This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Experimenting with tree inliner parameter settings
- From: Gerald Pfeifer <pfeifer at dbai dot tuwien dot ac dot at>
- To: gcc at gcc dot gnu dot org, Steven Bosscher <s dot bosscher at student dot tudelft dot nl>
- Cc: dave at hiauly1 dot hia dot nrc dot ca, rguenth at tat dot physik dot uni-tuebingen dot de,aj at suse dot de
- Date: Mon, 5 May 2003 14:32:40 +0200 (CEST)
- Subject: Re: Experimenting with tree inliner parameter settings
- References: <1052079526.731.235.camel@steven>
On Sun, 4 May 2003, Steven Bosscher wrote:
> So the question is: What do these numbers do for your code, both in
> terms of compile time and in runtime performance? Richard, Dave,
> Gerald, can you please test the settings below on your favorite code
> and on the test case for PR 10160 and PR 8361 (and maybe PR 10316)?
>
> max-inline-insns-single=230
> max-inline-insns-auto=230
> max-inline-insns=500
> min-inline-insns=80
> max-inline-slope=20
>
> If these settings produce improved compile times with no or a small
> runtime performance degradation, then I would like to propose these
> settings as the defaults [...]
Build time of the whole package improves from 3:37 to 3:00, but bench-
marks, which exercise different parts (and code bases) of the system
decrease.
Times in [s], average over three runs, std.dev. in parentheses.
| original | --params |
---------------------+---------------+---------------+
STRATCOMP1-ALL | 20.07 (0.04) | 18.62 (0.04) |
STRATCOMP-770.2-Q | 0.87 (0.00) | 0.90 (0.00) |
2QBF1 | 21.49 (0.05) | 22.30 (0.03) |
PRIMEIMPL2 | 10.45 (0.01) | 10.74 (0.03) |
ANCESTOR | 10.56 (0.05) | 10.82 (0.02) |
3COL-SIMPLEX1 | 7.74 (0.02) | 8.00 (0.03) |
3COL-LADDER1 | 49.72 (0.10) | 52.39 (0.36) |
3COL-N-LADDER1 | 25.84 (0.11) | 26.12 (0.04) |
3COL-RANDOM1 | 13.70 (0.03) | 13.94 (0.04) |
HP-RANDOM1 | 17.99 (0.07) | 18.40 (0.06) |
HAMCYCLE-FREE | 1.80 (0.00) | 1.88 (0.00) |
DECOMP2 | 21.02 (0.06) | 22.49 (0.09) |
BW-P4-Esra-a | 77.36 (0.09) | 78.56 (0.20) |
BW-P5-nopush | 6.55 (0.02) | 6.64 (0.03) |
BW-P5-pushbin | 6.24 (0.04) | 6.31 (0.01) |
BW-P5-nopushbin | 2.03 (0.00) | 2.10 (0.02) |
3SAT-1 | 33.38 (0.21) | 34.25 (0.06) |
3SAT-1-CONSTRAINT | 19.94 (0.09) | 20.50 (0.05) |
HANOI-Towers | 4.61 (0.02) | 5.03 (0.02) |
RAMSEY | 9.28 (0.00) | 9.78 (0.03) |
CRISTAL | 13.34 (0.03) | 13.70 (0.08) |
HANOI-K | 40.25 (0.21) | 41.83 (0.11) |
21-QUEENS | 9.34 (0.06) | 9.44 (0.03) |
MSTDir[V=13,A=40] | 19.74 (0.00) | 20.22 (0.01) |
MSTDir[V=15,A=40] | 19.97 (0.05) | 20.49 (0.04) |
MSTUndir[V=13,A=40] | 10.81 (0.01) | 11.11 (0.01) |
TIMETABLING | 11.12 (0.05) | 12.00 (0.02) |
---------------------+---------------+---------------+
The descrease is not drastic, but in fact GCC already is not too good
on that code base, and by reducing these parameters I feel we're mostly
papering over problem elsewhere (and should do that as a measure of last
resort only).
Gerald