Re: An sqrt() experiment

["Andrei A. Dergatchev" <A Dergatchev tn utwente nl>]

Hi,

Thanks for pointing to the library.


> Here are the timing results on my 533MHz 21164a (UX2 Ruffian):
>
> y = 30865358776.919712, time = 2.037109 secs. [sqrt]   <-- libm
> y = 30865358776.919712, time = 0.470703 secs. [sqt]
>
> y = 30865358776.919712, time = 0.303711 secs. [sqrt]   <-- libffm
> y = 30865358776.919712, time = 0.475586 secs. [sqt]

Have you used any optimization ? I've got .453125 with -O5(6) on SX.

Any expert out here wants to explain why I've got 0.450195 secs for [sqt]
running under MC?

>  * WARNING: Special cases such as 0, +/-inf, +-NaN are ignored

Btw, is it fair to compare with a code which tests for special cases ?

Rgds,

Andrei