IEEE standard arithmetic revisions for bignums

[Richard Fateman]

(I'm not on this list, so make sure any replies you
want me to see are cc'd to me.)

There is a committee revising the IEEE floating-point
standard, including other radix (e.g. 10) arithmetic,
and various other topics, including bigfloats. I think
they mean like (doubled double), and doubling that...

I was asked if I could advise on whether
there should be an IEEE standard for integer bignums.


My response to the committee chair was to look at
the software around (like GMP), and consider
what architectural assist might work to speed
up the time sinks.  (I recall from my own experience
in the past, that having a 32X32->64 bit multiply,
with carry, would have been very handy.)

What say you all about this?  (Note, the IEEE
standards do not have to be implemented in hardware
especially.)

Presumably also of concern is that a design would have
to work with languages.  I pointed out that of the
ANSI languages, Common Lisp and its predecessors have
had integer bignums for about 40 years1


Regards,
 Richard Fateman
UC Berkeley