SQL & PL/SQL :: Assigning ID (Fill SID Column)

I have a table with 200k entries, which has ~12k unique combination of columns that I wish to assign an ID to. My real world problem is with addresses and street components (of which nulls are allowed); but this should be a reasonable example.

DROP TABLE tt;

CREATE TABLE tt AS
SELECT ROWNUM id,
CASE
WHEN MOD(ROWNUM, 107) = 0 THEN NULL
ELSE MOD(ROWNUM, 101)
END
c1,
CASE

[Code]...

Now I have 3 ways to assigned an ID (fill the SID column).

The first is congruent with the method most commonly used, but seems especially slow in this case. I think this is due to the COALESCE/TO_CHAR usage. Can this be optimized?

DECLARE
CURSOR c1 IS
SELECT a.*, sid_seq.NEXTVAL sid
FROM (SELECT DISTINCT c1, c2 FROM tt) a;
BEGIN
FOR s IN c1 LOOP
UPDATE tt d

[Code]...

The second I've made relies on the usage of a temp table, which I really don't like, but it is faster; so that is something at least.

CREATE TABLE ttt AS
SELECT a.*, sid_seq.NEXTVAL sid
FROM (SELECT DISTINCT c1, c2 FROM tt) a;

[Code]...

The third discards the sequence, and uses dense_rank. This allows us to use the ID (which might have an index, whereas c1 and c2 will not).

MERGE INTO tt d
USING (SELECT id, (SELECT COALESCE(MAX(sid), 317) FROM tt) + DENSE_RANK() OVER (ORDER BY c1, c2) sid FROM tt) s
ON (d.id = s.id)
WHEN MATCHED THEN
UPDATE SET d.sid = s.sid

The ideal solution would use the SEQUENCE, not use a temporary table and match/merge/update based on ID, and be 'fast'; but I cannot conceive of such a solution; does one exist? What would you do?