Re: Intermediate step for new arch for lc040 compatibility

Le sam. 3 mai 2025 à 17:22, Nat Sloss <[email protected]> a écrit :
> So this is really me asking the remaining m68k users if there is interest in
> this approach and eventually (hopefully sooner than later) a new arch m68klc.

While I love the idea of demonstrating a fix for that particular
issue, I don't see much of a practical point at the distribution
level.

m68k CPUs need all the speed they can get, so extra nops everywhere
are a big no-no for systems with FPU and systems with non-buggy F-trap
support.

* Systems with FPU don't need it / don't want it
* Systems without FPU and no bugs ('020, '030, debugged MC68LC040)
don't need it / don't want it
* Systems with the buggy XC68LC040 are the only one where it offers a solution

However, this is only for those *using hard-float binaries*. And
that's where I'm confused. While I do see why one would want to run a
standard distribution on XC68LC040, why hard-float? It makes a lot
more sense to me to just run a soft-float distribution on any FPU-less
system. And it will be faster, as the FP emulation code doesn't need
the overhead of the F-traps.

I must confess I don't even know if there's a soft-float variant of
NetBSD/m68k (all my systems running NetBSD have FPU for now, though
I'm considering a home-made one with a FPU-less MC68040V), but for me
it would make a lot more sense to use that (and add it if it's
missing) rather than a XC68LC040-specific variant. Sure a bunch of Mac
LC475 shipped with those, but ultimately it's probably easier and more
convenient to either upgrade them to full '040 or bug-free MC68LC040,
or to run soft-float on them.

Cordially,

-- 
Romain Dolbeau