[patch] rs6000: default for -D_SOFT_FLOAT

[DJ Delorie]

> There's something I don't understand about this.  It's not introduced
> by your patch, but you preserve it.  This makes --without-fp change
> -D_SOFT_FLOAT - but nothing else!
> 
> Shouldn't it also set the mask so that we actually don't generate
> floating point?  In other words, act like -msoft-float?

Perhaps.  I supported it only because the original bug report lists
"-nfp" as one of the options used, and that just says "obsolete, use
--without-fp" so I supported that.

I would be more willing to remove that in the patch, and simplify it
to just use --with-cpu, but then there's no way to get bugwards
compatibility.