Fixes#95900
Using the following repro as guide:
```python
import torch
import torch._dynamo
from torch._subclasses import fake_tensor
from torch.fx.experimental.symbolic_shapes import ShapeEnv
from torch._dynamo.output_graph import config
class Model(torch.nn.Module):
def __init__(self) -> None:
super().__init__()
self.linear = torch.nn.Linear(2, 2)
self.linear2 = torch.nn.Linear(2, 2)
def forward(self, x):
out = self.linear(x)
out = self.linear2(out)
return out
fake_mode = fake_tensor.FakeTensorMode(allow_non_fake_inputs=False,
allow_fallback_kernels=True,
shape_env=ShapeEnv(
allow_scalar_outputs=config.capture_scalar_outputs,
allow_dynamic_output_shape_ops=config.capture_dynamic_output_shape_ops,
frame_id=0
),
)
# Fakefying input/model before calling torch._dynamo.export
with fake_mode:
fake_x = torch.rand(5, 2, 2)
model = Model()
# Calling torch._dynamo.export without active fake mode
graph_module, guards = torch._dynamo.export(
model,
fake_x,
aten_graph=True,
fake_mode=fake_mode
)
graph_module.print_readable()
graph_module.graph.print_tabular()
```
Summary of changes:
* Plumb fake_mode through torch.export API. When specified, it
replaces the creation of a new FaketendorMode at InstructionTranslator on behalf of OutputGraph
Hacks FakeTensor.__new__ to prevent a
torch.tensor._make_subclass call for inputs that are already fakefied by
user. This probably need to be fixed in a nicer way. Any idea?
* Removed a few asserts that didn't want faked tensors coming
from user script
* Added torch._subclasses.fake_tensor.FakeTensor to type list on a few
asserts check to allow fake inputs
The changes above allowed symbolic tracing with both static and dynamic shapes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100017
Approved by: https://github.com/ezyang
Originally, my goal for this PR was to remove the `dynamic_shapes` tests in torch/_dynamo/variables/builder.py. However, one thing lead to another, and it turns out that it was easiest to do all of the following in one go:
* Unconditionally allocate a ShapeEnv, no matter if dynamic_shapes is enabled or not (torch/_dynamo/output_graph.py). There is a small adjustment to export torch/_dynamo/eval_frame.py to account for the fact that a ShapeEnv always exists, even if you're not doing symbolic export.
* Remove dynamic_shapes test from unspec logic (torch/_dynamo/variables/builder.py), the original goal
* Specialize strides and storage offset if all sizes are dynamic (torch/fx/experimental/symbolic_shapes.py). This is required to deal with unconditional ShapeEnv: if a ShapeEnv exist, fake tensor-ification may choose to allocate symbols. The idea is that with `automatic_dynamic_shapes == False`, Dynamo should never request dynamic sizes, but this invariant was not upheld for nontrivial strides/offset.
The rest are just auxiliary fixups from the above:
* Workaround bug in FakeTensorProp where sometimes it doesn't return a FakeTensor (torch/fx/passes/fake_tensor_prop.py), see https://github.com/pytorch/pytorch/pull/103395 for follow up
* Make ShapeProp correctly handle int inputs (torch/fx/passes/shape_prop.py)
* Disable indexing strength reduction if `assume_static_by_default` is False (torch/_inductor/codegen/triton.py)
* Fix hf_T5_generate to NOT toggle `assume_static_by_default` if dynamic shapes is not enabled (benchmarks/dynamo/common.py); technically this is not necessary anymore but it's in for safety.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103302
Approved by: https://github.com/voznesenskym
We do not raise constraint violations for complex binary conditions, such as conditions involving `%`. Moreover, while these constraints are discovered by our solver, the solver does not inject new constraint violations. This can result in cases where export passes, appropriate assertions are not added, and we get runtime crashes.
Now, when the solver discovers constraints that are too complex, we force-specialize the involved dimensions and raise a constraint violation when such dimensions are marked dynamic. This forces the user to remove the dynamic marking, and causes the appropriate specialization assertions to be added.
Differential Revision: [D46415786](https://our.internmc.facebook.com/intern/diff/D46415786/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102897
Approved by: https://github.com/tugsbayasgalan
We discussed in a composability meeting a few weeks ago that `pre_autograd` should probably be renamed to `pre_dispatch`.
One question in this PR was: should I re-use a dispatch key? Or should I create a new dispatch key (that yet again corresponds to "top of the dispatcher")?
~~For now, I ended up sticking our proxy mode on the mode stack corresponding to `PythonTLSSnapshot`, because it was simple and it works. It looks like one of the functorch dispatch keys has higher priority though, so it's possible that functorch will end up running first. Open to options, but we can consider adding a new dispatch key later if that becomes a problem~~
Update: I added a dedicated dispatch key, `PreDispatch`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101818
Approved by: https://github.com/ezyang, https://github.com/Neilblaze, https://github.com/albanD, https://github.com/zou3519
Manually generate guards for optimizer rather than use variable builder, which can be slow with lots of params.
This is the reason for ~10s compile slowdown
Redisable `_init_group`. This is important, because if for any reason a frame which calls `_init_group` is run in the python interpreter, we will trace it, which we don't want to do. We only want to call it when it is accessed via the fast path implemented with the optimizer variable during symbolic interpretation.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103121
Approved by: https://github.com/jansel
- Disables dynamo on the differentiable optimizer tests
- Disables dynamo on some test methods which expose a very rare dynamo edge case
- Disables dynamo on export/save optimizer state methods because it shouldn't trace those anyway.
I have a draft PR to fix the two tests marked skip due to unsupported mutation of step.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103066
Approved by: https://github.com/janeyx99, https://github.com/malfet
On calls to `_init_group` rather than tracing through it, extract python values from the arguments, and call the initialization. This avoids having to trace this function which is very slow with large parameters, and also avoids graph breaking on it. This is sound in this case because the state is only initialized once in the eager case. Guards on the state and params are generated explicitly rather than via tracing the initialization.
Caveats:
`_init_group` also gathers various state tensors into lists via mutating list arguments to pass to the functional optimizer implementation. These state tensors exist on the optimizer itself, but we don't know exactly how the gathering is done and which tensors correspond to which attributes of the optimizer module (each optimizer has different states). To rectify this, we keep weak_ptrs to all of the tensors collected in the lists in globals (similar to how parameter keys are stored for dictionaries). These pointers are guaranteed to be alive as long as the optimizer object is alive if the internal state is not interfered with and they are guarded with weakref guards
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102640
Approved by: https://github.com/jansel
This PR adds aot_export_module as the lowering path from torch.level graph to aten graph. Some known limitations that need to be addressed in the follow up PRs:
1. Store param/buffer data in ExportedProgram
2. Fully support torch.cond with params/buffers
3. Making sure no duplicated ExportMetaData entry
4. This API will break Executorch if used on PyE, we will figure out a plan internally.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101490
Approved by: https://github.com/avikchaudhuri
Previously we had runtime asserts for range constraints. This diff adds runtime asserts for equality constraints.
This requires a bit of refactoring that is worth calling out.
1. [Minor] Some of the data structures produced by export and consumed by the runtime assertion pass need to be broadened. This is a WIP. There are some associated code improvements that are included in this diff, but by and large the structures are similar to what exists now. Meanwhile @angelayi and I are chatting about how to make it qualitatively better: briefly, we want to index everything by symbols, which are 1-1 with (name, dim) pairs.
2. [Major] The order in which runtime asserts are emitted is changed. Previously we used to do the work in `placeholder`, now this diff adds a hook for "post-processing" after processing of all placeholders is done. This is needed because equality constraints can mention different placeholders. This change also opens the way to optimizing codegen: e.g., each (name, dim) pair should correspond to a single intermediate variable that is reused across runtime asserts. This is future work.
Differential Revision: [D46177642](https://our.internmc.facebook.com/intern/diff/D46177642/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102256
Approved by: https://github.com/tugsbayasgalan, https://github.com/angelayi
Summary: We don't need to leak matched input positions from dynamo anymore if we can just populate all args with corresponding fake tensors.
Test Plan: CI
Differential Revision: D46131556
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102129
Approved by: https://github.com/angelayi
Fixes https://github.com/pytorch/pytorch/issues/100415
Results in the following error:
```
Traceback (most recent call last):
File "/scratch/angelayi/work/pytorch/test/export/test_export.py", line 572, in test_export_constrain_static
export(f, example_inputs, constraints)
File "/scratch/angelayi/work/pytorch/torch/_export/__init__.py", line 348, in export
method_name_to_graph_module[compile_spec.method_name] = _export(
File "/scratch/angelayi/work/pytorch/torch/_export/__init__.py", line 119, in _export
raise UserError(UserErrorType.CONSTRAIN_VIOLATION, str(e))
torch._dynamo.exc.UserError: File "/scratch/angelayi/work/pytorch/test/export/test_export.py", line 561, in f
constrain_as_value(c, min=1, max=3)
It appears that you're trying to set a constraint on a value which we evaluated to have a static value of 3. Scroll up to see where this constraint was set.
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101655
Approved by: https://github.com/avikchaudhuri
Notes:
- No segfaults observed in any CI tests: dynamo unittests, inductor unittests, dynamo-wrapped pytorch tests. So we remove the warning that using dynamo 3.11 may result in segfaults.
- Some dynamo-wrapped pytorch tests hang. They will be skipped in the dynamo-wrapped test suite and will be addressed in a future PR
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99180
Approved by: https://github.com/malfet
