Re: [PATCH] perf: make perf.data more self-descriptive (v5)

[David Ahern <dsahern@gmail.com>]

On 09/23/2011 07:40 AM, Stephane Eranian wrote:
> 
> 
> On Fri, Sep 23, 2011 at 3:36 PM, David Ahern <dsahern@gmail.com
> <mailto:dsahern@gmail.com>> wrote:
> 
> 
> 
>     On 09/23/2011 03:04 AM, Stephane Eranian wrote:
>     > On Fri, Sep 23, 2011 at 10:03 AM, Pekka Enberg
>     <penberg@cs.helsinki.fi <mailto:penberg@cs.helsinki.fi>> wrote:
>     >> Hi Stephane!
>     >>
>     >> On Fri, Sep 23, 2011 at 10:31 AM, Stephane Eranian
>     <eranian@google.com <mailto:eranian@google.com>> wrote:
>     >>>> So how important is this information? The output is going to be
>     somewhat
>     >>>> awkward for very large CPU counts... :-)
>     >>>>
>     >>> It is useful to determine how CPUs share caches for instance.
>     >>> It can get large but large, but the meta-data header is not
>     printed by
>     >>> default, you need to request it with the -I option.
>     >>
>     >> Well, sure but it blocks rest of the interesting information too.
>     It seems to me
>     >> that the CPU information could be truncated to some sane limit by
>     default and
>     >> introduce a command line option for users that really want to see
>     all of it.
>     >>
>     > Ok, so here is a proposal:
>     > - reorder the info so one liners appear first
>     > - display the "truncated" info by default (no option)
>     > - truncated: numa topo, cpu topo, stop after 4 cpus/nodes, print msg
> 
>     Earlier I gave an example for a 2 socket, quad-core with hyperthreading
>     (16 cpus total):
>        https://lkml.org/lkml/2011/9/6/355
>     The information is repetitive. It would be better to devise a way to
>     reduce the repetition versus truncate the information.
> 
> 
> I have modified the patch to NOT print the CPU, NUMA topology by
> default (but mentioned they are available with the -I option). The
> other bits of information are displayed systematically (no truncation).
> 
> What ways would you propose to still print the info is a less-verbose
> fashion?
> 

for example, sibling cores:
# CPU0 sibling cores  : 0,2,4,6,8,10,12,14

you don't need to print that info for CPUs 0,2,4,6,8,10,12,14. Just
once. Maybe just:

# CPU sibling cores  : 0,2,4,6,8,10,12,14

Similarly for the threads:
# CPU sibling threads: 0,8


That drops the output from 32 lines to 10.

# CPU sibling cores  : 0,2,4,6,8,10,12,14
# CPU sibling cores  : 1,3,5,7,9,11,13,15
# CPU sibling threads: 0,8
# CPU sibling threads: 1,9
# CPU sibling threads: 2,10
# CPU sibling threads: 3,11
# CPU sibling threads: 4,12
# CPU sibling threads: 5,13
# CPU sibling threads: 6,14
# CPU sibling threads: 7,15