Parametrized Reals

Since a floating point number is represented by mantissa and exponent, we need to select two parameters

P

the precision, giving the number of decimal digits allowed in the mantissa, and

R

the range of the decimal exponent.

The selector function

selected_real_kind(P=7, R=12)

returns a kind value that supports numbers with 7 decimal digits and exponent magnitudes in the range .

Consider the following declarations

integer, parameter :: single = selected_real_kind(p=6, r=37), &
                      double = selected_real_kind(p=14,r=307), &
         	      quad   = selected_real_kind(p=33,r=4931)
real(kind=single) :: x
real(kind=double) :: y
real(kind=quad)   :: z

On an Alpha 21264 processor the kind values are single = 4, double = 8, quad = 16. The variables x, y, z are of type REAL, but have different kinds; using parametrized declarations we can work at the level of accuracy desired. Since the kinds single, double and quad are defined via the selector function, we can count on the precision and range desired, regardless of the machine. The code is therefore portable.