This PR:
- disallows FakeTensor.data_ptr when it is called inside PT2 or fx tracing.
- disallows FunctionalTensor.data_ptr (python FunctionalTensor is only used in
PT2)
The motivation behind this is that the leading cause of segfaults when
using custom ops with PT2 is calling .data_ptr on FunctionalTensor or
FakeTensor.
This change is BC-breaking. If your code broke as a result of this, it's
because there was a bug in it (these .data_ptr should never be
accessed!). You can either fix the bug (recommended) or get the previous
behavior back with:
```
from torch._subclasses.fake_tensor import FakeTensor
from torch._subclasses.functional_tensor import FunctionalTensor
data_ptr = 0 if isinstance(tensor, (FakeTensor, FunctionalTensor)) else tensor.data_ptr()
```
Test Plan:
- existing tests
Differential Revision: [D55366199](https://our.internmc.facebook.com/intern/diff/D55366199)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/122514
Approved by: https://github.com/ezyang, https://github.com/albanD, https://github.com/yifuwang, https://github.com/kurtamohler
Make `torch.__future__.get_swap_module_params_on_conversion() == True` account for `assign` argument to `nn.Module.load_state_dict`
Similar to when `torch.__future__.set_swap_module_params_on_conversion()` is `False`, `assign=True` means that we do not incur a `self.copy_(other)` and the properties of `other` will be preserved
Pull Request resolved: https://github.com/pytorch/pytorch/pull/121158
Approved by: https://github.com/albanD
ghstack dependencies: #121157
Added a `torch.Tensor` method that defines how to transform `other`, a value in the state dictionary, to be loaded into `self`, a param/buffer in an `nn.Module` before swapping via `torch.utils.swap_tensors`
* `param.module_load(sd[key])`
This method can be overridden using `__torch_function__`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/117913
Approved by: https://github.com/albanD
By finally breaking FC promise on new dtypes by serializing untyped
storage and tensor dtypes
- Add `_rebuild_tensor_v3` that takes an extra dtype argument
- In `Tensor.__reduce_ex__` serialize tensor using untyped storage for
v3_dtypes (which are at the moment limited to float8 dtypes)
Test plan: `python -c "import torch;x=torch.arange(10).to(dtype=torch.float8_e4m3fn);torch.save(x, 'pt.pt');print(torch.load('pt.pt'))"`
Fixes https://github.com/pytorch/pytorch/issues/114634
Pull Request resolved: https://github.com/pytorch/pytorch/pull/114662
Approved by: https://github.com/ngimel
Notes:
* `debug_insert_nops` in testing.py was passing `None` to the compiler_fn
parameter of `OutputGraph`, hence the modifications there.
* I added `disable-error-code="method-assign"` to debug_utils.py as it
does several such assignments. I guess mypy doesn't like it because it
makes code near-impossible to safely typecheck.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/113519
Approved by: https://github.com/Skylion007
ghstack dependencies: #113413, #113518
I am trying to turn on `follow_imports=silent` for MYPYNOFOLLOW.
However, this requires a huge number of changes, so I am breaking it
down to a per-file basis.
Unfortunately, we will not be able to turn on `follow_imports` until all
files are fixed, so there is no way to stop regressions. So I hope to get
these fixes in as fast as possible.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/113304
Approved by: https://github.com/Skylion007
Summary:
This is a stride based attribute for a tensor available in Python.
This can help inspect tensors generated using `torch.empty_permuted(.., physical_layout, ...)`, where physical_layout should match the dim_order returned here. `empty_permuted` will be renamed to use dim_order as the param name in the future. And also help Executorch export pipeline with implementing dim_order based tensors.
Differential Revision: D48134476
Pull Request resolved: https://github.com/pytorch/pytorch/pull/106835
Approved by: https://github.com/ezyang
It seems that some legacy default stream logic (e.g., present in a8ff647e42/torch/utils/dlpack.py (L114) ) is not handled on the potential receiving end in `torch/_tensor.py`.
Open to suggestions on how to make the test case less clunky, as this was the combination we arrived at after discovering flakiness in alternate versions.
Thanks to Olga Andreeva for surfacing this issue and providing a repro.
CC @Aidyn-A @ngimel
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101318
Approved by: https://github.com/ngimel
Fixes#72428 according to decision reached in comments.
I've left other instances of `w.r.t.` in tact (e.g. in parameter/return descriptions, in comments, etc) because there were many, and I didn't' want to go out-of-scope. That being said, I'm happy to change those as well if we'd prefer the consistency!
I've also fixed a typo that I came across while grepping for instances.
Will update with screenshots once docs are built.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100028
Approved by: https://github.com/albanD
# Motivation
The DLPack device type kDLOneAPI stands for the Unified Shared Memory allocated on a oneAPI device. The corresponding Pytorch backend type is XPU.
Support to export/import the Pytorch XPU tensor as a DLPack tensor of kDLOneAPI device.
# Solution
1. Update the DLPack protocol to v0.7.
2. Add the XPU hooks to map the Aten device and DLPack device with the address value and device information.
# Additional Context
Reopen (#82867)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94968
Approved by: https://github.com/kit1980
Currently it falls through to a call to `storage()`, which the IPU doesn't support.
I've made the minimal change here for ease of merging (this'd help us if it was in for 1.13.1), however...
**QUESTION**: Is there any reason why `not torch._C._has_storage(self)` needs to *also* be guarded on `self.device.type == privateuseone`? in other words, could the condition for using `clone` not be this?
```python
self.is_sparse
or self.device.type
in ["lazy", "xla", "mps", "ort", "meta", "hpu", "ipu"]
or not torch._C._has_storage(self)
or (type(self) is not Tensor and self.data_ptr() == 0)
```
If the condition fails, the very next thing is a call to `self._typed_storage()` which will fail, so it feels to me like *any* case without storage shouldn't fall through to the `storage()` call.
The original PR for adding the 'no storage and device is `PrivateUse1`' condition ([86557](https://github.com/pytorch/pytorch/pull/86557)) doesn't discuss whether this could be broadened.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/89129
Approved by: https://github.com/albanD
Fixes#81690
TODO:
* [x] C++ Unpickler Fix (locally tested pickled in Python and unpickled in C++)
* [x] C++ Pickler Fix (locally tested pickled in C++ and unpickled in Python)
* [x] Do quant_tensor, sparse_tensor, etc require similar changes? (Sparse and Quant don't need this)
* [x] Add Comments
* [x] How to make sure C++ and Python are in sync? (Functions in `pickler.h` help in getting and setting Tensor Metadata (math-bits for now) on a tensor. They are the only place which should handle this.)
Notes:
Quant Tensor don't support complex dtypes and for float they segfault with `_neg_view` : https://github.com/pytorch/pytorch/issues/88484
Sparse Tensor:
```python
>>> a = torch.tensor([[0, 2.], [3j, 0]]).to_sparse()
>>> a.conj().is_conj()
False
>>> a._neg_view()
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
NotImplementedError: Cannot access storage of SparseTensorImpl
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88182
Approved by: https://github.com/ezyang, https://github.com/anjali411
