Floating Point Problem ffast-math Complex divide

[Tobias Schlüter]

FX Coudert wrote:
>> The following floating point complex divide doesn't work with -ffast-math.
>>[...]
>> This works correctly with Intel and Pathscale compilers.
> 
> 
> Some optimization options can cause incorrect behaviour ("incorrect" in 
> the sense of "not respecting the IEEE rules"). This is known and 
> documented, as for example (in the gcc manual):
> 
> `-ffast-math'
>       Sets `-fno-math-errno', `-funsafe-math-optimizations',
>       `-fno-trapping-math', `-ffinite-math-only', `-fno-rounding-math',
>       `-fno-signaling-nans' and `fcx-limited-range'.
> 
>       This option causes the preprocessor macro `__FAST_MATH__' to be
>       defined.
> 
>       This option should never be turned on by any `-O' option since it
>       can result in incorrect output for programs which depend on an
>       exact implementation of IEEE or ISO rules/specifications for math
>       functions.
> 
> Please note that the same kind of options enabling agressive 
> optimization (and sometimes leading to incorrect FP behaviour) exist in 
> other compilers too:
> 
> $ ifort a.f90 -O2 -complex-limited-range && ./a.out
>    1.389121338912134E+276 -0.166000000000000      -1.195000000000000E-277
>   (Infinity,NaN) (-0.166000000000000,0.000000000000000E+000)
>   (-1.195000000000000E-277,0.000000000000000E+000)

While you're right in principle, FX, maybe there should be a more fine-grained
switch controlling whether the unsafe division algorithm is used or not, jsut
as there is in ifort.  Ron, can you open a bug report for this, after
double-checking the documentation that there is no such option?

Thanks,
- Tobi