[L0 v2] initial implementation of queueFinish and enqueueKernelLaunch

[igchor]

No description provided.

[pbalcer]

this seems overcomplicated.

Let's do something like:

class ur_wait_list {
std::pair<ze_event_handle_t *, size_t> getArray(ur_event_handle_t  *urHandles, size_t size, ze_event_handle_t extra) {
  size_t real_size = size + (extra ? 1 : 0);
  handles.reserve(real_size);
  for (size_t i = 0; i < size; ++i) {
     handles[i] = urHandles[i]->zeHandle;
  }
  
  if (extra) 
    handles[size] = extra;

  return (handles.data(), real_size );
}

private:
  std::vector<ze_event_handle_t *> handles; //preallocated, never shrinks
}

This would then be stored in the queue and reused for every enqueue.

[igchor]

Good idea. Done.

[igchor]

I will close this and open a new PR rebased on top of #1893 once it's merged