This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Getting glibfortran to build on newlib targets aka. missingstdint.h (PR 14325)
On Mon, 8 Nov 2004, Zack Weinberg wrote:
> Richard Henderson <rth@redhat.com> writes:
>
> > On Mon, Nov 08, 2004 at 01:40:58PM -0700, Tom Tromey wrote:
> >> How about defining these types using gcc-specific mode attributes?
> >> This is what libgcj does for jni.h.. E.g.:
> >>
> >> typedef int GFC_INTEGER_4 __attribute__ ((__mode__ (__SI__)));
> >>
> >> This is portable and has the advantage of not needing any configury.
> >
> > It is *not* portable. This is a 128-bit type on c4x.
> >
> > Perhaps we should introduce __attribute__((__precision__(BITS))) to
> > solve this problem, but that's a different question.
>
> Or maybe we should change c4x to use SImode for its 32-bit type, and
> not define QI/HImode operations. (Yes, I know this has its own set of
> problems.)
You've expressed this suggested change to c4x before, but I
think it's not doable as memory is fundamentally addressed in
increments of QImode, which *is* 32 bits on c4x. I doubt you
can change the semantics of a memory address in GCC that way.
Anyway, I somewhat hope that this approach, changing
libgfortran.h, is done, because it seems so much simpler and readable than the
configury stuff - not to mention the changes to loop over
[QHSDT]I modes requested by pbrook. ;-)
BTW, there are int8_t and int16_t uses in e.g. io/read.c which
need to be eliminated (unless they're #defined of course).
brgds, H-P