Re: soft-fp: Refactor exception handling for comparisons

["Joseph S. Myers" <joseph at codesourcery dot com>]

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

> > +/* Helper for comparisons.  EX is 0 not to raise exceptions, 1 to
> > +   raise exceptions for signaling NaN operands, 2 to raise exceptions
> > +   for all NaN operands.  */
> > +
> > +#define _FP_CMP_CHECK_NAN(fs, wc, X, Y, ex)		\
> > +  do							\
> > +    {							\
> > +      if (ex)						\
> 
> Does the soft-fp code care about implicit boolean coersion?

Not based on existing handling of rsigned in _FP_TO_INT.  I've made a note 
to review that as a coding style issue (I have a few other residual coding 
style issues in this code to deal with, although lots have been fixed).

> If I had to write this it would have written:
> 
> if ((ex) == 1 
>     && (_FP_ISSIGNAN (fs, wc, X) 
>         || _FP_ISSIGNAN (fs, wc, Y))
>   FP_SET_EXCEPTION (FP_EX_INVALID);
> 
> if ((ex) == 2)
>   FP_SET_EXCEPTION (FP_EX_INVALID);
> 
> Since this is *very* easy to read and understand given the description,
> but I don't know if the compiler generates better or worse code given
> the above.

The code is written to make it obvious that things can be folded for 
constant arguments to the macros, rather than based on what code a 
particular compiler version does or does not generate.  (If it's 
convenient to do things in a way that doesn't change compiler code 
generation at all, so making it obvious that a patch is safe for existing 
users not using the new features, that's a bonus, though it wasn't 
possible for this patch.)

-- 
Joseph S. Myers
joseph@codesourcery.com