[altq 1118] Re: Floating point operations

[Kenjiro Cho <kjc@csl.sony.co.jp>]

Nicolas Christin wrote:
> > You are not allowed to use floating-point operations in the kernel.
> > It is a design decision; the kernel code can be written without
> > floating-point and the overhead isn't justified.
> 
> 	That's what I thought, but I wanted to be sure. Is that true for
> all *BSD OS's?

There are 2 different flavors to support floating-points in the
kernel.
 (1) FPU support: the overhead of storing/restoring the FPU state
     is prohibitive.  I believe no OS supports it.
 (2) compiling the kernel without FPU support (-msoft-float for gcc)
     and linking the FPU emulation library.
     This should work for any OS.
Still, it's the best practice not to use floating-points inside the
kernel.

> > So, you have to use other methods (e.g., fixed-point).
> 
> 	Yes, that's exactly what I did for my queuing scheme. As an aside,
> for everyone interested, I think the HFSC code in ALTQ is quite
> educational in that respect. HFSC is a very theoretical scheduler, yet,
> the code processes everything with u_int64_t's... (and scaling).
> Pretty nice.

Thanks.  I agree that the HFSC code is the best reference for
translating theory into implementation.

-Kenjiro