Xerox 530 and SEL 810

[Quadibloc]

On my web page at

http://www.quadibloc.com/comp/cp0201.htm

I've added to my illustration of Group III floating-point formats the
floating-point format of the SEL 810 computer - which happens to be
the only 16-bit computer I know of with a Group III floating-point
format.

This is my own classification of old floating-point formats.

Group I is what we're most familiar with - a format with the structure

(sign) (exponent) (mantissa)

or

(sign) (exponent sign) (exponent) (mantissa)

These formats usually have the mantissa in sign-magnitude format and
the exponent in excess-n format.

Group II is found on many computers that do floating-point in
software; these formats have the structure

(exponent sign) (exponent) (sign) (mantissa)

or

(sign) (mantissa) (exponent sign) (exponent)

and both the mantissa and the exponent are usually in two's complement
form - although they can be in other forms; they're in the same form
as is used for integers.

The exponent field could be somewhat on the large side, in order that
the boundary between the exponent and the mantissa could fall on a
word boundary, but putting it on a byte boundary is also possible.

Group III is generally found on computers which did have hardware
multiply but not hardware floating-point. This form went like this:

First word:
(sign) (start of mantissa)

Second word:
(ignored bit) (rest of mantissa) (exponent sign) (exponent)

This is why it tended to be associated with 24-bit computers, because
at the time under consideration, they were medium-range systems.

John Savard

[Joe Chisolm]

Just to clarify a few things on the X530. For a time I was
responsible for all software maintenance for the Xerox 16 bit
products after Honeywell took over Xerox.

The 530 was a LSI/MSI micro programmed follow on to the Sigma 2/3 lines
(in a similar way that the 550/560 were LSI/MSI micro programmed
Sigma 9s). The 530 did have a optional floating point hardware. There
was a bit in the program status word that made the fixed point
instructions (load,store,add,sub,mul,div,compare) operate as floating
point instructions. The floating point numbers were 48 bits long (3
16 bit words). The first 2 words were a signed 2 comp mantissa and
the signed 2 comp exponent was the 3rd word. Note that the exponent
was not biased by 0x40 like they were on the 32 bit product line. There
were several ways you could flip between floating point and fixed point
mode.

--
Chisolm
Republic of Texas

[Christian Brunschen]

In article <3fdc49e2-acb2-4c46...@a3g2000vbr.googlegroups.com>,

Quadibloc <jsa...@ecn.ab.ca> wrote:
>On my web page at
>
>http://www.quadibloc.com/comp/cp0201.htm
>
>I've added to my illustration of Group III floating-point formats the
>floating-point format of the SEL 810 computer - which happens to be
>the only 16-bit computer I know of with a Group III floating-point
>format.
>
>This is my own classification of old floating-point formats.

SMIL <http://en.wikipedia.org/wiki/SMIL_(computer)>, in its upgraded form,
used a floating-point format of:

* mantissa, 32 bits, in 2's complement form
* exponent, 8 bits, in excess-128 form - i.e., 80(hex) + actual exponent.

40 bits in total, as you'd expect of an IAS machine design!

See also <http://www.df.lth.se/~cb/SMIL/prog_for_smil.pdf>, Chapter 1
("Talrepresentation") Section 4 ("Maskintal, packad form").

>John Savard

Best wishes,

// Christian

[Robin Vowels]

Type IV ?

First word:
(sign) exponent
Second word:
(sign) mantissa

as 32 bits for each word, stored in twos complement notation.

[Quadibloc]

On May 15, 6:56 am, Robin Vowels <robin.vow...@gmail.com> wrote:

> Type IV ?
>
> First word:
>      (sign) exponent
> Second word:
>      (sign) mantissa
>
> as 32 bits for each word, stored in twos complement notation.

I'd class that as Type II, as the classifications are fairly broad.
And, of course, such formats were used in real life. With 40 bits for
each word, the RECOMP II ended up celebrating this compromise as a
feature.

John Savard

[Walter Bushell]

In article
<d64a117b-7c93-4072...@li6g2000pbb.googlegroups.com>,

Robin Vowels <robin....@gmail.com> wrote:

> Type IV ?
>
> First word:
> (sign) exponent
> Second word:
> (sign) mantissa
>
> as 32 bits for each word, stored in twos complement notation.

Very inefficient as to storing numbers. If you're devoting 64 bits for
each float you'd want more precision.

--
Gambling with Other People's Money is the meth of the fiscal industry.
me -- in the spirit of Karl and Groucho Marx

[Quadibloc]

On May 15, 8:55 am, I wrote:

> With 40 bits for
> each word, the RECOMP II ended up celebrating this compromise as a
> feature.

Finally,

http://www.popscreen.com/p/MTIwNzI1MDkw/Amazoncom-1960-Autonetics-RECOMP-Digital-Computer-Print-Ad-

I found an example of one of those advertisements online.

John Savard