Adapting the GA Approach to Solve Traveling Salesman Problems on CUDA Architecture