FORTRAN 77 had a problem with numeric portability, the precision (and exponent
range) of a data type
on one processor would not necessarily be the same on another. For
example, a default REAL may be able to support numbers up to (say)
on one machine and up to 
on another. One of
the goals of Fortran 90 was to overcome this portability problem.
Fortran 90 implements a portable precision selecting mechanism, it supports
types which can be parameterised by a kind value (an integer).
The kind value is used to select a representation model for a type
and, for numeric types, can be specified in terms of the required
precision. A processor can
support different precisions (or representations) of
INTEGER, REAL and COMPLEX, different CHARACTER sets (for example,
arabic, musical notation and cryllic script)
and different ways of representing LOGICAL objects. DOUBLE PRECISION does not
have different precisions and its use is not
recommended -- use a parameterised REAL type instead.
An example of the parameterisation of an intrinsic type is,
INTEGER(KIND=1) :: ik1
REAL(4) :: rk4
The kind parameters correspond to differing precisions supported by the compiler (details in the compiler manual).
Objects of different kinds can be mixed in arithmetic expressions and rules exist for type coercion, procedure arguments, however, must match in type and kind. There is no type coercion across procedure boundaries.
Return to corresponding overview page 