Re: [avr-libc-dev] fmod() accuracy

[Eric Weddington]

Dmitry K. wrote:
> Hi.
>
> The fmod() function is not absolutely accurate.
> For example:
>    fmod (1000001, 1000000) --> 9.536743e-01
>    fmod (1000001, 10)      --> 1.015625e+00
> The reason is too simple algorithm with float point
> arithmetic usage.
>
> fmod() is an important function. It is used in
> trigonometric: sin/cos/tan .
>
> After replacement of algorithm (to integral division)
> the fmod() became absolutely exact. Speed is up
> considarable. Size is increased a little (but the stack
> usage is reduced).
> Testing in progress...

Fantastic!

I'd rather take an accurate function over a slight increase in size any day.

--
Eric Weddington