[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

A little timing data

Below is some data (Nqthm time triples) for the first nontrivial Nqthm
proveall.  A triple [x y z] means that x seconds were used in
miscellaneous processing, y seconds in proving, and z seconds in io.
Generally, the middle term completely dominates the others.  (To be
specific, the test is the evaluation of the third form in the file
basic.events in the standard Nqthm distribution available by anonymous
ftp from cli.com.)

Crudely, the data suggest that, for Nqthm running purposes, funware (a
Sparc 2) is 4 times faster than cli.com (a Sun 3-280) and AKCL is two
times faster than CMU Lisp.

To even begin to be remotely fair it should be pointed out that (a)
CMU Lisp has not been tuned for SPARC yet and (b) ACKL was tuned for
Nqthm to some extent, and Nqthm for AKCL to a very small extent, to
wit, putting in proclamations and declaring type sets to be fixnums.

1.  Time to do an ancient, archived run on cli.com in 1987 under an
AKCL built Sun Dec 13 12:23:02 CST 1987 (no version number I could

     [ 68.1 2129.1 245.3 ]

2. Time on funware under ACKL 1.530.

     [ 8.7 507.1 44.9 ]

3. Time on funware under CMU Lisp (beta).

     [ 10.8 1084.1 55.4 ]

For running CMU Lisp, it is worth noting that it does not default to
the maximum speed.  But the declaration I used was the following,
which I think gets maximum speed.

  (proclaim '(optimize (safety 0) (space 0) (speed 3)
                       (debug 0) (compilation-speed 0)))