How would it be if we modified MATHS (and only MATHS) to do an explicit
check for NaNs in the output NDF? This would be pretty simple to do,would
not involve any extra parameters or environment variables, and would not
have too much of an impact on the performance of KAPPA as a whole.
David
On Fri, 16 Jun 2006, Brad Cavanagh wrote:
> I've discovered that KAPPA/MATHS doesn't generate bad values when it
> divides by zero, whereas KAPPA/DIV does. Which is the correct behaviour?
>
> This can be easily tested by creating an NDF of all zeros, then doing
> either:
>
> % div in1=orig in2=zero out=div_by_zero
>
> ...or:
>
> % maths exp='IA/IB' IA=orig IB=zero out=div_by_zero_maths
>
> The div_by_zero NDF will be filled with bad values, whereas the
> div_by_zero_maths NDF will be filled with INFs.
>
> I've updated KAPPA and TRN (which is where it looks like the relevant
> MATHS work is done) and get the behaviour described above.
>
> Brad.
>
|