The "I-Love-Kd-trees'' annual con is receiving too many applicants so they decided to complicate a bit the task used to select participants. (They realized some people were using other data structures to solve their problems, so they designed this problem, almost only solvable with Kd-trees).

You are given a list of N numbers and Q queries.

Each query can be of two types.

Type-0 queries (marked with 0 in the input), consist of three integers: i, l and k ; let d be the k-th smallest element until the index i (i.e. if the first i +1 elements were sorted in non-descending way, d would be the element at index k - 1 ). Then, the answer to each query is the index of the l-th occurrence of d in the array. If there's no such index, the answer is -1.

Type-1 queries (marked with 1 in the input) are  contiguous-swap update-queries, and consists of a single integer i. When a type-1 query is executed the elements at index i and i +1 in the list must be swapped.

You have to consider that all indexes are counted starting with 0.

Input

Input consists of one test case.

The first line contains two integers, N ( 1 ≤  N  ≤  10⁶) and Q (1 ≤ Q  ≤ 10⁵).

The next line contains N possible distinct integers a_i ( -10⁹  ≤ ai ≤ 10⁹).

Then Q lines follow. Each of them starts with an integer which can be 0 or 1, denoting the type of the query. If it’s 0, then three integers i, l and k follow (0 ≤ i < N, 1 ≤ k  ≤  i+1, 1 ≤ l ≤ N).

If it’s 1, then an integer i follows, meaning that you have to swap the elements at indexes i and i+1 (0 ≤ i ≤  N-1).

Output

For each query of type-0 (in the same order as the input) output a single line with the answer to that query.

Example

Input:
10 6
2 3 1 1 2 7 9 1 2 6
0 2 3 2
1 1 
1 2 
0 2 3 2 
1 0 
0 0 2 1
Output:
8
7
2