Vectorized unix_timestamp is inconsistent with non-vectorized counterpart

create table t (d string);
insert into t values('2020-11-16 22:18:40 UTC');

select
  '>' || d || '<' , unix_timestamp(d), from_unixtime(unix_timestamp(d)), to_date(from_unixtime(unix_timestamp(d)))
from t
;

set hive.fetch.task.conversion=none;

select
  '>' || d || '<' , unix_timestamp(d), from_unixtime(unix_timestamp(d)), to_date(from_unixtime(unix_timestamp(d)))
from t
;

-- std udf:
>2020-11-16 22:18:40 UTC<       1605593920      2020-11-16 22:18:40     2020-11-16
-- vectorized udf
>2020-11-16 22:18:40 UTC<       NULL    NULL    NULL