This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: TFmode support.


On Sun, Sep 08, 2002 at 05:21:02PM +0200, Toon Moene wrote:
> I have been asked to find out whether (and under which circumstances) 
> GCC supports TFmode computations.

Right now, the answer is

  LONG_DOUBLE_TYPE_SIZE == 128 && !INTEL_EXTENDED_IEEE_FORMAT

This includes sparc64, pa64, ia64-hpux (only), alpha-osf5.

There is room for improvement here, since the definitions we have
are only valid for C.  For instance, I understand that some versions
of alpha-osf4 do have the TFmode arithmetic routines, but are not
used for C.

So what you _actually_ want to see if there is any proper TFmode
support is to test !INTEL_EXTENDED_IEEE_FORMAT, then see if you've
got add_optab->handlers[TFmode] etc.

Further in the future, it would be nice if the INTEL_EXTENDED_IEEE_FORMAT
128-bit type were distinguished from the true IEEE quad format.
We could do this possibly by defining an XXFmode, and chose between
XXFmode and TFmode in mode_for_size based on ... something.

It would be possible for us to make sure that there is a version
of the TFmode routines in libgcc.  I believe the current fp-bit.c
would have to be extended, at which point we should probably change
to use Torbjorn's ieeelib.c (posted to gcc-patches some years ago).


r~


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]