Performance-Monitoring Counters Library, for Intel/AMD Processors and Linux

This example introduces
   rabbit command-line option -Stats

Return to Main Menu


Command-Line Options

-Stats 0

  The total number of events, and the average number of events per second,
  are reported.  This is the default.

-Stats 1

  The report also includes, for the number of cycles and events, the minimum,
  maximum, mean and standard deviation over all the measured time intervals,
  and similarly for the number of events per cycle over all intervals with a
  nonzero number of cycles.

-Stats 2

  The report also includes, for the ratio of event 0 to event 1, the minimum,
  maximum, mean and standard deviation over all the measured time intervals
  where event 1 has occurred, and similarly for the ratio of event 1 to event
  0 where event 0 has occurred.

  On the Athlon, where four counters are available, all 12 combinations of
  events are reported (0:1, 0:2, 0:3, etc.).


Examples

  rabbit -g 0 date
  rabbit -g 0 -Stats 0 date
  rabbit -g 0 -Stats 1 date
  rabbit -g 0 -Stats 2 date


Example

holmes% rabbit -g 0 -Stats 2 foo d
<output from foo omitted>

---------------------------   performance counters   ---------------------------

Host processor: holmes.scl.ameslab.gov, Pentium II, 450 MHz
Command executed: foo d
  command exited with non-zero status 33
Options: --duration 0,0 --user 1,1 --os 1,1
Options: --compare 0,0 --invert 0,0
Options: --mesi 0xf,0xf --bus_agent 1,1 --MMX 0x3f,0x3f
Sampling: rate = 100 per second, 1403 taken

Event                                                 Events          Events/sec
----------------------------------------    ----------------    ----------------
0x10  16  fp_comp_ops_exe                         1800041849        128091208.62
0x43  67  data_mem_refs                           2232693146        158878730.36

  -group 0
  0: 0x10  floating-point unit, computational FP operations executed (counter 0 only)
  1: 0x43  L1 d-cache, all memory references, both cacheable and non-cacheable

  Events 0x10 0x43  intervals           cycles          event 0          event 1
  total                  1403       6323766016       1800041849       2232693146
  minimum                              2963102           784715          1021242
  maximum                              6326578          1350274          1987445
  mean                             4507317.189      1282994.903      1591370.738
  std.dev.                           66839.983        14255.175        21547.643

  Events per cycle                                   0.28464713       0.35306385
  minimum                1403                        0.21342881       0.31414218
  maximum                                            0.28502803       0.35666604
  mean                                               0.28466519       0.35307363
  std.dev.                                           0.00218924       0.00225280

  Ratio, event 0/1, event 1/0
  minimum                1403             1403       0.67940195       1.20431043
  maximum                                            0.83035069       1.47188274
  mean                                               0.80625664       1.24035970
  std.dev.                                           0.00538332       0.00899868

resource usage:
  time = 14.03 sec user, 0.00 sec sys, 14.06 sec real, 99.77% of cpu
  page reclaims, faults = 11, 92
holmes% 


Notes

  The output format makes it possible to write awk or perl scripts for more
  data-processing, but the raw data files provided with the -output option
  can also be used.  Be aware that, because of the file activity, the
  measurements may differ somewhat.