https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67790
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue here is that
/* If we detected "res -= x[i]" earlier, rewrite it into
"res += -x[i]" now. If this turns out to be useless reassoc
will clean it up again. */
if (orig_code == MINUS_EXPR)
{
tree rhs = gimple_assign_rhs2 (def_stmt);
tree negrhs = make_ssa_name (TREE_TYPE (rhs));
gimple *negate_stmt = gimple_build_assign (negrhs, NEGATE_EXPR, rhs);
gimple_stmt_iterator gsi = gsi_for_stmt (def_stmt);
set_vinfo_for_stmt (negate_stmt, new_stmt_vec_info (negate_stmt,
loop_info));
gsi_insert_before (&gsi, negate_stmt, GSI_NEW_STMT);
gimple_assign_set_rhs2 (def_stmt, negrhs);
gimple_assign_set_rhs_code (def_stmt, PLUS_EXPR);
update_stmt (def_stmt);
}
actually inserts stmts into the IL which assigns UIDs to them (via
set_vinfo_for_stmt) which later confuses get_later/earlier_stmt.
The "proper" way of doing the above is by using a pattern which avoids
this issue.
