> With a full 68040 you are using the FPSP support in the kernel which is
> based on code provided by Motorola.  With a 68LC040 you are using the FPE
> (Floating Point Emulator) written by Ken Nakata for MacBSD.  FPE is
> probably still a bit buggy, at least on 68040's and has the main advantage
> of just allowing the kernel to boot at all.

Ok, here I'm summoned :-) Actually, "just allowing the kernel to boot
at all" is not the main advantage of FPE.  It is the main advantage of
compiling the kernel with -msoft_float, so I haven't done anything to
contribute to it :-)

> Once you are up you can use the libm-nomc68881 mentioned above and again
> provided by Ken.  It is based on GNU code and is probably pretty reliable.

This latter sentence is not true.  The libm-nomc68881 is a standard
NetBSD libm which is compiled without any 68881 specific library
routines (which are mostly assembly routines).  So, it's not from FSF
but from Sun.  Since it does not execute any 68881 instructions except
the ones generated by GCC (such as FADD, etc), it has the advantage of
not making FPE choke on some FPU instructions which are unsupported or
buggy under the current implementation of FPE.

> It's also probably a touch faster since it gets called directly instead of
> through interrupt service routines.

Probably yes, but I've never measured the performance gain.

> I believe Ken Nakata is the person currently posting from ken@net5.co.jp.
> If he corrects me on any of these points then believe him, not me ;-)

Well, I'm awfully out of date in terms of -current development, so I
think Scott knows about the current shape of the m68k FPE much better
than I do.

I hope to fix the FPE bug(s) I've created, but at the moment, I don't
have much time to hack... (a usual excuse...)

ken