[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

SRFI-77 with more than one flonum representation



I think that the flonum-specific parts of SRFI-77 has some issues when 
there is more than one flonum representation[*]. For example, consider 
an implementation with IEEE 754 single-precision and double-precision 
flonums.

(a) Does (fixnum->flonum 0) return a single-precision or a 
double-precision flonum?

(b) If x and y are single-precision flonums whose sum cannot be 
represented as a finite single-precision flonum, does (fl+ x y) return a 
single-precision infinity or a double-precision finite value?

(c) How do I convert between single-precision and double-precision flonums?

I suspect there may be other issues.

Regards,

Alan

[*] Why support anything other than double? Well, long double can be 
useful in certain algorithms. Furthermore, external data is often 
represented as shorts: if such data is represented internally as 
doubles, the additional precision and range can cause problems. Finally, 
on a 64-bit machine you can probably unbox shorts, which may make 
arithmetic with shorts significantly faster than arithmetic with 
doubles. I'm sure other people can come up with other justifications.