https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109141
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2023-03-15
Status|UNCONFIRMED |NEW
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Eh, and we diagnose late because of a bug in pass_waccess::use_after_inval_p.
We have
<bb 4> [local count: 118111600]:
# q_12 = PHI <q_9(8), p_4(D)(7)>
free (p_4(D));
late but early the PHI is in an earlier block. With the above we run into
if (bitmap_set_bit (m_bb_uids_set, inval_bb->index))
/* The first time this basic block is visited assign increasing ids
to consecutive statements in it. Use the ids to determine which
precedes which. This avoids the linear traversal on subsequent
visits to the same block. */
for (auto si = gsi_start_bb (inval_bb); !gsi_end_p (si);
gsi_next_nondebug (&si))
{
gimple *stmt = gsi_stmt (si);
unsigned uid = inc_gimple_stmt_max_uid (m_func);
gimple_set_uid (stmt, uid);
}
return gimple_uid (inval_stmt) < gimple_uid (use_stmt);
but we never set UIDs for PHIs so the answer here is random.
