REG_UNUSED notes problem

[Joern Rennecke]

> Since there is no REG_UNUSED note attached to insn 47, the m68k backend
> will generate code which sets %d0.  And we lose since (reg 0) is actually
> expected to hold part of an address (remember, we had a REG_UNUSED note for
> the original pseudo in the divmod insn, so the compiler can and does assume
> that nothing is written to that unused value).

Huh?  That is wrong.  The compiler has to assume that the register is
clobbered.
So it is a reload bug that reg 0 was used for the reload.
I guess that the problem stems from single_set returning the set of reg
42, and the caller then checking only for clobbers, but not for unused
sets.