Re: -O generates NaN

[Phil Edwards <pedwards at disaster dot jaj dot com>]

On Sat, Mar 03, 2001 at 03:01:58PM -0300, Alexandre Oliva wrote:
> On Mar  2, 2001, Haruhiko Okumura <okumura@matsusaka-u.ac.jp> wrote:
> 
> > Could someone tell why this code generates lots of NaN's when compiled
> > with any optimization flag (-O, -O2 etc) on?
> 
> Probably some of the math optimizations in GNU libc.  Not libstdc++.

I concur.  On Solaris 8 -- therefore not using GNU libc -- I always see
the same results:

     2: 85.9715
     6: 85.9715
     18: 85.9715
     2: 86.5915
     6: 86.5915
     18: 86.5915
    len1 = 85.9715, len2 = 86.5915

That is the output with no optimization, with -O, and with -O3.


Phil

-- 
pedwards at disaster dot jaj dot com  |  pme at sources dot redhat dot com
devphil at several other less interesting addresses in various dot domains
The gods do not protect fools.  Fools are protected by more capable fools.