Ian Lance Taylor <[email protected]> writes:
> Denis Chertykov <[EMAIL PROTECTED]> writes:
>
> > > > I think that better to call update_eliminables() somewhere after
> > > > setup_save_areas()
> > >
> > > Exactly. We do that. About 15 lines after the lines you quoted
> > > above.
> > >
> > > What am I missing?
> >
> > I'm (exactly AVR port) need in call to update_eliminables() somewhere
> > between setup_save_areas() and calculate_needs_all_insns()
> > (Not "about 15 lines after" ;) because all bad things happened inside
> > calculate_needs_all_insns().
> >
> > calculate_needs_all_insns() collect wrong reloads because
> > reg_eliminate structure for FP->SP have wrong can_eliminate field.
>
> But then we go around the loop again, so everything should get
> recalculated based on the new assumptions. Doesn't that happen for
> you?
If you mean the "continue" here:
if (caller_save_needed)
setup_save_areas ();
/* If we allocated another stack slot, redo elimination bookkeeping. */
if (starting_frame_size != get_frame_size ())
continue;
-------^^^^^^^^^
then answer no because only set_initial_elim_offsets() will be called.
set_initial_elim_offsets() doesn't change reg_eliminate[].can_eliminate
flag.
If you mean a call update_eliminables() after
calculate_needs_all_insns() then answer no (again :) because all wrong
things already happened inside calculate_needs_all_insns() and
compiler will call
select_reload_regs() -> find_reload_regs() -> spill_failure()
Denis.
