Fix: #115792
This PR implements 2 virtual functions of `TensorImpl` that are called when setting the
`tensor.data`:
- `shallow_copy_from`: which calls `copy_tensor_metadata`; and
- `copy_tensor_metadata`: which copies all `FunctionalTensorWrapper` metadata and ~calls
`dest->value_.set_data(src->value_)`~ assigns `dest->value_ = src->value_`, so as to copy also the inner tensor using the same
method
Before this PR, the inner tensor of a `FunctionalTensorWrapper` was being ignored.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/118783
Approved by: https://github.com/bdhirsh
In certain edge cases when using lazy tensors, the base tensor stored in the `FunctionalStorageImpl` and the `value_` tensor stored in the `FunctionalTensorWrapper` diverge. For instance, take this simple example
```python
class Model(torch.nn.Module):
def __init__(self):
super().__init__()
self.fc1 = torch.nn.Linear(4, 2, bias=False)
def forward(self, x):
return x @ self.fc1.weight.transpose(0, 1)
with torch.device("lazy"):
model = Model()
x = torch.ones(4)
out = model(x)
```
The call to `transpose` on the lazily initialized weight `fc1.weight` applies a view op on the functional tensor which only gets propagated to the functional tensor wrapper and not the base tensor in the storage. Thus, causing them to diverge.
To fix this behaviour, we need to reset the functional tensor's storage. To facilitate this, we add a `reset_storage` method to `FunctionalTensorWrapper` which clears away the old storage and view metas.
CC: @behzad-a @GlebKazantaev @wconstab @bdhirsh
Pull Request resolved: https://github.com/pytorch/pytorch/pull/115235
Approved by: https://github.com/bdhirsh
