[patch, Fortran] Make REAL(KIND=16) detection more robust

[FX]

Hi,

Right now, the logic in libgfortran for the detection of REAL(KIND=16) is in kinds-override.h:

/* What are the C types corresponding to the real(kind=10) and
   real(kind=16) types? We currently rely on the following assumptions:
     -- if real(kind=10) exists, i.e. if HAVE_GFC_REAL_10 is defined,
        then it is necessarily the "long double" type
     -- if real(kind=16) exists, then:
         * if HAVE_GFC_REAL_10, real(kind=16) is "__float128"
        * otherwise, real(kind=16) is "long double"
   To allow to change this in the future, we create the
   GFC_REAL_16_IS_FLOAT128 macro that is used throughout libgfortran.  */


Well, this may not be true of all platforms, and it’s possible to have other combinations. On the aarch64-apple-darwin port, I’m currently playing with enabling a binary128 floating-point mode, and that target has double == long double… so the assumptions above are not true.

Funnily, we already have more fine-grained logic in the mk-kinds-h.sh script, where we actually check the Fortran kind corresponding to C’s long double. We just have to use it, and emit the GFC_REAL_16_IS_FLOAT128 / GFC_REAL_16_IS_LONG_DOUBLE macros there.


Bootstrapped and regtested on x86_64-linux, checked that no symbols were introduced or removed.
(and tested on a port to aarch64-apple-darwin).

OK to commit?

FX

-------------- next part --------------
A non-text attachment was scrubbed...
Name: libgfortran.patch
Type: application/octet-stream
Size: 2645 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/fortran/attachments/20211207/e219a9f7/attachment.obj>