Re: [PATCH v5 0/5] perf stat: Support overall statistics for interval mode

[Jin, Yao]

Hi Kajoljain,

On 5/14/2020 5:53 PM, kajoljain wrote:
> On 5/14/20 11:06 AM, Jin Yao wrote:
> > Currently perf-stat supports to print counts at regular interval (-I),
> > but it's not very easy for user to get the overall statistics.
> >
> > With this patchset, it supports to report the summary at the end of
> > interval output.
> >
> > For example,
> >
> >   root@kbl-ppc:~# perf stat -e cycles -I1000 --interval-count 2
> >   #           time             counts unit events
> >        1.000412064          2,281,114      cycles
> >        2.001383658          2,547,880      cycles
> >
> >    Performance counter stats for 'system wide':
> >
> >            4,828,994      cycles
> >
> >          2.002860349 seconds time elapsed
> >
> >   root@kbl-ppc:~# perf stat -e cycles,instructions -I1000 --interval-count 2
> >   #           time             counts unit events
> >        1.000389902          1,536,093      cycles
> >        1.000389902            420,226      instructions              #    0.27  insn per cycle
> >        2.001433453          2,213,952      cycles
> >        2.001433453            735,465      instructions              #    0.33  insn per cycle
> >
> >    Performance counter stats for 'system wide':
> >
> >            3,750,045      cycles
> >            1,155,691      instructions              #    0.31  insn per cycle
> >
> >          2.003023361 seconds time elapsed
> >
> >   root@kbl-ppc:~# perf stat -M CPI,IPC -I1000 --interval-count 2
> >   #           time             counts unit events
> >        1.000435121            905,303      inst_retired.any          #      2.9 CPI
> >        1.000435121          2,663,333      cycles
> >        1.000435121            914,702      inst_retired.any          #      0.3 IPC
> >        1.000435121          2,676,559      cpu_clk_unhalted.thread
> >        2.001615941          1,951,092      inst_retired.any          #      1.8 CPI
> >        2.001615941          3,551,357      cycles
> >        2.001615941          1,950,837      inst_retired.any          #      0.5 IPC
> >        2.001615941          3,551,044      cpu_clk_unhalted.thread
> >
> >    Performance counter stats for 'system wide':
> >
> >            2,856,395      inst_retired.any          #      2.2 CPI
> >            6,214,690      cycles
> >            2,865,539      inst_retired.any          #      0.5 IPC
> >            6,227,603      cpu_clk_unhalted.thread
> >
> >          2.003403078 seconds time elapsed
>
> Hi Jin,
> 	Reporting the summary will be great for understanding overall stats. So, Before the
> patch where we are reseting rt_stat before read_counters to make sure, whatever printing
> in final aggregate is as per counts on that interval,

Yes, I had similar thoughts, so I posted following patch.

https://lore.kernel.org/lkml/20200420145417.6864-1-yao.jin@xxxxxxxxxxxxxxx/

> we used to update stats->means and other info as described in
>
> RFC: https://lkml.org/lkml/2020/3/24/158

I've checked your patch but sorry I'm also not very sure if it's the expected 
behavior.

> Now, stats->means is same as counts which we are using in generic_metric function. Is this expected behavior?
> I am not sure, if data like stats->means and all suppose to update per interval or we are using it somewhere else.

I just think it's easy to understand, that is the metric calculated by the 
counts per interval.

> So, As we call update_stats for each event and for each interval, can we somehow use that
> to print overall stats maybe by adding some var in `struct stats` to keep count of total counts for that event.
> Please let me know if my understanding is fine.

Adding var in 'struct stats' looks not enough (or more complicated), because 
perf-stat also needs to report some counts according to different aggregation 
modes (not only the metric). I just think copying total counts to current counts 
is a easy way because we can reuse most of existing non-interval processing code.

Thanks
Jin Yao

> Thanks,
> Kajol Jain
>
>
>   
> >   v5:
> >   ---
> >   1. Create new patch "perf stat: Save aggr value to first member
> >      of prev_raw_counts".
> >
> >   2. Call perf_evlist__save_aggr_prev_raw_counts to save aggr value
> >      to first member of prev_raw_counts for AGGR_GLOBAL. Then next,
> >      perf_stat_process_counter can create aggr values from per cpu
> >      values.
> >
> >   Following patches are impacted in v5:
> >      perf stat: Copy counts from prev_raw_counts to evsel->counts
> >      perf stat: Save aggr value to first member of prev_raw_counts
> >      perf stat: Report summary for interval mode
> >
> >   v4:
> >   ---
> >   1. Create runtime_stat_reset.
> >
> >   2. Zero the aggr in perf_counts__reset and use it to reset
> >      prev_raw_counts.
> >
> >   3. Move affinity setup and read_counter_cpu to a new function
> >      read_affinity_counters. It's only called when stat_config.summary
> >      is not set.
> >
> >   v3:
> >   ---
> >   1. 'perf stat: Fix wrong per-thread runtime stat for interval mode'
> >      is a new patch which fixes an existing issue found in test.
> >
> >   2. We use the prev_raw_counts for summary counts. Drop the summary_counts in v2.
> >
> >   3. Fix some issues.
> >
> >   v2:
> >   ---
> >   Rebase to perf/core branch
> >
> > Jin Yao (5):
> >    perf stat: Fix wrong per-thread runtime stat for interval mode
> >    perf counts: Reset prev_raw_counts counts
> >    perf stat: Copy counts from prev_raw_counts to evsel->counts
> >    perf stat: Save aggr value to first member of prev_raw_counts
> >    perf stat: Report summary for interval mode
> >
> >   tools/perf/builtin-stat.c | 101 ++++++++++++++++++++++++++------------
> >   tools/perf/util/counts.c  |   4 +-
> >   tools/perf/util/counts.h  |   1 +
> >   tools/perf/util/stat.c    |  43 +++++++++++++---
> >   tools/perf/util/stat.h    |   3 ++
> >   5 files changed, 113 insertions(+), 39 deletions(-)