Based on https://github.com/pytorch/pytorch/pull/126376, this PR tries to update all PT callers (e.g., `Tensor.is_pinned()`, `Tensor.pin_memory()`) to not pass `device` argument.
As for `storage/untyped_storage.is_pinned()/pin_memory()`, we keep the `device` argument but passing `device` is discouraged. And if not given, the default `device` is still 'cuda' for BC.
Additionally, based on device-agnostic pin_memory, `pin_memory_device` argument of `torch.utils.data.DataLoader` is discouraged now. For BC, explictly passing this argument is still effective. If not given, the default `device` will be the current accelerator.
Fixes#124908
Relates https://github.com/pytorch/pytorch/pull/126376
Pull Request resolved: https://github.com/pytorch/pytorch/pull/131858
Approved by: https://github.com/albanD
Co-authored-by: albanD <desmaison.alban@gmail.com>
Fixes#142144
A global x is saved in checkpoint as `GLOBAL x.__module__ x.__name__`. So , after allowlisting a GLOBAL it is expected to match any GLOBAL instruction of the form `GLOBAL x.__module__ x.__name__` but there are edge cases when for the same API from the same module, what `__module__` gives changes between versions which prevents users from allowlisting the global.
In this case, in numpy < 2.1
```
torch.save("bla", np_array)
# checkpoint has GLOBAL "np.core.multiarray" "_reconstruct"
```
In np version 2.1
```
with safe_globals([np.core.multiarray._reconstruct]):
torch.load("bla")
```
np.core.multiarray._reconstruct.__module__ gives "np._core.multiarray" (note the extra _ before core) and see what was done [here](https://github.com/numpy/numpy/blob/main/numpy/core/multiarray.py)
Since the dictionary to access safe globals is keyed on "{foo.__module__}.{foo.__name__}", __module__, __name__ will no longer match that in the checkpoint so "np.core.multiarray._reconstruct" can no longer be properly allowlisted (instead np._core.multiarray._reconstruct is a key in the dict).
We allow `add_safe_globals/safe_globals` to optionally take tuples of (global, str of module.name) to workaround such (odd/edge case) situations.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/142153
Approved by: https://github.com/albanD
Related: https://github.com/pytorch/xla/issues/7799#issuecomment-2375818263
Follow ups: Do the same for maia and mtia
## Motivation
With the move to `weights_only` by default, we are making an explicit decision not to allowlist GLOBALs required to deserialize `numpy` tensors by default. The implication is that backends relying on numpy for serialization will fail loudly when `torch.load` flips `weights_only`.
However, we make the observation that this dependency on numpy was legacy and is not actually needed anymore. So we can remove it, which aligns with our weights_only strategy.
## Why is this ok?
The following comment on why numpy is necessary for serialization is legacy
c87c9f0a01/torch/_tensor.py (L303-L312)
We no longer do the following, though it was the case 5 years ago in the PR that added this
> CPU storage is reconstructed with randomly initialized data, moved onto backend device, and then storage is updated to the serialized content
**Instead what now happens is that CPU storage is constructed with data from the file **and then** moved onto backend device.**
Old behavior (`legacy_load`): 67adda891a/torch/serialization.py (L620)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/137444
Approved by: https://github.com/albanD
## Semantic
The semantic is
(1) By default `torch.serialization.skip_data(materialize_fake_tensors=False)` will make `torch.save` skip writing storages (but reserve space for them in the checkpoint).
```python
import torch
import torch.nn as nn
sd = nn.Linear(3, 5).state_dict()
with torch.serialization.skip_data():
torch.save(sd, 'foo.pt')
print(torch.load('foo.pt', weights_only=True))
```
(2) With `torch.serialization.skip_data(materialize_fake_tensors=True)`If FakeTensor is passed to `torch.save` the pickler will treat these FakeTensors as being "materialized" space will be reserved in the checkpoint for the associated storage bytes, and when loading the type will be Tensor instead of FakeTensor)
```python
import torch
import torch.nn as nn
from torch._subclasses.fake_tensor import FakeTensorMode
with FakeTensorMode():
m = nn.Linear(3, 5, dtype=torch.float16, device='cuda')
sd = m.state_dict()
with torch.serialization.skip_data(materialize_fake_tensors=True):
torch.save(sd, 'bla.pt')
print(torch.load('bla.pt', weights_only=True))
# OrderedDict([('weight', tensor([[0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.]], device='cuda:0', dtype=torch.float16)), ('bias', tensor([0., 0., 0., 0., 0.], device='cuda:0', dtype=torch.float16))])
```
## Follow Ups
- [ ] `torch.load` semantic for skip_data context manager
- [ ] Mechanism for getting offsets of storages saved via this method (for writing in a separate pass)
Differential Revision: [D62238610](https://our.internmc.facebook.com/intern/diff/D62238610)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134504
Approved by: https://github.com/albanD
## Semantic
The semantic is
(1) By default `torch.serialization.skip_data(materialize_fake_tensors=False)` will make `torch.save` skip writing storages (but reserve space for them in the checkpoint).
```python
import torch
import torch.nn as nn
sd = nn.Linear(3, 5).state_dict()
with torch.serialization.skip_data():
torch.save(sd, 'foo.pt')
print(torch.load('foo.pt', weights_only=True))
```
(2) With `torch.serialization.skip_data(materialize_fake_tensors=True)`If FakeTensor is passed to `torch.save` the pickler will treat these FakeTensors as being "materialized" space will be reserved in the checkpoint for the associated storage bytes, and when loading the type will be Tensor instead of FakeTensor)
```python
import torch
import torch.nn as nn
from torch._subclasses.fake_tensor import FakeTensorMode
with FakeTensorMode():
m = nn.Linear(3, 5, dtype=torch.float16, device='cuda')
sd = m.state_dict()
with torch.serialization.skip_data(materialize_fake_tensors=True):
torch.save(sd, 'bla.pt')
print(torch.load('bla.pt', weights_only=True))
# OrderedDict([('weight', tensor([[0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.]], device='cuda:0', dtype=torch.float16)), ('bias', tensor([0., 0., 0., 0., 0.], device='cuda:0', dtype=torch.float16))])
```
## Follow Ups
- [ ] `torch.load` semantic for skip_data context manager
- [ ] Mechanism for getting offsets of storages saved via this method (for writing in a separate pass)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134504
Approved by: https://github.com/albanD
## Semantic
The semantic is
(1) By default `torch.serialization.skip_data(materialize_fake_tensors=False)` will make `torch.save` skip writing storages (but reserve space for them in the checkpoint).
```python
import torch
import torch.nn as nn
sd = nn.Linear(3, 5).state_dict()
with torch.serialization.skip_data():
torch.save(sd, 'foo.pt')
print(torch.load('foo.pt', weights_only=True))
```
(2) With `torch.serialization.skip_data(materialize_fake_tensors=True)`If FakeTensor is passed to `torch.save` the pickler will treat these FakeTensors as being "materialized" space will be reserved in the checkpoint for the associated storage bytes, and when loading the type will be Tensor instead of FakeTensor)
```python
import torch
import torch.nn as nn
from torch._subclasses.fake_tensor import FakeTensorMode
with FakeTensorMode():
m = nn.Linear(3, 5, dtype=torch.float16, device='cuda')
sd = m.state_dict()
with torch.serialization.skip_data(materialize_fake_tensors=True):
torch.save(sd, 'bla.pt')
print(torch.load('bla.pt', weights_only=True))
# OrderedDict([('weight', tensor([[0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.]], device='cuda:0', dtype=torch.float16)), ('bias', tensor([0., 0., 0., 0., 0.], device='cuda:0', dtype=torch.float16))])
```
## Follow Ups
- [ ] `torch.load` semantic for skip_data context manager
- [ ] Mechanism for getting offsets of storages saved via this method (for writing in a separate pass)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134504
Approved by: https://github.com/albanD
# Motivation
Structured codegen is beneficial for easier decoupling tensor meta setting and kernel implementation. At present, XPU operators need to handle tensor metas in hand-written way.
We plan to leverage the codegen system for auto generate structured operators. This PR facilitate the `DispatchStub` support for Intel GPUs. Based on that, XPU operators would have possibility to register kernel functor to operator stubs.
This is a prerequisite of PR #130082, where we will modify the codegen system to generate XPU needed source files and headers.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130019
Approved by: https://github.com/EikanWang, https://github.com/gujinghui, https://github.com/albanD
This PR re-implements pin memory aiming to get rid of the optional `device` argument and makes all related APIs to be device-agnostic. We add two new abstract APIs in [AcceleratorHooksInterface](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/detail/AcceleratorHooksInterface.h#L12) and redefine pin memory as: "Pin memory is always pinned for the current accelerator device". In detail, it uses [getAcceleratorHooksInterface](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/Context.h#L61) in pin_memory/is_pinned to get an appropriate device and invoke the corresponding overridden interfaces, instead of using BackendSelect and then dispatching to CUDA or other specific backends' implement methods.
Note: For new backends who want to implement and use pin memory, just inherit AcceleratorHooksInterface and overwrite the `isPinnedPtr` and `getPinnedMemoryAllocator` methods.
Additional context: To avoid BC-breaking, this PR just preserves the `device` arg of related APIs and would throw a deprecation warning if `device` arg is passed. Another PR will be submitted to update all PT callers (`Tensor.is_pinned()`, `Tensor.pin_memory()`...) not to pass this arg based on this PR. In future, `device` arg will be actually removed.
Relates #124908
Relates #14560
Pull Request resolved: https://github.com/pytorch/pytorch/pull/126376
Approved by: https://github.com/albanD
1) Add skip undefined tensor in cpu fallback when call _copy_from_and_resize;
2) Modify to_cpu function support optional tensor;
3) Add copy back to origin optional tensor when alias_info isWrite is true.
@ezyang @bdhirsh
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130237
Approved by: https://github.com/ezyang
This PR re-implements pin memory aiming to get rid of the optional `device` argument and makes all related APIs to be device-agnostic. We add two new abstract APIs in [AcceleratorHooksInterface](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/detail/AcceleratorHooksInterface.h#L12) and redefine pin memory as: "Pin memory is always pinned for the current accelerator device". In detail, it uses [getAcceleratorHooksInterface](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/Context.h#L61) in pin_memory/is_pinned to get an appropriate device and invoke the corresponding overridden interfaces, instead of using BackendSelect and then dispatching to CUDA or other specific backends' implement methods.
Note: For new backends who want to implement and use pin memory, just inherit AcceleratorHooksInterface and overwrite the `isPinnedPtr` and `getPinnedMemoryAllocator` methods.
Additional context: To avoid BC-breaking, this PR just preserves the `device` arg of related APIs and would throw a deprecation warning if `device` arg is passed. Another PR will be submitted to update all PT callers (`Tensor.is_pinned()`, `Tensor.pin_memory()`...) not to pass this arg based on this PR. In future, `device` arg will be actually removed.
Relates #124908
Relates #14560
Pull Request resolved: https://github.com/pytorch/pytorch/pull/126376
Approved by: https://github.com/albanD
1. Fix the wrong tests about lazy init for PrivateUse1 named foo
2. Refactor the tests and make it more flexible
3. Disable the two tests temporarily
- test_open_device_faketensor
- test_open_device_scalar_type_fallback
Pull Request resolved: https://github.com/pytorch/pytorch/pull/125572
Approved by: https://github.com/albanD
Fixes#100152
1. Fix the wrong tests about lazy init for PrivateUse1 named foo
2. Fix wrong backend meta registry mechanism when compiling with clang++( compiling with g++ work well)(introduced by static variable in inline function)
3. Refactor the tests and make it more flexible
4. Disable the two tests temporarily
- test_open_device_storage_pin_memory
- test_compile_autograd_function_aliasing
Pull Request resolved: https://github.com/pytorch/pytorch/pull/124712
Approved by: https://github.com/albanD, https://github.com/malfet
1) add operand and get_dim_names API;
2) set will_resize to true when output tensor is undefined;
3) add abs_stub for dummy device and calculate on cpu device;
4) support dummy device copy with stride;
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120792
Approved by: https://github.com/ezyang
Fixes https://github.com/pytorch/pytorch/issues/102970. See the comment [here](https://github.com/pytorch/pytorch/issues/102970#issuecomment-1577223773) for details.
We normally treat "outputs that alias inputs" specially in AOTAutograd, by replaying the views at runtime, instead of baking them into the graph. For views that are part of custom autograd functions though, we can't do that view-replay, since it will clobber the backwards function that the user specified in their custom autograd.Function.
Right now in this PR, I distinguish between "aliased inputs that are normal views" vs. "aliased inputs that are views that came from an autograd.Function call" by checking the outputs `.grad_fn` field, to see if it inherits from our custom CBackward function class. Then I added a new `OutputType` enum value, that we effectively treat the "normal" way (the same way that we treat ordinary, non-aliased outputs). The new enum val is mostly for debugging - so we can print it and know that our graph had custom autograd.Function aliased outputs in it.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102992
Approved by: https://github.com/ezyang, https://github.com/zou3519
Fixes #ISSUE_NUMBER
as the title, add context support for custom device and testcase.
And in the future, we may want to refactor these hooks for different device to unify the APIs, would you agree my
idea？ @albanD
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105056
Approved by: https://github.com/albanD
Fixes #ISSUE_NUMBER
Now for the custom device, we use `getattr` and `setattr` to run the func defined in custom device module in some files, such as `AMP`, `random`, `DDP` and so on. So I want to add a generic func to get these funcs more friendly, could you take a look? @bdhirsh @albanD
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99048
Approved by: https://github.com/bdhirsh
Fixes #ISSUE_NUMBER
Add the serialization logic of backend metadata to the serialization of tensor, which is implemented through custom registration functions.
In #97429 , the structure backendMeta is provided in TensorImpl, and we think that this part of information may also need to be serialized for custom.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99808
Approved by: https://github.com/ezyang, https://github.com/huydhn
