My GPU coder app does not work well

I'm trying to accelerate my code with GPU coder app.

However, when I change my code into MEX file, It takes 136s on Running test file with MEX procedure because of one of an array which I want to have as output.

This array is a triangular matrix which its elements are in a range of 1~1000.

However, when I change this array into a triangular matrix which its elements are only 1, GPU coder works well and it takes only 0.14s on Running test file with MEX.

I wonder why this problem happened and what is the difference between an array which has only 1 as its elements and an array which has 1~1000 as its elements.

Also, another matrix has also various values, and it has no problem(matrix D in my code), and I wonder why this matrix makes any problems.

I attach all my codes. The array which has the problem in my code is N.

I hope someone can help my problem. I hold this problem for a month.. and I solve my problem quickly..