How to improve performance in the following PL/SQL

Question

I'm running this PL/SQL in a Job step inside a UNIX job, is taking too long to finish and is a simple process to do but definetely there is something that I need to do.
I have to find records that are not included in my T_XREF table from my temporary table T_TEMP, identify those records and insert a new record with values located in the T_TEMP table into the T_XREF table.
T_TEMP has 2 millions records and T_XREF has 12,000 records, I have a Unique IDX in T_TEMP (claim,num_dtl) and a NON UNIQUE IDX by (7 fields).
on the T_XREF table I have 1 UNIQUE IDX (comp_key).
this is part of the PL/SQL:
DECLARE
v_rowid ROWID;
CURSOR c_FixComp1 IS
SELECT DISTINCT sak_acct,ind_split,cde_state,cde_fed,med_a,med_b,ind_cnty
FROM T_TEMP
WHERE claim >= 0;
BEGIN
FOR r_FixComp1 IN c_FixComp1 LOOP
BEGIN
SELECT ROWID
INTO v_rowid
FROM T_XREF x
WHERE x.sak_acct = r_FixComp1.sak_acct
AND x.ind_split = r_FixComp1.ind_split
AND x.cde_state = r_FixComp1.cde_state
AND x.cde_fed = r_FixComp1.cde_fed
AND x.med_a = r_FixComp1.med_a
AND x.med_b = r_FixComp1.med_b
AND x.ind_cnty = r_FixComp1.ind_cnty;
EXCEPTION
WHEN NO_DATA_FOUND THEN
INSERT INTO T_XREF
(comp_key,
sak_acct,
ind_split,
cde_state,
cde_fed,
med_a,
med_b,
ind_cnty)
VALUES
(seq_xref.Nextval,
r_FixComp1.sak_acct,
r_FixComp1.ind_split,
r_FixComp1.cde_state,
r_FixComp1.cde_fedt,
r_FixComp1.med_a,
r_FixComp1.med_b,
r_FixComp1.ind_cnty);
WHEN OTHERS THEN
cnt_notinserted := cnt_notinserted + 1;
END;
END LOOP;
COMMIT;
EXCEPTION
WHEN OTHERS THEN
ROLLBACK;
END;
in my Cursor's query I'm forcing to do a IDX (range scan) and the result of that query is only 1300 rec after doing the DISTINCT of those fields. When starts analyzing record by record this simple script takes more than 10 hrs to run!!!
I'll appreciate any advice you can give me. Thanks

Answer 1

The reason your code is slow is that you are doing something that wasn't even best practice in version 7.0.12. Chris Poole is absolutely correct: There is no reason for any procedural code.
Even if there were a valid reason for PL/SQL the row-by-row cursor fetch construct has been obsolete for 6 years.
Do what Chris is suggesting here and then put away the cursor loops and learn how to write this code:
http://www.psoug.org/reference/array_processing.html
Check out the SLOW_WAY and FAST_WAY demos.