https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116627
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |aoliva at gcc dot gnu.org
--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Ah, the thing is that one of the VALUEs is moved from cselib_hash_table to
cselib_preserved_hash_table.
cselib_find_slot looks for VALUEs first from the latter and only if not found,
in the former, which means that the
FOR_EACH_MODE_UNTIL (narrow_mode_iter, int_mode)
{
PUT_MODE_RAW (copy, narrow_mode_iter);
cselib_val *v = cselib_lookup (copy, narrow_mode_iter, 0, VOIDmode);
if (v)
{
rtx sub = lowpart_subreg (narrow_mode_iter, e->val_rtx,
int_mode);
if (sub)
new_elt_loc_list (v, sub);
}
}
code finds here a preserved VALUE in cselib_preserved_hash_table table and
new_elt_loc_list adds another loc_list to it with a SUBREG of non-preserved
VALUE.
remove_useless_values then discards locs referencing useless VALUEs from the
cselib_hash_table and then actually discards the useless VALUEs, but nothing
discards useless locs from cselib_preserved_hash_table table.
So, I wonder if we shouldn't do:
--- gcc/cselib.cc.jj 2024-04-26 11:46:54.960269768 +0200
+++ gcc/cselib.cc 2024-09-11 10:54:05.018242593 +0200
@@ -751,6 +751,11 @@ remove_useless_values (void)
}
*p = &dummy_val;
+ if (cselib_preserve_constants)
+ cselib_preserved_hash_table->traverse <void *,
+ discard_useless_locs> (NULL);
+ gcc_assert (!values_became_useless);
+
n_useless_values += n_useless_debug_values;
n_debug_values -= n_useless_debug_values;
n_useless_debug_values = 0;
to discard references to useless VALUEs even from the preserved VALUEs.
With the above patch the ICE is gone, but completely untested.
