Re: profiling emacs-23.1 vs emacs-22.3

[Dan Nicolaescu]

Kenichi Handa <address@hidden> writes:

  > In article <address@hidden>, Dan Nicolaescu <address@hidden> writes:
  > 
  > > Kenichi Handa <address@hidden> writes:
  > > In article <address@hidden>, Dan Nicolaescu <address@hidden> writes:
  > > 
  > > > Could you try the attached patch?  If it improves the
  > > 
  > > > It does improve performance:
  > > 
  > > How much in real time?
  > 
  > > from 22 seconds to 19 (vs about 16 for 22.3 )
  > 
  > I see.  Then it is worth installing that patch.  I've just
  > done it.

Thanks.

  > > [...]
  > > > It's still slower than 22.3 though.
  > > 
  > > > One big difference is then time/number of calls to mark_objects 
  > > > 129733 vs 18834514, so 145 times more calls to mark_object.
  > > > Do you know where do those come from?
  > > 
  > > > The number of Fgarbage_collect calls does not increase that much: 
  > > > from 37 (for 22.3) to  43 (for 23.1).
  > > 
  > > It seems that c-indent-region has been changed a lot.  I'm
  > > not sure, but perhaps that is the reason.  How do the other
  > > people think?
  > 
  > > I set the load path to cc-mode from 22.3, it did not make any significant
  > > difference in the amount of mark_object and Fgarbage_collect calls.
  > 
  > Ok, then I suspect that the slowness is because of newly
  > introduced char-tables.  I've just installed the attached
  > change too to improve the performance of object marking in
  > GC.
  > 
  > Please try again with the latest code.

It's better, but still not too good:

  %   cumulative   self              self     total           
 time   seconds   seconds    calls   s/call   s/call  name    
 43.03      8.55     8.55  4864529     0.00     0.00  forw_comment
 17.16     11.96     3.41    32860     0.00     0.00  scan_sexps_forward
  7.90     13.53     1.57    17306     0.00     0.00  scan_lists
  7.20     14.96     1.43    50443     0.00     0.00  re_search_2
  4.43     15.84     0.88 33382219     0.00     0.00  lookup_char_property
  2.42     16.32     0.48 18504890     0.00     0.00  next_interval
  2.21     16.76     0.44  8177518     0.00     0.00  mark_object
  1.86     17.13     0.37 14382565     0.00     0.00  previous_interval
  1.66     17.46     0.33  3329754     0.00     0.00  re_match_2_internal
  1.26     17.71     0.25  2156399     0.00     0.00  update_syntax_table
  1.16     17.94     0.23    13472     0.00     0.00  
Fprevious_single_property_change
  1.01     18.14     0.20 23703332     0.00     0.00  Fassq
  1.01     18.34     0.20 57233041     0.00     0.00  Fcdr
  0.81     18.50     0.16   201989     0.00     0.00  Fbyte_code
  0.70     18.64     0.14       43     0.00     0.02  Fgarbage_collect
  0.45     18.73     0.09  3943360     0.00     0.00  mark_vectorlike
  0.45     18.82     0.09  1976191     0.00     0.00  update_interval
  0.40     18.90     0.08   103511     0.00     0.00  skip_chars
  0.40     18.98     0.08     3111     0.00     0.00  mark_char_table
  0.35     19.05     0.07 33382219     0.00     0.00  textget
  0.25     19.10     0.05  1076559     0.00     0.00  find_interval
  0.25     19.15     0.05   893966     0.00     0.00  re_iswctype
  0.25     19.20     0.05     1294     0.00     0.00  Fsetcar
  0.20     19.24     0.04    41380     0.00     0.00  re_compile_pattern
  0.15     19.27     0.03  3613707     0.00     0.00  buf_bytepos_to_charpos
  0.15     19.30     0.03   589363     0.00     0.00  set_point_both
  0.15     19.33     0.03   329641     0.00     0.00  validate_interval_range
  0.15     19.36     0.03    15287     0.00     0.00  adjust_for_invis_intang
  0.15     19.39     0.03                             interval_start_pos
  0.10     19.41     0.02   987278     0.00     0.00  Fcons
 

compare to doing the same thing in an emacs -Q -nw session:

  %   cumulative   self              self     total           
 time   seconds   seconds    calls   s/call   s/call  name    
 46.55      9.10     9.10  4864529     0.00     0.00  forw_comment
 15.40     12.11     3.01    32860     0.00     0.00  scan_sexps_forward
  7.31     13.54     1.43    17306     0.00     0.00  scan_lists
  7.16     14.94     1.40    27938     0.00     0.00  re_search_2
  4.40     15.80     0.86 33368915     0.00     0.00  lookup_char_property
  2.35     16.26     0.46 18498165     0.00     0.00  next_interval
  1.74     16.60     0.34    13472     0.00     0.00  
Fprevious_single_property_change
  1.64     16.92     0.32  3319319     0.00     0.00  re_match_2_internal
  1.53     17.22     0.30  5830028     0.00     0.00  mark_object
  1.53     17.52     0.30  2155779     0.00     0.00  update_syntax_table
  1.48     17.81     0.29 14382565     0.00     0.00  previous_interval
  1.28     18.06     0.25 23687804     0.00     0.00  Fassq
  0.72     18.20     0.14 57187184     0.00     0.00  Fcdr
  0.66     18.33     0.13   198744     0.00     0.00  Fbyte_code
  0.51     18.43     0.10 33368915     0.00     0.00  textget
  0.36     18.50     0.07    15287     0.00     0.00  adjust_for_invis_intang
  0.36     18.57     0.07     2744     0.00     0.00  Frassq
  0.36     18.64     0.07       38     0.00     0.01  Fgarbage_collect
  0.31     18.70     0.06   103511     0.00     0.00  skip_chars
  0.26     18.75     0.05 10044672     0.00     0.00  Fcar
  0.26     18.80     0.05  1976191     0.00     0.00  update_interval
  0.26     18.85     0.05     1862     0.00     0.00  mark_char_table
  0.20     18.89     0.04  1066193     0.00     0.00  find_interval
  0.20     18.93     0.04    73761     0.00     0.00  Flocal_variable_p
  0.20     18.97     0.04    20489     0.00     0.00  re_compile_pattern
  0.20     19.01     0.04     9351     0.00     0.00  back_comment


the -nw -Q session has fewer charsets loaded, the -Q version loads the
charset for BIG5 (and others) at startup even though it's running in a
LANG=C environment.

It seems that the charset data structure is too GC intensive, maybe it's
time to rethink it to make it less so.