[cfp-interest 3418] Re: TC1/SC22/WG14/CFP 2025/03/12 Meeting minutes

[Damian McGuckin]

On Thu, 13 Mar 2025, Jerome Coonen wrote:

>     nexttoward exceptions
>         Ly: The issue is raising exceptions for 0 or subnormal results. It's
> not clear that one or both of ERRNO and the fp exceptions must be indicated
> if the result is subnormal or zero. Whether it's raised is
> implementation-defined in the standard. Annex F does not have that latitude.
>        Ly: In Linux, we don't want to spend the execution time to signal the
> exception.. Linux has the newer forms nextup and nextdown but they're not
> generally available in programming environments.

Ly, can you please send me all the clauses that cause the lack of clarity 
including that which (I cannot fnd and) refers to implementation-defined. 
I will note that the range error mentioned in 7.12.12.3 has been ignored 
in more than one of the common LIBMs out there. I will add  that the GNU 
libm shipped with Linux is not one of those.

Ly, is it possible to share the code which has the undue overhead please
or is it the code in FPUtil. Note sure which version of that I am looking 
at.

Thanks - Damian