Re: Distributed compilers

w.purvis@daresbury.ac.uk (Bill Purvis, ext 3357)
Thu, 2 Dec 1993 14:58:51 GMT

          From comp.compilers

Related articles
Distributed compilers karl@hsr.no (1993-11-26)
Re: Distributed compilers roth@helena.cs.rice.edu (1993-12-01)
Re: Distributed compilers andy@research.canon.oz.au (1993-12-01)
Re: Distributed compilers w.purvis@daresbury.ac.uk (1993-12-02)
| List of all articles for this month |
Newsgroups: comp.compilers
From: w.purvis@daresbury.ac.uk (Bill Purvis, ext 3357)
Keywords: parallel
Organization: Daresbury Laboratory, UK
References: 93-12-011
Date: Thu, 2 Dec 1993 14:58:51 GMT

Andy Newman wrote:
> The original Inmos occam compiler, written in occam, was structured
> as a pipeline of processes, eight if I remember correctly.


I worked a port of the Occam 2 compiler a few years back. In fact the
whole developement system runs as a suite of processes, including the
editor, the folded file manager, the compiler and linker. In total there
must have been well over twenty processes involved! Of course since it was
all written in Occam, that was no problem. The actual compiler split into
10 well defined each being a high-level process, that is there was a
10-way PAR in the main compiler routine, but then several of these
processes then had nested PAR's within them so that processes sprang up
and died as the compilation progressed. The organisation was fairly
typical for a `multi-pass' compiler except that the output from each pass
was fed over a channel to the next pass directly. Dictionary management
was a single process that receive update and query requests and sent the
info back to the caller. It took a little bit of adjusting to the parallel
approach, but once you got the feel of it it was quite natural to work
with.


Bill Purvis
--


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.