On Fri, 24 Aug 2001, Matthias Pfaller wrote:

> In article <Pine.LNX.4.33.0108241413080.10077-100000@denkigama.nat.shibuya.blink.co.jp>,
> 	cjs@cynic.net (Curt Sampson) writes:
> > (Summary: storing and reloading FPU state is expensive, and most code,
> > including the kernel, makes little to no use of the FPU. So it's
> > cheaper to do any floating point in the kernel in software than it is
> > to store/reload the FPU state with every entrance to the kernel.)
>
> As far as I remember using -msoftfloat isn't used to force the kernel to
> do floating point in software, but to give you a linktime error if you'd
> accidently use floating point inside the kernel.

	On some CPUs it also keeps the compiler from using floating
	point registers as workspace for speeding up some (non
	float) operations.

-- 
		David/absolute		-- www.netbsd.org: No hype required --