after reading a nice conversation on comp.lang.lisp I decided to play a little bit with SBCL and Debian’s Shootout Benchmarks.
I considered the spectral norm benchmark and compared the C and SBCL implementations.
I started re-running the tests without any further optimization and the results were similar to the ones reported with C beating SBCL by a factor of 2.
(for the C optimization used, see here).
then, I noticed that there was no (declaim (optimize (speed 3) (safety 0) (space 0))) in the SBCL file!
I’ve added it and rerun the test. here are my results:
N=500
gcc 0.15u 0.00s 0.17r
sbcl 0.08u 0.02s 0.21rN=3000
gcc 5.60u 0.00s 5.69r
sbcl 5.18u 0.01s 5.41rN=5500
gcc 18.81u 0.01s 19.12r
sbcl 17.42u 0.02s 17.76r
the SBCL implementation is actually faster than C!
you can find the code for the tests here.
lorenzo bolla\'s blog 
February 8, 2011 at 10:40 am
Did you consider submitting your benchmark to shootout?
February 8, 2011 at 11:19 am
Actually, I did: spectralnorm, fasta and pidigits.
February 8, 2011 at 2:10 pm
This post triggered a nice discussion, with further analysis, on Reddit.
February 8, 2011 at 3:23 pm
I believe you are shooting yourself in the foot. You should not use -O3 with -Os. You should only use -O3.
February 8, 2011 at 4:20 pm
I’ve just re-run the C benchmark without -Os (only -O3) but the results are the same.
February 8, 2011 at 3:24 pm
[…] Read more here Posted in Uncategorized , interesting, science, tech | No Comments » […]
February 8, 2011 at 4:42 pm
C rules.
February 8, 2011 at 5:58 pm
Are you considering libc load time?
Try strace both process to see how many libs, and it’s sizes, each link to.
February 8, 2011 at 6:27 pm
benchmarks game
N CPU Elapsed
5,500 11.14 2.80 spectral-norm C GNU gcc #4
5,500 15.25 3.97 Lorenzo Bolla 2010-12-06 Lisp SBCL #2
http://shootout.alioth.debian.org/u64q/benchmark.php?test=spectralnorm&lang=all
February 8, 2011 at 7:23 pm
The numbers shown on the post are for the benchmark run on my box.
In fact, after seeing the results “locally”, I submitted the code to the shootout, getting the numbers you (Isaac) correctly show.
So, different numbers on different boxes, which is not at all unexpected.
February 8, 2011 at 7:47 pm
Do you get similar measurements when time isn’t a makefile action?
February 9, 2011 at 11:55 am
yes, same results. why would they be different?
February 10, 2011 at 8:23 pm
Why would the results be any different on your box?
You haven’t provided basic information that would allow readers to understand what’s the same about the way you made your measurements and what’s different about the way you made your measurements.
You haven’t provided basic information about “your box” or the language tools on “your box”.