Project 37: Array processing 

Pasadena 1984:
==============

(Rice): This activity is separate from the Fortran 8x activity and
focuses on a higher level language suitable for a variety of
architectures. The approach is experimental in nature and involves
writing various algorithms in various "languages". These programs are
then examined to see (A) if the computation is clearly expressed, and
(B) if any barriers seem to have been placed in the way of easily
recognizing computations that can be efficiently implemented on
parallel architectures.

Wright asks what sort of Fortran is suitable for parallel architecture. 
Rice thinks Fortran 8x is a good example of a language which does not 
require to rewrite the program for different architectures. Ford feels 
that lack of transportability is an enormous burden. Rice says that 
Fortran 8x does not address the non array processing parallelism.
Reid and Lawson think that the subject is not ready for standardization. 
Ford says there are two approaches (i) decompose the domain or (ii)
write an intelligent program on a Cray (say). There is yet no way to know 
what is the best approach Reid feels the user wants his program to run 
without change efficiently on the new architecture. Rice answers that 
the user should express the natural parallelism in the program.

Sophia-Antipolis 1985:
======================

Rice said nothing about Fortran 8x or the BLAS because they
were on the agenda later. He had written a report (Problems to 
Test Parallel and Vector Languages) which will be sent to each 
member. The problems are in five forms: 

(1) English summary, 
(2) Fortran 77, 
(3) CDC Fortran (optimized), 
(4) Fortran 8x (extended for parallelism) and 
(5) Vector PROTRAN (PROTRAN extended for parallelism and 
    Fortran 8x capabilities).

A student, Jairo Panetta, is doing a Ph.D. thesis on the 
possibility of a language-independent and machine-independent 
specification of a set of linear algebra modules. He is testing 
his design on four architectures: sequential (VAX), vector (Cyber 
205), MIND (FLEX 32) and systolic-like (Pringle). The work should 
be completed this year.

Rice will send members a copy of the report of the workshop held 
at Purdue last October about the BLAS. Its interest is to see 
ideas for BLAS extensions that were not included by the BLAS 
extension group in their proposal published in the SIGNUM 
newsletter.	

Stanford 1988:
==============

The following is the summary of the report presented by G. Paul .

Paul said that he believed that the removal of the IDENTIFY statement
has severely crippled the array processing capability of FORTRAN 8X.
With its removal, FORTRAN 8X can no longer manipulate skewed sections,
diagonals, etc. IDENTIFY, of course, has been a major point of
contention between members of the X3J3 committee, and in the general
public community.  Nevertheless, the requirement to handle skewed
arrays must be addressed.

On the positive side the current proposals to re-introduce
vector-valued subscripts, if adopted, will enhance the array
processing ability although this will not solve the above problem.
Similarly the proposals to introduce pointers into the language will
significantly enhance array processing and could potentially solve the
skewed section problem in the pointer facility includes a mapping
capability.

Jerusalem 1990:
===============

Rice moved that the project be closed (seconded by Feldman). This was
unanimously accepted.