Juan,

> >   I agree with Prof. Richard O'Keefe's comment here.  Most of the
> > space/time consuming floats are in Arrays, so optimizing the boxed
> > representation, or the rare case, wouldn't give me too much good.
> >
> That's not always the case. For example, I'm currently working with a
> simulation system that on a modest simulation uses over 30000 Floats and
> millions of float operations (in VSE). The simulation is made of lots of
> small objects, and we could not make much use of FloatArrays. I believe that
> an inmediate float representation (perhaps 32 bit, and not 61, to use
> standard IEEE and C Floats) could make float operations faster, and avoiding
> the creation of lots of instances would save memory and GC time.

  Ah, ok.  But I know that you know how to write primitives well^^; So
probably you might want to stick them into an array and write
primitives to manipulate them in some way?  Even we store a 32 bit
value in a 64 bit word, we need to shift and mask to align it...

-- Yoshiki