Re: [MIPS] PR target/22209: Support TImode arithmetic on MIPS

[Roger Sayle <roger at eyesopen dot com>]

Hi Richard,

On Mon, 6 Feb 2006, Richard Sandiford wrote:
> Thanks for looking at this.  As before, I don't have access to any
> 64-bit OS targets at the moment, so I can't really help much, either
> with testing or with conceptual stuff (in the latter case because it's
> hard to get configury right without access to a running system).

I was wondering whether you could test of a 32bit MIPS platform,
just to confirm that I don't break anything on machines that you
do have access to?  Or agree to kick of a build once this patch
is committed, if thats more convenient.


> Still, the MIPS parts of your patch look good to me.  I suppose not
> defining MIN_UNITS_PER_WORD make sense for libgcc2.c, which is always
> compiled with a fixed word size.  But that makes me wonder why we use
> MIN_UNITS_PER_WORD rather than UNITS_PER_WORD at all in libgcc2.c.
> Why doesn't the same logic apply for all targets?

Ulrich Weigand's patch to enable TImode arithmetic on s390 is here:
http://gcc.gnu.org/ml/gcc/2003-03/msg01365.html  I won't pretend to
understand why MIN_UNITS_PER_WORD has the semantics that it does in
libgcc2.c, but rs6000, i386 and sparc, all use this idiom of avoiding
the definition of MIN_UNITS_PER_WORD when IN_LIBGCC2 is defined, in
order to provide their TImode arithmetic.  Interestingly, pa/pa.h
has a FIXME pointing out that it defines MIN_UNITS_PER_WORD to be
four, precisely to disable libgcc providing TImode implementations!?


I tried the alternate approach of defining the a MIPS specific
target_mode_supported_p hook that claimed not to support arithmetic
for TImode, but as predicted in the bugzilla PR, TImode operations
are assumed to be available in order to support TFmode long doubles.


> Since you have access to an affected system, I'm happy for you to use
> your discretion and middle-end approval rights to apply this patch to
> whatever branches you see fit.

Thanks for your confidence!  I'll wait a day or two to see whether a
global maintainer or someone more familiar with libgcc cares to comment,
or whether there are any volunteers to try it on other MIPS backend
configurations.

Roger
--