Re: Floating point computations in kernel modules

["Markus Rechberger" <mrechberger@gmail.com>]

Hi,

On 9/17/07, Ram <vshrirama@gmail.com> wrote:
> Hi,
>
>    I am trying to write a driver which uses double, float. I am using
> an arm11 with gcc 3.4.4
>
>    When i try to compile my modules (with float variables) i get the error
>
>    WARNING: "__extendsfdf2" undefined!
>    WARNING: "__mulsf3"undefined!
>    WARNING: "__fixsfsi"undefined!
>    WARNING: "__floatsisf"undefined!
>
>    Can we use float, double in kernel modules?.
>

in general you shouldn't use floating point in the kernel since there's no
portable way which will work among different architectures.

>   Obviously it is a linking error, I am using soft-float. I tried with
> -msoft-float but got the same   result.
>
>   I suspect that its a toolchain issue.  However i am able to compile
> my applications
>   with float, double variables.
>
>
>   Please do advice.
>

if you need a solution inkernel best would be to try to get around by converting
your algorithm to integer.

although you might have a look at:
http://www.fi.muni.cz/~xslaby/unpr.html
http://www.mail-archive.com/linuxppc-embedded%40ozlabs.org/msg01291.html

Markus