This adds some utilities for conveniently working with fast combined CapturedTraceback from Python. The main goal of these utilities is to make it easier for people to use CapturedTraceback as a drop-in replacement for `traceback.extract_stack`, which is 20x slower than CapturedTraceback.
I port symbolic shapes to use the new CapturedTraceback code, to validate that the APIs work and are useful.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107358
Approved by: https://github.com/zdevito, https://github.com/albanD
ghstack dependencies: #107438
Adds API to mark tensor as a static input -
To make this trigger recompiles properly, I'll need to update tensor match checks to also check for this new attribute
Additional concern is memory - the tensors will be kept alive, but this is the current behavior for nn modules and parameters.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107154
Approved by: https://github.com/eellison
RFC: https://github.com/pytorch/rfcs/pull/54
First commit is the contents of https://github.com/Quansight-Labs/numpy_pytorch_interop/
We have already been using this in core for the last few months as a external dependency. This PR pulls all these into core.
In the next commits, I do a number of things in this order
- Fix a few small issues
- Make the tests that this PR adds pass
- Bend backwards until lintrunner passes
- Remove the optional dependency on `torch_np` and simply rely on the upstreamed code
- Fix a number dynamo tests that were passing before (they were not tasting anything I think) and are not passing now.
Missing from this PR (but not blocking):
- Have a flag that deactivates tracing NumPy functions and simply breaks. There used to be one but after the merge stopped working and I removed it. @lezcano to investigate.
- https://github.com/pytorch/pytorch/pull/106431#issuecomment-1667079543. @voznesenskym to submit a fix after we merge.
All the tests in `tests/torch_np` take about 75s to run.
This was a work by @ev-br, @rgommers @honno and I. I did not create this PR via ghstack (which would have been convenient) as this is a collaboration, and ghstack doesn't allow for shared contributions.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/106211
Approved by: https://github.com/ezyang
This PR adds a new configuration that enables shapes of torch.nn.Parameter to be treated as dynamic in order to avoid extensive recompilation when Paramters are used instead of Tensor.
This features addresses part of issue #105279
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105855
Approved by: https://github.com/ezyang
Since Python 3.11 bytecode contains endline and column information, for each bytecode, we attribute the source code corresponding to the bytecode in a more accurate way. For example, we can highlight a function call in a series of nested function calls, or highlight a function call spanning multiple lines.
Sample:
```python
import torch
import torch._dynamo
from functorch.experimental.control_flow import cond
def h(x):
return x * 5
def true_fn(x):
return x * 2
def false_fn(x):
return x * 3
def f(pred, x):
x = h(
h(h(x))
)
x = x[1:][:2]
torch._dynamo.graph_break()
x = cond(pred, true_fn, false_fn, [x])
opt_f = torch.compile(f, backend="eager")
opt_f(torch.tensor(True), torch.randn(3, 3, 3, 3))
```
Output:
```
$ TORCH_LOGS="trace_call" python playground9.py
TRACE inlined call h from f /scratch/williamwen/work/pytorch/playground9.py:16
h(h(x))
~^^^
TRACE FX call mul from h /scratch/williamwen/work/pytorch/playground9.py:6 (inline depth: 1)
return x * 5
~~^~~
TRACE inlined call h from f /scratch/williamwen/work/pytorch/playground9.py:16
h(h(x))
~^^^^^^
TRACE FX call mul_1 from h /scratch/williamwen/work/pytorch/playground9.py:6 (inline depth: 1)
return x * 5
~~^~~
TRACE inlined call h from f /scratch/williamwen/work/pytorch/playground9.py:15
x = h(
~^
h(h(x))
^^^^^^^
)
^
TRACE FX call mul_2 from h /scratch/williamwen/work/pytorch/playground9.py:6 (inline depth: 1)
return x * 5
~~^~~
TRACE FX call getitem from f /scratch/williamwen/work/pytorch/playground9.py:18
x = x[1:][:2]
~^^^^
TRACE FX call getitem_1 from f /scratch/williamwen/work/pytorch/playground9.py:18
x = x[1:][:2]
~~~~~^^^^
TRACE inlined call true_fn from <resume in f> /scratch/williamwen/work/pytorch/playground9.py:20
x = cond(pred, true_fn, false_fn, [x])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TRACE FX call mul from true_fn /scratch/williamwen/work/pytorch/playground9.py:9 (inline depth: 1)
return x * 2
~~^~~
TRACE inlined call false_fn from <resume in f> /scratch/williamwen/work/pytorch/playground9.py:20
x = cond(pred, true_fn, false_fn, [x])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TRACE FX call mul from false_fn /scratch/williamwen/work/pytorch/playground9.py:12 (inline depth: 1)
return x * 3
~~^~~
TRACE FX call cond from <resume in f> /scratch/williamwen/work/pytorch/playground9.py:20
x = cond(pred, true_fn, false_fn, [x])
~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/104676
Approved by: https://github.com/ezyang
This PR adds necessary plumbing through torchdynamo to allow tensor
subclasses with certain contract (i.e. with `__tensor_flatten__` and
`__tensor_unflatten__`) to goes through the dynamo fakification pass by
fakifying the tensor subclass internal components.
Some of the tensor subclass contract logic mostly borrowed from
https://github.com/pytorch/pytorch/pull/97540
Added some tests to verify simply passing through a tensor subclass
(i.e. DTensor) through dynamo eager works as expected.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105308
Approved by: https://github.com/ezyang
Add similar semantics for creating a buffer object similar to creating a parameter. This is done by introducing a new `Buffer` class that can be used for type disambiguation. The underlying functionality of registering a buffer remains the same as the `register_buffer` method has not been changed. The `persistent` parameter in the `Buffer` type is to indicate whether a buffer object should be persistent or not. Other non-test changes have to do with getting the new `Buffer` type recognized by inductor and dynamo. Remaining changes are test changes to make sure that the `Buffer` type can be used as a drop in replacement for `register_buffer` as it just leads to `register_buffer` being called. The addition of this new functionality still allows for normal tensors to be used as buffers so these changes are intended to be backwards compatible.
Fixes#35735
Pull Request resolved: https://github.com/pytorch/pytorch/pull/104069
Approved by: https://github.com/mikaylagawarecki
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
These are the numbers with this PR
There are 3 main followups
* A naive partitioner gives better memory footprint than min-cut partitioner here. Currently, we are using min-cut partitioner. Waiting for @Chillee to discuss this further to either modify min-cut or add a naive partitioner.
* aot_eager is < 1x memory footprint. This is true even for non AC models. This could hide some inefficiency somewhere.
* inductor is giving very different memory numbers between AOT-traced-AC (duplicate early) vs this implementation. This leads to some inefficiency in inductor that we need to resolve.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102935
Approved by: https://github.com/jansel
This PR adds universal support for ndarray methods. After #100839 each `NumpyNdarrayVariable` should wrap a `torch.Tensor`. This PR adds a `numpy_method_wrapper` which converts the `torch.Tensor` to `torch_np.ndarray` and then call the numpy ndarray method. Then we also try to return a `torch.Tensor` (return as-is if the value is not ndarray-like)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97537
Approved by: https://github.com/ezyang
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
Issue: #93684
In previous PRs #95849#99560 we redirect `numpy.*`, `<tensor>.numpy()` calls to `torch_np.*` methods and attributes, by creating `NumpyNdarrayVariable` for those calls.
We need to handle `NumpyNdarrayVariable` when graph break happens.
This PR did 2 things:
1. In `codegen.py` we made sure we can reconstruct the value wrapped by `NumpyNdarrayVariable`, to be `torch_np.ndarray` in the stack whenerver we recompiles the subgraph.
2. In `builder.py` we can wrap the value to be `NumpyNdarrayVariable` and save it as graph input.
-----
Starting from commit 6:
## A new design for supporting numpy in dynamo
In short the core concept doesn't change: we still convert `numpy` API calls to `torch_np` API calls. However, instead of wrapping a `torch_np.ndarray` in `NumpyNdarrayVariable`, the new design wraps a `torch.Tensor`.
The reason for doing this change is because we need to keep `torch.Tensor` everywhere in the captured graph, so that it works well with the backend of dynamo. See discussions in https://github.com/Quansight-Labs/numpy_pytorch_interop/issues/142 for details.
### Flow
This is an example showing how do we think about dynamo working on a simple function:
```python
def f(x: torch.Tensor, y: torch.Tensor):
a, b = x.numpy(), y.numpy()
c = np.add(x, y)
return torch.from_numpy(c)
```
```
+------------+ +------------+
torch.Tensor | |numpy.ndarray| |
-------------- .numpy() --------------| |
| | | | +------------------+
+------------+ | numpy.add |numpy.ndarray| |torch.Tensor
+------------+ | --------------| torch.from_numpy --------------
torch.Tensor | |numpy.ndarray| | | |
-------------- .numpy() --------------| | +------------------+
| | | |
+------------+ +------------+
+------------+ +----------------+
torch.Tensor | |torch.Tensor | |
-------------- .detach() --------------| |
| | | | +----------------+ +------------+
+------------+ | |torch_np.ndarray| |torch.Tensor| |torch.Tensor
| torch_np.add -----------------| util.to_tensor -------------| .detach() --------------
+------------+ | | | | | |
torch.Tensor | |torch.Tensor | | +----------------+ +------------+
-------------- .detach() --------------| |
| | | |
+------------+ | +----------------+ |
| wrapper on torch_np.add |
+--------------------------------------------------------+
```
### Approach
`torch_np` APIs can take both `torch_np.ndarray` as well as `torch.Tensor`. What we need to do is to have a wrapper for these APIs to convert the return value back to `torch.Tensor`. This way only the wrapper is showing up in the captured graph, with `torch.Tensor`s as input and `torch.Tensor` as output.
If we have a graph break or we've traced to the end of the program, we need to inspect all the `NumpyNdarrayVariable` in the stack and convert them back to `numpy.ndarray`, to make sure the compiled version is still behaving the same as the eager version.
### Examples
Here's an example of the graph generated:
```python
def fn(x: np.ndarray, y: np.ndarray):
a = x.real
b = y.real
torch._dynamo.graph_break()
return np.add(a, 1), np.add(b, 1)
```
Graph generated:
```
[2023-05-16 10:31:48,737] torch._dynamo.output_graph.__graph: [DEBUG] TRACED GRAPH
__compiled_fn_0 <eval_with_key>.0 opcode name target args kwargs
------------- -------------- ---------------------------------------------------------- ---------------------- --------
placeholder l_x_ L_x_ () {}
placeholder l_y_ L_y_ () {}
call_function from_numpy <built-in method from_numpy of type object at 0x12b1fdc80> (l_x_,) {}
call_function from_numpy_1 <built-in method from_numpy of type object at 0x12b1fdc80> (l_y_,) {}
call_function attr_wrapper <function attr_wrapper at 0x12e8693a0> (from_numpy, 'real') {}
call_function attr_wrapper_1 <function attr_wrapper at 0x12e8693a0> (from_numpy_1, 'real') {}
output output output ((),) {}
[2023-05-16 10:31:48,908] torch._dynamo.output_graph.__graph: [DEBUG] TRACED GRAPH
__compiled_fn_2 <eval_with_key>.1 opcode name target args kwargs
------------- ------------- ---------------------------------------------------------- ------------------------------- --------
placeholder l_a_ L_a_ () {}
placeholder l_b_ L_b_ () {}
call_function from_numpy <built-in method from_numpy of type object at 0x12b1fdc80> (l_a_,) {}
call_function from_numpy_1 <built-in method from_numpy of type object at 0x12b1fdc80> (l_b_,) {}
call_function wrapped_add <Wrapped function <original add>> (from_numpy, 1) {}
call_function wrapped_add_1 <Wrapped function <original add>> (from_numpy_1, 1) {}
output output output ((wrapped_add, wrapped_add_1),) {}
```
### Changes
* `codegen.py`: reconstruct `numpy.ndarray` from `NumpyNdarrayVariable` by adding bytecode to call `utils.to_numpy_helper()`.
* `output_graph.py`: getting rid of legacy code that does exactly what `codegen.py` does, which only handling return case but not graph break case.
* `utils.py`: added helpers to convert `numpy.ndarray` to `torch.Tensor` and vice versa. Also adding a wrapper class that takes in a function. In `__call__` it calls the function and converts its out to `torch.Tensor` (or a list of it).
* `builder.py`: add method to wrap `numpy.ndarray` graph inputs into `NumpyNdarrayVariable`, by calling `torch.numpy` in the proxy.
* `misc.py`: `numpy` API calls goes into `NumpyVariable` and we find the function with the same name in `torch_np` module, then wrap it with the wrapper defined in `utils.py`.
* `tensor.py`, `torch.py`: proxy `tensor.numpy()` to be `torch.detach()` but wrap it with `NumpyNdarrayVariable`. Similarly, `torch.from_numpy()` -> `torch.detach()` but wrap it with `TensorVariable`. In `NumpyNdarrayVariable`, do the similar `torch_np.ndarray` to `torch.Tensor` wrapping for attributes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100839
Approved by: https://github.com/ezyang
Summary:
This PR adds support for folding bn weights into conv for QAT flow, this is equivalent
to the QAT branch of `from_float` in eager mode quantized conv module: https://github.com/pytorch/pytorch/blob/main/torch/ao/nn/quantized/modules/conv.py#L223
Items that needs followup:
* there are some workaround I did because quantize_per_tensor is using float/int args and dynamo does not support these args, need to fix after we change the quantized model representation and also change these args to Tensor
Test Plan: buck2 test @//mode/opt //caffe2/test:quantization_pt2e -- --exact 'caffe2/test:quantization_pt2e - test_convert_qat_conv_bn_fusion (quantization.pt2e.test_quantize_pt2e.TestQuantizePT2E)'
Reviewed By: andrewor14
Differential Revision: D45344281
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100442
Approved by: https://github.com/kimishpatel
This adds a new operator debugprims::load_storage which does the unusual thing of loading a tensor from disk (via ContentStoreReader). This will be used in a later PR to implement delta debugging in the minifier, even when the repro is too big to fit into memory. The way it works is that you specify a name of the tensor you want to load, as well as enough metadata to reconstruct the tensor, if the store isn't available. If there is an active content store, we read and return the tensor from that store; otherwise we use `rand_strided` to create it.
I needed some infra improvements to do this:
* `custom_op` now supports factory functions. Factory functions have to be registered specially via `impl_factory`
* I modified `clone_input` to also support dtype conversion, which I use to change the dtype of a loaded tensor if necessary.
* ContentStore needs to work with a device argument, so we torch.load directly to the correct device. This is for fake tensor support.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100519
Approved by: https://github.com/zou3519, https://github.com/anijain2305
The changes:
* Add config knob `same_two_models_use_fp64` for toggling whether or not to use fp64
* Add a test showing that RMSE is superior to atol/rtol
* Add `--strict-accuracy` options, which allows for testing against integral/boolean accuracy. Regular accuracy by default now ONLY. There's a test which exercises this, it's a little delicate but I had trouble thinking of a good test otherwise.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100447
Approved by: https://github.com/voznesenskym
This PR splits OutputGraph into two classes:
- SubgraphTracer (handles FX-tracing)
- OutputGraph (handles Dynamo-specific output graph logic, like
tracking graph inputs, compiling the graph, and executing it).
The motivation behind this is in the next PR up in the stack.
TL;DR is: in order to do higher-order operators, we need nested
SubgraphTracer, one for each level of nesting of the higher-order
operators.
I'm happy to flatten the stack into a single PR, but this separate made
it easier for me to test. Lmk if you want the stack flattened.
Test Plan:
- existing tests
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99987
Approved by: https://github.com/anijain2305, https://github.com/voznesenskym
This is a two part PR; I can split it if you really want me to.
The first part is a refactor of the after aot repro/minifier scripts to come with a command line interface. I maintain exact BC with the previous interface (so, e.g., you still get a repro.py and a run_minifier.py that do the same thing as before), but each of these scripts also take command line arguments now which you can use to customize what actually happens. Check `run_repro` for full documentation on the arguments.
The second part of this is an implementation of `analyze` subcommand on the new CLI for any repro.
<img width="1277" alt="image" src="https://user-images.githubusercontent.com/13564/235045677-8545aab7-5e83-4813-bbec-47783dc60122.png">
This facility is oriented towards accuracy debugging. It does several things:
1. It will run your model twice and check for nondeterminism in inductor/float64, *even* on intermediate inputs (our benchmarking nondeterminism test only checks for nondeterminism on the final output). This makes localizing which operator is nondeterministic easy.
2. It will run your compiled model side-by-side with eager and float64 variants, and then report when things diverge too far from RMSE delta from float64.
Importantly, it does all this without requiring every intermediate to be held in memory (which will cause an OOM on large repros, such as the one I tested this on.)
Some other minor improvements:
* MinifierTestBase now has an easy to comment out spot that you can use to retain the temporary directory; good for debugging
* We print "running minifier" and "running repro" in MinifierTestBase to make it easier to orient where logs are coming from
* same takes a `log_error` optional argument which you can use to reroute the error logs when things mismatch
* counters["inductor"]["intermediate_hooks"] tracks the number of intermediate hooks we've codegen'ed; good for populate the tqdm interface
* torch.fx.interpreter gets an official `boxed_run` interface which uses the boxed arguments calling convention and doesn't retain inputs unnecessarily long
* torch.utils._content_store gets compute_tensor_metadata/read_tensor_metadata helper functions for computing tensor information without serializing it
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100226
Approved by: https://github.com/bertmaher, https://github.com/bdhirsh, https://github.com/anijain2305
On top of #95849 this PR is trying to handle the special case when dealing with numpy.
Consider the following example:
```
def f(x: torch.Tensor) -> np.ndarray:
a = x.numpy()
return a.T
```
In previous PR this will error out because we translate `a.T` to be a method call on `torch_np.ndarray.T` which is also a `torch_np.ndarray`.
This PR handles this case, by conditionally converting a `torch_np.ndarray` to `np.ndarray` before returning, to match the original behavior.
The compiled version will be:
```
def f(x):
___tmp_0 = __compiled_fn_0(x)
if isinstance(___tmp_0, torch_np.ndarray):
return ___tmp_0.tensor.numpy()
else:
return ___tmp_0
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99560
Approved by: https://github.com/jansel, https://github.com/yanboliang
Issue: #93684
# Problem
Reduce graph breaks when dynamo compiles python functions containing numpy functions and ndarray operations.
# Design (as I know it)
* Use torch_np.ndarray(a wrapper of tensor) to back a `VariableTracker`: `NumpyTensorVariable`.
* Translate all attributes and methods calls, on ndarray, to torch_np.ndarray equivalent.
This PR adds `NumpyTensorVariable` and supports:
1. tensor to ndarray, ndarray to tensor
2. numpy functions such as numpy.meshgrid()
3. ndarray attributes such as `itemsize`, `stride`
Next PR will handle returning `np.ndarray` and add support for ndarray methods
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95849
Approved by: https://github.com/ezyang
Before this PR, if users call ```Conv2d(x)```, dynamo handles it well(no graph break) and puts a ```call_module``` op in the FX graph. However, if users explicitly call ```Conv2d.forward(x)``` in another ```forward``` function, the inlining would be failed(caused graph break). This PR fixed this issue by translating the explicit ```Conv2d.forward(x)``` to ```Conv2d(x)```.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99015
Approved by: https://github.com/jansel, https://github.com/wconstab
Wrapper for users to insert constraints into model code.
The constraints will not be maintained in the graph after tracing through make_fx so retracing with dynamo/make_fx will not work. This will be supported after torch._assert supported is implemented. Then we can convert the constrain_range calls to torch._asserts.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98433
Approved by: https://github.com/avikchaudhuri, https://github.com/tugsbayasgalan
Wrapper for users to insert constraints into model code.
The constraints will not be maintained in the graph after tracing through make_fx so retracing with dynamo/make_fx will not work. This will be supported after torch._assert supported is implemented. Then we can convert the constrain_range calls to torch._asserts.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98433
Approved by: https://github.com/avikchaudhuri, https://github.com/tugsbayasgalan
Summary:
Replace _dynamo.config with an object instead of module
Current usage patterns of setting and reading fields on config will work
unchanged.
Only changes needed going forward:
1. import torch._dynamo.config will not work. However, just doing
import torch._dynamo is sufficient to access dynamo config
as torch._dynamo.config.
2. Files inside of _dynamo folder need to access config via
from torch._dynamo.config_util import config instead of
from torch._dynamo import config. Because _dynamo/__init__.py
imports some of the files so it would be circular import.
Test Plan:
Reviewers:
Subscribers:
Tasks:
Tags:
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/96455
Approved by: https://github.com/williamwen42
This PR makes basic nnmodule forward hooks work by default, without any overhead. But it leaves silent correctness issues if users modify/remove their hooks later, thus also emits a warning.
- the usual case is to not use hooks, so avoid guard overhead here
- registering any hook before compile will trigger a warning about hook support
- registering a hook later (or removing one) requires user knowledge and opting in,
currently this isn't warnable (but maybe we can observe compiled nnmodules to make it
warnable).
Why skip hook guards by default instead of not tracing __call__/hooks by default?
- avoid having a mode flag that alters dynamo tracing behavior (harder to test both codepaths
in CI with full coverage)
- the most basic hook usecase (registering a hook before compile, and never removing it)
will work by default with this PR, while it would require enablement and incur overhead
in the 'not tracing __call__' proposal.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98371
Approved by: https://github.com/jansel
This replaces fake_mode_from_tensors but it preferentially looks for
fake_mode in TracingContext and also if there is an active fake mode
on the dispatch stack, before groveling in tensors to find it.
This advances PegasusForCausalLM, which was previously failing because
we generated a graph that had a parameter (non-fake) and a SymInt,
and thus previously we failed to detect the correct fake mode.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98321
Approved by: https://github.com/voznesenskym
**Summary**: profiler.record_function inserts an event into the chrome trace generated by the pytorch profiler. This PR adds record_function everywhere that @dynamo_timed is annotated.
dynamo_timed and the CLI viewer torch._dynamo.utils.compile_times() are already useful on their own; but for identifying _when_ these get called, it's nice to be able to view in the profiler chrome trace.
Why not just turn on python stack traces in the profiler to get this information? Dynamo compilation is implemented in python and therefore produces a huge amount of events when it records compilation steps. The resulting trace files are often too large to load in chrome://tracing, and they take a long time to generate. Additionally, the stack traces are deep enough that they are often hard to read. This approach produces much more readable traces with lower overhead.
**Tests**:
- Added in test/dynamo/test_profiler.py. Verified in https://github.com/pytorch/pytorch/actions/runs/4559322864/jobs/8043307798?pr=96495 that the tests are actually running.
- Performance run with `ciflow/inductor-perf-compare` shows no noticeable change in compilation time or speedup numbers. Geomean speedup changes from 1.275 -> 1.277. Geomean compilation times change from 54.2s -> 53.8s. That's likely just due to noise. All individual benchmark numbers regressed by no more than 5% between the two runs; and we see improvements of around the same magnitude, suggesting this is, again, just noise. For meta employees, you can see the results in a google sheets here: https://docs.google.com/spreadsheets/d/1Ki69XvcgxcA3ZnqC5n_jav5KiD4u7Wojlad3VTnIdlk/edit?usp=sharing
**Example**:
Run this:
```python
import torch
def gn(x):
return x.sin().cos()
def fn(x, y):
return x.sin() * y.cos()
x, y = [torch.rand((2, 2), device='cuda') for _ in range(2)]
# just to clear out any lazy initialization
with torch.profiler.profile() as prof:
torch.compile(gn)(x)
with torch.profiler.profile() as prof:
torch.compile(fn)(x, y)
prof.export_chrome_trace("./dynamo_timed_profile.json")
```
and we can see that the resulting trace shows important dynamo steps, even when python tracing is turned off.
<img width="867" alt="Screenshot 2023-03-29 at 7 26 15 PM" src="https://user-images.githubusercontent.com/5067123/228712263-8ae67ab9-1a52-4765-a9c2-7c5cf0abe2f5.png">
Pull Request resolved: https://github.com/pytorch/pytorch/pull/96495
Approved by: https://github.com/ngimel, https://github.com/mlazos
Summary:
Adds NNC-like logging that is configured through an env var `TORCH_COMPILE_LOGS`
Examples:
`TORCH_LOGS="dynamo,guards" python script.py` - prints dynamo logs at level INFO with guards of all functions that are compiled
`TORCH_LOGS="+dynamo,guards,graph" python script.py` - prints dynamo logs at level DEBUG with guards and graphs (in tabular) format of all graphs that are compiled
[More examples with full output](https://gist.github.com/mlazos/b17f474457308ce15e88c91721ac1cce)
Implementation:
The implementation parses the log settings from the environment, finds any components (aot, dynamo, inductor) or other loggable objects (guards, graph, etc.) and generates a log_state object. This object contains all of the enabled artifacts, and a qualified log name -> level mapping. _init_logs then adds handlers to the highest level logs (the registered logs), and sets any artifact loggers to level DEBUG if the artifact is enabled.
Note: set_logs is an alternative for manipulating the log_state, but if the environment contains TORCH_LOGS, the environment settings will be prioritized.
Adding a new log:
To add a new log, a dev should add their log name to torch._logging._registrations (there are examples there already).
Adding a new artifact:
To add a new artifact, a dev should add their artifact name to torch._logging._registrations as well.
Additionally, wherever the artifact is logged, `torch._logging.getArtifactLogger(__name__, <artifact_name>)` should be used instead of the standard logging implementation.
[design doc](https://docs.google.com/document/d/1ZRfTWKa8eaPq1AxaiHrq4ASTPouzzlPiuquSBEJYwS8/edit#)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94858
Approved by: https://github.com/ezyang
OK, so this PR used to be about reducing the number of constants we specialize on, but it turns out that unspecialization was ~essentially never used (because we still constant specialized way too aggressively) and I ended up having to fix a bunch of issues to actually get tests to pass. So this PR is now "make int unspecialization actually work". As part of this, I have to turn off unspecialization by default, as there are still latent bugs in inductor.
The general strategy is that an unspecialized int is represented as a SymInt. Representing it as a 0d tensor (which is what the code used to do) is untenable: (1) we often need unspecialized ints to participate in size computations, but we have no way of propagating sympy expressions through tensor compute, and (2) a lot of APIs work when passed SymInt, but not when passed a Tensor. However, I continue to represent Numpy scalars as Tensors, as they are rarely used for size computation and they have an explicit dtype, so they are more accurately modeled as 0d tensors.
* I folded in the changes from https://github.com/pytorch/pytorch/pull/95099 as I cannot represent unspecialized ints as SymInts without also turning on dynamic shapes. This also eliminates the necessity for test_unspec.py, as toggling specialization without dynamic shapes doesn't do anything. As dynamic shapes defaults to unspecializing, I just deleted this entirely; for the specialization case, I rely on regular static shape tests to catch it. (Hypothetically, we could also rerun all the tests with dynamic shapes, but WITH int/float specialization, but this seems... not that useful? I mean, I guess export wants it, but I'd kind of like our Source heuristic to improve enough that export doesn't have to toggle this either.)
* Only 0/1 integers get specialized by default now
* A hodgepodge of fixes. I'll comment on the PR about them.
Fixes https://github.com/pytorch/pytorch/issues/95469
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95621
Approved by: https://github.com/jansel, https://github.com/Chillee
This PR allows us to reuse the static per tensor decision making we make at fake tensorification time. We can use this to avoid setting up dynamic dim guards later if the tensor was never a candidate.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95566
Approved by: https://github.com/ezyang
**Summary**: torch.nn.Module implementations previously did not support custom implementations of `__getattr__`; if a torch.nn.Module subclass implemented `__getattr__` and we tried to access an attribute that was expected to be present in `__getattr__`, dynamo would not check `__getattr__` and would error out with an AttributeError. This PR copies the functionality from UserDefinedObjectVariable into torch.nn.Module so that it also supports `__getattr__`
Example of a module which previously would fail:
```python
class MyMod(torch.nn.Module):
def __init__(self):
super().__init__()
self.custom_dict = {"queue": [torch.rand((2, 2)) for _ in range(3)]}
self.other_attr = torch.rand((2, 2))
def __getattr__(self, name):
custom_dict = self.custom_dict
if name in custom_dict:
return custom_dict[name]
return super().__getattr__(name)
def forward(self, x):
return x @ self.other_attr + self.queue[-1]
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94658
Approved by: https://github.com/yanboliang, https://github.com/jansel
Fixes https://github.com/pytorch/pytorch/issues/93890
We do the following:
1. fix __init__constructor for `AutocastModeVariable` with exisiting `mode` while copying
2. `resume_execution` is made aware of constant args (`target_values`), by storing said args in `ReenterWith`. To propagate between subgraphs (in straightline code), we also store the constant args in the downstream's `code_options["co_consts"]` if not already.
---
Future work:
1. handle instantiating context manager in non-inlineable functions. Simultaneously fix nested grad mode bug.
2. generalize to general `ContextManager`s
3. generalize to variable arguments passed to context manager, with guards around the variable.
---
Actually, if we look at the repro: 74592a43d0/test/dynamo/test_repros.py (L1249), we can see that the method in this PR doesn't work for graph breaks in function calls, in particular, in function calls that don't get inlined.
Why inlining functions with graph breaks is hard:
- When we handle graph breaks, we create a new code object for the remainder of the code. It's hard to imagine doing this when you are inside a function, then we need a frame stack. And we just want to deal with the current frame as a sequence of straight line codes.
Why propagating context manager information is hard:
- If we do not inline the function, the frame does not contain any information about the parent `block_stack` or `co_consts`. So we cannot store it on local objects like the eval frame. It has to be a global object in the output_graph.
---
Anyway, I'm starting to see clearly that dynamo must indeed be optimized for torch use-case. Supporting more general cases tends to run into endless corner-cases and caveats.
One direction that I see as viable to handle function calls which have graph breaks and `has_tensor_in_frame` is stick with not inlining them, while installing a global `ContextManagerManager`, similar to the `CleanupManager` (which cleans up global variables). We can know which context managers are active at any given point, so that we can install their setup/teardown code on those functions and their fragments.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94137
Approved by: https://github.com/yanboliang
Applies the remaining flake8-comprehension fixes and checks. This changes replace all remaining unnecessary generator expressions with list/dict/set comprehensions which are more succinct, performant, and better supported by our torch.jit compiler. It also removes useless generators such as 'set(a for a in b)`, resolving it into just the set call.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94676
Approved by: https://github.com/ezyang
I applied some flake8 fixes and enabled checking for them in the linter. I also enabled some checks for my previous comprehensions PR.
This is a follow up to #94323 where I enable the flake8 checkers for the fixes I made and fix a few more of them.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94601
Approved by: https://github.com/ezyang
Summary:
There are a few races/permission errors in file creation, fixing
OSS:
1. caffe2/torch/_dynamo/utils.py, get_debug_dir: multiple process may conflict on it even it's using us. Adding pid to it
2. caffe2/torch/_dynamo/config.py: may not be a right assumption that we have permission to cwd
Test Plan: sandcastle
Differential Revision: D42905908
Pull Request resolved: https://github.com/pytorch/pytorch/pull/93407
Approved by: https://github.com/soumith, https://github.com/mlazos
Previously, Dynamo faked support for item() when `capture_scalar_outputs` was True by representing it internally as a Tensor. With dynamic shapes, this is no longer necessary; we can represent it directly as a SymInt/SymFloat. Do so. Doing this requires you to use dynamic shapes; in principle we could support scalar outputs WITHOUT dynamic shapes but I won't do this unless someone hollers for it.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Differential Revision: [D42885775](https://our.internmc.facebook.com/intern/diff/D42885775)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/93150
Approved by: https://github.com/voznesenskym
for some tensor x, x.type(torch.FloatTensor) will essentially do the same thing as x.to(torch.float). x.type can be called with at least 3 types of inputs:
* a string "torch.FloatTensor"
* a dtype torch.float
* a tensor type torch.FloatTensor
the third option (torch.FloatTensor) fails in fx, because fx cannot trace torch.FloatTensor objects. So this PR will replace the torch.FloatTensor type with a string "torch.FloatTensor"
Why not fix this in fx? Well, it's possible, but I'm not sure a nice way to do it. We would want to update [torch.fx.node.BaseArgumentTypes](d88bc38b0c/torch/fx/node.py (L17)) to contain torch.FloatTensor etc. We could hard-code a list of tensor types there (the types vary depending on build type, e.g. whether or not cuda tensors are available), but that's not great in case our hardcoded list differs from the actual list registered by python_tensor.cpp. Another option is to dynamically populate the list of types with `Union[tuple(...)])`, and fill the tuple with `torch._tensor_classes` (which is directly populated by python_tensor.cpp), but apparently this breaks most typecheckers.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/93043
Approved by: https://github.com/jansel
The output of Torchbench model `doctr_det_predictor` on CPU is a `numpy ndarray`. When running the accuracy benchmark of this model, the below error is raised: `RuntimeError: unsupported type: ndarray`.
Repro CMD:
```bash
python benchmarks/dynamo/torchbench.py --accuracy --float32 -dcpu -n50 --inductor --no-skip --dashboard --only doctr_det_predictor --batch_size 1 --threads 1
```
This PR adds the support to compare `numpy ndarray` in the dynamo utils.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/91870
Approved by: https://github.com/jgong5, https://github.com/Chillee
When we run the node with fake value for tensor.item, it would previously error because the utility method doesn't know how to handle placeholder node. The tensor we are calling item can be input from user will be placeholder in the graph.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/91668
Approved by: https://github.com/voznesenskym
GraphArgs worked fairly well, but it was still missing sources
sometimes. Now, we maintain an auxiliary data structure which we
MUST populate whenever we fakeify a tensor / allocate a bare SymInt.
This should guarantee once and for all that every symbol is available.
Should fix swin_base_patch4_window7_224.
While I was at it, I moved fakeification utility back to builder
as it was only used at once call site.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90911
Approved by: https://github.com/voznesenskym
Instead of inferring shape mappings from a bunch of data structures that were plumbed in InstructionTranslator, we instead work out mappings by just iterating over the GraphArgs and mapping symbols to arguments as they show up. If multiple argument sizes/strides/offset map to the same symbol, this means they are duck sized, so we also generate extra equality tests that they must be equal. Finally, we generate 0/1 specialization guards. The resulting code is much shorter, and I think also easier to understand.
TODO: Delete all the tensor ref tracking code, it's unnecessary
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90528
Approved by: https://github.com/voznesenskym
Wow, I had to sweat so much to get this PR out lol.
This PR enforces the invariant that whenever we allocate SymInts as part of fakeification, the SymInt is associated with a Source, and in fact we store the string source name on SymbolWithSourceName. We use 'sname' as the shorthand for source name, as 'name' is already used by sympy to name symbols.
In order to store source names, we have to plumb source names from Dynamo to PyTorch. This made doing this PR a bit bone crushing, because there are many points in the Dynamo codebase where we are improperly converting intermediate tensors into fake tensors, where there is no source (and there cannot be, because it's a frickin' intermediate tensor). I've fixed all of the really awful cases in earlier PRs in the stack. This PR is just plumbing in source names from places where we do have it.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90295
Approved by: https://github.com/voznesenskym
The old code didn't actually fakeify traceable tensor subclasses at the
time they are added as a GraphArg to the module; now we do, by ignoring
the subclass during fakeification and relying on Dynamo to simulate
the subclass on top. See comments for more details.
BTW, this codepath is super broken, see filed issues linked on the
inside.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/90009
Approved by: https://github.com/wconstab, https://github.com/voznesenskym
In #87741 we added the inference support for dynamo/torchxla integration. Later on in #88449 we attempt to add the training support. That attempt is not smooth because
- we try 2 things together
1. let dynamo trace the model on xla rather than eager
2. enable training
- It turns out neither of these two tasks are trivial enough.
Furthermore, item 2 (enable training) depends on item 1 (tracing on xla). We enable training via AOTAutograd. AOTAutograd lift all model parameters/buffers as graph inputs. Without item 1 being done, we would need copy all graph inputs (including model parameters/buffers) from eager device to xla devices. That hurts performance a lot. Have a cache to map eager parameter to XLA parameter does not solve the problem since the update on either will not sync automatically to the other. They will easily go out of sync.
This PR let dynamo trace the model on XLA rather than eager. This is a preparation step to enabling training.
Also, tracing on XLA makes the data movement more efficient. We see 1.5x geomean speedup compared to previous 1.38x.
```
+-------------------------+--------------------+-------------------------+
| Model | XLA (trace once) | XLA (trace everytime) |
+=========================+====================+=========================+
| resnet18 | 1.38 | 1.008 |
+-------------------------+--------------------+-------------------------+
| resnet50 | 1.227 | 0.998 |
+-------------------------+--------------------+-------------------------+
| resnext50_32x4d | 1.544 | 1.008 |
+-------------------------+--------------------+-------------------------+
| alexnet | 1.085 | 1.045 |
+-------------------------+--------------------+-------------------------+
| mobilenet_v2 | 2.028 | 1.013 |
+-------------------------+--------------------+-------------------------+
| mnasnet1_0 | 1.516 | 0.995 |
+-------------------------+--------------------+-------------------------+
| squeezenet1_1 | 0.868 | 1.01 |
+-------------------------+--------------------+-------------------------+
| vgg16 | 1.099 | 1.008 |
+-------------------------+--------------------+-------------------------+
| BERT_pytorch | 3.26 | 1.027 |
+-------------------------+--------------------+-------------------------+
| timm_vision_transformer | 2.182 | 1.015 |
+-------------------------+--------------------+-------------------------+
| geomean | 1.50389 | 1.01261 |
+-------------------------+--------------------+-------------------------+
```
Example command
```
GPU_NUM_DEVICES=1 python benchmarks/dynamo/torchbench.py --randomize-input --performance --trace-on-xla --only resnet18 --backend=torchxla_trace_once
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88904
Approved by: https://github.com/wconstab, https://github.com/JackCaoG, https://github.com/jansel
**Introduces symbolic shape guards into dynamo.**
In this PR, we take the existing fake tensor infra and plumbing in dynamo and we start passing a shape_env around. This shape_env does not get plumbed down to middle layers / backend yet - it only collects expressions from frontend invocations at the moment. We then translate these expressions into guards at the point where we take other guards installed throughout dynamo - and add them to check_fn.
Part 1 of https://docs.google.com/document/d/1QJ-M4zfMkD-fjHIqW089RptjLl9EgozZGCceUbvmgfY/edit#
cc @jansel @lezcano @fdrocha @mlazos @soumith @yanboliang @penguinwu @anijain2305
Pull Request resolved: https://github.com/pytorch/pytorch/pull/87570
Approved by: https://github.com/ezyang
