unreviewed patch: opt/7409

[Ulrich Weigand]

Richard Henderson wrote:

> Really?  Hmm.  How are global_regs handled?

It looks like they are marked as 'may_not_optimize',
because they are in regs_invalidated_by_call.

This doesn't apply to the case in opt/7409, because
our register 6 is call-saved, but at the same time
used to pass an argument (this is a distinctly weird
aspect of our ABI, but there's nothing now we can do
about it ...).

However, maybe it would be better to leave single_usage alone 
and simply mark all registers in CALL_INSN_FUNCTION_USAGE as
'may_not_optimize'?  Most of them will be marked such anyway
as they are usually call-clobbered.

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  weigand@informatik.uni-erlangen.de