Very Simple Problem

During a preparation of programming contest, its jury is usually faced
with many difficult tasks. One of them is to select a problem simple
enough for most, if not all, contestants to solve. 

The difficulty here lies in diverse meanings of the term "simple"
amongst the jury members. So, the jury uses the following procedure to
reach a consensus: each member assigns each proposed problem a
positive integer "complexity rating" (not necessarily different for
different problems). The jury member calls "simplest" those problems
that he gave the minimum complexity rating, and "hardest" those
problems that he gave the maximum complexity rating. 

The ratings received from all jury members are then compared, and a
problem is declared as "very simple", if it was called a "simplest" problem
by more than half of the jury, and was called a "hardest" problem by
nobody.

Input:

The input will consist of multiple problem instances.  The first
line will contain a single positive integer, giving the number of
instances.

The first line of input file contains integers N and P, the number of
jury members and the number of problems. The following N lines contain
P integers in range from 0 to 1000 each - the complexity ranks. 1 <=
N, P <= 100

Output:

For each input instance, the output should produce a single line of 
output containing an ordered list of the "very simple" problems, 
separated by spaces. If there are no such problems, the output
must contain a single integer 0 (zero).

Sample Input:

3
2 3
1 2 1
4 6 4
4 4
1 1 1 2
5 900 21 40
10 10 9 10
3 4 3 5
2 2
1 10
800 5

Sample Output:

1 3
3
0