[Date Prev] [Date Next] [Thread Prev] [Thread Next] Date Index Thread Index Search archive:
Date:24 Dec 2004 15:02:45 -0000 
Subject:Re: Problems with linux dynamic linker and separately compiled modules 
From:Aaron Sloman 
Volume-ID: 

On Fri, 24 Dec 2004, Jonathan L Cunningham wrote:

> On Thu, 23 Dec 2004 19:24:33 +0000 (UTC), Aaron Sloman
> <A.Sloman@cs.bham.ac.uk> wrote:
>
> (snip)
> >Poplog web site around 1989. My hunch is that the same change should
> >be made to the OSF version (defined in the file as the Linux version
> >was previously), but I doubt that anyone is now using an OSF version
> >of poplog.
>
> I couldn't actually remember what OSF was, so I googled to find out.
>
> No success. The second page of results (searching for OSF), gave
> me a page of links, a few of which mentioned OSF/Motif, but the
> first link I tried was broken.

I gave google 'osf unix' and near the top of the list was

    http://kb.indiana.edu/data/agka.html?cust=023362.83043.131

Which includes

   In the late 1980s, Sun Microsystems, the major BSD vendor at the
   time, entered into a partnership with System V developer AT&T.
   Together, they integrated major BSD and SunOS features into System
   V to create System V, revision 4 (SVR4). In response, a number of
   other Unix vendors, including IBM, Hewlett-Packard, and
   DEC, formed the Open Software Foundation (OSF). OSF was
   dedicated to the creation of open architecture systems (specifications
   that were not under the control of any single company).

   OSF/1 was one of the fruits of this effort. Based on Carnegie
   Mellon's BSD-derived Mach and integrating features of both major
   Unix flavors, it was released as an alternative to SVR4. It conforms
   to Unix standards and is also compliant with the System V Interface
   Definitions (SVID). OSF/1 is not a complete operating system but
   specifies many operating system components that may be used when
   developing an operating system. As such, OSF/1 is as much a
   specification as it is an implementation. Operating systems that it
   has influenced include HP-UX, AIX, and especially Tru64
   UNIX (formerly known as Digital UNIX and DEC OSF/1).

I think the OSF1 code in Poplog is for the version that runs on unix on
Alphas (Digital Unix was an implementation of OSF1 as explained above.)

It's the only working 64bit implementation of poplog, as far as I know.
In principle Solaris poplog could be expanded to use 64bit addresses and
instructions but I don't know if anyone would use it.

64bit poplog running on AMD Opteron might be very useful. One of my
former colleagues (Riccardo Poli, now a professor at Essex university)
obtained a huge speedup in some of his evolutionary computation software
by using using 64bit integers as representing membership in a 64 element
set. Then machine operations on bit patterns could perform very fast set
operations. I think his programs ran more than 10 times faster as a
result.

I suspect all the required code fo r74-bit achines is there in the Alpha
Poplog sources (labelled as OSF ?). If I remember correctly Alphas are
same-endian as PC cpus, unlike Sparc/solaris.

> Not that it's a high priority, because it does no real harm, but I
> suspect there's a lot of legacy code that could be usefully expunged,
> as part of a general tidy-up.

In a system like poplog, it's possible that quite a lot of harmless
legacy code is like unused material in a gene pool: you never know when
you are going to need it?

> Jonathan
> (Hoping that the brain cells which used to know what OSF were have
> been reallocated to something more recent, rather than merely dying
> off.)

Or temporarily hidden because of reorganisation of search priorities
following disuse. My bet is that you will discover that you knew most of
what is stated above about OSF, but simply could not recall it on
demand.

Aaron
===
Aaron Sloman, ( http://www.cs.bham.ac.uk/~axs/ )