OCaml machine (was: Re: speed versus C)

OCaml machine (was: Re: speed versus C)

[Juergen Pfitzenmaier]

Alain Frisch wrote:
> Do you think it would be easy to design processors with built-in support
> for boxed values, GC tags, OO, etc ... that is, a concrete OCaml machine ?

Hey that's exactly what I was talking about with some friends a week ago.
One evening we were sitting over some coffe and beer and talking what we
would like to do in the near future and how to make a living of it.
One idea that came up was the design of ML in hardware just like the
long gone Lisp machines. Our idea was to build it because it would mean
some value to the collector/FP programmer/freak and not because of
some imagined gain in execution speed. And the first reason is: I would
like to have a ML machine sitting on my desk (others collect tea spoons,
stamps ... I collect hardware).

At this time that idea is not considered seriously but next year my company
might have some extra money that we could spend in some VHDL design just
for fun and there a some cool people around who are willing to join in.

ciao pfitzen

Re: speed versus C

[William Chesters]

Alain Frisch writes:
 > Do you think it would be easy to design processors with built-in support
 > for boxed values, GC tags, OO, etc ... that is, a concrete OCaml machine ?

This was tried in the 80s on quite a large scale by Symbolics, with
the Lisp machine (to which you will find 1000s of references on the
net).  It implemented some of the Lisp runtime, which is pretty
similar in conception to the ocaml runtime, in hardware.  It was all
very nicely done, and apparently the machines and OS were pretty
wonderful in many ways.  Enough examples were sold to upmarket
(e.g. AIish) firms and labs over a period of years to make it into a
bit of legend.

But of course they ran into a problem: the big firms spend billions on
developing sophisticated classical chips with incredibly large numbers
of transistors, to serve their vast market.  So the cost of fixing
your software performance hit just by buying a faster, but entirely
standard, computer is simply not big enough to support the development
of a whole different architecture.

Maybe things have matured and opened up now to the point where one
could take a readily available SPARC or StrongARM core and tack some
GC support onto it, I don't know.  Certainly Sun are hyping their MAJC
Java chip pretty strongly.

Ocaml Machine (was Re: speed versus C)

[Alain Frisch]

On Mon, 11 Oct 1999, William Chesters wrote:

> This was tried in the 80s on quite a large scale by Symbolics, with
> the Lisp machine (to which you will find 1000s of references on the
> net).  

Yes, I heard about it (how could I miss it : I'm working at the MIT AI
Lab :) ).

> could take a readily available SPARC or StrongARM core and tack some
> GC support onto it, I don't know.  Certainly Sun are hyping their MAJC
> Java chip pretty strongly.

Yes, I was thinking about modifying an existing chip. For instance, adding
a tag bit to every machine word should not be so difficult.


Alain Frisch

Re: Ocaml Machine (was Re: speed versus C)

[David Monniaux]

On Tue, 12 Oct 1999, Alain Frisch wrote:
> Yes, I was thinking about modifying an existing chip. For instance, adding
> a tag bit to every machine word should not be so difficult.

This is so true that the Sparc architecture already supports tagged
arithmetic (with two bits of tag per word, if I remember well).

---
David Monniaux     Tel: +33 1 44 32 20 66    Fax: +33 1 44 32 20 80 
Laboratoire d'informatique de l'École Normale Supérieure,
45 rue d'Ulm - 75230 PARIS cedex 5 - FRANCE

Re: Ocaml Machine (was Re: speed versus C)

[Alain Frisch]

> This is so true that the Sparc architecture already supports tagged
> arithmetic (with two bits of tag per word, if I remember well).

Does the ocaml native compiler use this feature ? (int will be 32 or 64
bits, not 31 or 63).