NetNews Usenet Archive 1992 #18/NN_1992_18.iso/spool/comp/arch/8941

Internet Message Format | 1992-08-17 | 3.1 KB

Xref: sparky comp.arch:8941 alt.folklore.computers:12689
Path: sparky!uunet!dtix!darwin.sura.net!Sirius.dfn.de!math.fu-berlin.de!news.uni-stuttgart.de!ifi!news
From: [e-mail] (Hans-Georg Zipperer)
Newsgroups: comp.arch,alt.folklore.computers
Subject: Re: Babbage books (was: Proposal: Computer History Project)
Message-ID: <[e-mail]>
Date: 18 Aug 92 07:44:34 GMT
References: <samw.713392337@bucket> <[e-mail]> <[e-mail]>
Sender: [e-mail]
Organization: Informatik, Uni Stuttgart, W. Germany
Lines: 47

In article <[e-mail]> [e-mail] (Dave Martindale) writes:
>In article <[e-mail]> [e-mail] writes:
...
>>[about the adding mechanism of the Babbage engines]
...
>This is not true of the Difference Engine, at least.  The basic addition
>is done with all digits in parallel, but that's a relatively constant load.
>Any carries generated just result in a little lever being moved to a
>new position (Babbage called it "warned"), but no carry propagation happens
>at this stage.
>
>Then, when the main addition is done, another piece of mechanism comes
>into play to handle carry propagation, and that is done serially.
>Each digit wheel is rotated one place if a carry is pending in that
>position, and any "carry in" that creates a "carry out" at this stage
>simply sets the "carry in" lever for the next higher digit, before the
>carry happens in that place.  So the more digits you have the longer the
>carry process takes, but at any given time there are at most two or three
>pieces of hardware moving.
>
>I don't know what the Analytical Engine did for addition, though.
>
>    Dave

The Difference Engine indeed used the carry mechanism described above.
Babbage needed 10 'time steps' for the main addition and spent a lot
of time to speed up the following carry propagation. He used his
'mechanical notation' (a formal language for the description of
mechanical motions) to reason about this and found a way to do the
carry propagation for his 20 or 30 digits in 10 time steps. Babbage then
used some kind of pipelining in the Difference Engine. One basic cycle
of the engine was divided in 4 subcycles. In the first subcycle, the
even-numbered columns (registers) were added to the odd-numbered columns.
In the second subcycle, the carry propagation for this additions was
done. In the third and fourth subcycles, the odd-numbered columns were
added to the even-numbered columns in the same fashion.

For the Analytic Engine, Babbage wanted 50 decimal places. This would
make the carry propagation slower than the main addition and would slow
down the whole system. In 1834, he invented the 'anticipating carriage'
(look-ahead carry), which solved the problem. The new adder still needed
10 time steps to do the main addition, but the carry propagation for an
unlimited number of digits could be done in a single time step afterwards.

--
  Hans-Georg Zipperer     Universit"at Stuttgart, Institut f"ur Informatik
  Tel. [phone]    e-mail: [e-mail]