This is the mail archive of the gcc-patches@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: MIPS64 soft-float: missing definition for __floatdidf and friends


Roger Sayle <roger@eyesopen.com> writes:
> Pragmatically, the more immediately pressing issue is what to do about
> the MIPS64 -msoft-float regressions during the 4.2 timeframe, i.e.
> during stage3.
>
> My personal preference of only affecting the MIPS64 targets whilst
> continuing the evolution away from a monolithic libgcc2.c seems like
> a reasonable "software engineering" compromise.  However, the other major
> possibilities are to either just not fix this problem for 4.2, or go for
> the more significant reorganization now (probably using macros akin
> to GLIBC's soft-fp as you suggest above).
>
> I don't have maintainership over libgcc, so it's not my call.  But
> there's currently only one patch/suggestion on the table which limits
> the options.

In case you're wondering, I haven't proposed by own patch because
Adam said:

Adam Nemet <anemet@caviumnetworks.com> writes:
> Richard Sandiford writes:
>> Perhaps I'm going over old ground here, but do we really want to
>> cut-&-paste bits of libgcc2.c like this?  The templates there are
>> fine (and indeed Adam's versions seem to be copied from there, with
>> appropriate macro substitutions).
>
> Yes but this seems to be the approach Roger had taken in his original
> patch which you have positively commented on.  But anyway the problem
> of duplicate DI conversion routines for MIPS32 seems to a be a real
> one and it is probably easier to deal with that in libgcc2.c where the
> definition of MIN_UNITS_PER_WORD is available.
>
> Let me see if I can come up with a patch and then we'll decide if that
> is safe for the 4.1 release branch.

...which gave me the impression there was another patch pending.
That's also why I've been quiet about later comments on the thread.

As Adam says, adding these routines to libgcc2.c for mixed 32-bit/64-bit
combinations might lead to duplicate definitions.  Note that both
mips-sgi-irix6.5 and mips64-linux-gnu fall into this category.
The patch can't go in for that reason alone, ignoring the pragmaticism
or otherwise of duplicating code.

If in fact there are no WIP patches, I'll look at it.  This is clearly
the a show-stopping bug for MIPS targets.  Note that if there's no simple,
non-invasive solution, I think we'll have to revert your patch on 4.1.

Richard


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