Re: Target FUNCTION_{PRO,EPI}LOGUE question

["Joseph S. Myers" <jsm28 at cam dot ac dot uk>]

On Thu, 28 Jun 2001, Richard Kenner wrote:

>     But it isn't the widest type - that's HOST_WIDEST_INT.  
> 
> Why was *that* added?  And whoever added it didn't modify the documentation
> in hwint.h saying that HOST_WIDE_INT is the widest host type.

It supports 64-bit preprocessor arithmetic without the inefficiency of
making HOST_WIDE_INT 64-bit on 32-bit systems.  This is slightly broken,
since in general you want (per C99) the *target* intmax_t and uintmax_t,
not the host ones.

Fundamentally, a typedef relating to the HOST should not be used for a
TARGET type.  I agree that we need a multiple precision arithmetic library
for use for target arithmetic.  ISTR that the Fortran 95 front end uses
GMP for its own purposes; that it has been mentioned that the Ada front
end includes a multiple precision arithmetic implementation; and that the
COBOL front end needs 128-bit arithmetic on 32-bit systems.  I tend to
think we should just use GMP - with wrappers for functions on target types
(with and without overflow checking) so we can change the underlying
multiple precision library if necessary - including it in the source tree
and providing a --with-system-gmp configure option - though Torbjorn
Granlund thinks GMP is inappropriate for this
<URL:http://gcc.gnu.org/ml/gcc/2000-11/msg00725.html>.  Then, where
genuinely dealing with the host, types such as long, int and size_t could
be used.

-- 
Joseph S. Myers
jsm28@cam.ac.uk