Path: utzoo!attcan!uunet!samsung!sdd.hp.com!wuarchive!mit-eddie!uw-beaver!cornell!wilk
From: wilk@fife.cs.cornell.edu (Michael Wilk)
Newsgroups: comp.lang.lisp
Subject: Loop speeds
Message-ID: <47103@cornell.UUCP>
Date: 15 Oct 90 13:42:37 GMT
Sender: nobody@cornell.UUCP
Reply-To: wilk@cs.cornell.edu (Michael Wilk)
Distribution: comp
Organization: Cornell Univ. CS Dept, Ithaca NY
Lines: 43

After puzzling over the results of the speed tests I performed on the "mirror"
program, I discovered a disturbing discrepancy between the loop macro and the
mapping functions.  Consider the performance gap between running these two
versions of function foo on a list of 30000 items:

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

(defun foo (x)
  (loop for item in x
	collect (atom item)))

Elapsed Real Time = 7.86 seconds
Total Run Time    = 7.84 seconds
User Run Time     = 7.84 seconds
System Run Time   = 0.00 seconds
Process Page Faults    =          7
Dynamic Bytes Consed   =          0
Ephemeral Bytes Consed =    240,176

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

(defun foo (x)
  (mapcar #'atom x))

Elapsed Real Time = 0.14 seconds
Total Run Time    = 0.14 seconds
User Run Time     = 0.14 seconds
System Run Time   = 0.00 seconds
Process Page Faults    =          0
Dynamic Bytes Consed   =          0
Ephemeral Bytes Consed =    240,008

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

(These tests were run in Lucid Common Lisp on a DECstation 5000.  The functions
were compiled and optimized for speed.  I ran the tests several times and chose
the run with the fewest page faults.  The predicate #'atom was chosen
arbitrarily.)

Would you conclude from this that one should always use mapcar instead of
loop/for/collect in simple cases?  Why isn't macro expansion smarter?

-Michael Wilk (wilk@cs.cornell.edu)