Re: [PATCH 3/9] perf, tools, stat: Move noise/running printing into printout

2016-01-28 Thread Jiri Olsa 
On Wed, Jan 27, 2016 at 12:57:43AM +0100, Andi Kleen wrote:
> On Sat, Jan 23, 2016 at 05:19:58PM +0100, Jiri Olsa wrote:
> > On Sat, Jan 23, 2016 at 04:43:05AM +0100, Andi Kleen wrote:
> > > > > - if (run == 0 || ena == 0) {
> > > > > - fprintf(output, "CPU%*d%s%*s%s",
> > > > > - csv_output ? 0 : -4,
> > > > > - perf_evsel__cpus(counter)->map[cpu], 
> > > > > csv_sep,
> > > > > - csv_output ? 0 : 18,
> > > > > - counter->supported ? CNTR_NOT_COUNTED : 
> > > > > CNTR_NOT_SUPPORTED,
> > > > > - csv_sep);
> > > > 
> > > > this hunk is not preserved in the new code.. I guess the output is
> > > > different for -A if counter wasn't meassure?
> > > 
> > > The code for this is common in printout() now.
> > 
> > but what will be printed for -A not counted counters?
> > not the "CPU%*d%s%*s%s" ...
> 
> 
> CPU0 stalled-cycles-backend
>   
> CPU1 stalled-cycles-backend
>   
> CPU2 stalled-cycles-backend
>   
> CPU3 stalled-cycles-backend
>   
> CPU4 stalled-cycles-backend
>   
> CPU5 stalled-cycles-backend
>   
> CPU6 stalled-cycles-backend
>   
> CPU7 stalled-cycles-backend
>   
> CPU8 stalled-cycles-backend
>   
> CPU9 stalled-cycles-backend
>   
> CPU10stalled-cycles-backend
>   
> CPU11stalled-cycles-backend
>   

ah the aggr_printout provides that now.. which wasn't
the case for the removed code

thanks,
jirka

Re: [PATCH 3/9] perf, tools, stat: Move noise/running printing into printout

2016-01-26 Thread Andi Kleen 
On Sat, Jan 23, 2016 at 05:19:58PM +0100, Jiri Olsa wrote:
> On Sat, Jan 23, 2016 at 04:43:05AM +0100, Andi Kleen wrote:
> > > > -   if (run == 0 || ena == 0) {
> > > > -   fprintf(output, "CPU%*d%s%*s%s",
> > > > -   csv_output ? 0 : -4,
> > > > -   perf_evsel__cpus(counter)->map[cpu], 
> > > > csv_sep,
> > > > -   csv_output ? 0 : 18,
> > > > -   counter->supported ? CNTR_NOT_COUNTED : 
> > > > CNTR_NOT_SUPPORTED,
> > > > -   csv_sep);
> > > 
> > > this hunk is not preserved in the new code.. I guess the output is
> > > different for -A if counter wasn't meassure?
> > 
> > The code for this is common in printout() now.
> 
> but what will be printed for -A not counted counters?
> not the "CPU%*d%s%*s%s" ...


CPU0 stalled-cycles-backend  

CPU1 stalled-cycles-backend  

CPU2 stalled-cycles-backend  

CPU3 stalled-cycles-backend  

CPU4 stalled-cycles-backend  

CPU5 stalled-cycles-backend  

CPU6 stalled-cycles-backend  

CPU7 stalled-cycles-backend  

CPU8 stalled-cycles-backend  

CPU9 stalled-cycles-backend  

CPU10stalled-cycles-backend  

CPU11stalled-cycles-backend  


-Andi

Re: [PATCH 3/9] perf, tools, stat: Move noise/running printing into printout

2016-01-26 Thread Andi Kleen 
On Fri, Jan 22, 2016 at 10:41:53AM +0100, Jiri Olsa wrote:
> On Tue, Jan 19, 2016 at 06:26:46PM -0800, Andi Kleen wrote:
> 
> SNIP
> 
> > uval = avg * counter->scale;
> > -   printout(-1, 0, counter, uval, prefix);
> > -
> > -   print_noise(counter, avg);
> > -
> > -   print_running(avg_running, avg_enabled);
> > +   printout(-1, 0, counter, uval, prefix, avg_running, avg_enabled, avg);
> > fprintf(output, "\n");
> >  }
> >  
> > @@ -1037,36 +1006,8 @@ static void print_counter(struct perf_evsel 
> > *counter, char *prefix)
> > if (prefix)
> > fprintf(output, "%s", prefix);
> >  
> > -   if (run == 0 || ena == 0) {
> > -   fprintf(output, "CPU%*d%s%*s%s",
> > -   csv_output ? 0 : -4,
> > -   perf_evsel__cpus(counter)->map[cpu], csv_sep,
> > -   csv_output ? 0 : 18,
> > -   counter->supported ? CNTR_NOT_COUNTED : 
> > CNTR_NOT_SUPPORTED,
> > -   csv_sep);
> 
> this hunk is not preserved in the new code.. I guess the output is
> different for -A if counter wasn't meassure?

It's the same, at least in the final state.

-Andi

Re: [PATCH 3/9] perf, tools, stat: Move noise/running printing into printout

2016-01-23 Thread Jiri Olsa 
On Sat, Jan 23, 2016 at 04:43:05AM +0100, Andi Kleen wrote:
> > > - if (run == 0 || ena == 0) {
> > > - fprintf(output, "CPU%*d%s%*s%s",
> > > - csv_output ? 0 : -4,
> > > - perf_evsel__cpus(counter)->map[cpu], csv_sep,
> > > - csv_output ? 0 : 18,
> > > - counter->supported ? CNTR_NOT_COUNTED : 
> > > CNTR_NOT_SUPPORTED,
> > > - csv_sep);
> > 
> > this hunk is not preserved in the new code.. I guess the output is
> > different for -A if counter wasn't meassure?
> 
> The code for this is common in printout() now.

but what will be printed for -A not counted counters?
not the "CPU%*d%s%*s%s" ...

jirka

Re: [PATCH 3/9] perf, tools, stat: Move noise/running printing into printout

2016-01-22 Thread Andi Kleen 
> > -   if (run == 0 || ena == 0) {
> > -   fprintf(output, "CPU%*d%s%*s%s",
> > -   csv_output ? 0 : -4,
> > -   perf_evsel__cpus(counter)->map[cpu], csv_sep,
> > -   csv_output ? 0 : 18,
> > -   counter->supported ? CNTR_NOT_COUNTED : 
> > CNTR_NOT_SUPPORTED,
> > -   csv_sep);
> 
> this hunk is not preserved in the new code.. I guess the output is
> different for -A if counter wasn't meassure?

The code for this is common in printout() now.

-Andi

Re: [PATCH 3/9] perf, tools, stat: Move noise/running printing into printout

2016-01-22 Thread Jiri Olsa 
On Tue, Jan 19, 2016 at 06:26:46PM -0800, Andi Kleen wrote:

SNIP

>   uval = avg * counter->scale;
> - printout(-1, 0, counter, uval, prefix);
> -
> - print_noise(counter, avg);
> -
> - print_running(avg_running, avg_enabled);
> + printout(-1, 0, counter, uval, prefix, avg_running, avg_enabled, avg);
>   fprintf(output, "\n");
>  }
>  
> @@ -1037,36 +1006,8 @@ static void print_counter(struct perf_evsel *counter, 
> char *prefix)
>   if (prefix)
>   fprintf(output, "%s", prefix);
>  
> - if (run == 0 || ena == 0) {
> - fprintf(output, "CPU%*d%s%*s%s",
> - csv_output ? 0 : -4,
> - perf_evsel__cpus(counter)->map[cpu], csv_sep,
> - csv_output ? 0 : 18,
> - counter->supported ? CNTR_NOT_COUNTED : 
> CNTR_NOT_SUPPORTED,
> - csv_sep);

this hunk is not preserved in the new code.. I guess the output is
different for -A if counter wasn't meassure?

thanks,
jirka