lkml.org 
[lkml]   [2012]   [Sep]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    Subject[RFC 00/12] perf diff: Factor diff command
    Date
    hi,
    this patchset factors the perf diff command to be usable for
    differential profiling following paper from Paul McKenney:
    (thanks to Arnaldo for sharing it with me).

    http://www2.rdrop.com/users/paulmck/scalability/paper/profiling.2002.06.04.pdf

    The 'perf diff' and 'std/hist' code is now changed to allow computations
    mentioned in the paper. Two of them are implemented within this patchset:
    1) ratio differential profiling
    2) weighted differential profiling

    The standard ratio delta computation stays as default.

    To sum it up:
    - perf diff displays output for matching event pairs within 2 given perf.data files
    - stdio ui code is factored to allow easy insertion of new data column
    - added perf diff '-b' option to display only matched hist entries
    (hist entries found in both files)
    - added perf diff '-c' option to choose diff computation,
    support for:
    delta: the current default one
    ratio: ratio differential profile
    wdiff: weighted differential profile
    - added perf diff '-c+' option to sort entries based on the computation data
    - added perf diff '-F' option to show formula used to compute the data
    - added perf diff '-p' option to display hist entries periods


    Attached patches:
    01/12 perf diff: Make diff command work with evsel hists
    02/12 perf tools: Replace sort's standalone field_sep with symbol_conf.field_sep
    03/12 perf hists: Add struct hists pointer to struct hist_entry
    04/12 perf diff: Refactor diff displacement possition info
    05/12 perf diff: Refactor stdio ui data columns output
    06/12 perf diff: Add -b option for perf diff to display paired entries only
    07/12 perf diff: Add ratio computation way to compare hist entries
    08/12 perf diff: Add option to sort entries based on diff computation
    09/12 perf diff: Add weighted diff computation way to compare hist entries
    10/12 perf diff: Add -p option to display period values for hist entries
    11/12 perf diff: Add -F option to display formula for computation
    12/12 perf diff: Add -F option for ratio computation


    I'm still testing this, trying to find out useful outputs/computations/options,
    so looking for any ideas and recommendations ;)

    thanks,
    jirka


    Eamples:

    display default profile
    -----------------------------------------------------------------------------------
    $ ./perf diff
    # Event 'cache-misses:u'
    #
    # Baseline Delta Shared Object Symbol
    # ........ ........ .................. .................................
    #
    0.00% +63.54% libc-2.15.so [.] __dcigettext
    0.00% +5.38% libc-2.15.so [.] _dl_addr
    0.00% +5.30% libc-2.15.so [.] __register_atfork
    0.31% +3.94% [kernel.kallsyms] [k] page_fault
    0.00% +4.07% ld-2.15.so [.] check_match.11335
    0.00% +3.65% ld-2.15.so [.] version_check_doit
    0.00% +3.56% ld-2.15.so [.] _dl_fixup
    0.00% +3.05% ld-2.15.so [.] _dl_map_object
    0.00% +2.90% [kernel.kallsyms] [k] system_call
    3.94% -1.53% [kernel.kallsyms] [k] device_not_available
    0.00% +1.21% libc-2.15.so [.] __GI___libc_write
    0.00% +0.54% libc-2.15.so [.] __memcpy_ssse3_back
    0.00% +0.11% libc-2.15.so [.] execvp
    7.71% -7.69% ld-2.15.so [.] _dl_start
    0.03% -0.02% libpthread-2.15.so [.] __read_nocancel
    0.20% -0.18% perf [.] perf_evlist__prepare_workload



    display ratio profile
    -----------------------------------------------------------------------------------
    $ ./perf diff -cratio
    # Event 'cache-misses:u'
    #
    # Baseline Ratio Shared Object Symbol
    # ........ .............. .................. .................................
    #
    0.00% 0.000 libc-2.15.so [.] __dcigettext
    0.00% 0.000 libc-2.15.so [.] _dl_addr
    0.00% 0.000 libc-2.15.so [.] __register_atfork
    0.31% 15.450 [kernel.kallsyms] [k] page_fault
    0.00% 0.000 ld-2.15.so [.] check_match.11335
    0.00% 0.000 ld-2.15.so [.] version_check_doit
    0.00% 0.000 ld-2.15.so [.] _dl_fixup
    0.00% 0.000 ld-2.15.so [.] _dl_map_object
    0.00% 0.000 [kernel.kallsyms] [k] system_call
    3.94% 0.678 [kernel.kallsyms] [k] device_not_available
    0.00% 0.000 libc-2.15.so [.] __GI___libc_write
    0.00% 0.000 libc-2.15.so [.] __memcpy_ssse3_back
    0.00% 0.000 libc-2.15.so [.] execvp
    7.71% 0.002 ld-2.15.so [.] _dl_start
    0.03% 0.500 libpthread-2.15.so [.] __read_nocancel
    0.20% 0.077 perf [.] perf_evlist__prepare_workload



    display ratio profile only with entries matched in both files
    -----------------------------------------------------------------------------------
    $ ./perf diff -cratio -b

    # Event 'cache-misses:u'
    #
    # Baseline Ratio Shared Object Symbol
    # ........ .............. .................. .................................
    #
    0.31% 15.450 [kernel.kallsyms] [k] page_fault
    3.94% 0.678 [kernel.kallsyms] [k] device_not_available
    7.71% 0.002 ld-2.15.so [.] _dl_start
    0.03% 0.500 libpthread-2.15.so [.] __read_nocancel
    0.20% 0.077 perf [.] perf_evlist__prepare_workload



    display ratio profile only with entries matched in both files and sorted
    -----------------------------------------------------------------------------------
    $ ./perf diff -c+ratio -b

    # Event 'cache-misses:u'
    #
    # Baseline Ratio Shared Object Symbol
    # ........ .............. .................. .................................
    #
    0.31% 15.450 [kernel.kallsyms] [k] page_fault
    3.94% 0.678 [kernel.kallsyms] [k] device_not_available
    0.03% 0.500 libpthread-2.15.so [.] __read_nocancel
    0.20% 0.077 perf [.] perf_evlist__prepare_workload
    7.71% 0.002 ld-2.15.so [.] _dl_start



    display weighted profile with weights w1=1 w2=2, with formula, sorted, matching
    entries only and with periods displayed
    -----------------------------------------------------------------------------------
    $ ./perf diff -c+wdiff:1,2 -F -b -p

    # Baseline Weighted diff Formula Baseline Period Period Shared Object Symbol
    # ........ ............. .................................................. ............... ............ .................. .................................
    #
    0.31% +598 (309 * 2) - (20 * 1) 20 309 [kernel.kallsyms] [k] page_fault
    3.94% +92 (175 * 2) - (258 * 1) 258 175 [kernel.kallsyms] [k] device_not_available
    0.03% +0 (1 * 2) - (2 * 1) 2 1 libpthread-2.15.so [.] __read_nocancel
    0.20% -11 (1 * 2) - (13 * 1) 13 1 perf [.] perf_evlist__prepare_workload
    7.71% -503 (1 * 2) - (505 * 1) 505 1 ld-2.15.so [.] _dl_start


    Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
    Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
    Cc: Ingo Molnar <mingo@elte.hu>
    Cc: Paul Mackerras <paulus@samba.org>
    Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
    Cc: Frederic Weisbecker <fweisbec@gmail.com>
    Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
    Cc: Andi Kleen <andi@firstfloor.org>
    Cc: David Ahern <dsahern@gmail.com>
    Cc: Namhyung Kim <namhyung@kernel.org>

    ---
    tools/perf/Documentation/perf-diff.txt | 63 ++++++++++
    tools/perf/builtin-diff.c | 488 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++------
    tools/perf/builtin-report.c | 6 +-
    tools/perf/builtin-top.c | 6 +-
    tools/perf/ui/stdio/hist.c | 574 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-----------------------------------
    tools/perf/ui/stdio/hist.h | 26 ++++
    tools/perf/util/evsel.h | 7 ++
    tools/perf/util/hist.c | 7 +-
    tools/perf/util/hist.h | 24 +++-
    tools/perf/util/session.h | 4 +-
    tools/perf/util/sort.c | 6 +-
    tools/perf/util/sort.h | 22 +++-
    12 files changed, 957 insertions(+), 276 deletions(-)


    \
     
     \ /
      Last update: 2012-09-06 18:21    [W:4.108 / U:0.276 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site