Striding in numpy and pytorch, How force writing to an array or "input tensor and the written-to tensor refer to a single memory location"?

Experimenting with both numpy arrays and pytorch tensors I have came across a difference in an attempt to "do black magic" related to time series that rely on previous time steps

A key feature of the numpy-esque libraries is that they often don't copy data but manipulate it in different views. one of the most interesting tools to do that in numpy is using stride_tricks's functions.

As a simpler problem I was trying to make an "inplace cumsum" but i stumbled upon the fact that numpy doesn't error out and a seemingly equivalent code of pytorch.
I don't really need an in place cumsum implementation but rather inplace broadcasting operation that will use inbroadcast-calculated data.

So my questions are; How can I force pytorch to allow the sum?; How can I make numpy not make a copy of my data and sum over it?

import numpy as np
# array = np.arange(6.)
array = np.arange(6.).reshape(2,3) + 3
# array([[3., 4., 5.],
#       [6., 7., 8.]])
s = np.lib.stride_tricks.as_strided(array, shape=(1,3)) # array([[3., 4., 5.]])
array += s
array
# array([[ 6.,  8., 10.],
#        [ 9., 11., 13.]])

import torch
t = torch.arange(6.).reshape(2,3) + 3
# tensor([[3., 4., 5.],
        # [6., 7., 8.]])
s = t.as_strided(size=(1,3),stride=(1,1)) # tensor([[3., 4., 5.]])
t += s
t
# RuntimeError                              Traceback (most recent call last)
#       3 # tensor([[3., 4., 5.],
#       4         # [6., 7., 8.]])
#       5 s = t.as_strided(size=(1,3),stride=(1,1)) # tensor([[3., 4., 5.]])
# ----> 6 t += s
#       7 t

# RuntimeError: unsupported operation: some elements of the input tensor and the written-to tensor refer to a single memory location. Please clone() the tensor before performing the operation.

Desired:

#sum the first row to itself inplace
#sum the 2nd row with the inplace modified 1st row
array([[6., 8., 10.],
      [12., 15., 18.]])

In [129]: arr=np.arange(3,9).reshape(2,3);
In [130]: s=arr[None,0]  # or the strided

In [131]: np.add.at(arr,((0,1,1),slice(None)),s)

In [132]: arr
Out[132]: 
array([[ 6,  8, 10],
       [12, 15, 18]])

In [133]: s
Out[133]: array([[ 6,  8, 10]])

add.at is most used to handled duplicated indices.

While the numbers match, I'm not sure it's doing your cumsum.