What is the use of tensor.share_memory_()?

1

I have two codes

import torch
import multiprocessing

def worker(tensor):
    tensor += 1
    print(f"Worker tensor: {tensor}")

if __name__ == '__main__':
    tensor = torch.zeros(3)
    p = multiprocessing.Process(target=worker, args=(tensor,))
    p.start()
    p.join()
    print(f"Main tensor: {tensor}")

import torch
import multiprocessing

def worker(tensor):
    tensor += 1
    print(f"Worker tensor: {tensor}")

if __name__ == '__main__':
    tensor = torch.zeros(3)
    tensor.share_memory_()
    p = multiprocessing.Process(target=worker, args=(tensor,))
    p.start()
    p.join()
    print(f"Main tensor: {tensor}")

Only difference between two is the use of tensor.share_memory_(). My output is same in both cases

Worker tensor: tensor([1., 1., 1.])
Main tensor: tensor([1., 1., 1.])

Can anyone explain if tensors are shared directly, why do we even need shared memory?