From: Pekka Niiranen
Subject: Processor for Lisp machine
Date: 
Message-ID: <s6qqc.313$Xe5.18@read3.inet.fi>
Just thoughts I got while visiting www.netbsd.org
and seeing the list of supported platforms:

If one where to design his own Lisp Machine which
one of those platforms is hardware-wise closest to those
80s machine? I mean which platform requires the least amount of
additional special hardware?

-pekka-

From: Joe Marshall
Subject: Re: Processor for Lisp machine
Date: 
Message-ID: <pt91iu89.fsf@ccs.neu.edu>
Pekka Niiranen <··············@wlanmail.com> writes:

> Just thoughts I got while visiting www.netbsd.org
> and seeing the list of supported platforms:
>
> If one where to design his own Lisp Machine which
> one of those platforms is hardware-wise closest to those
> 80s machine? I mean which platform requires the least amount of
> additional special hardware?

Symbolics ported Genera to the DEC Alpha, so that one is at least
doable.

I can't even recognize the half the platforms on there, so there may
be some that are just about perfect (but it isn't likely!)

A real lisp machine would be ideal, of course, but the next best thing
would probably be a PDP-10 (which isn't on the list).  If there is any
way to associate a tag with a machine word such that the tag does not
normally participate in address generation or alu operations, then
that would be a good candidate.
From: Svein Ove Aas
Subject: Re: Processor for Lisp machine
Date: 
Message-ID: <Haqqc.2257$eH3.48827@news4.e.nsc.no>
Pekka Niiranen wrote:

> Just thoughts I got while visiting www.netbsd.org
> and seeing the list of supported platforms:
> 
> If one where to design his own Lisp Machine which
> one of those platforms is hardware-wise closest to those
> 80s machine? I mean which platform requires the least amount of
> additional special hardware?

Said special hardware includes such things as hardware GCs, so I doubt
anything we'd get today comes close.

That said, from what I gather, the more registers the better - anything
but x86, probably. Others can probably give you more specific
information.
From: Christopher Browne
Subject: Re: Processor for Lisp machine
Date: 
Message-ID: <2h2p6eF856p2U1@uni-berlin.de>
After a long battle with technology, Pekka Niiranen <··············@wlanmail.com>, an earthling, wrote:
> Just thoughts I got while visiting www.netbsd.org
> and seeing the list of supported platforms:
>
> If one where to design his own Lisp Machine which one of those
> platforms is hardware-wise closest to those 80s machine? I mean
> which platform requires the least amount of additional special
> hardware?

Regrettably, nothing seems to be as seeming-suitable as the processors
that are getting obsolete.  

The thing about the LMs was that they had hardware support for GC.
You'd presumably need a system supporting microcode to do that now.
Digital Alpha had microcode created by Genera/Symbolics to cope with
this.

The typical recent 'somewhat-RISC-variants' don't generally grok
microcode, and the one architecture being actively worked on,
"Itanic," looks to be pretty Anti-Lisp by virtue of preferring heavily
statically-optimized code...
-- 
wm(X,Y):-write(X),write(·@'),write(Y). wm('cbbrowne','acm.org').
http://www3.sympatico.ca/cbbrowne/advocacy.html
As of next Monday, MACLISP will no longer support list structure.
Please downgrade your programs.
From: Carl Shapiro
Subject: Re: Processor for Lisp machine
Date: 
Message-ID: <ouyad03bpi6.fsf@panix3.panix.com>
Christopher Browne <········@acm.org> writes:

> The thing about the LMs was that they had hardware support for GC.
> You'd presumably need a system supporting microcode to do that now.
> Digital Alpha had microcode created by Genera/Symbolics to cope with
> this.

This last statement has no grounding in fact.  As far as the Alpha is
concerned Open Genera is an ordinary application.  There is no use of
custom microcode or PAL code.
From: Barry Margolin
Subject: Re: Processor for Lisp machine
Date: 
Message-ID: <barmar-23E3DC.12021320052004@comcast.dca.giganews.com>
In article <···············@panix3.panix.com>,
 Carl Shapiro <·············@panix.com> wrote:

> Christopher Browne <········@acm.org> writes:
> 
> > The thing about the LMs was that they had hardware support for GC.
> > You'd presumably need a system supporting microcode to do that now.
> > Digital Alpha had microcode created by Genera/Symbolics to cope with
> > this.
> 
> This last statement has no grounding in fact.  As far as the Alpha is
> concerned Open Genera is an ordinary application.  There is no use of
> custom microcode or PAL code.

Right.  Open Genera was basically an Ivory emulator, analogous to 
Virtual PC running on a Macintosh.

-- 
Barry Margolin, ······@alum.mit.edu
Arlington, MA
*** PLEASE post questions in newsgroups, not directly to me ***
From: Duane Rettig
Subject: Re: Processor for Lisp machine
Date: 
Message-ID: <48yfnxh22.fsf@franz.com>
Christopher Browne <········@acm.org> writes:

> After a long battle with technology, Pekka Niiranen <··············@wlanmail.com>, an earthling, wrote:
> > Just thoughts I got while visiting www.netbsd.org
> > and seeing the list of supported platforms:
> >
> > If one where to design his own Lisp Machine which one of those
> > platforms is hardware-wise closest to those 80s machine? I mean
> > which platform requires the least amount of additional special
> > hardware?
> 
> Regrettably, nothing seems to be as seeming-suitable as the processors
> that are getting obsolete.  

Coming from a "GP Hardware" point of view, I would tend to go
the opposite way, and say that none of these processors were
ever suitable, and the GP-style route is the only way to go,
because no such support is ever going to be true "support" in
GP hardware.

> The thing about the LMs was that they had hardware support for GC.
> You'd presumably need a system supporting microcode to do that now.
> Digital Alpha had microcode created by Genera/Symbolics to cope with
> this.

This illustrates my point perfectly.  Carl Shapiro has refuted
that Genera had any PAL code, but suppose it did anyway?  One
looks at PAL code and says "excellent; this is what I need for
a LM".  But if you dig deeper, you find that PAL code is extremely
bulky; it is in effect a trap for every instruction decode (think
of a user-defined PAL code as an emulated instruction), and
that is likely why Genera didn't use it.  When we were working
on the Alpha port of Allegro CL, we consulted with DEC (at the
time) developers because we needed to find a conditional-trap
instruction, which the Alpha doesn't provide, in order to do our
interrupt checking.  I had seen the definition of PAL, and had
thought that PAL code would be perfect for implementing such a
conditoonal trap.  But the DEC developers told me not to use
these; the latency on a PAL instruction was worse than a light
subroutine call.  I finally settled on a load instruction, which
would keep loading the car of a register that has nil in it until
the condition is met, at which time the register gets some
impossible value and the load generates a SIGSEGV or SIGBUS.
A poorman's conditional trap, but effective, and many tens of
times faster than a PAL instuction would have taken.

> The typical recent 'somewhat-RISC-variants' don't generally grok
> microcode, and the one architecture being actively worked on,
> "Itanic," looks to be pretty Anti-Lisp by virtue of preferring heavily
> statically-optimized code...

That doesn't make it Anti-Lisp - it only makes it Anti-LM-Style-Lisp
(like all the others :-).  I'm sure it would be a decent GP-Lisp
port, other factors considered...

-- 
Duane Rettig    ·····@franz.com    Franz Inc.  http://www.franz.com/
555 12th St., Suite 1450               http://www.555citycenter.com/
Oakland, Ca. 94607        Phone: (510) 452-2000; Fax: (510) 452-0182