https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116374
Michael Matz <matz at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |matz at gcc dot gnu.org
--- Comment #1 from Michael Matz <matz at gcc dot gnu.org> ---
Try the below. m68k has the property that sometimes the initial elimination
offset between %sp and argptr is zero. But zero is also the initialization
value of the elimation table offsets, and it's _changes_ in those offsets that
trigger various state tracking. And if the initial "change" is ignored
everything goes downhill from there: in the first_p lra_eliminate() pass
insns_with_changed_offsets will remain empty because "nothing changed", and
hence the initial setup via process_insn_for_elimination()
(with first_p==true) is never carried out.
I guess also -1 is not a very good value (though probably acceptable for all
targets with stack-downwards or wordsize!=1) and should rather be replaced with
some proper "this is the first init" flag. But ... well. Hack.
diff --git a/gcc/lra-eliminations.cc b/gcc/lra-eliminations.cc
index 5bed259cffe..86741aa5a92 100644
--- a/gcc/lra-eliminations.cc
+++ b/gcc/lra-eliminations.cc
@@ -1283,7 +1283,7 @@ init_elim_table (void)
for (ep = reg_eliminate, ep1 = reg_eliminate_1;
ep < ®_eliminate[NUM_ELIMINABLE_REGS]; ep++, ep1++)
{
- ep->offset = ep->previous_offset = 0;
+ ep->offset = ep->previous_offset = -1;
ep->from = ep1->from;
ep->to = ep1->to;
value_p = (targetm.can_eliminate (ep->from, ep->to)
