Check if tensor_a is view of tensor_b

1

Was wondering if there is a way to check tensor_a.is_view_of(tensor_b).

parent = torch.arange(4 * 5).view(4, 5)
child1 = parent[[0, 1]]
child2 = parent[:2]
child3 = parent[2:]

child2 and child3 are views of parent because of the slice-style indexing, while child1 is detached. I tried child2.data_ptr() == parent.data_ptr() and this correctly determines that child2 shares memory with parent, but will not work for child3 because of the memory offset (maybe with some pointer arithmetic).

2

Maybe This thread can help you.

BTW, this snippet works for your case:

import torch
def same_storage(x, y):
	x_ptrs = set(e.data_ptr() for e in x.view(-1))
	y_ptrs = set(e.data_ptr() for e in y.view(-1))
	return (x_ptrs <= y_ptrs) or (y_ptrs <= x_ptrs)

same_storage(parent, child1) # False
same_storage(parent, child2) # True
same_storage(parent, child3) # True