This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Improve PR28796 (inconsistend __builtin_nan() and __builtin_unordered()) further


On Sun, 3 Sep 2006, Toon Moene wrote:

> Richard Guenther wrote:
> 
> > but at that point we didn't convert all of the remaining "-ffast-math"
> > (which just became flag_unsafe_math_optimizations check) to either
> > flag_finite_math_only or flag_unsafe_math_optimizations.
> > 
> > So I don't think we can say that all these bugs are regressions.  They
> > are technically still wrong-code, so maybe applicable for fixing in
> > mainline and branches once they are open for bugfixes again.
> 
> [Sorry for the late reply - catching up after holidays]
> 
> I meant flag_finite_math to be a sub-flag of -ffast-math.  Most people (at
> least in the hard sciences) know whether their models behave "correctly"
> without generating NaNs or Infinities.  That's what flag_finite_math is
> supposed to govern: The person compiling the code knows that his/her code
> won't generate NaNs or Infinities, unless there's an (unrelated) bug in the
> code.

There is no flag for -ffast-math.  -ffast-math activates a bunch of
flags, some of which have very strict definitions, one for all the rest
(-funsafe-math-optimizations).  Apart from this - what is your comment
suggesting?  That -funsafe-math-optimizations is supposed to include
-ffinite-math-only?  I agree that -funsafe-math-optimizations is a bad
name and should be dropped in favor of -fassociative-math -frecip-math.

Richard.

--
Richard Guenther <rguenther@suse.de>
Novell / SUSE Labs


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]