Re: as.exe 2.14 -mips3 with -n32 option gives Invalid bfd target

[Thiemo Seufer <ica2_ts at csv dot ica dot uni-stuttgart dot de>]

Ken Faiczak wrote:
> 
> > > thanks,
> > > i changed bfd/config.bfd and got the correct result
> > > 
> > > 
> > > and I think this is right since for vxworks/rtem/elf it 
> > should support
> > > all ABI since you have no idea how the os may be compiled 
> > (it is shipped
> > > with as one typically may be rebuilt by some customers) so it should
> > > support all ABI
> > 
> > This can't be the full truth since the OS will need some more support
> > for n32, e.g. NewABI calling conventions and 64 bit register width.
> > It's not just a matter of compilation. OTOH, it won't do much harm to
> > include the n32 stuff.
> > 
> 
> We have been using abi=o64 anyway so the registers have been 64 bit for a
> while and a few specific changes were made to support n32 so now it 
> supports both o64/n32

Ok, then mips*-*-vxworks* should support the n32 BFD vector, and
the generic mips*-*-elf* as well.

I'm not sure about mips*-*-rtems* and mips*-*-none.

> in order to test out whether some of
> the n32 ABI changes would benefit us in terms of performance (more args,
> less stack)
> It currently works fine under gcc 3.2.3/ binutils 2.13.2.1
> though showed no marked performance difference.

Floating point performance should profit most, as well as some
(e.g. crypto-) algorithms which use 'long long' heavily.


Thiemo