PATCH COMMITTED: More lower-subreg.c patches

[Ian Lance Taylor]

DJ Delorie <dj@redhat.com> writes:

> > Sure.  This still seems to me like a case of lying to the compiler.
> > We have to expect problems in this area, especially as the compiler
> > continues to get more aggressive.
> 
> I don't *want* to lie to the compiler.  I'm just finding it more and
> more difficult to determine which truth the compiler wants at any
> given time, and how to provide that truth.

I agree that it can be very ambiguous.  Still, this particular case
has been a rule for a long time: if the insn predicate accepts the
operand, there must be a constraint which matches the operand, or one
for which the operand can be reloaded.

Ian