Thanks for STAR-100 info + new question :-)
ulm...@fafner.zdv.uni-mainz.de
I wish to thank all of you who helped me find information about the CDC-STAR
100. Thank you for giving me hints, for copying articles, etc.!
In one posting the pages of Gordon Bell (THE Gordon Bell :-) ) was mentioned
and on these pages I found a manual for the CDC-8600 multiprocessor machine
which is a very interesting lecture. But - as with all interesting things -
there arose some questions - mainly the following:
The machine used a modified version of ones-complement in which the problem of
two representations of the value zero (0...0 and 1...1 respectively) is avoided
by some additional logic which forces a word consisting of ones only to a word
with the value 00000...0. It may be a silly question but currently I am
wondering why this scheme has been abandoned by Cray (and the rest of the
computer architects) - to me it seems to be quite elegant - there is no problem
of different representations of zero, calculating the complement of a number is
quite simple (only inverting of all bits with checking for 1111...1, so no
carry logic etc. is required).
Have a nice day - best regards,
Bernd Ulmann. :-)
Douglas A. Gwyn
> two representations of the value zero (0...0 and 1...1 respectively) is avoided
> by some additional logic which forces a word consisting of ones only to a word
> with the value 00000...0. It may be a silly question but currently I am
> wondering why this scheme has been abandoned by Cray (and the rest of the
> computer architects) - to me it seems to be quite elegant - there is no problem
> of different representations of zero, calculating the complement of a number is
> quite simple (only inverting of all bits with checking for 1111...1, so no
> carry logic etc. is required).
Ah, ones-complement, a specialty of mine..
There really wasn't much of a practical problem with other CDC
implementations of ones-complement; for example, the 1700 had
a ones-complement subtractive adder, and the system documentation
spelled out the resulting conditions under which an arithmetic
operation could produce a -0 result; apart from exceptional
conditions, nearly all of them involved having a -0 as one of
the operands. Under normal circumstances, -0 didn't get in the
way; for the few cases that it might, macroassembler programmers
and compiler code generators would insert an extra "normalizing"
instruction to turn -0 into +0. My guess is that for the 8600
it was thought worthwhile to add logic to eliminate this
occasional code inefficiency.
Complementing a number doesn't require carry logic anyway.
I think you had in mind *negating* using a *twos-complement*
representation. Indeed, the main reason CDC used ones-complement
was that it made some common arithmetic operations slightly
faster (fewer gates).
Virtually all modern architectures use twos-complement, for no
overwhelming reason other than that it is what people expect.
Jim Stockard
"Douglas A. Gwyn" <DAG...@null.net> wrote in message
news:3AFFFEBA...@null.net...
As an ex-CDC logic designer (30 years), who worked on the 7400, a TCS (CDC
version of MECL) version of the 7600, and the 7700, a dual headed 7600,
which were all forunners of the 8600 (I actually did a minor amount of work
on the 8600 before Cray left CDC), As I recall, the output result of the
subtraction pyramid was blocked if all sections had a borrow. An easy
implementation. Did not account for some special cases which could generate
an -0 though. Sorry if details are sketchy, time does erase some of the
finer points of some machine designs.
The STAR-100 design did not share much with the Cray machines (1604, 160,
6600, 7600) They were done by another independent design group, which was
somewhat competing with Cray at CHOPS.
Old Logic Designer