Re: [perfmon2] pfm_start/pfm_stop needed for thread-based self-monitoring?

[stephane e. <er...@go...>]

Hello,


On Mon, Dec 22, 2008 at 5:50 AM, Andrej van der Zee
<and...@gm...> wrote:
> Hi,
>
> I am using perfmon2 for thread-based self-monitoring. The thread
> executes many requests and I want to monitor each request separately.
> To make it little intrusive as possible, I am trying to minimize code
> for re-initialization of the PMU for each request as much as possible.
> For each request, I execute the following code now:
>
You don't really need to reset for each request. You can just read at the
beginning and end of each request and compute delta.

As for start/stop, it all depends on the level of granularity. For
self-monitoring,
and if all PMD fit within one syscall, I think there is little reason
for stopping.
The read syscall will readout the hardware values fairly closely from each
others.

> for (i=0; i < thd->mbrace_data.output_param.pfp_pmd_count; i++) {
>    thd->mbrace_data.pmd[i].reg_value = 0;
>  }
>
>  if (pfm_write_pmds(thd->mbrace_data.context_fd, thd->mbrace_data.pmd,
>        thd->mbrace_data.output_param.pfp_pmd_count)) {
>    general_log_print(thd, COM_MBRACE, "mBrace: pfm_write_pmds error
> errno %d", errno);
>    return;
>  }
>
>  if (pfm_start(thd->mbrace_data.context_fd, NULL)) {
>    general_log_print(thd, COM_MBRACE, "mBrace: pfm_start error errno
> %d", errno);
>    return;
>  }
>
>
> But I was wondering, for thread-based self-monitoring, do I really
> need to call pfm_start and pfm_stop for every request? Or can I
> read/write the PMU without stopping pfm?
>
> Thank you,
> Andrej
>
> ------------------------------------------------------------------------------
> _______________________________________________
> perfmon2-devel mailing list
> per...@li...
> https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
>