Re: One more oddity with gcc and vax floats...

[Dave McGuire <mcguire%neurotica.com@localhost>]

On 6/13/26 08:26, Johnny Billquist wrote:
> > > > We had 128 bit floats on a VAX, so I'm sure somebody running funny 
> > > > things on it :)
> > >
> > > 40 years ago, sure... Today? I doubt it.
> >
> >    That's a popular assumption, but I just took a contract to maintain 
> > several Modcomp systems (1978) that are running in production at a 
> > company, and have just been approached about two more. (apparently the 
> > word has spread that there's someone who can work on those machines 
> > who isn't 80..)
> >
> >    At least this installation IS using their EAU boards.. ;)
> >
> >    Ancient hardware doing real work is shockingly common.  It's just 
> > not popular to talk about.
>
> You are misinterpreting my answer.
> I still sometimes help out on PDP-11 systems out there in the field. 
> Yes, there are plenty still running.
>
> But here we're talking about a specific kind of usage and application, 
> which is high precision math. Anyone into that is almost certainly also 
> into high performance computing. These kind of systems and applications 
> are not running on 40 year old platforms. No matter the fact that VAX 
> have 128 bit FP. IEEE FP have 256 bit format. So not only does IEEE 
> formats allow for larger ranges and higher precision, but it also does 
> it on hardware that runs way faster than any VAX. Anyone interested in 
> either very high precision, or very high speed math is not on a VAX 
> anymore, and haven't been for a long time.
>
> There are certainly other fields and types of application where the age 
> of the hardware matters very little. But where tested, reliable and cost 
> matter more, and where it makes sense to continue with what already 
> works, and does the work as well as it could ever be done.

  I didn't misinterpret your answer.  I know you're aware that there is 
ancient hardware still running.  My implied statement, which I guess I 
should've made more explicit, is that we can never really know what a 
particular application is running.  In the case of these Modcomps, they 
are doing fairly heavy FP in real time (I'd not have done it that way, 
but this has been a debugged piece of code since the early 1980s) for 
control of some very large equipment.

  High speed is a requirement for their application, but that's 
subjective.  As I understand it, the Modcomps are "fast enough".  Only 
barely, but that's no motivation to change a complex installation.

  They don't use IEEE FP either, so I shudder to think of what will 
happen when some salesman tries to sell them some Windows toy to replace 
it all.  Apparently there have been some attempts, but the big iron is 
still there.

  Anyway, it's a big world.

             -Dave

--
Dave McGuire, AK4HZ
New Kensington, PA