The current code yields this error:

```
[rank0]: Traceback (most recent call last):
[rank0]:   File "/home/devuser/pytorch/test_symmem.py", line 98, in <module>
[rank0]:     do_mem_pool()
[rank0]:   File "/home/devuser/pytorch/test_symmem.py", line 81, in do_mem_pool
[rank0]:     reduced = torch.ops.symm_mem.one_shot_all_reduce(x, "sum", dist.group.WORLD.group_name)
[rank0]:               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/devuser/pytorch/torch/_ops.py", line 1350, in __call__
[rank0]:     return self._op(*args, **kwargs)
[rank0]:            ^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: NotImplementedError: "one_shot_all_reduce" not implemented for 'Long'
```

As a side question, is a rendezvous call needed here? After the dtype update the code ran successfully.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/191893
Approved by: https://github.com/soulitzer