Sun is working on making computers do floating point math correctly.
“The problem that Gustafson and Papadopoulos referred to stems from the fact that the binary mathematics employed by computers has a hard time accurately representing certain numbers. Fractions, for example, are particularly tough, because they often involve non-terminating numbers that are impossible to accurately express in binary format.”
Muli Ben-Yehuda's journal 
Oh, man.
This has got to be the biggest hype for something everyone knew for a long time. “Sun Implements Interval Arithmetic in Hardware” would have been a more accurate title. Also, it’s not meant to make things more exact — it’s meant to make the inexactness more noticeable. But then, calling Java “a modern object-oriented language” already showed what the Sun PR machine is about 🙂
Comment by moshez — December 19, 2003 @ 10:49 AM |
Well, I know almost nothing at all about floating point arithmetic. We don’t do no FP in the kernel 😉 References to interval arithmetic?
Comment by mulix — December 19, 2003 @ 12:40 PM |
SICP has a reference, at least you start implementing some preliminary interval arithmetic stuff. The trick Sun seems to be playing is that occasionally they’ll artificially worsen the boundaries to keep storage to a minimum. You could probably implement a quick prototype of that with the rational class in Python’s sandbox — it has a “approximate to best within given storage”, so all you’d have to do is do the approximation, except make sure it’s from the bottom for the lower-bound and from above for the upper bound. Implementing it with floats might be a bit trickier, but doesn’t sound that earth shattering.
Comment by moshez — December 19, 2003 @ 3:21 PM