To: J3                                                     J3/18-257
From: Dan Nagle
Subject: log<n> and friends
Date: 2018-October-01


I Introduction

At 215, J3 agreed, and at 216 WG5 concurred, to add the following
to the iso_fortran_env intrinsic module.

log8 - an integer constant whose value is the kind value
       of an 8-bit logical

log16 - an integer constant whose value is the kind value
       of a 16-bit logical

log32 - an integer constant whose value is the kind value
        of a 32-bit logical

log64 - an integer constant whose value is the kind value
        of an 64-bit logical

real16 - an integer constant whose value is the kind value
         of a 16-bit real


II Use-cases

At times, one wants the smallest, or the fastest, logical.
Presently, there is no way to get these kind values, other
than trying to decode the logical_kinds array.  However,
this array is unsorted, and so it is useless for the purpose.

Newer hardware now supports a 16-bit real (for use in AI work).
This is missing from the current real32, real64, real128 set.


III What I have in mind

Add log[8-64] and real16 to iso_fortran_env.

Straw Vote the (presumably negative) value if none is available:

If no real16 is available, real16 ==
1.  -6 (the next available negative number) 'no such real that small' ?
2.  any other negative number but not (-1, -2, -3, -4, -5)
3.  undecided



IV (Rough) Requirements

iso_fortran_env is expanded to include:

log8 - a default integer constant whose value is the kind value
       of an 8-bit logical, or -1 if no such kind is available

log16 - a default integer constant whose value is the kind value
        of an 16-bit logical, or -1 if no such kind is available

log32 - a default integer constant whose value is the kind value
        of an 32-bit logical, or -1 if no such kind is available

log64 - a default integer constant whose value is the kind value
        of an 64-bit logical, or -1 if no such kind is available

real16 - a default integer constant whose value is the kind value
         of a 16-bit real kind, or a negative (-6?) value if no such
         kind is available.