Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1342121 > unrolled thread

[PATCHSET 00/18] perf tools: Add support for hierachy view (v7)

Started byNamhyung Kim <namhyung@kernel.org>
First post2016-02-24 16:20 +0100
Last post2016-02-25 03:30 +0100
Articles 4 on this page of 24 — 3 participants

Back to article view | Back to linux.kernel


Contents

  [PATCHSET 00/18] perf tools: Add support for hierachy view (v7) Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:20 +0100
    [PATCH v7 16/18] perf report: Add --hierarchy option Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:20 +0100
      [tip:perf/core] perf report: Add --hierarchy option tip-bot for Namhyung Kim <tipbot@zytor.com> - 2016-02-25 08:50 +0100
    [PATCH v7 12/18] perf hists browser: Support collapsing/expanding whole entries in hierarchy Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:20 +0100
      [tip:perf/core] perf hists browser: Support collapsing/expanding  whole entries in hierarchy tip-bot for Namhyung Kim <tipbot@zytor.com> - 2016-02-25 08:50 +0100
    [PATCH v7 11/18] perf hists browser: Count number of hierarchy entries Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:20 +0100
      [tip:perf/core] perf hists browser: Count number of hierarchy  entries tip-bot for Namhyung Kim <tipbot@zytor.com> - 2016-02-25 08:50 +0100
    [PATCH v7 13/18] perf hists browser: Implement hierarchy output Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:20 +0100
      [tip:perf/core] perf hists browser: Implement hierarchy output tip-bot for Namhyung Kim <tipbot@zytor.com> - 2016-02-25 08:50 +0100
    [PATCH v7 07/18] perf hists: Resort after filtering hierarchy Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:30 +0100
      [tip:perf/core] perf hists: Resort after filtering hierarchy tip-bot for Namhyung Kim <tipbot@zytor.com> - 2016-02-25 08:50 +0100
    [PATCH v7 05/18] perf hists: Introduce hist_entry__filter() Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:30 +0100
      Re: [PATCH v7 05/18] perf hists: Introduce hist_entry__filter() Arnaldo Carvalho de Melo <acme@kernel.org> - 2016-02-25 00:30 +0100
        Re: [PATCH v7 05/18] perf hists: Introduce hist_entry__filter() Namhyung Kim <namhyung@kernel.org> - 2016-02-25 03:10 +0100
      [tip:perf/core] perf hists: Introduce hist_entry__filter() tip-bot for Namhyung Kim <tipbot@zytor.com> - 2016-02-25 08:50 +0100
    [PATCH v7 03/18] perf hists: Resort hist entries with hierarchy Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:30 +0100
      [tip:perf/core] perf hists: Resort hist entries with hierarchy tip-bot for Namhyung Kim <tipbot@zytor.com> - 2016-02-25 08:50 +0100
    [PATCH v7 02/18] perf hists: Basic support of hierarchical report view Namhyung Kim <namhyung@kernel.org> - 2016-02-24 16:30 +0100
      [tip:perf/core] perf hists: Basic support of hierarchical report  view tip-bot for Namhyung Kim <tipbot@zytor.com> - 2016-02-25 08:50 +0100
    Re: [PATCHSET 00/18] perf tools: Add support for hierachy view (v7) Arnaldo Carvalho de Melo <acme@kernel.org> - 2016-02-24 21:10 +0100
    Re: [PATCHSET 00/18] perf tools: Add support for hierachy view (v7) Arnaldo Carvalho de Melo <acme@kernel.org> - 2016-02-25 00:30 +0100
      Re: [PATCHSET 00/18] perf tools: Add support for hierachy view (v7) Namhyung Kim <namhyung@kernel.org> - 2016-02-25 03:20 +0100
    Re: [PATCHSET 00/18] perf tools: Add support for hierachy view (v7) Arnaldo Carvalho de Melo <acme@kernel.org> - 2016-02-25 02:30 +0100
      Re: [PATCHSET 00/18] perf tools: Add support for hierachy view (v7) Namhyung Kim <namhyung@kernel.org> - 2016-02-25 03:30 +0100

Page 2 of 2 — ← Prev page 1 [2]


#1342518

FromArnaldo Carvalho de Melo <acme@kernel.org>
Date2016-02-25 00:30 +0100
Message-ID<r5R2y-3qM-25@gated-at.bofh.it>
In reply to#1342121
Em Thu, Feb 25, 2016 at 12:13:32AM +0900, Namhyung Kim escreveu:
> Hello,
> 
> This patchset implements a new feature that collects hist entries in a
> hierachical manner.  That means lower-level entries belong to an
> upper-level entry.  The entry hierachy is built on the sort keys
> given, so users can set it whatever they want.  It only shows
> top-level entries first, and user can expand/collapse it dynamically.
> 
>  * Changes from v6)
>   - pass srcline, srcfile and trace_output of hist entry properly  (Jiri)
>   - use hist_entry__snprintf_alignment()  (Jiri)

Suggestion for further improvement:

perf top --hierarchy -e sched:sched_switch -s prev_comm,prev_pid,next_comm,next_pid

Will have four levels, would be nice to have just two, perhaps expressed
via:

perf top --hierarchy -e sched:sched_switch -s prev_comm+prev_pid,next_comm+next_pid

Where '+' would work as sort order separator, just like ',' but would
have special meaning for --hierarchy.

What do you think?

- Arnaldo

[toc] | [prev] | [next] | [standalone]


#1342618

FromNamhyung Kim <namhyung@kernel.org>
Date2016-02-25 03:20 +0100
Message-ID<r5TH4-5ok-9@gated-at.bofh.it>
In reply to#1342518
On Wed, Feb 24, 2016 at 08:29:32PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Thu, Feb 25, 2016 at 12:13:32AM +0900, Namhyung Kim escreveu:
> > Hello,
> > 
> > This patchset implements a new feature that collects hist entries in a
> > hierachical manner.  That means lower-level entries belong to an
> > upper-level entry.  The entry hierachy is built on the sort keys
> > given, so users can set it whatever they want.  It only shows
> > top-level entries first, and user can expand/collapse it dynamically.
> > 
> >  * Changes from v6)
> >   - pass srcline, srcfile and trace_output of hist entry properly  (Jiri)
> >   - use hist_entry__snprintf_alignment()  (Jiri)
> 
> Suggestion for further improvement:
> 
> perf top --hierarchy -e sched:sched_switch -s prev_comm,prev_pid,next_comm,next_pid
> 
> Will have four levels, would be nice to have just two, perhaps expressed
> via:
> 
> perf top --hierarchy -e sched:sched_switch -s prev_comm+prev_pid,next_comm+next_pid
> 
> Where '+' would work as sort order separator, just like ',' but would
> have special meaning for --hierarchy.
> 
> What do you think?

Looks like a good addition.  To do that, a hist entry has to have its
own perf_hpp_list instead of a single perf_hpp_fmt.  I'll try to
implement it..

Thanks,
Namhyung

[toc] | [prev] | [next] | [standalone]


#1342568

FromArnaldo Carvalho de Melo <acme@kernel.org>
Date2016-02-25 02:30 +0100
Message-ID<r5SUG-4Ka-7@gated-at.bofh.it>
In reply to#1342121
Em Thu, Feb 25, 2016 at 12:13:32AM +0900, Namhyung Kim escreveu:
> Hello,
> 
> This patchset implements a new feature that collects hist entries in a
> hierachical manner.  That means lower-level entries belong to an
> upper-level entry.  The entry hierachy is built on the sort keys
> given, so users can set it whatever they want.  It only shows
> top-level entries first, and user can expand/collapse it dynamically.

Some oddity: look at the line with 3.92%, it shows just the DSO... Ok, there
probably are no symbols above the specified threshold (--percent-limit 1), but
it is strange none the less, perhaps we can add something like:

#       Overhead  Command / Shared Object / Symbol      
# ..............  ..............................................................................................................................................
#
    72.83%        cc1       
       60.88%        cc1                    
           1.92%        [.] bitmap_set_bit
        7.37%        libc-2.22.so           
           2.90%        [.] _int_malloc
            |          
             --2.04%--0
                       _int_malloc

           1.13%        [.] _int_free
        3.92%        [kernel.vmlinux]       
           (no symbols >= 1%)


Original output:


[acme@jouet linux]$ perf report --hierarchy --percent-limit 1 | head -30
no symbols found in /usr/bin/gcc, maybe install a debug package?
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 10K of event 'cycles:ppp'
# Event count (approx.): 6703622666
#
#       Overhead  Command / Shared Object / Symbol      
# ..............  ..............................................................................................................................................
#
    72.83%        cc1       
       60.88%        cc1                    
           1.92%        [.] bitmap_set_bit
        7.37%        libc-2.22.so           
           2.90%        [.] _int_malloc
            |          
             --2.04%--0
                       _int_malloc

           1.13%        [.] _int_free
        3.92%        [kernel.vmlinux]       
     5.96%        as        
        3.19%        [kernel.vmlinux]       
        1.31%        as                     
     5.33%        conf      
        2.39%        conf                   
        1.52%        libc-2.22.so           
        1.37%        [kernel.vmlinux]       
     4.88%        ld        
[acme@jouet linux]$ 

[toc] | [prev] | [next] | [standalone]


#1342670

FromNamhyung Kim <namhyung@kernel.org>
Date2016-02-25 03:30 +0100
Message-ID<r5TQM-5rV-67@gated-at.bofh.it>
In reply to#1342568
On Wed, Feb 24, 2016 at 10:22:07PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Thu, Feb 25, 2016 at 12:13:32AM +0900, Namhyung Kim escreveu:
> > Hello,
> > 
> > This patchset implements a new feature that collects hist entries in a
> > hierachical manner.  That means lower-level entries belong to an
> > upper-level entry.  The entry hierachy is built on the sort keys
> > given, so users can set it whatever they want.  It only shows
> > top-level entries first, and user can expand/collapse it dynamically.
> 
> Some oddity: look at the line with 3.92%, it shows just the DSO... Ok, there
> probably are no symbols above the specified threshold (--percent-limit 1), but
> it is strange none the less, perhaps we can add something like:
> 
> #       Overhead  Command / Shared Object / Symbol      
> # ..............  ..............................................................................................................................................
> #
>     72.83%        cc1       
>        60.88%        cc1                    
>            1.92%        [.] bitmap_set_bit
>         7.37%        libc-2.22.so           
>            2.90%        [.] _int_malloc
>             |          
>              --2.04%--0
>                        _int_malloc
> 
>            1.13%        [.] _int_free
>         3.92%        [kernel.vmlinux]       
>            (no symbols >= 1%)

Hmm.. makes sense.  But the problem is that these non-entry lines
hinder counting the number of (printed) entries which might break the
output in some way.  Maybe I can use a mock entry to work around it.
I'll try to do something. :)

Thanks,
Namhyung


> 
> 
> Original output:
> 
> 
> [acme@jouet linux]$ perf report --hierarchy --percent-limit 1 | head -30
> no symbols found in /usr/bin/gcc, maybe install a debug package?
> # To display the perf.data header info, please use --header/--header-only options.
> #
> #
> # Total Lost Samples: 0
> #
> # Samples: 10K of event 'cycles:ppp'
> # Event count (approx.): 6703622666
> #
> #       Overhead  Command / Shared Object / Symbol      
> # ..............  ..............................................................................................................................................
> #
>     72.83%        cc1       
>        60.88%        cc1                    
>            1.92%        [.] bitmap_set_bit
>         7.37%        libc-2.22.so           
>            2.90%        [.] _int_malloc
>             |          
>              --2.04%--0
>                        _int_malloc
> 
>            1.13%        [.] _int_free
>         3.92%        [kernel.vmlinux]       
>      5.96%        as        
>         3.19%        [kernel.vmlinux]       
>         1.31%        as                     
>      5.33%        conf      
>         2.39%        conf                   
>         1.52%        libc-2.22.so           
>         1.37%        [kernel.vmlinux]       
>      4.88%        ld        
> [acme@jouet linux]$ 

[toc] | [prev] | [standalone]


Page 2 of 2 — ← Prev page 1 [2]

Back to top | Article view | linux.kernel


csiph-web