Re: [PATCH v3 3/4] perf stat: Copy counts from prev_raw_counts to evsel->counts

["Jin, Yao" <yao.jin@linux.intel.com>]

Hi Jiri,

On 5/7/2020 11:19 PM, Jiri Olsa wrote:
> On Thu, May 07, 2020 at 02:58:21PM +0800, Jin Yao wrote:
>> It would be useful to support the overall statistics for perf-stat
>> interval mode. For example, report the summary at the end of
>> "perf-stat -I" output.
>>
>> But since perf-stat can support many aggregation modes, such as
>> --per-thread, --per-socket, -M and etc, we need a solution which
>> doesn't bring much complexity.
>>
>> The idea is to use 'evsel->prev_raw_counts' which is updated in
>> each interval and it's saved with the latest counts. Before reporting
>> the summary, we copy the counts from evsel->prev_raw_counts to
>> evsel->counts, and next we just follow non-interval processing.
> 
> I did not realize we already store the count in prev_raw_counts ;-) nice catch!
> 

Thanks! :)

>>
>> In evsel__compute_deltas, this patch saves counts to the position
>> of [cpu0,thread0] for AGGR_GLOBAL. After copying counts from
>> evsel->prev_raw_counts to evsel->counts, we don't need to
>> modify process_counter_maps in perf_stat_process_counter to let it
>> work well.
> 
> I don't understand why you need to store it in here.. what's the catch
> in process_counter_maps?
> 

Sorry, I didn't explain clearly.

You know the idea is to copy evsel->prev_raw_counts to evsel->counts before 
reporting the summary.

But for AGGR_GLOBAL (cpu = -1 in perf_evsel__compute_deltas), the 
evsel->prev_raw_counts is only stored with the aggr value.

if (cpu == -1) {
	tmp = evsel->prev_raw_counts->aggr;
	evsel->prev_raw_counts->aggr = *count;
} else {
	tmp = *perf_counts(evsel->prev_raw_counts, cpu, thread);
	*perf_counts(evsel->prev_raw_counts, cpu, thread) = *count;
}

So after copying evsel->prev_raw_counts to evsel->counts, 
perf_counts(evsel->counts, cpu, thread) are all 0.

Once we go to process_counter_maps again, in process_counter_values, count->val 
is 0.

case AGGR_GLOBAL:
	aggr->val += count->val;
	aggr->ena += count->ena;
	aggr->run += count->run;

And the aggr->val is 0.

So this patch uses a trick that saves the previous aggr value to cpu0/thread0, 
then above aggr->val calculation can work correctly.

Thanks
Jin Yao

> thanks,
> jirka
>