Re: [gnugo-devel] Regression results for the large scale patch

[Arend Bayer]

Stéphane wrote:

> Here are the regression results (against 3.4) for the large
> scale capture patch posted in
> <http://mail.gnu.org/archive/html/gnugo-devel/2003-07/msg00209.html>

As it introduces a couple of owl_does_attack calls, one might wonder
about its performance penalties. I've tested against 12 gtp files, each
consisting of 20 moves of a game against a computer opponent (from Gifu
and Edmonton), and it shows some slowdown as shown below.

It varies between <1% and 19.5%

Arend

First result is always CVS, second CVS+large_scale_patch.

GnuGo-Aya.gtp:
real    1m33.120s
user    1m32.990s
sys     0m0.130s
GnuGo-Aya.gtp:

real    1m45.286s
user    1m43.660s
sys     0m0.100s
11.5%
GnuGo-GORO.gtp:

real    2m51.659s
user    2m51.510s
sys     0m0.110s
GnuGo-GORO.gtp:

real    3m1.084s
user    3m0.970s
sys     0m0.110s
5.4%
GnuGo-Go4.gtp:

real    8m9.073s
user    8m8.940s
sys     0m0.120s
GnuGo-Go4.gtp:

real    8m15.219s
user    8m15.060s
sys     0m0.150s
2.9%
GnuGo-GoInt.gtp:

real    6m24.995s
user    6m24.870s
sys     0m0.130s
GnuGo-GoInt.gtp:

real    7m40.372s
user    7m40.210s
sys     0m0.160s
19.5%
GnuGo-GoeMate.gtp:

real    17m1.804s
user    17m1.290s
sys     0m0.190s
GnuGo-GoeMate.gtp:

real    17m28.310s
user    17m27.960s
sys     0m0.210s
2.5%
GnuGo-Haruka.gtp:

real    1m4.071s
user    1m3.930s
sys     0m0.140s
GnuGo-Haruka.gtp:

real    1m4.715s
user    1m4.550s
sys     0m0.170s
<1%
GnuGo-KCC.gtp:

real    9m1.307s
user    9m1.200s
sys     0m0.110s
GnuGo-KCC.gtp:

real    9m14.256s
user    9m14.120s
sys     0m0.140s
2.3%
GnuGo-Katsunari.gtp:

real    2m1.220s
user    2m1.150s
sys     0m0.070s
GnuGo-Katsunari.gtp:

real    2m10.090s
user    2m9.950s
sys     0m0.140s
7.2%
GnuGo-Martha.gtp:

real    7m58.066s
user    7m57.930s
sys     0m0.140s
GnuGo-Martha.gtp:

real    8m39.837s
user    8m39.710s
sys     0m0.120s
8.7%
golois-gnugo.gtp:

real    7m6.355s
user    7m6.240s
sys     0m0.110s
golois-gnugo.gtp:

real    9m3.186s <-- interrupted
user    7m53.840s
sys     0m0.150s
11%
great5-gnugo.gtp:

real    5m28.786s
user    5m28.630s
sys     0m0.160s
great5-gnugo.gtp:

real    6m24.997s
user    6m24.840s
sys     0m0.160s
17%
manyfaces-gnugo.gtp:

real    2m54.713s
user    2m54.610s
sys     0m0.100s
manyfaces-gnugo.gtp:

real    2m59.819s
user    2m59.670s
sys     0m0.150s
2.5%


Sum:
User time CVS:
User time patch: