Re: Moving VAX into 21 century :-)

[Paul Koning <paulkoning%comcast.net@localhost>]

> On Aug 26, 2019, at 1:25 PM, Mark Pizzolato <Mark%infocomm.com@localhost> wrote:
> 
> 
> 
> On Aug 26, 2019 10:14 AM, Anders Magnusson <ragge%ludd.ltu.se@localhost> wrote:
> Den 2019-08-26 kl. 17:38, skrev Paul Koning:
> >>...
> > VAX does have NaN, but not Inf or denormal.  I can't see why real float applications have a need for Inf, though.
> >
> Well, not really NaN, but a floating point fault will be generated. 
> Handling SIGFPE from a user application is quite uncommon these days, 
> you get a core file instead which is not what you want when running a 
> program just because it uses floating point.

Right, so it has a "signalling NaN" but not a "quiet NaN".  Interesting, because the PDP-11, which is where all this came from, has explicit control as to whether a "reserved" float operand (NaN) produces an exception, or not.

> If you're going to implement new primitive data types, why not pick a range of unused VAX opcode's and handle the needed operations in the trap handler.  Sure, you'd have to teach gcc about these 'instructions', but you are probably going to have to do much of that anyway.  Once these new instructions are implemented, SIMH can be extended to actually execute these extra instructions directly with the host provided floating point instead of faulting.
> 
> - Mark

Nice idea.  It mimics what was done when G and H float were added to the VAX architecture.

	paul