18 bit CPU; was: Speed now & then

[Bill Gunshannon]

On 04/12/2018 09:55 AM, Toby Thain via cctalk wrote:
> On 2018-04-12 7:48 AM, Bill Gunshannon via cctalk wrote:
>>
>> On 04/12/2018 02:45 AM, Chuck Guzis via cctalk wrote:
>>> On 04/11/2018 06:38 PM, Chuck Guzis via cctalk wrote:
>>>> On 04/11/2018 02:48 PM, ben via cctalk wrote:
>>>>
>>>>> I have a nice 18 bit cpu here, with only a few hardware bugs.
>>>>> Hmm would it work better if I change that around ideas.
>>>>>
>>>>> Care to point to a nice 18 bit version of unix or C.
>>>>> BTW The cpu has a frame pointer S but no S++ --S operations
>>>>> so pushing and popping wild data is not a option.
>>>> Well, the Univac 1100/2200 series mainframes ran V7 Unix--and they're
>>>> 36-bit machines, so probably not far from your 18-bitter--and they're
>>>> ones' complement machines.
>>>>
>>>> Univac called it "SX1100", so you have a search term.
>>> Remarkably, Unisys keeps a lot of old documentation around.  Here's the
>>> reference manual for their "C" on th 1100:
>>>
>>> https://public.support.unisys.com/2200/docs/cp14.0/pdf/78310422-011.pdf
>>>
>>> Looks pretty much like standard C until you get into the minutiae, such
>>> as "A character constant is 1 to 4 characters" ...
> Apple also used this extension in their Mac C compilers.
>
>>> (9 bit characters and 36 bit ints and 18 bit short ints).
>>>
>>> So, it should be pretty straightforward unless you assume that a char is
>>> 8 bits, with a signed char having a range of +/-255.
>>>
>> Signed 8 bits would be  -128 to +127.
> Yes, but Chuck was describing the Univac, where char is a 9-bit type,
> and a signed char is 1 sign bit and 8 magnitude bits, giving +/- 255.
>
I'm familiar with Univac's having worked on the 1100 many moons ago,
But look at the line above my comment:
  "you assume that a char is 8 bits, with a signed char having a range 
of +/-255".

An 8 bit signed char has the values -128 to +127, as I stated.  even a 9 bit
signed char would not be +/-255 but -256 to +255.

bill