Timings

[boyer@rascal.ics.utexas.edu (Bob Boyer)]

Several people who ought to know better have claimed to me recently
that KCL is considerably slower than some other Lisps around.  I find
this contention hard to take, and so I thought I'd broadcast some
current Gabriel benchmark stats.  These results seem to me to show
that KCL (not to mention the several improvements to it in AKCL) is
pretty competitive.

Comparisons of some Lisps on a Sun-3/280, namely rascal.ics.utexas.edu,
on May 3, 1988: (times in seconds)

              AKCL     LUCID     KCL

BOYER         7.700   11.480     9.067         
BROWSE       12.717   21.900    21.650        
CTAK          3.417    1.440     3.883         
DDERIV        3.350    7.320     4.300         
DERIV         2.550    5.680     3.433         
DESTRU        1.371    1.140     1.992         
DIV2          3.217    6.460     3.217         
FFT          34.700   36.000     error      
FPRINT        0.696    0.990     0.883         
FREAD         1.288    2.220     1.487         
FRPOLY       48.683   32.980    52.000        
PUZZLE        5.200    3.580     6.900         
STAK          1.817    1.485     2.183         
TAK           3.817    3.900     3.400         
TAKL          1.417    1.600     1.650         
TAKR          0.521    0.455     0.758         
TPRINT        1.079    2.480     1.371         
TRAVERSE     27.133   31.100    38.883        
TRIANG       82.300   85.960   113.017 
								                      

AKCL = Bill Schelter's improvements to Kyoto Common Lisp
 Version 1.27 Apr 25 1988 (based on KCL below)

Lucid = Sun Common Lisp from Lucid, Inc.
 Development Environment 2.1.1 5-Jan-88 

KCL = Kyoto Common Lisp
 June 3, 1987

If anyone wants to reproduce these sorts of figures, they can fetch
the file rascal.ics.utexas.edu:/pub/gabriel.tar.Z by anonymous ftp.
(It's about 100K bytes.)  After uncompressing and untarring, all you
need to do is:

% make -e "LISP=foo"  >> #errs &

The times will be recorded in a file called times.  All files will be
proclaimed and compiled automatically.  (This automation of running of
the Gabriel test suite was put together by Schelter.)

Bob

P.S.  Schelter also notes that additional or different declarations
more suitable to KCL than Lucid reduce the time in FFT to 5.35 sec,
TRIANG to 63.5 sec, and DESTRU to 1.0 seconds.  Oddly the same
declarations sometimes increase the time in Lucid (by a factor of 3.3
in the case of triang!).  These alternate declarations are in the
files fft-mod.cl, triang-mod.cl, ...  which are included in the tar
file.