PR 37159

[Tobias Burnus]

Hi Dennis,

Dennis Wassel wrote:
> My troubles are with HAVE_GFC_REAL_16 and {
> dg-require-effective-target fortran_large_real }.
>   
There are machines (esp. going towards embeeded hardware) where real(8) 
is the largest floating point number. For others fortran_large_real is 
defined; on x86/x86-64 the largest real number is REAL(10) while e.g. on 
PowerPC the largest number is REAL(16) [and REAL(10) is unsupported].


They are supposed to be used as
   integer, parameter :: large_real_kind = 
selected_real_kind(precision(0.0_8)+1)
That of cause will not work for for your specific case, where you need 
to destinguish between having REAL_16 or not. I think you have to live 
with testing a too short seed (all platforms) without having a special 
version for HAVE_GFC_REAL_16 systems as test case.

Tobias