Messages in this thread | | | Date | Wed, 6 Jan 2016 15:38:53 +0100 | From | Jiri Olsa <> | Subject | Re: [PATCHv9 0/3] perf stat: Add scripting support |
| |
On Wed, Jan 06, 2016 at 11:18:02AM -0300, Arnaldo Carvalho de Melo wrote: > Em Wed, Jan 06, 2016 at 11:49:54AM +0100, Jiri Olsa escreveu: > > hi, > > sending another version of stat scripting. > > > v9 changes: > > - rebased on top of accepted patches > > - desribed CPI metric in changelog [Arnaldo] > > - fixed cpu conversion [Arnaldo] > > Thanks, applied. Testing the cpi script with an endless: > > perf stat record -I 1000 -a | perf script -s ~acme/git/linux/tools/perf/scripts/python/stat-cpi.py > 187.151796: cpu -1, thread -1 -> cpi 0.797917 (2568467461/3218963700) > 188.151979: cpu -1, thread -1 -> cpi 0.734628 (2714373371/3694892981) > 189.152212: cpu -1, thread -1 -> cpi 0.753332 (2958819204/3927644236) > 190.152975: cpu -1, thread -1 -> cpi 1.587754 (202360895/127451009) > 191.153486: cpu -1, thread -1 -> cpi 1.558219 (290557309/186467579) > > in one monitor while trying various workloads in another, to see how it reacts, > and after a while those raw numbers at the end of each line become just noise, > I think they should be shown only under 'stat-cpy -v' (I think we can pass args > to the 'perf script' scripts, right? > > Also the 'cpu', 'thread' and '-> cpi' could be turned into headers, like 'perf > stat' does? > > Anyway, you called it 'example script' and we can improve it on top of what I > pushed to perf/core, i.e. this patchkit, unchanged, thanks!
yep, it's just noise and it's just an example to show how the stat scripting works.. we can tune it up with more ratios ;-)
jirka
| |