PDP-11/83 w/FPU?

[allison]

On 04/18/2019 10:56 AM, Noel Chiappa via cctalk wrote:
>     > From: W2HX
>
>     > i have a few CPUs available to me, a 11/23+, an 11/73 and I also have
>     > available to me an 11/83
>     > I would like to try to run as many different OS's as may interest me,
>     > including some unixes as possible (bsd...etc).
>
> Early Unixes in general will run on those machines - but not straight off the
> tape (since they didn't exist then, and have quirks which aren't supported).
>
> I've brought up V6 on a /23 (which must have the KTF11-A MMU chip); here:
>
>   http://gunkies.org/wiki/Running_UNIX_V6_on_an_-11/23
>
> are instructions on exactly what (minor) changes need to be made for it to
> run.
>
> The /73 and /83 should be subsets of that, although you'll want to start with
> m45.s, because those machines support the split-I-D MMU of the -11/45. (A /23
> Unix binary would boot/run on them, if you don't feel like doing a special one
> for them.) I haven't yet tried V6 on them; if you want me to, and do a
> writeup, let me know. The /73 and /83 have LTC registers, so on those you
> won't need the LTC hack.
>
> Also, you may know this already, but if not, note that the /83 is a PMI:
>
>   http://gunkies.org/wiki/Private_Memory_Interconnect
>
> machine, and _MUST_ be plugged into a Q/CD backplane _only_; plugging into
> a standard Q/Q backplane will _damage_ it.
>
>
>     > would I see any improvement in performance with the FPU compared to
>     > without it? Or does the application running need to be something like
>     > fortran to see any perceivable difference?
>
> As someone noted, the /73 and /83 implement thefloating point instructions in
> microcode, so the code can't tell if the optional FPJ11 FP hardware
> accelerator is plugged in or not. In general, only on applications (the
> language is not relevant) which are heavy users of FP would you see any
> difference.
>
> On the /23, with no KEF11-A FPU chip plugged in, there are no floating point
> instructions at all, so any application which tries to use them will blow out
> (although under V6 there's an emulator); see here:
>
>   http://gunkies.org/wiki/Setting_up_UNIX_-_Sixth_Edition
>
> and search for 'floating point' to see discussion of it).
>
>
>     > From: Ethan Dicks
>
>     > v5, v6, and v7 UNIX shouldn't require any sort of math hardware.
>
> Don't know v5/v7 in detail, but AFAIK that's accurate. V6 can _support_
> FP hardware on machines which have it, and is otherwised prepared to
> emulate those instructions (see above).
>
>     > From: Paul Koning
>
>     > I think that was typically called "EAE" (extended arithmetic element),
>     > a Unibus peripheral that implemented integer mul/div ... It only
>     > applies to 11/20 and 11/05 since all the other machines have the
>     > relevant instructions built into the CPU.
>
> Also the -11/04 and -11/03 were both missing the EIS; the former could use
> the EAE, for the latter the optional KEV11-A or KEV11-B microcode chips both
> provide it.
>
>     > From: Josh Dersch
>
>     > The EAE was also an option on the 11/40.
>
> Technically, on any UNIBUS machine; on the /40, the EIS (added instructions,
> not the device model of the EAE) was available via an optional board in
> the CPU.
>
> 	Noel
All,


Experience is that an 11/23 or 23+ will run V6 as mine does.  REason it
does is V6 does not require I&D support

The usual issue is not FPU for Unix as questioned but if there is a need
for I&D spaces.  The 11/23 (f11 chipset)
does not support that but the J11 (11/73 and 11/83) do support that.  I
have a 11/73 so I could run BSD and a few
others commonly found that require I&D support.

There may be other versions that place less of a burden on requiring I&D.
However I've not encountered a need for FPU connected to OS.  Also
the assumption for many unix is MMU support but not all DEC OS have
that requirement.

Allison