Arrays, Functional Languages, and Parallel Systems

318 Pages·1991·3.16 MB·English

by  Klaus Berkling (auth.)

ARRAYS, FUNCTIONAL
LANGUAGES, AND
PARALLEL SYSTEMS
ARRAYS, FUNCTIONAL
LANGUAGES, AND
PARALLEL SYSTEMS
Editedby
Lenore M. R. Mullin
University ofVermont
Michael Jenkins
Queen's University
Gaetan Hains
University of Montreal
Robert Bemecky
Snake Island Research
GuangGao
McGill University
SPRINGER SCIENCE+BUSINESS MEDIA, LLC
Library ofCongress Cataloging-in-Publication Data
Arrays, functionallanguages, and parallel systems / edited by Lenore Mullin ... [et al.].
p. cm.
"Representative of talks given at the First International Workshop on Arrays,
Functional Languages, and Parallel Systems held 12-15 June 1990 in Montreal,
Quebec" --Perf.
Includes index.
ISBN 978-1-4613-6789-5 ISBN 978-1-4615-4002-1 (eBook)
DOI 10.1007/978-1-4615-4002-1
1. Functional programming languages --Congresses. 2. Parallel processing
(Electronic computers) --Congresses. 3. Array processors --Congresses.
1. Mullin, Lenore. II. International Workshop on Arrays, Functional Languages,
and Parallel Systems (1st : 1990 : Montreal, Quebec)
QA76.7.A7 1991
004' .35 --dc20 91-27527
CIP
Copyright © 1991 by Springer Science+Business Media New York
Originally published by Kluwer Academic Publishers in 1991
Softcover reprint ofthe hardcover Ist edition 1991
AlI rights reserved. No part of this publication may be reproduced, stored in a retrieval
system or transmitted in any form orby any means, mechanical, photo-copying, recording,
or otherwise, without the prior written permission of the publisher, Springer Science+
Business Media, LLC.
Printed on acid-free paper.
Dedication
Alan J. Perlis
1922 -1990
We dedicate this workshop to the memory of
Alan J. Perlis and his commitment to research in
arrays and programming languages.
The Organizing Committee:
R. Bernecky, G. Gao, G. Hains, M. Jenkins, L. Mullin
A Mentor
Arrays, how you lived them
You supported their research before their time
APL and STA PL thanks to you
You advised and mentored students everywhere
Perdue, Yale, and CMU
Thank you for your encouragement, my mentor
For your help and guidance, my advisor
We will all miss you
We will all remember you
a student
Table or Contents
Preface .............................................................................................................. ix
1 Arrays and the Lambda Calculus -Klaus Berlding ....•............................................
2 Compiling APL -Robert Bernecky .....•.....•...•..••....••.•••...•.......••...•......••...•........... 19
3 Acorn Run-Time System for the CM-2 -Walter Schwarz ......................................... 35
4 Compiling Issues of Monolithic Arrays -GUilng Gao .......•...................................... 59
5 Arrays in Sisal -lohn T. Feo ..... ...........• ....... ••.... ....•........... ........ ... .... ....... ......... 93
6 Falafel: Arrays in a Functional Language -Carl McCrosky, Kanad Roy, KenSailor
107
7 Arrays in HaskeIl -Guy Lapalme ....................................................................... 125
8 Array Theory and Knowledge Representation -lanice Glasgow .... ........ ....... ............. 133
9 A Parallel Intermediate Representation Based on Lambda Expressions -
Timothy A. Budd ........•.............•................................................................... 145
10 Arrays in Fidil -Luigi Semanzato, Paul Hilfinger ...... .............. ....... ....... ....... ....... 155
11 Structured Data-Types in the Reduction System 'IT-RED -
Claudia Schmittgen, Harald BI/jdorn, Werner Kluge ...... ............... .............. ......... 171
12 Psi, The Indexing Function: A Basis for FFP with Arrays-
Lenore M. Restifo Mullin ................................................................................ 185
13 Genotype-A Pure Functional Array Language -Sebastian Shaumyan ....... ................. 201
14 A Comparison of Array Theory and A Mathematics of Arrays -
Michael A. lenkins, Lenore M. Restifo Mullin ................................................... 237
15 Alternative Evaluation of Array Expressions -Cory F. Skutt ........... .............. ......... 269
16 Matrix Inverion in 3 Dimensions -Gaetan Hains .................................................. 295
17 Fuzzy Inference Using Array Theory and Nial -Ian lantzen ................................... 303
PREFACE
During a meeting in Toronto last winter, Mike Jenkins, Bob Bernecky and I
were discussing how the two existing theories on arrays influenced or were in
fluenced by programming languages and systems. More's Army Theory was
the basis for NIAL and APL2 and Mullin's A Mathematics of A rmys(MOA) ,
is being used as an algebra of arrays in functional and A-calculus based pro
gramming languages. MOA was influenced by Iverson's initial and extended
algebra, the foundations for APL and J respectively.
We discussed that there is a lot of interest in the Computer Science
and Engineering communities concerning formal methods for languages that
could support massively parallel operations in scientific computing, a back
to-roots interest for both Mike and myself. Languages for this domain can
no longer be informally developed since it is necessary to map languages
easily to many multiprocessor architectures. Software systems intended for
parallel computation require a formal basis so that modifications can be
done with relative ease while ensuring integrity in design. List based lan
guages are profiting from theoretical foundations such as the Bird-Meertens
formalism. Their theory has been successfully used to describe list based
parallel algorithms across many classes of architectures.
We decided that it would prove fruitful to get together the research
community investigating formalisms for array based languages designed for
parallel machines with the community actually building and using func
tional parallel languages. Many of the colleagues we know individually did
not know each other's work, yet they were addressing topics with much
commonality.
For example, some of our colleagues that have worked with language
communities such as FORTRAN, NIAL, and APL, have developed arrays
for scientific computation since the early 60's. The language development
groups in these areas have spent years investigating design methodology,
efficient implementations, and compiler technology for array operations and
recently, parallelizing compilers to classes of multiprocessor architectures.
Although issues of functional programming and constructive derivations are
not the focus of these communities, issues on array compilation to various
architectures, performance and scientific applications are.
Other colleagues have been directly involved in the design and extension
of list-oriented functional programming languages to include array mech
anisms. There has been very little communication between the language
design and implementation groups in these different communities. As a
x
result many ideas in the functional programming community have been re
discovered independently.
We decided to hold an invitational workshop in Montreal as a first step
to bring these communities together. The list of attendees was drawn up
from colleagues who had an interest in array and list algebras, functional lan
guages, parallel algorithms, multiprocessor architectures, intelligent compil
ers, and scientific applications. The papers presented herein are representa
tive of the talks given at the First International Workshop on Arrays, Func
tional Languages, and Parallel Systems held 12-15 June 1990 in Montreal,
Quebec.
We are very grateful to everyone who helped us achieve our goals, in par
ticular, CRIM( Centre de recherche injormatique de MontreaQ, ITRC(Injormati07
Technology Research Centre), NSERC, and IBM Corporation for financial
support. We would like to thank Kluwer for its assitance in publishing this
record of the conference.
Lenore Mullin
ARRAYS, FUNCTIONAL
LANGUAGES, AND
PARALLEL SYSTEMS
1
ARRAYS AND THE
LAMBDA CALCULUS
Kla.us Berk ling
CASE Center
and
School of Computer and Information Science
Syracuse University
Syracuse, NY 13244-1190
May 1990
Keywords: Array processing, functional programming, lambda cal
culus, reduction machines.
INTRODUCTION
Why do functional languages have more difficulties with arrays than
procedural languages? The problems arising in the designing offunc
tionallanguages with arrays and in their implementations are man
ifold_ They can be classified according to 1) first principles, 2) se
mantics, 3) pragmatics, and 4) performance. This paper attempts
to give an outline of the issues in this area, and their relation to
the lambda calculus. The lambda calculus is a formal system and as
such seemingly remote from practical applications. However, specific
representations and implementations of that system may be utilized
to realize arrays such that progress is made towards a compromise
of 1) adhering to first principles, 2) clear semantics, 3) obvious prag
matics, and 4) high performance. To be more specific, a form of the
lambda calculus which uses a particular representation of variables,
namely De Bruijn indices, may be a vehicle to represent arrays and
the functions to manipulate them.
PROBLEM
Serious computational problems come almost always in terms of ar
rays. Nothing seems to be more appropriate to array processing than
the von Neumann Computer Architecture. Its linear memory struc
+ +
ture, its addressing methods (absolute address offset index, base
registers, offset registers, index registers), and its instruction set ob
viously constitute a perfect base to implement arrays. Together with
2
language features originating from FORTRAN, the system indeed
fills basic needs in all four categories stated above. Why would or
should one try to improve on it? Because only very basic needs are
filled.
For the sake of efficiency, the higher level concepts of program
ming languages had to be compromised. Real problems are dealing
with very large matrices and/or very frequently used matrices, for ex
ample for pointwise transformations in image processing. There is no
time (or space!) for costly procedure or function entries (prologues)
and exits (epilogues), for scoping of variables and blockstructures,
but there is a tendency for rather flat programs and code. This be
comes manifest in terms of complicated nests of DO-loop constructs
and difficult to follow GOTO's. A premium is on the removal of
code from more frequently executed innermost DO-loops to less fre
quently executed outer DO-loops or initialization pieces by program
transformations.
The combination of flat code, flat name space, static memory lay
out, and the assignment statement as basic programming tool causes
very unpleasant properties. Since index variables are usually con
sidered first class citizens, occurrences of, for example, A[J, J(] may
refer to different matrix elements depending on the code executed be
tween these occurrences. Assignments to J and J( may destroy the
referential transparency. The scope structure is intricately mapped
onto sequencing. As a consequence, variables can be easily confused,
alterations become a costly, error prone, and time consuming enter
prise.
It is particularly difficult to transform such flat programs into
"parallel" programs, which make inherent concurrency explicit. Ob
viously, scoping cannot be expelled without penalty. If it is not part
of the language, the compiler must symbolically execute the code to
reconstruct dependencies.
Many of the problems to produce efficient and accurate code
without the base of higher level concepts arise from the utilization of
memory locations. The conventional viewpoint of a memory location
is best described as a "box-variable," as opposed to a "place-holder
variable," which is the viewpoint taken by functional languages. This
difference makes it difficult for functional languages to adopt easily
array features from conventional languages. Four operations are as
sociated to box variables: allocate (a name to refer to the box is
created), deallocate, fill the box, read the contents of the box. In