Newsgroups: comp.lang.apl
Path: watmath!ljdickey
From: ljdickey@watmath.waterloo.edu (L. J. Dickey)
Subject: Re: performance figures
Message-ID: <1991Nov16.122148.22094@watmath.waterloo.edu>
Organization: University of Waterloo
References: <4172@mrsvr.UUCP>
Date: Sat, 16 Nov 1991 12:21:48 GMT
Lines: 56

In article <4172@mrsvr.UUCP> gurr@mrsvr.UUCP (David Gurr 4-xxxx) writes:
>I work with ~2 meg arrays of complex 16bit and 32 bit integers (chemical
>shift images from nuclear magnetic resonance spectroscopy).  We have
>various interactive programs for analysing and displaying these arrays,
>but none of them are as well designed as APL.  However nobody around
>here believes that APL can handle arrays this large with any speed. Does
>anyone have facts to prove my naysayers wrong?
>
>The facts need to be performance figures for Sun SPARCs on problems like
>3D and 4D  FFT on 2 and 4 meg arrays of 16 or 32 bit complex integers
>(16+16 or 32+32 bit), and tensor multiplys of vectors with these arrays.
>You are not only allowed to cheat by calling external fortran and C
>programs, you are incouraged to do so if you can tell me how it is done.


The folks at the German National Cancer Research Centre make extensive
use of APL2 for their work.  Have you seen any of it?  I saw a
marvelous presentation at APL 90 in Copenhagen.  In it, we were shown
the limitations of CAT scans, with boundaries between hard and soft
tissue clearly visible, and then we saw the advantage provided by NMR
images, especially for discriminating soft tissues.  Part of their image
processing includes the use cloud techniques to "look through" other
soft tissue (clouds) and to distinguish between various layers of
soft tissue.  The main presentation, the highlight as far as I was
concerned, was a smooth, continuous sequence of images of a rotating
human head that started by showing the soft tissue outside the skull,
that effectively "peeled" away layer after layer, first skin, then
bone, revealing the surface of the brain, the folds clearly visible,
and then moving inside, and eventually finishing with a view of an
(unwanted) growth inside the head.  The visual presentation, rotating
and sharp, was so clear that even lay person such as me, with no
medical training, could plainly see that the tumour was not embedded
inside a lobe of the brain, but was between between certain folds and
was probably removable.

The director of the project, who gave the talk, said simply that it
could not have been done without APL.

I think that there is at least one version of APL for the SUN, but I
don't know if it has complex numbers built in.  However, J does have
complex numbers, and there is an implementation of J for the SUN
SPARCstation which you can pick up by ftp for free from
"watserv1.waterloo.edu".  Why not do that for youself and do some
experiments?  After you get familiar with J, challenge your nay-saying
friends to a contest.  Be sure to include program developemnt time in
the contest.  After you win the first race, tell your boss that you
challenge the other group to a second race that modifies the first
challenge in some way.  I don't know what kinds of images you are
making, but I think I know which language wins both races.  Let your
boss be the judge.

-- 
Leroy J. (Lee) Dickey, Faculty of Mathematics, U of Waterloo, Canada  N2L 3G1
Internet:      ljdickey@watmath.UWaterloo.ca
UUCP:          ljdickey@watmath.UUCP             ...!uunet!watmath!ljdickey
Obsolescent?:  ljdickey@watmath.waterloo.edu     BITNET/EARN: ljdickey@watdcs
