Hardware STG?

Seth Kurtzberg seth@cql.com
Tue, 27 Aug 2002 22:38:33 -0700


Simon,

I agree about the Tcode and GHCi byte code.  On the hardware level, however, I 
think the requirements for success have changed substantially since the 
research you spoke about occurred.  In many designs, more powerful processors 
are used than necessary, because they are mass produced and have cost 
advantages.  The cost of course is always an issue, but my point is that the 
difference in performance may in many cases be insignificant.

I see a definite place for this hardware, if it can be engineered 
successfully.  For example, I could build a firewall that is provably 
inpenetrable.  I could build embedded medical devices that I can prove will 
not fail mysteriously.  I can build devices for spacecraft that don't behave 
strangely in orbit (this last example happens frequently, and I've spend many 
hours hacking these things by radio over long distances, which as you can 
imagine is not terribly amusing.  When it fails, the loss of a $20 million 
dollar spacecraft is also not amusing.)

There is also 20 additional years of additional experience with conventional 
architectures which lends significant support to the proposition that such 
hardware can never be "fixed" sufficiently for critical situations. 

On Tuesday 27 August 2002 09:36, Simon Peyton-Jones wrote:
> Think carefully first!  Lots of people tried this in the 80's.
> To beat generic (but high-investment) microprocessors you
> need to find a factor or 10 or so improvement in the basic
> architecture, and that is hard to find.
>
> The STG machine is designed to be ok for stock hardware,
> but it's not specialised for stock hardware, so there's no reason
> it's a bad base for specialised hardware.
>
> No, Tcode is very old.  We use some different byte-code for
> the GHCi byte code, but it is designed for easy compilation
> rather than fast execution.  For fast execution we compile to
> native code, or to C, or (soon I hope) C--.
>
> Simon
>
> | -----Original Message-----
> | From: MR K P SCHUPKE [mailto:k.schupke@ic.ac.uk]
> | Sent: 21 August 2002 16:32
> | To: glasgow-haskell-users
> | Subject: Hardware STG?
> |
> |
> | Hi,
> |     I am looking into starting a project on special hardware to run
> | haskell - (or perhapse any
> | lazy functional hardware)... If I go ahead the project will
> | result in an
> | open VHDL implementation.  I would like to ask a couple of questions.
> |
> |     Is there any point in doing this as the STG-machine is
> | designed to
> | make running haskell
> |         efficient on standard hardware?
> |
> |     is TCode and the byte code used by GHCi the same?
> |
> |     finally, does anyone know of any implementations that
> | have already
> | been attempted,
> |         and how successful were they? Any general comments on how
> | worthwhile this might
> |         be?
> |
> |     Regards,
> |     Keean Schupke
> |     Department of Electrical & Electronic Engineering,
> |     Imperial College London.
> |
> |
> | _______________________________________________
> | Glasgow-haskell-users mailing list
> | Glasgow-haskell-users@haskell.org
> | http://www.haskell.org/mailman/listinfo/glasgow-| haskell-users
>
> _______________________________________________
> Glasgow-haskell-users mailing list
> Glasgow-haskell-users@haskell.org
> http://www.haskell.org/mailman/listinfo/glasgow-haskell-users

-- 
-----------------------------------
Seth Kurtzberg
M. I. S. Corp.
1-480-661-1849