[Sbcl-devel] Performance

[Christophe R. <cs...@ca...>]

Attached is the report generated by Eric Marsden's performance
benchmarking code, referenced at
<URL:http://ww.telent.net/cliki/Performance%20Benchmarks>. I've run it
using Debian's CMUCL as the "reference implementation"; the numbers in
the other columns are relative times (so 1.00 would imply the same
amount of time, 2.00 would imply twice as long [worse!] and 0.5 would
imply half the time [better!]).

The labelling isn't too clear, but the numbers are for (in order)
CMUCL time in seconds, CLISP-cvsish relative time, SBCL-0.pre7.132
relative time, 0.6.13 relative time.

Things to note include the fact that, while most performance is the
same or better than 0.6.13, array performance (examine for instance
the numbers for 1D-ARRAYS, 2D-ARRAYS and FILL-STRINGS) seems
noticeably worse -- BITVECTORS seem particularly slower on 0.pre7.132
than 0.6.13...

I spent some time while the benchmarks were running reading the CVS
commit logs; I couldn't see anything that immediately explained this
performance degradation, which means it's likely to be something
terribly subtle. Any ideas?

I don't want to hold up 0.7.0 by any means, of course... since it's
probably something subtle it might take ages to uncover. This is what
point releases are for :)

Cheers,

Christophe
-- 
Jesus College, Cambridge, CB5 8BL                           +44 1223 510 299
http://www-jcsu.jesus.cam.ac.uk/~csr21/                  (defun pling-dollar 
(str schar arg) (first (last +))) (make-dispatch-macro-character #\! t)
(set-dispatch-macro-character #\! #\$ #'pling-dollar)