The KIND Number

• makes DOUBLE PRECISION redundant;
  
  
• Fortran standard requires processor to offer:
  • at least one representation for INTEGERs, LOGICALs, CHARACTERs and
  • at least two representations for REALs, with different precisions.
  There are usually more.
  
  
• a KIND number is associated with each representation of each type;
  
  
• one can use KIND to write portable code --- code in which the representation of integers and reals is guaranteed to be the same on different machines.

Integers

Suppose a processor supports three different range values for integers, -128--127, -65536--65535 and a 32-bit representation (-4.29*10^9--4.29*10^9):

• these might have KIND numbers 1,2 and 4;

    INTEGER(KIND=1) :: a, b
    INTEGER(2)      :: c, d
    INTEGER(4)      :: e

    INTEGER, PARAMETER :: short = 1, medium = 2, long = 4

    INTEGER(short)  :: a, b
    INTEGER(medium) :: c, d
    INTEGER(long)   :: e

SELECTED_INT_KIND

SELECTED_INT_KIND(r)

returns the KIND value of the integer representation with the smallest range that can represent values n such that

-10^r < n < 10^r

Therefore we have:

INTEGER, PARAMETER :: short = SELECTED_INT_KIND(2), & medium = SELECTED_INT_KIND(4), & long = SELECTED_INT_KIND(10) INTEGER(short) :: a, b INTEGER(medium) :: c, d INTEGER(long) :: e

Reals

• here we are concerned not only with range of values possible for a variable, but also precision;

    REAL, PARAMETER :: single = 4, double = 8, extended = 16;

    REAL(single)   :: x, y
    REAL(double)   :: xx, yy
    REAL(extended) :: eee

SELECTED_REAL_KIND

SELECTED_REAL_KIND(p,r)

where p is the required precision and r is as before. Therefore we have:

REAL, PARAMETER :: single = SELECTED_REAL_KIND(5,20), & double = SELECTED_REAL_KIND(10,40), & extended = SELECTED_REAL_KIND(20,100) REAL(single) :: a, b REAL(double) :: c, d REAL(extended) :: e

COMPLEX

Is analogous to that for REAL --- same KIND values.

CHARACTER

Skipped!

LOGICAL

Skipped!