Newsgroups: comp.lang.apl
Path: watmath!ljdickey
From: ljdickey@math.uwaterloo.ca (L.J. Dickey)
Subject: Re: Questions of the Day (On the identity matrix)
Message-ID: <C3y3GF.FC5@math.uwaterloo.ca>
Organization: University of Waterloo
References: <1993Mar6.062420.17271@csus.edu> <1993Mar12.201727.19981@Lehigh.EDU> <1993Mar13.135350.468@fid.morgan.com>
Date: Mon, 15 Mar 1993 19:29:03 GMT

In article <1993Mar13.135350.468@fid.morgan.com> hui@fid.morgan.com (Roger Hui) writes:
>re: identity matrix.  There is another "old fashioned" way which
>is faster, and a "new fashioned" way which is nearly as fast
>but is shorter.
>
>   Defn              6!:2 'fx 1000' on a Sparc10
>
>f1 =. e.@i.                5.466
>f2 =. (=/ ])@i.            0.490
>f3 =. =/~@i.               0.501
>
>f4 =. =@i.                 0.507
>f5 =. ,~ $ {.&1@>:         0.076

I tried each of these functions on a MIPS RC6280, with Risc/OS 4.52,
under several different conditions in a time sharing environment,
and found these numbers, analogous to those Roger found above:

        f1      4.92956         7.8578          4.05832
        f2      1.1711          1.40028         0.936725
        f3      1.08965         1.57383         0.807325
        f4      1.98449         1.60935         0.867658
        f5      0.0626061       0.066965        0.08725



-- 
Prof. Leroy J. Dickey, Faculty of Mathematics, U of Waterloo, Canada  N2L 3G1
   Internet:      ljdickey@math.UWaterloo.ca
                  ljdickey@math.waterloo.edu
   UUCP:          ljdickey@watmath.UUCP
