rok commented on PR #34797:
URL: https://github.com/apache/arrow/pull/34797#issuecomment-1499479421
A change introduced late in #8510 (and then removed) was that
`ComputeRowMajorStrides` is replaced with a general `ComputeStrides` to
calculate strides of permuted tensors which should help matching pytorch
behavior:
```python
import torch
def print_strides(shape, permutation):
permutation2 = [p - 1 for p in permutation[1:]]
shape2 = shape[1:]
x = torch.randn(shape).permute(permutation)
y = torch.randn(shape2).permute(permutation2)
strides = [z * 8 for z in x.stride()]
strides2 = [z * 8 for z in y.stride()]
print("[full tensor] shape:", shape, "permutation:", permutation,
"strides:", strides)
print("[cell tensor] shape:", shape2, "permutation:", permutation2,
"strides:", strides2)
shapes = [
(3, 3, 4),
(3, 3, 4),
(3, 4, 3),
(3, 4, 3),
]
permutations = [
(0, 1, 2),
(0, 2, 1),
(0, 1, 2),
(0, 2, 1),
]
for shape, permutation in zip(shapes, permutations):
print_strides(shape, permutation)
```
```
[full tensor] shape: (3, 3, 4) permutation: (0, 1, 2) strides: [96, 32, 8]
[cell tensor] shape: (3, 4) permutation: [0, 1] strides: [32, 8]
[full tensor] shape: (3, 3, 4) permutation: (0, 2, 1) strides: [96, 8, 32]
[cell tensor] shape: (3, 4) permutation: [1, 0] strides: [8, 32]
[full tensor] shape: (3, 4, 3) permutation: (0, 1, 2) strides: [96, 24, 8]
[cell tensor] shape: (4, 3) permutation: [0, 1] strides: [24, 8]
[full tensor] shape: (3, 4, 3) permutation: (0, 2, 1) strides: [96, 8, 24]
[cell tensor] shape: (4, 3) permutation: [1, 0] strides: [8, 24]
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
