Re: [RESEND PATCH v2] perf stat: improve readability of shadow stats

[Changbin Du <changbin.du@gmail.com>]

On Tue, Mar 16, 2021 at 02:53:41PM +0100, Jiri Olsa wrote:
> On Mon, Mar 15, 2021 at 10:30:47PM +0800, Changbin Du wrote:
> 
> SNIP
> 
> > diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c
> > index 6ccf21a72f06..3f800e71126f 100644
> > --- a/tools/perf/util/stat-shadow.c
> > +++ b/tools/perf/util/stat-shadow.c
> > @@ -9,6 +9,7 @@
> >  #include "expr.h"
> >  #include "metricgroup.h"
> >  #include "cgroup.h"
> > +#include "units.h"
> >  #include <linux/zalloc.h>
> >  
> >  /*
> > @@ -1270,18 +1271,15 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
> >  		generic_metric(config, evsel->metric_expr, evsel->metric_events, NULL,
> >  				evsel->name, evsel->metric_name, NULL, 1, cpu, out, st);
> >  	} else if (runtime_stat_n(st, STAT_NSECS, cpu, &rsd) != 0) {
> > -		char unit = 'M';
> > -		char unit_buf[10];
> > +		char unit = ' ';
> > +		char unit_buf[10] = "/sec";
> >  
> >  		total = runtime_stat_avg(st, STAT_NSECS, cpu, &rsd);
> > -
> >  		if (total)
> > -			ratio = 1000.0 * avg / total;
> > -		if (ratio < 0.001) {
> > -			ratio *= 1000;
> > -			unit = 'K';
> > -		}
> > -		snprintf(unit_buf, sizeof(unit_buf), "%c/sec", unit);
> > +			ratio = convert_unit_double(1000000000.0 * avg / total, &unit);
> > +
> > +		if (unit != ' ')
> > +			snprintf(unit_buf, sizeof(unit_buf), "%c/sec", unit);
> >  		print_metric(config, ctxp, NULL, "%8.3f", unit_buf, ratio);
> 
> hum, does this still change the metric unit in the csv output? 'perf -a -x,'
>
The unit is changed in csv format, too. See below.

before:
$ sudo ./perf stat -a -x, -- sleep 1
8037.85,msec,cpu-clock,8037851596,100.00,7.999,CPUs utilized
714,,context-switches,8037838466,100.00,0.089,K/sec
11,,cpu-migrations,8037832590,100.00,0.001,K/sec
71,,page-faults,8037824974,100.00,0.009,K/sec
84033551,,cycles,8037750471,100.00,0.010,GHz
22563553,,instructions,8037733879,100.00,0.27,insn per cycle
4685736,,branches,8037708301,100.00,0.583,M/sec
356327,,branch-misses,8037667950,100.00,7.60,of all branches

after:
$ sudo ./perf stat -a -x, -- sleep 1
8026.19,msec,cpu-clock,8026194365,100.00,7.983,CPUs utilized
621,,context-switches,8026178186,100.00,77.372,/sec
16,,cpu-migrations,8026172135,100.00,1.993,/sec
73,,page-faults,8026142626,100.00,9.095,/sec
92645028,,cycles,8026066285,100.00,0.012,GHz
56268285,,instructions,8026048894,100.00,0.61,insn per cycle
10979859,,branches,8026022127,100.00,1.368,M/sec
441719,,branch-misses,8025981169,100.00,4.02,of all branches

But is this a real problem?

> jirka
> 
> >  	} else if (perf_stat_evsel__is(evsel, SMI_NUM)) {
> >  		print_smi_cost(config, cpu, out, st, &rsd);
> > diff --git a/tools/perf/util/units.c b/tools/perf/util/units.c
> > index a46762aec4c9..32c39cfe209b 100644
> > --- a/tools/perf/util/units.c
> > +++ b/tools/perf/util/units.c
> 
> SNIP
> 

-- 
Cheers,
Changbin Du