Re: mips* port plans for the squeeze cycle
On Thu, Aug 20, 2009 at 11:27:31PM +0200, Florian Lohoff wrote:
> On Wed, Aug 19, 2009 at 02:25:13PM +0200, Aurelien Jarno wrote:
> > > Do we have anything to say here? I don't think there are any major
> > > plans with regards to MIPS, are there?
> >
> > * Switch the default ABI to mips2 instead of the current mips1 ABI. The
> > mips1 ABI corresponds to R3000 CPUs that we don't support anymore for
> > a lot of time. This lead us to writing ugly patches using ".set mips
> > 2" when using ll/sc instructions for atomic code in some packages.
> >
> > This should only be only change in GCC that I plan to do unless someone
> > else volunteer.
>
> You mean switching to N32 instead O32? There is no point in switching to mips2
> just for dropping the little patches. If we start changing something we should
> drop mips1 aka r3k/32bit support (And early R4k with 64bit bugs) and switch to
> n32 using a more efficient ABI.
No, staying with o32, but "dropping" mips1 support from the toolchain and
switching to mips2 by default. That's the ABI we are using in practice,
as we have a lot of ".set mips2" in our archive. That's just changing
GCC, nothing more.
The goal with those small patch is actually not to be able to drop
them, but to stop writing new ones and having people complaining about
MIPS because of that.
Switching to N32 is something really interesting, but that requires
creating a new port. Moreover it does not run on MIPS32 CPU, so we
may have to keep the two ports in parallel.
--
Aurelien Jarno GPG: 1024D/F1BCDB73
aurelien@aurel32.net http://www.aurel32.net
Reply to: