[Simh] Non-2^n architectures - Re: PIP10 on PDP-8 SIM

[Rob Doyle]

On 3/19/2013 7:27 PM, Toby Thain wrote:
> On 19/03/13 9:43 AM, Johnny Billquist wrote:
>> On 2013-03-19 14:30, Armistead, Jason wrote:
>>> I had trouble with Timothe’s link to the USPTO, but found this
>>> same patent in PDF form at
>>>
>>> http://www.brouhaha.com/~eric/retrocomputing/dec/dectape/3387293.pdf
>>>
>>> As a relative newbie who started my serious journey into
>>> computing with an Apple ][ I’ve never fully understood DEC’s
>>> fascination with word lengths that weren’t multiples of 2 ...
>>
>> It wasn't just DEC. Back in the day, most everyone used various
>> word lengths that wasn't a power of two. I can't really make many
>> comments on why other word lengths were more popular.
>
> 12, 15, 18, 36, 60 ...
>
> 24 is still common in DSPs (and maybe 48 and 56?) - and even word
> addressing is still used.

24-bit data is common for cost sensitive audio applications because
16-bit data is insufficient and 32-bit data is overkill.

GPUs typically have odd data widths for similar reasons.

In these types of applications, the notion of 8-bit bytes is mostly
irrelevant. As stated above, many of these devices can't even address
bytes. Interestingly enough, none of these /limitations/ preclude having
a compliant "C" compiler.

>> I've seen mentioned that floating point formats was pretty nice to
>> do with something like 60 or 72 bits. Reason being that you had
>> large enough exponents for useful things, and enough precision for
>> most calculations. So a word length that related to this made
>> sense.
>>
>> Number of bits being a power of two started with IBM in the 60s,
>> and became common with the PDP-11 in the 70s. (Or so I'd like to
>> think.)
>
>
>
> And, not insignificantly, the rise of 8-bit microprocessors.
>
> (Though octal was still standard notation for addresses and many
> constants on PDP-11, hence C's octal literals.)
>
> --Toby
>
>>
>> Johnny
>>
>
> _______________________________________________ Simh mailing list
> Simh at trailing-edge.com
> http://mailman.trailing-edge.com/mailman/listinfo/simh
>