literal-kind (PORT011)

This rule is turned on by default.

What it does

Checks for use of raw number literals as kinds

Why is this bad?

Rather than setting an intrinsic type's kind using an integer literal, such as real(8) or integer(kind=4), consider setting kinds using parameters in the intrinsic module iso_fortran_env such as real64 and int32. For C-compatible types, consider instead iso_c_binding types such as real(c_double).

Although it is widely believed that real(8) represents an 8-byte floating point (and indeed, this is the case for most compilers and architectures), there is nothing in the standard to mandate this, and compiler vendors are free to choose any mapping between kind numbers and machine precision. This may lead to surprising results if your code is ported to another machine or compiler.

For floating point variables, we recommended using real(sp) (single precision), real(dp) (double precision), and real(qp) (quadruple precision), using:

use, intrinsic :: iso_fortran_env, only: sp => real32, &
                                         dp => real64, &
                                         qp => real128

Or alternatively:

integer, parameter :: sp = selected_real_kind(6, 37)
integer, parameter :: dp = selected_real_kind(15, 307)
integer, parameter :: qp = selected_real_kind(33, 4931)

Some prefer to set one precision parameter wp (working precision), which is set in one module and used throughout a project.

Integer sizes may be set similarly:

integer, parameter :: i1 = selected_int_kind(2)  ! 8 bits
integer, parameter :: i2 = selected_int_kind(4)  ! 16 bits
integer, parameter :: i4 = selected_int_kind(9)  ! 32 bits
integer, parameter :: i8 = selected_int_kind(18) ! 64 bits

Or:

use, intrinsic :: iso_fortran_env, only: i1 => int8, &
                                         i2 => int16, &
                                         i4 => int32, &
                                         i8 => int64