HP calculators and extended precision <9103181630.AA03422adgh.Eng.Sun.COM>

[sun!Eng!khb]

>can be gleaned, I think, from the HP calculators that have exploited it

Calculators are _primarily_ used one instruction at a time. Even those
of us with highlyprogrammable ones (hp 41c|71b, TI 56) spend most of
our time in single step mode. Computers are _primarily_ used in a more
batch oriented mode. This difference in interface has important
consequences; not the least of which is that all calculator
optimization is done by hand (or not at all), and that nearly all
computer optimization is done automatically....

Also, intermediate results are not seen by most computer users ....
whilst most caclulator computations are quite visible to the user.

It does not seem strange to me, that one would want the underlying
design to reflect these differences.

Extended arithmetic is very handy for hand coding .... and seems to
violate user sensibilities whenever automatic optimization is employed
.... for at least some interesting cases. Since automatic optimization
is quite effective (on many interesting machines it is good for
factors of 4x and better in application performace speedup) it seems
to me that we should design high performance machines w/o "extra"
features which tend to cause more trouble than benefit.

There is no reason to remove such features from calculators; but the
fact that such are useful in that domain doesn't seem to suggest much
about other domains.