I have a program that I want to recompile to use different floating point
precisions.  A lot of compilers support a command-line option to switch
the default real type from single to double or quad, e.g.

Intel: -real_size 32,64,128
GNU Fortran: -fdefault-real-4,8
Portand Group: -r4,8,16
Sun F95: -xtypemap=real:32,64

so until now, I have been using the default real kind and compiler
switches.  However, the time has come to use some of DH Bailey and Yozo
Hida's work, which means substituting (for example) type(qd_real) for real
in the declarations.

If I'm going to go through the code to do this, what I would really like
to do is something like "typedef" in C, e.g. something like

typedef float my_real

for single precision

typedef double my_real

for double precision

typedef qd_real my_real

for quad-double precision.

All the declarations would use the "my_real" type instead of an explicit
type.

Can this sort of thing be done in Fortran 95/2003?

Thanks,

Chip

You define a type parameter (which is an integer - some of us don't like
that, but that's the way it is). Best to do it in a module, such as

  module precisions
    integer, parameter :: my_real = ....an_appropriate value
  end module precision

There are several options for selecting the appropriate value. Often the
selected_real_kind intrinsic does just what you want. Or if you know
that you want single/double/quad, you can use something like kind(0.0)
or kind(0.0d0) or kind(non_portable_quad_literal). Or you can just
hard-wire the compiler dependent values. Hard-wiring does require
changing the code for different compilers, but it is only a single line
of code for the whole program.

Elsewhere in the program, do things like

  subroutine whatever
    use precisions
    real(my_real) :: x

Note that you do have to say real(my_real) instead of just my_real. Note
also that you can then use this kind parameter for other things such as
literal constants. For example 1.2345678_my_real is a literal constant
of the appropriate kind.