Re: [0/18] new argv_ref branch for m4 speedup

[Ralf Wildenhues]

Hi Eric,

* Eric Blake wrote on Tue, Nov 20, 2007 at 05:12:27AM CET:
> 
> I've been working on this for over a month, now, and am finally happy
> enough with my series of patches to post it as a new branch named
> 'argv_ref'.  I still need to write the ChangeLogs, perform additional
> regression testing (such as the complete autoconf testsuite, and
> bootstrapping libtool),

Cool.  A very nice job you've done there.  Libtool bootstrap, which
contains 11 autoreconf's, goes from

79.92user 8.14system 1:41.50elapsed 86%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+1504396minor)pagefaults 0swaps

to

69.96user 8.07system 1:31.28elapsed 85%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+1509603minor)pagefaults 0swaps

with identical configure output files.  The Autoconf testsuite passes
over here.  A m4_join with 4000 iterations now takes 0.43 s, compared
with 14 s before, and the scaling remains strictly linear, see this
table of argument number N, timing [s], and relative time increase for
doubling of N:

   1000   0.25   
   2000   0.33   1.32
   4000   0.43   1.30
   8000   0.66   1.53
  16000   1.14   1.72
  32000   2.09   1.83
  64000   3.98   1.90
 128000   7.84   1.97

With N >= 32000, I needed to fix a couple of very minor issues first
(followup patches on m4-patches only).

Cheers,
Ralf