On Thursday,  7 March 2002 at 10:51:34 +0000, David Laight wrote:
> On Wed, Mar 06, 2002 at 04:54:38PM -0800, Derek Peschel wrote:
>> On Wed, Mar 06, 2002 at 01:42:38AM -0500, der Mouse wrote:
>>>> how cross compiler generate 36-bit code on 32-bit machine?
>>>
>>> Same way any foreign code is generated - mostly up to the assembler and
>>> linker.  Define a way of representing 36-bit words (as, say, 5 8-bit
>>> bytes with 4 bits wasted, or two 36-bit words in 9 8-bit bytes, or even
>>> in extremis one 36-bit word stored in a 64-bit space with 28 bits of
>>> padding).  Setting up the first boot may be a bit interesting,
>>> depending on the boot methods and media, but that's about it.
>>
>> The designers of FTP already solved that problem.  If a file contains
>> 8-bit bytes, you can transfer it between a 16- or 32-bit machine and a 36-bit
>> machine.  Probably the same is true for files containing 36-bit words.
>> And it may even be true for word sizes other than 36.
>
> What about 18 bit words?
>
> The only system I had anything to do with (in this case reading
> mag tapes it had written) that had 18/36 bit words used 6 bit
> characters with an EBCDIC sytle encoding - think it was a UNIVAC.

The UNIVAC 1100 series answer to this description, modulo EBCDIC.
EBCDIC is an 8 bit code.  The older 1100s used a 6 bit encoding called
Fieldata, but later they moved to 9 bit ASCII.

> The system words were written to 9 track (8 + parity) mag tape
> without any wasted space - so one mag tape 'byte' could hold parts
> of 2 different 36 bit words.  Fortunately the characters were
> sensibly ordered (as if the system had 24 bit words).

The UNIVAC 494 had 30 bit words.  Otherwise it was quite like the
1100s, but 9 trace tapes had a number of different encodings,
including 4 words spread over 15 frames.

Greg
--
Finger grog@lemis.com for PGP public key
See complete headers for address and phone numbers