C++ front end slow on MicoORB's current.cc

C++ front end slow on MicoORB's current.cc

[Steven Bosscher]

Hi,

Karel Gardas's MicoORB has already shown to be a real compiler killer, see
for example ir.ii which exposed the exponential behavior in synth_mult, or
typecode.ii which demonstrated a few other less serious bottlenecks.

And we have a new winner.  I have this profile from gprof for current.cc:

current.cc.ii.profile

  %   cumulative   self              self     total
 time   seconds   seconds    calls   s/call   s/call  name
  5.30      1.41     1.41  1281802     0.00     0.00  lookup_fnfields_1
  4.59      2.63     1.22  2501425     0.00     0.00  comptypes
  3.57      3.58     0.95   266122     0.00     0.00  dfs_walk_all
  2.71      4.30     0.72   395340     0.00     0.00  make_node_stat
  2.33      4.92     0.62  1574958     0.00     0.00  ggc_alloc_stat
  2.18      5.50     0.58  3887115     0.00     0.00  dfs_find_final_overrider_pre
  1.99      6.03     0.53     2350     0.00     0.00  store_bindings
  1.92      6.54     0.51   191164     0.00     0.00  copy_node_stat
  1.54      6.95     0.41   143221     0.00     0.00  walk_tree
  1.43      7.33     0.38  1040141     0.00     0.00  look_for_overrides_here

Perhaps a C++ maintainer could investigate this?  I've put the file up
for download on http://hekje1.shuis.tudelft.nl/steven/current.cc.ii.gz
for the rest of the week (i686 preprocessed source).

Gr.
Steven

Re: C++ front end slow on MicoORB's current.cc

[Steven Bosscher]

On Thursday 25 November 2004 00:32, Steven Bosscher wrote:
> And we have a new winner.  I have this profile from gprof for current.cc:

Hate to reply to my own mails, but...

typecode_seq.cc.ii.profile

  %   cumulative   self              self     total
 time   seconds   seconds    calls   s/call   s/call  name
  5.31      1.66     1.66  1282843     0.00     0.00  lookup_fnfields_1
  3.65      2.80     1.14   270238     0.00     0.00  dfs_walk_all
  3.52      3.90     1.10  2538083     0.00     0.00  comptypes


timebase.cc.ii.profile

  %   cumulative   self              self     total
 time   seconds   seconds    calls   s/call   s/call  name
  5.38      1.44     1.44  1281932     0.00     0.00  lookup_fnfields_1
  4.90      2.75     1.31  2502192     0.00     0.00  comptypes
  4.34      3.91     1.16   266693     0.00     0.00  dfs_walk_all


tckind.cc.ii.profile

  %   cumulative   self              self     total
 time   seconds   seconds    calls   s/call   s/call  name
  6.35      1.68     1.68  1281503     0.00     0.00  lookup_fnfields_1
  4.27      2.81     1.13  2513728     0.00     0.00  comptypes
  4.23      3.93     1.12   265840     0.00     0.00  dfs_walk_all

Note that these profiles are at -O2, but it also happens at -O0.

Gr.
Steven