Re: 36 vs 32 bit architecture, was Fat references

glen herrmannsfeldt <>
Sun, 3 Jan 2010 21:10:29 +0000 (UTC)

          From comp.compilers

Related articles
Fat references (Jon Harrop) (2009-12-29)
Re: Fat references (Kaz Kylheku) (2009-12-30)
Re: Fat references (glen herrmannsfeldt) (2009-12-31)
Re: Fat references (2010-01-02)
Re: Fat references (Hans-Peter Diettrich) (2010-01-03)
Re: 36 vs 32 bit architecture, was Fat references (glen herrmannsfeldt) (2010-01-03)
| List of all articles for this month |

From: glen herrmannsfeldt <>
Newsgroups: comp.compilers
Date: Sun, 3 Jan 2010 21:10:29 +0000 (UTC)
Organization: California Institute of Technology, Pasadena
References: 09-12-045 09-12-055 10-01-003 10-01-008 10-01-012
Keywords: architecture, history
Posted-Date: 04 Jan 2010 11:19:06 EST

Hans-Peter Diettrich <> wrote:
(John wrote)

> [Early DEC machines had 18, 12, and 36 bit words, but once the PDP-11
> became a success in the early 1970s, byte sizes other than 8 and word
> sizes other than 16 and 32 soon withered away. -John]

There is a discussion on comp.sys.pdp10 on whether the PDP-10
architecture could have survived into the 1980's and 1990's if
DEC had tried to keep it alive.

Many of the IBM scientific machines before S/360, including
the 704 that started Fortran, were 36 bit machines.
There are many stories about the significance of the
precision loss going from the 36 bit 7090 to 32 bit S/360.
IBM's assumption was that people would go to 64 bit double
precision for most computation, which is probably about right.

I do believe that the continued use of the PDP-10 hardware would
have needed a file system based on at least 8 bit characters.
(TOPS-10 stores five 7 bit ASCII characters in a 36 bit word.)

I have heard that C compilers for the PDP-10 use a CHAR_BIT of 9,
though the architecture makes addressing 18 bit halfwords a
little easier than 9 bit bytes.

-- glen
[The 360 had well documented design errors in its floating point which
made the useful precision three bits less than it should have been, so
it wasn't really 32 vs. 36 bits. But the real fatal flaw on the PDP-10
was the 18 bit addressing, and the indirect segmented indirect addressing
kludge they added to later DEC-20s just wasn't good enough. -John]

Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.