soft-fp: Support more precise "invalid" exceptions

[Joseph S. Myers]

On Wed, 8 Oct 2014, Carlos O'Donell wrote:

> I agree that other code in soft-fp is *much* more complicated, but 
> this code is new and you have asked for review. We are not reviewing
> the other code.

I am (well, have done - I've reviewed all the existing soft-fp code 
multiple times).

> As I mentioned before there is a lack of reviewers precisely because
> the code is difficult to review and understand even if you have read
> the IEEE standard and understand the underlying binary formats.

I think the actual lack is more to do with limited interest in this code, 
given that on the most widely used platforms it's not used for standard 
floating-point types.  (In the GCC context, it seems to be the Fortran 
community that's most interested in __float128.)

-- 
Joseph S. Myers
joseph@codesourcery.com