CODEWITHSUNDEEP
floating-pointfortranprecision
SKPS
SKPS

Reputation: 5836

Avoid declaring "_8" for fortran real variables?

I assigned a real variable in Fortran code as a=0.00001. On printing the variable on following line gave a=9.9999997473787516E-006 instead of 1E-5.

The fix for this problem is declaring a=0.00001_8 or a=0.00001_rk with rk defined with value 8.

I want to avoid this underscore style in my programming. Is there a way to do this without underscores as is in C or C++

Upvotes: 0

Views: 1094

Answers (2)

Ross
Ross

Reputation: 2179

I'll list a couple options here, but ultimately the answer is use the underscore notation. Especially when developing new code, all real variables should use a kind parameter. There are several good many ways to define a kind parameter:

  • ISO_FORTRAN_ENV has reals and integers defined by their size: REAL32, REAL64, INT32, INT64, and others.
  • Use the intrinsic function SELECTED_REAL_KIND to obtain a kind with the specified precision and exponent range
  • IEEE_SELECTED_REAL_KIND for IEEE compatability
  • ISO_C_BINDING contains C_DOUBLE and others for C interoperability

This way, you can achieve the desired precision on any machine. For example,

use ISO_FORTRAN_ENV, only : wp => REAL64
real(wp) :: x

The important part is to have wp be explicitly declared. Then, if you want x to have the value 0.1, you set it equal to 0.1 represented using the wp kind using

x = 0.1_wp

This is the fortran notation. You should not reject it simply because you think it's less readable.

However, I will include a couple other options. They're not as good though!

  • Declare constants using d instead of e, which requests double precision, for example x = 0.1d0

  • Use a compiler flag to change the default real kind. In gfortran, you would typically use -fdefault-real-8 -fdefault-double-8. This compiler option doesn't always exist (certainly not in the exact same way), and also limits the possible real kinds you can declare.

Upvotes: 2

Francois Jacq
Francois Jacq

Reputation: 1274

You may use the exponent D instead of E :

a=0.00001d0

which is only useful if your variable a has been declared either double precision of real(rk) wih rk defined as :

integer,parameter :: rk=selected_real_kind(15,100)

In addition, don't use _8 but _rk (_8 is not portable).

Upvotes: 0

Related Questions