Path: utzoo!attcan!uunet!dalsqnt!pollux!ti-csl!pf@csc.ti.com
From: pf@csc.ti.com (Paul Fuqua)
Newsgroups: comp.lang.lisp
Subject: Re: Summing a list
Message-ID: <62445@ti-csl.CSNET>
Date: 1 Nov 88 03:57:02 GMT
References: <29718@think.UUCP>
Sender: news@ti-csl.CSNET
Organization: TI Computer Science Center, Dallas
Lines: 41


    Date: Wednesday, October 26, 1988  11:23am (CDT)
    From: barmar at think.COM (Barry Margolin)
    Subject: Re: Summing a list
    Newsgroups: comp.lang.lisp
    
    I decided to time all the various methods that have been discussed.  I
    tried them on a Symbolics 3640 running Genera 7.2 and a Sun 3/280
    running Lucid CL 2.0.3.  I used the default OPTIMIZE parameters.

To add another implementation to the pot, I ran roughly the same code on
a TI Explorer 2 running release 4.1+, also with the default
optimisations.  I included Mark Johnson's tail-recursive versions, and
both helper-function and LABELS versions of the original recursive code.

The fastest were DO-LOOP, APPLY-+, and TRECURSIVE (tail-recursion using
LABELS).  The latter is up there because the compiler managed to
open-code the recursive calls into a loop, so all three are running
essentially the same code.

Next in line are RECURSIVE1 and TRECURSIVE1 (two helper-function
recursive versions), and REDUCE-+, all about 7 times as long.  No
open-coding here, it's all function-calling speed.

The slowpoke of the bunch was RECURSIVE (Barry Margolin's original
code), at about 13 times as long, because it's forming and discarding
lexical closures on each function call (all stack-consing, though).  It
was this surprising result that led me to examine the code.

[I was unable to run EVAL-+, because the evaluator spread the whole list
onto the stack and exceeded the 256-word limit on stack frames.
(APPLY-+ doesn't do that, because it's applying a function that takes a
&REST argument.)  On a shorter list, it was about 6 times as long as
DO-LOOP.]

                              pf

Paul Fuqua
Texas Instruments Computer Science Center, Dallas, Texas
CSNet:  pf@csc.ti.com (ARPA too, sometimes)
UUCP:   {smu, texsun, cs.utexas.edu, im4u, rice}!ti-csl!pf