Why increasing number of threads/blocks in my Numba CUDA program not showing any improvement?

Question

I have a large complex data structure, comparing and searching integer sequences using suffix array. So, my @jit kernel function is taking three Numpy arrays

1. suffix