[PATCH]: fix conversion of nearbyint/rint -> lrint

[Richard Guenther]

On 3/5/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> On Mon, 5 Mar 2007, Richard Guenther wrote:
>
> > > Okay for mainline/4.2/4.1?
> >
> > I think this is not correct, as lrint will get extra exceptions compared
> > to using conversion and rint (the invalid exception in case the result
> > doesn't fit the integer type). It looks like now we are only trying to
> > not remove traps and do not care about introducing more, which is wrong
> > as you noted.  I fear the only valid solution is to make both
> > transformations conditional under flag_no_trapping_math.
> >
> > Or am I missing something?
>
> True, plain rint won't generate FE_INVALID.  But we're not converting
> plain rint, we're converting e.g. (long)rint -> lrint.
>
> AFAICT, the cast-to-long will generate FE_INVALID given an argument that
> doesn't fit into a long.  Try it with e.g. DBL_MAX.

Hm, indeed that is what IEC60559 and IEEE754 say.  So the patch is ok
for mainline.
If it is a regression, it is also ok for the branches (like, if we
didn't to the transformation
in some ancient gcc).

Thanks,
Richard.