On 1/3/2018 3:58 PM, Anton Shterenlikht wrote:
> Tim
>
>> Definitely of interest. I didn't see a statement about whether the
>> results were affected by violating the documented requirements of
>> certain compilers on command line options for USE IEEE_ARITHMETIC.
>>
>> --
>> Tim Prince
>
> Have I violated the documented requirements?
> I'd be grateful for any details.
>
> Thanks
>
> Anton
>
Intel says that ifort requires the option -fp-model precise for full
support of IEEE_ARITHMETIC. That option would be unpopular for
performance. Which IEEE_ARITHMETIC features will work with the option
you chose is not documented.
gfortran may not say even that much; popular optimizations enabled by
-O3 and -ffast-math may be questionable for IEEE_ARITHMETIC, but
defaults should be OK.
--
Tim Prince
|