Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!unmvax!gatech!hubcap!sc
From: sc@vlsi.cs.cmu.edu (Siddhartha Chatterjee)
Newsgroups: comp.parallel
Subject: Re: "Molecule: Language Construct for Development Parallel Programs"
Message-ID: <5654@hubcap.clemson.edu>
Date: 31 May 89 12:17:46 GMT
Sender: fpst@hubcap.clemson.edu
Lines: 33
Approved: parallel@hubcap.clemson.edu

In article <5633@hubcap.clemson.edu>, dciem!array!len@uunet.UU.NET (Leonard Vanek) writes:
> Check out "The Parallation Model: Architecture-Independent Parallel
> Programming" by Gary Sabot. This book, published by the MIT Press and
> based on the author's thesis, describes a model and implementation of
> a simulator for this model. The prototyping work was all done in LISP
> so LISP is used for most of the examples in the book. This tends to
> support the use of dataflow notation in Molecule. However, Sabot
> claims that the model is language-independent and to illustrate that
> point does include some examples in C.

The Paralation stuff is fundamentally different from Molecule in that it is
not process-oriented, but rather data-oriented.  It has strict
synchronization guarantees: barrier synchronization semantics are guaranteed
at the end of all elwise loops.  There are two things I don't like about the
language: the same-paralation restriction in elwise loops, and the
communication mechanism (match and move are a pain).  The nice thing about
the language is that it allows nested elwise's.  We used it here last
semester in an undergraduate course, and it went down quite well with the
students.

There is a compiler for a subset of the language for the Connection Machine,
built by Guy Blelloch (see his thesis for details), but it seems very
unlikely there will ever be a compiler for the full language.  The
interaction between the paralation constructs and the type mechanisms in
Common Lisp is sufficiently complex, for one.  I've seen the Paralation C
example in the book, but I don't think anything's coming out on that front.
It's not that you couldn't port the three constructs to C, just that I think
no one's doing it.
-- 
----
ARPA:	Siddhartha.Chatterjee@CS.CMU.EDU
UUCP:	{seismo,decvax,allegra}!rochester!cmu-cs-pt!vlsi!sc
----