time
>
>        0.005047008 seconds time elapsed
>
>
> with the current perf it is printed on all CPUs, but not counted on most
> of them:
>
> % perf stat -e cycles,duration_time true
>
> # started on Tue Aug  1 13:29:26 2023
>
>  Performance counter stats for 'system wide':
>
> CPU0               61,301      cycles
> ... print cycles for all CPUS ...
> CPU0            5,343,556 ns   duration_time
> CPU1        <not counted> ns   duration_time
> CPU2        <not counted> ns   duration_time
> CPU3        <not counted> ns   duration_time
> CPU4        <not counted> ns   duration_time
> ... repeat the same for all CPUs ...
>
> which is fairly useless.

Agreed, I sent out:
https://lore.kernel.org/lkml/20230801205452.3380578-1-irogers@google.com/

Thanks,
Ian

> -Andi

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2023-08-01 21:48 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2023-08-01 20:32 duration_time regression Andi Kleen
2023-08-01 21:47 ` Ian Rogers

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).