/************************************************************************/
/*									*/
/*  Notes:								*/
/*	1) Rational approximations make especially good sense, as	*/
/*		the divide time is not much longer than add or multiply	*/
/*									*/
/*	2) Floating point compares can be implemented as integer	*/
/*		compares when both numbers are positive by 		*/
/*		doing an ordinary compare.				*/
/*		When both numbers are negative, the sense of the	*/
/*		compare is reversed.					*/
/*									*/
/*	3) Someone with more time and ambition might want to examine	*/
/*		the fixed point algorithms in Cody & Waite.  In any	*/
/*		case, the call overhead could be reduced enormously	*/
/*		by providing versions of the arithmetic routines that	*/
/*		would expect their arguments in registers.		*/
/*		(However, this seems likely to make only about a 5%	*/
/*		difference in r_exp, for example.)			*/
/*									*/
/************************************************************************/
