Newsgroups: comp.lang.apl
Path: watmath!watserv2.uwaterloo.ca!torn!cs.utexas.edu!usc.edu!elroy.jpl.nasa.gov!jato!csi!sam
From: sam@csi.jpl.nasa.gov (Sam Sirlin)
Subject: Re: Quality of Implementation
Message-ID: <1993Feb3.020417.2045@csi.jpl.nasa.gov>
Originator: sam@kalessin
Keywords: Needless copying, one data point.
Sender: usenet@csi.jpl.nasa.gov (Network Noise Transfer Service)
Nntp-Posting-Host: kalessin
Organization: Jet Propulsion Laboratory, Pasadena, CA
References:  <1539@kepler1.rentec.com>
Date: Wed, 3 Feb 1993 02:04:17 GMT
Lines: 31


In article <1539@kepler1.rentec.com>, andrew@rentec.com (Andrew Mullhaupt) writes:
|> Can anyone suggest a small suite of language independent benchmarks
|> which might make for a useful comparison? I would be surprised if
|> Splus turned out to be very fast compared to APL, but I wonder now
|> how much difference is left...

Sounds like something that should exist. In the comparison of APLs
done on the BBS\APL, these were used (on a 386):

   Benchmark B1: Q*9; where Q is (2.5+iota 1800): 1800 exponentiations.
   Benchmark B2: (iota 3000) divided by 44.3: 3000 f.p. divisions.
   Benchmark B3: +/iota 15000: 15000 integer additions.
   Benchmark B4: 20!40, looped 1000 times (no, looping doesn't cost
                 all that much time).
   Benchmark B5: length 10,000 boolean vectors, and-ed and or-ed 10
                 times (100,000 boolean operations).

I'd list the times, but only relative numbers were given for a
particular machine. These also seem to simple for fast machines,
though the loop is slow in J. They also don't test for copying of
arrays... I suspect you want a real program example.  Nevertheless
these seem a reasonable starting point.

-- 
Sam Sirlin
Jet Propulsion Laboratory         sam@kalessin.jpl.nasa.gov




