Re: [arm] Implement soft-float abi

[Richard Earnshaw <rearnsha at gcc dot gnu dot org>]

On Tue, 2004-08-24 at 10:58, Wouter van Heyst wrote:

> > Note that I'd like to work towards the point where libcalls always use
> > the same (soft-float) calling convention, regardless of the ABI. 
> > However, that's going to require some work to ensure that we don't break
> > backwards compatibility on Linux (we can do it, but it's not entirely
> > trivial).
> 
> What is involved in not breaking backwards compatibility?

Hand-coding in assembler the small number of functions that are needed
for FPA-based systems and that currently return results in f0 and making
them return results in both f0 and r0/r1.  We can then make the compiler
assume that the results will come back in r0/r1, but legacy code will
continue to recover the value from f0.

I think there's just two such routines: floatdidf and floatdisf.  I've
got a patch for the latter, but not the former yet.

R.