In article <4a52ff2bdcdave_daniels@argonet.co.uk> you write:
>In article
><Pine.NEB.4.30.0102251814340.22587-100000@crowley.our-flat.net>,
>   David Forbes <david@flossy.u-net.com> wrote:
>> What was your test program?
>
>It was actually written in Basic and run using my BBC Basic
>interpreter. I recompiled the interpreter after building the new
>kernel to be on the safe side.

Did you compile it with -mhard-float?  Did you recompile libc (and libm)
with -mhard-float as well?

> Rather
>than writing some test code in assembler, I thought that using a
>real program would tell me what I wanted to know, especially as
>this is the sort of context where the floating point hardware
>would be of use to me. Perhaps I misunderstood what was needed, or
>maybe it was never going to work because more work was needed on
>NetBSD itself to enable the use of the hardware.

More work is probably needed.  At the moment, the kernel support is
probably basically OK, but I suspect that libc doesn't react automatically
to changing the mode of the compiler, so things like setjmp and fpsetmask
won't work correctly.  The reason for suggesting using assembler was that
that will test the kernel support independently of the C library.

> On the other
>hand, the time it took to run the program under RISC OS (80
>seconds) compared to the time it took under NetBSD (well over
>1,300) in my view shows that the effort would not be wasted. The
>only snag is that I doubt that many people would benefit from it.

That's never stopped me.

-- 
Ben Harris                                                   <bjh21@netbsd.org>
Portmaster, NetBSD/arm26               <URL:http://www.netbsd.org/Ports/arm26/>