> I don't see how much we gain from the separation.  If it does much good,
> though, maybe we should go the other way, and separate things out a bit
> more.

One thing a separate math library buys us is the ability to have
different implementations from which the user can select.  There
could be the generic version plus one each for the various math
coprocessors that might be available to a system.

	--jtc