From @IU.AI.SRI.COM,@sunset.ai.sri.com:mcw@rice.edu  Tue Jan 16 14:06:25 1990
Received: by iris.ucdavis.edu (5.57/UCD.EECS.2.0)
        id AA19036; Tue, 16 Jan 90 14:06:25 PST
Received: from hydra.ucdavis.edu by clover.ucdavis.edu (5.59/UCD.EECS.1.11)
        id AA07047; Tue, 16 Jan 90 14:11:20 PST
Received: from IU.AI.SRI.COM (a050cc0) by hydra (4.12/3.14)
        id AA01643; Tue, 16 Jan 90 13:31:49 pst
Received: from sunset.ai.sri.com by IU.AI.SRI.COM via SMTP with TCP;
          Tue, 16 Jan 90 12:43:25-PST
Received: from cam.sri.com by sunset.ai.sri.com (4.0/4.16) id AA00277
          for info-hol@clover.ucdavis.edu; Tue, 16 Jan 90 12:43:19 PST
Received: from drakes.ai.sri.com by cam.sri.com (4.0/4.16) id AA11939
          for info-hol@clover.ucdavis.edu; Tue, 16 Jan 90 20:43:16 GMT
Received: from Warbucks.AI.SRI.COM by drakes.ai.sri.com (4.0/4.16) id
          AA16311 for info-hol%edu.ucdavis.clover@cam.sri.com; Tue,
          16 Jan 90 12:42:55 PST
Received: from rice.edu by Warbucks.AI.SRI.COM with INTERNET ; Tue,
          16 Jan 90 12:11:18 PST
Received: from iapetus.rice.edu by rice.edu (AA09772); Tue,
          16 Jan 90 14:09:24 CST
Received: by iapetus.rice.edu (AA07376); Tue, 16 Jan 90 14:09:24 CST
Date: Tue, 16 Jan 90 14:09:24 CST
From: mcw@rice.edu (Michael Wirth)
Message-Id: <9001162009.AA07376@iapetus.rice.edu>
To: info-hol%edu.ucdavis.clover%cam.sri.com@Warbucks.ai.sri.com,
        mjcg%uk.ac.cam.cl%uk.ac.nsfnet-relay@NSS.Cs.Ucl.AC.UK
Subject: Speed of Lisp systems on the Macintosh
Cc: D0966@applelink.apple.com, alms@cambridge.apple.com, chewy@apple.com,
        clark.l@applelink.apple.com, jaj@cambridge.apple.com,
        mcw@iapetus.rice.edu

Re a recent msg on "Apple Macintosh prices to run HOL" which contains the
statement:
Apple themselves market a version of Common Lisp for the Macintosh. It is
slower (by about 50%) than Procyon Common Lisp, and so is not really
recommended for running HOL. Procyon Common Lisp is available from:
...

We have benchmarked Macintosh Allegro Common Lisp 1.2.2 against Procyon 2.1
(or whatever the new release was in August 1989) using the KSL 89-02 (Stanford's
Knowledge Systems Lab report of that number) benchmarks.  The results contra-
dicted the "conventional wisdom" that Procyon is faster than Allegro.  On the
BB1 benchmark,  ACL was comparable to Procyon on compile times and 10% faster
on execution.  On the SOAR benchmark, Procyon failed to compile the program
and we couldn't find a quick fix.  ACL compiled and ran the program fine.

We tested Procyon's native CLOS performance against ACL + Victoria Day PCL (a
subset of a machine-portable CLOS).  We were particularly concerned with the
performance hit we might be taking with ACL+PCL.  To our surprise, ACL+PCL was
comparable to Procyon CLOS on generic function calls and significantly faster
on slot accesses.

We have communicated these results to Expertelligence (the US marketer for
Procyon) and to Apple.  Since then, Apple has released ACL 1.3, which is
noticeably faster.  Xerox PARC is also about ready to release a new version of
PCL.  It would be interesting to run this suite of benchmarks (and additional
ones again).


