array syntax subgroup evaluation criteria

[David Alpern]

First, to introduce myself, I work for Ken Hansen in the languages group
at MasPar Computer Corporation.  I am the project lead for our effort to
develop a virtualized data-parallel dialect of C and its compiler.  Ken
presented many of the core concepts of our language approach at the last
NCEG meeting.  I will most likely be attending the January NCEG meeting
and the Monday array syntax subgroup meeting instead of him.

In reference to Frank Farance's recent document on evaluation criteria for
proposals to the array syntax subgroup:

What I don't see are questions that define the problem(s) or need(s) that
should be satisfied by any proposal found acceptable by the subgroup, and
ask of each proposal "does it accomplish ...?" or "does it provide ...?".

I would expect that the group might make good progress if we clearly defined
the needs/problems that proposals to it should satisfy, and then considered
proposals in terms of their ability to satisfy these needs.

 From MasPar's point of view, we need a data-parallel dialect of C that does
not create too many implementation problems.  I believe the top-most needs of
such a language to be:

1) Arrays of some sort need to be provided, and to be treated as first-class
   objects: assigned to, returned from functions, supported by elementwise
   operations, passed by-value or by-reference to subprograms.

2) Subset selection on arrays needs to be supported, with a selected subset
   of an array being a first-class object as well.

3) The sort of arrays being discussed here needs to not be burdened farther
   than reasonable compiler technology is capable of handling by ANSI C's
   definition of array usage in terms of pointer equivalents.  In particular,
   it is necessary that a reasonable compiler be capable of recognizing
   just about all cases when elementwise operations can be compiled to take
   advantage of distributed memory and parallel hardware (SIMD, MIMD,
   vector pipe, whatever).
   
I would like to raise the issue of whether the subgroup has a consensus that
these needs should be satisfied by any proposal it considers.  If so, what
other needs are there?  If not, what requirements of this sort can be agreed on?

- Dave