Dmitry wrote:
>
> Fellows,
> if somebody uses float point, could you check that everything works Ok.
> I am just trying to find rounding errors which make result of some operations
> slightly different then it should be.
>
> For example sin(2) with gcc's FP lib is about 0.878928721
> whereas with libfp it is 0.909314692 which is more precise.
>
> So, please mail me if you have some unusual gccs tricks there.
That's about a 3 degree error, making it more than slightly different.
Can you point me closer to the suspect code?
Does cos(2) give a similar error?
What do you get for e^-.1?
Do you have Stegun & Abramowitz handy?
If not, I will send some of their approximations.
Cheers,
Garst
