Subject: Re: MACHINE_ARCH on mips
To: None <tech-toolchain@NetBSD.ORG>
From: Scott Reynolds <scottr@og.org>
List: tech-toolchain
Date: 07/26/1998 00:37:10
while i haven't formed an opinion on this issue, i do want to set the
record straight one one point, and add just a bit to the discussion. 

On Sat, 25 Jul 1998, Todd Vierling wrote:

> It's not a direct superset, but the default output of the m68k compilers is
> MC68020 code, and all m68k kernels can emulate the rest of the unimplemented
> 68020 insns.

particularly with kernel code, this is not true.  while some of the
68020/68851-specific instructions have analogues in the 030/040/060,
others do not.  this happens to work in favor of your argument because the
toolchain is careful to either not produce the instructions that don't run
on the later processors, or knows how to DTRT if it wants to use them. 
either way, though, the kernel itself doesn't emulate any 020/851
instructions.[*]

relying heavily on m68k for precedent is probably unwise, anyway, most
importantly because of the fact that it's strictly big-endian.
extrapolating from this foundation is perilous at best.

--scott

[*] you may notice that i'm ignoring floating point emulation.  that's
because the m68k architectures completely avoid floating point in the
kernel.