Open wookjeHan opened 3 weeks ago
https://martinlwx.github.io/en/how-to-reprensent-a-tensor-or-ndarray/ contains a detailed discussion. stride is related to memory layout. if shape and type are same, stride can still be different. Sride can be derived from shape if the tensor is contiguous (but not in general).
Dear team,
I was wondering how the
stride
(e.g.,x.stride(0)
) is decided. I thought if the shape and dtype of two tensors are same, the stride would also be the same.However, I noted that in some cases, strides are different even with the same shape and dtype.
What makes the difference of stride in such cases and how the strides can be calculated?
Thanks in advance,