// Old code
         unsigned int nodes_start = usm_nodes_start[idx];
        unsigned int nodes_end = usm_nodes_start[idx +1 ];
        // New code with split device pointers
        device_ptr<unsigned int> DevicePtr_start(usm_nodes_start+offset);  
        device_ptr<unsigned int> DevicePtr_end(usm_nodes_start + 1+offset);
        unsigned int nodes_start = DevicePtr_start[idx];
        unsigned int nodes_end = DevicePtr_end[idx];

// Vector code
std::vector<event> e_vec;
e_vec.push_back(e1);
e_vec.push_back(e2);
e_vec.push_back(e3);
auto e =q.single_task<class Task>( e_vec, [=](){
...
// Seperate Events
 auto e =q.single_task<class Task>( {e1,e2,e3}, [=]() {
 ...

  q.single_task<class SingleTask>(  [=]() [[intel::kernel_args_restrict]] {
           #pragma unroll 16
           for(int tid =0; tid < no_of_nodes; tid++){
             unsigned int condition = usm_updating_mask[tid];
             if(condition){
               usm_updating_mask[tid]=0;  
            }
           }
         });
  q.parallel_for<class ParallelFor>(no_of_nodes, [=]() [[intel::kernel_args_restrict]] {
             unsigned int condition = usm_updating_mask[tid];
             if(condition){
               usm_updating_mask[tid]=0;  
            }
         });
  int BLOCK_SIZE = 512;
  int global_work_size = (no_of_nodes + BLOCK_SIZE - 1) / BLOCK_SIZE * BLOCK_SIZE;
  range<1> gws (global_work_size);
  range<1> lws (BLOCK_SIZE);
  q.parallel_for<class NDRange>(nd_range<1>(gws, lws), [=] (nd_item<1> item) [[intel::kernel_args_restrict]]  {
              int gid = item.get_global_id();
              if(gid<no_of_nodes){
             unsigned int condition = usm_updating_mask[tid];
             if(condition){
               usm_updating_mask[tid]=0;  
            }
           }
         });