Re: perf stat stddev reporting broken since "perf stat: Introduce read_counters function"

[Jiri Olsa]

On Mon, Jan 18, 2016 at 05:59:46PM +0000, Mel Gorman wrote:
> I noticed that perf stat in 4.4 is not printing stddev figures as
> expected and bisected it to commit 106a94a0f8c2 ("perf stat: Introduce
> read_counters function").
> 
> # Kernel 4.4
> # make clean && make prefix=/usr && ./perf stat -r 5 sleep 1
>  Performance counter stats for 'sleep 1' (5 runs):
> 
>           0.466683      task-clock (msec)         #    0.000 CPUs utilized
>                  1      context-switches          #    0.002 M/sec
>                  1      cpu-migrations            #    0.002 M/sec
>                 66      page-faults               #    0.144 M/sec
>          1,630,220      cycles                    #    3.562 GHz
>    <not supported>      stalled-cycles-frontend
>    <not supported>      stalled-cycles-backend
>            902,686      instructions              #    0.66  insns per cycle
>            185,692      branches                  #  405.786 M/sec
>              8,640      branch-misses             #    4.63% of all branches
> 
>        1.000855883 seconds time elapsed           ( +-  0.01% )
> 
> # git checkout 106a94a0f8c207ef4113ce7e32f34a00b3b174e7^
>  Performance counter stats for 'sleep 1' (5 runs):
> 
>           0.397939      task-clock (msec)         #    0.000 CPUs utilized            ( +-  3.37% )
>                  1      context-switches          #    0.003 M/sec
>                  0      cpu-migrations            #    0.000 K/sec
>                 67      page-faults               #    0.169 M/sec                    ( +-  0.56% )
>          1,360,763      cycles                    #    3.420 GHz                      ( +-  2.14% )
>    <not supported>      stalled-cycles-frontend
>    <not supported>      stalled-cycles-backend
>            903,901      instructions              #    0.66  insns per cycle          ( +-  0.24% )
>            186,472      branches                  #  468.595 M/sec                    ( +-  0.29% )
>              8,599      branch-misses             #    4.61% of all branches          ( +-  0.77% )
> 
>        1.000761882 seconds time elapsed                                               ( +-  0.00% )
> 
> # git checkout 106a94a0f8c207ef4113ce7e32f34a00b3b174e7
> 
>  Performance counter stats for 'sleep 1' (5 runs):
> 
>           0.583178      task-clock (msec)         #    0.001 CPUs utilized
>                  1      context-switches          #    0.002 M/sec
>                  0      cpu-migrations            #    0.000 K/sec
>                 67      page-faults               #    0.141 M/sec
>          1,224,416      cycles                    #    2.585 GHz
>    <not supported>      stalled-cycles-frontend
>    <not supported>      stalled-cycles-backend
>            892,568      instructions              #    0.68  insns per cycle
>            184,975      branches                  #  390.554 M/sec
>              8,512      branch-misses             #    4.57% of all branches
> 
>        1.000889249 seconds time elapsed                                          ( +-  0.01% )
> 
> I checked and current master in Linus' tree is also broken. Is this
> change in documented perf stat behaviour intentional?

nope, not intentional.. I'll check on that, thanks for reporting

jirka