About the REAL constants in iso_fortran_env

[Angelo Graziosi]

Il 09/07/2015 01:08, Steve Kargl ha scritto:
>
> FX's choice is fine, and actually has a performance consequence.
> The Intel 80-bit extended format has FPU support.  The REAL(16)
> type, which is also 128 bits, is a software implementation of
> IEEE binary128.

I know that FX does the right things..

>
> So, to answer the stack overflow question.  Learn how to
> use SELECTED_REAL_KIND(P=foo, R=bar) to get exactly what
> you want.
>

All my applications are based on this module (at least since 
quad-precision was added to GCC):


module kind_consts
   implicit none
   private

   integer, parameter :: QP = selected_real_kind(30,300)

   !integer, parameter :: DP = kind(1.D0)
   ! DP: Following A. Miller definitions...
   integer, parameter :: DP = selected_real_kind(12,60)

   integer, parameter :: I1B = selected_int_kind(2)
   !integer, parameter :: I8B = selected_int_kind(10)
   integer, parameter :: I8B = selected_int_kind(18)

   public :: DP, I1B, I8B, QP

end module kind_consts

which over the years it has undergone some evolution, but recently I 
have found discussions (as that on stack overflow) in which people uses 
REAL?? from iso_fortran_env module to define precision. Some test case 
suggested that something does not add up..

Thanks for the detailed explanation.


Angelo