On Fri, 9 Sept 2022 at 16:28, Bakul Shah <bakul@iitbombay.org> wrote:

> On Sep 9, 2022, at 12:39 PM, Nelson H. F. Beebe <beebe@math.utah.edu>
> wrote:
> >
> > Paul Winalski and Bakul Shah commented on bit addressable machines
> > on the TUHS list recently.  From Blaauw and Brooks' excellent
> > Computer Architecture book
> >
> >       http://www.math.utah.edu/pub/tex/bib/master.html#Blaauw:1997:CAC
> >
> > on page 98, I find
> >
> >>> ...
> >>> The earliest computer with bit resolution is the [IBM 7030] Stretch.
> >>> The Burroughs B1700 (1972) and CDC STAR100 (1973) are later examples.
> >>>
> >>> Bit resolution is costly in format space, since it uses a maximum
> >>> number of bits for address and length specification.  Sharpening
> >>> resolution from the byte to the bit costs the same as increasing
> >>> address-space size eight-fold.
> >>>
> >>> Since almost all storage realizations are organized as matrices,
> >>> bit resolution is also expensive in time or equipment.
> >>> ...
>
> And yet according to Wilner's article "the B1700 appears to
> require less than half the memory needed by byte-oriented
> systems to represent programs. Comparisons with word-oriented
> systems are even more favorable."
>
> Figure 9 shows sample sizes for Cobol, Fortran and RPG II programs
> comparing B1700 code sizes with other systems. I was surprised to
> see this but didn't look further.
>
> https://dl.acm.org/doi/pdf/10.1145/1479992.1480060
>
> From the same paper
>
>   DESIGN OBJECTIVE
>
>   Burroughs B1700 is a protean attempt to completely vanquish
>   procrustean structures, to give 100 percent variability, or
>   the appearance of no inherent structure. Without inherent
>   structure, any definable language can be efficiently used
>   for computing. There are no word sizes or data
>   formats—operands may be any shape or size, without loss of
>   efficiency; there are no a priori instructions—machine
>   operations may be any function, in any form, without loss
>   of efficiency; configuration limits, while not totally
>   removable, can be made to exist only as points of "graceful
>   degradation" of performance; modularity may be increased,
>   to allow miniconfigurations and supercomputers using the
>   same components.
>
>
The level of florid language in that paper is truly impressive.

This appears to be an early implementation of intermediate language
representation.  I gather by its relative level of success (I had not heard
of it until now) that it suffered from many of the common performance
problems of such machines (Java bytecode, the Transmeta CPU, etc.) and did
not succeed in the marketplace.

-Henry