The Lisp Machine
P. T. Withington
Mon, 6 Apr 1998 06:14:01 -0400
On 4/4/98 16:08, David Gadbois wrote:
> Date: Fri, 3 Apr 1998 19:27:53 -0500
> From: "P. T. Withington" <firstname.lastname@example.org>
> I've put two old (unpublished, until now) papers, "The Lisp
> Machine: Noble Experiment or Fabulous Failure?", and "The
> Symbolics Virtual Lisp Machine (extended abstract)", which might be
> amusing to this group, at <http://www.pobox.com/~ptw>.
>Wow, great papers! Any chance of ever seeing the full VLM paper?
>It occurs to me that there may be a couple of alternatives to doing
>the emulator in assembly. (This may have relevance if the MIT lispm
>sources ever get released.) Clearly, if performance permits, it would
>be better to do an emulator in a high-level language so as to allow
>portability and to take advantage of new implementations that have
>difference optimization characteristics: larger, differently mapped
>caches, more functional units, register renaming, etc.
>1. Reschedule the output of the compiler to better take advantage of
>the chip's instruction issue regime, and move stuff around (using
>knowledge of the code) to get better instruction cache behavior.
>2. Fix the compiler to do the right thing.
>Were these (and perhaps other) approaches rejected because doing it by
>hand made a big enough difference, or was it a time-to-market thing?
Strictly time-to-market. We didn't have time to build or wait for better
compilers. The program to build the VLM was a last ditch effort to save
the company. It exceeded our expectations, but it was not enough to
overcome the outstanding debts.