simple minded mapping on darwin?

[Jack Howarth]

Andrew,
    It turns out that the hack to gcc/fortran/f95-lang.c
is insufficient to fix the failures of the large_real_kind_2
and large_real_kind_form_io gfortran tests on Darwin PPC.
I may take a last look at it to see if I can determine the
offending long double call that wasn't getting the suffix
and resulted in those failures.
    I've pulled down revision 108803 of gcc trunk and will
try a build with your original patch tonight just to verify
that it used to work as expected. Looking at the context
changes between r108803 and current gcc trunk, the patch
sections all still seems to be going in the correct places.
However for some reason it seems as if the call to
SUBTARGET_INIT_BUILTINS() in config/rs6000/rs6000.c is
either coming too late now or being ignored. The weird
part is that the sym shown by DECL_ASSEMBLER_NAME() shows
any changes made in the DO_DEFINE_MATH_BUILTIN macro.
I'll try expanding rs6000_darwin_long_double_builtins()
to reread the sym using DECL_ASSEMBLER_NAME() after it
gets set with SET_DECL_ASSEMBLER_NAME() just to make
sure these changes are really being made. Very puzzling.
                Jack