Haha Hihi

__global__ void odd_even_sort_kernel(int* a, int *nPtr)
{
    int n = *(nPtr);
    // comput tid from blockIdx & threadIdx & blockDim
    j = blockIdx.x; // tid is thread id [0...n/2-1]
    for i := 1 to n
    {
        if (i is odd) {
            compare-exchange(a[2j + 1], a[2j + 2]);

        } else {
            // j must be between [1...n/2-1]. So that's why thread-0 sits idle!
            if (j >= 1) {
                compare-exchange(a[2j], a[2j + 1]);
            }

        }

        // all the threads must complete before going to next iteration!
        __syncthreads();
    }
}

main()
{
    //allocate gpu memories

    // copy data(array=a, size=n) from cpu mem to gpu mem


    dim gridDim(1, 1);
    dim blockDim(1024, 1, 1);

    odd_even_sort_kernel<<<gridDim, blockDim>>>(a_gpu, n_ptr_gpu);

    // copy results(array) back

    //deallocate memories

}