Does perf record support event multiplexing?

Does perf record support event multiplexing?

[AmirReza Ghods]

Hi,

I noticed that "perf record" does not calculate the true value for the
number of event's counts in the case of having more events than
hardware counters.
But in pef stat it does and there is a time interpolation algorithm
for estimating and correcting the true value.
Also I saw a patch for Perf's source code that implimented time
interpolation for Perf in record mode but it did not include in the
next official realses of Perf. I just want to know that is there any
technical difficulty for implimenting this feature in "perf record" or
not?


Thanks in advanced,
Amir

Re: Does perf record support event multiplexing?

[Andi Kleen]

AmirReza Ghods <ar.ghods@gmail.com> writes:

> I noticed that "perf record" does not calculate the true value for the
> number of event's counts in the case of having more events than
> hardware counters.
> But in pef stat it does and there is a time interpolation algorithm
> for estimating and correcting the true value.
> Also I saw a patch for Perf's source code that implimented time
> interpolation for Perf in record mode but it did not include in the
> next official realses of Perf. I just want to know that is there any
> technical difficulty for implimenting this feature in "perf record" or
> not?

Why would you care about the exact sample count for perf report?

The only thing that counts in such a histogram is the percent
relationships, which would stay the same if the numbers
were scaled up.

-Andi
-- 
ak@linux.intel.com -- Speaking for myself only

Re: Does perf record support event multiplexing?

[AmirReza Ghods]

Thank you for your response, but that doesn't really answer my
question. I wanted to know why scaling of the event counts is
implemented in "perf stat" but NOT implemented in "perf record".

As for caring about the exact sample count of perf report, I don't think
the relative usefulness of the values is a good reason to leave users
with some verifiably false results, and expect them to figure out why
they got them. I know some users, including myself, will be confused to
measure event counts differing (sometimes 300% in my experience) from
what they might expect.

Thanks,
Amir
On Wed, Jan 22, 2014 at 11:51 PM, Andi Kleen <andi@firstfloor.org> wrote:
> AmirReza Ghods <ar.ghods@gmail.com> writes:
>
>> I noticed that "perf record" does not calculate the true value for the
>> number of event's counts in the case of having more events than
>> hardware counters.
>> But in pef stat it does and there is a time interpolation algorithm
>> for estimating and correcting the true value.
>> Also I saw a patch for Perf's source code that implimented time
>> interpolation for Perf in record mode but it did not include in the
>> next official realses of Perf. I just want to know that is there any
>> technical difficulty for implimenting this feature in "perf record" or
>> not?
>
> Why would you care about the exact sample count for perf report?
>
> The only thing that counts in such a histogram is the percent
> relationships, which would stay the same if the numbers
> were scaled up.
>
> -Andi
> --
> ak@linux.intel.com -- Speaking for myself only



-- 
Amir Reza Ghods
MASc Candidate, Department of Electrical and Computer Engineering
University of Waterloo
200 University Avenue West
Waterloo, Ontario, Canada, N2L 3G1

Re: Does perf record support event multiplexing?

[Namhyung Kim]

Hi AmirReza,

On Mon, 27 Jan 2014 16:14:21 -0500, AmirReza Ghods wrote:
> Thank you for your response, but that doesn't really answer my
> question. I wanted to know why scaling of the event counts is
> implemented in "perf stat" but NOT implemented in "perf record".

I think the reason is "perf record" lacks enabled/running time
information in the samples.

>
> As for caring about the exact sample count of perf report, I don't think
> the relative usefulness of the values is a good reason to leave users
> with some verifiably false results, and expect them to figure out why
> they got them. I know some users, including myself, will be confused to
> measure event counts differing (sometimes 300% in my experience) from
> what they might expect.

You might want to use "perf stat --no-scale" then. ;-)

Thanks,
Namhyung