Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37587
Lifting RecordFunction up into the dispatcher code
Test Plan: Imported from OSS
Differential Revision: D21374246
fbshipit-source-id: 19f9c1719e6fd3990e451c5bbd771121e91128f7
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/41507
These fields have always been a part of tensor types, this change just
makes them serializable through IR dumps.
Test Plan: Imported from OSS
Reviewed By: Krovatkin, ngimel
Differential Revision: D22563661
Pulled By: ZolotukhinM
fbshipit-source-id: f01aaa130b7e0005bf1ff21f65827fc24755b360
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39343
Building on top of previous PR that adds fused add_relu op, this PR adds
a JIT pass to transform input graph to find all fusable instancs of add
+ relu and fuses them.
Test Plan:
python test/test_jit.py TestJit.test_add_relu_fusion
Imported from OSS
Differential Revision: D21822396
fbshipit-source-id: 12c7e8db54c6d70a2402b32cc06c7e305ffbb1be
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40718
Currently only constant except tensor must be inlined during serialization.
Tensor are stored in the contant table. This patch generalizes this capability
to any IValue. This is particularly useful for non ASCII string literal that
cannot be inlined.
Test Plan: Imported from OSS
Differential Revision: D22298169
Pulled By: bzinodev
fbshipit-source-id: 88cc59af9cc45e426ca8002175593b9e431f4bac
Summary:
Noticed while trying to script one of the models which happened to have numpy values as constants. Lacking the numpy prefix in the error message was quite confusing.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/41024
Differential Revision: D22426399
Pulled By: dzhulgakov
fbshipit-source-id: 06158b75355fac6871e4861f82fc637c2420e370
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40902
See the bottom of this stack for context.
Test Plan: Imported from OSS
Reviewed By: eellison
Differential Revision: D22360210
Pulled By: suo
fbshipit-source-id: 4275127173a36982ce9ad357aa344435b98e1faf
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40939
Previously, when we would do shape analysis by running the op with representative inputs, we would always set the grad property to false. This led to a wrong static analysis when we would create differentiable subgraphs, and propagate shapes without also propagating requires_grad, and then uninline them.
Test Plan: Imported from OSS
Differential Revision: D22394676
Pulled By: eellison
fbshipit-source-id: 254e6e9f964b40d160befe0e125abe1b7aa2bd5e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40807
We pack a lot of logic into `jit/__init__.py`, making it unclear to
developers and users which parts of our API are public vs. internal. This
is one in a series of PRs intended to pull implementation out into
separate files, and leave `__init__.py` as a place to register the
public API.
This PR moves all the tracing-related stuff out, and fixes other spots up
as necessary. Followups will move other core APIs out.
The desired end-state is that we conform to the relevant rules in [PEP 8](https://www.python.org/dev/peps/pep-0008/#public-and-internal-interfaces). In particular:
- Internal implementation goes in modules prefixed by `_`.
- `__init__.py` exposes a public API from these private modules, and nothing more.
- We set `__all__` appropriately to declare our public API.
- All use of JIT-internal functionality outside the JIT are removed (in particular, ONNX is relying on a number internal APIs). Since they will need to be imported explicitly, it will be easier to catch new uses of internal APIs in review.
Test Plan: Imported from OSS
Reviewed By: eellison
Differential Revision: D22320645
Pulled By: suo
fbshipit-source-id: 0720ea9976240e09837d76695207e89afcc58270
Summary:
Define static script implementation of __len__ and __contains__ on any subclass derived from a type such as ModuleList, Sequential, or ModuleDict. Implement getitem for classes derived from ModuleDict.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40789
Reviewed By: eellison
Differential Revision: D22325159
Pulled By: wconstab
fbshipit-source-id: fc1562c29640fe800e13b5a1dd48e595c2c7239b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40753
The reference says that this op always returns a 1-D tensor, even if
the input and the mask are 0-D.
Test Plan: Imported from OSS
Reviewed By: eellison
Differential Revision: D22300354
Pulled By: ZolotukhinM
fbshipit-source-id: f6952989c8facf87d73d00505bf6d41573eff2d6
Summary:
Remove `skipIfRocm` from most jit tests and enable `RUN_CUDA_HALF` tests for ROCm.
These changes passed more than three rounds of CI testing against the ROCm CI.
CC ezyang xw285cornell sunway513
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40447
Differential Revision: D22190711
Pulled By: xw285cornell
fbshipit-source-id: bac44825a2675d247b3abe2ec2f80420a95348a3
Summary:
Partial support for slicing of Sequential containers.
- works around missing Sequential slice functionality
by converting to tuple
- only supports iteration of resulting tuple values,
not direct call() on the sliced sequential
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40445
Differential Revision: D22192469
Pulled By: wconstab
fbshipit-source-id: 61c85deda2d58f6e3bea2f1fa1d5d5dde568b9b5
Summary:
Currently, torchvision annotates `batched_nms` with `torch.jit.script` so the function gets compiled when it is traced and ONNX will work. Unfortunately, this means we are eagerly compiling batched_nms, which fails if torchvision isn't built with `torchvision.ops.nms`. As a result, torchvision doesn't work on torch hub right now.
`_script_if_tracing` could solve our problem here, but right now it does not correctly interact with recursive compilation. This PR fixes that bug.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40468
Reviewed By: jamesr66a
Differential Revision: D22195771
Pulled By: eellison
fbshipit-source-id: 83022ca0bab6d389a48a478aec03052c9282d2b7
Summary: NVIDIA's Apex is updating to no longer rely on this behavior, but we're reverting this Python2->Python3 update to unblock internal apex users.
Test Plan: Sandcaslte + OSS CI.
Reviewed By: ngimel
Differential Revision: D22146782
fbshipit-source-id: f9483d2cbf9dc3a469ad48a6c863edea3ae51070
Summary:
**Summary**
This commit adds support for with statements to PyTorch JIT. Each
of the with items in a with statement is represented in the JIT IR
as a pair of `prim::Enter` and `prim::Exit` nodes that call the
`__enter__` and `__exit__` methods defined on the context manager objects
returned by the expressions in the with item.
**Testing**
This commit adds unit tests for with statements with named with items,
nameless with items, and with statements that encounter exceptions.
```
$ python test/test_jit.py TestWith.test_with_as
Fail to import hypothesis in common_utils, tests are not derandomized
.
----------------------------------------------------------------------
Ran 1 test in 0.430s
OK
```
```
$ python test/test_jit.py TestWith.test_with_no_as
Fail to import hypothesis in common_utils, tests are not derandomized
.
----------------------------------------------------------------------
Ran 1 test in 0.264s
OK
```
```
$ python test/test_jit.py TestWith.test_with_exceptions
Fail to import hypothesis in common_utils, tests are not derandomized
Couldn't download test skip set, leaving all tests enabled...
.
----------------------------------------------------------------------
Ran 1 test in 1.053s
OK
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34705
Differential Revision: D22095945
Pulled By: SplitInfinity
fbshipit-source-id: f661565a834786725259b8ea014b4d7532f9419d
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40144
as title, split remaining quantization tests out of test_jit to reduce
the size of test_jit
Test Plan: Imported from OSS
Differential Revision: D22085034
Pulled By: wanchaol
fbshipit-source-id: 0c8639da01ffc3e6a72e6f470837786c73a6b3f0
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/40142
test_jit is becoming huge again, which makes editor hard to load and
write new tests, this split out the tracer related tests.
Test Plan: Imported from OSS
Reviewed By: ailzhang
Differential Revision: D22085035
Pulled By: wanchaol
fbshipit-source-id: 696bee84985ecfbfeac8e2ee5c27f1bdda8de394
Summary:
BC-breaking note:
If a user is using one of these dunders directly they will not longer be available. Users should update to Python3 compatible dunders.
Original PR note:
`__div__` (and `__idiv__` and `__rdiv__`) are no longer special dunders in Python3. This PR replaces them with the `__truediv__` (`__itrudediv__`, `__rtruediv__`) dunders, since we no longer support Python2.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39151
Differential Revision: D22075713
Pulled By: mruberry
fbshipit-source-id: d318b47b51f7cc4c3728b1606a34d81e49ba0fa1
Summary:
Remove PY3 and PY34 checks from `torch/testing/_internal/common_utils.py`
Remove PY35 global var from `torch.jit.annotations`
Always call `try_get_real_signature` in `torch/jit/annotations.py`
Use `map` instead of `imap`, since Python-2 is no longer support, so map is always lazy.
Remove all pre Python-3.6 checks from `torch/_six.py` and `torch/_appdirs.py`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39879
Differential Revision: D22037811
Pulled By: malfet
fbshipit-source-id: af0c79f976569c2059d39ecb49c6b8285161734f
Summary:
Clearly expressing a type is inferred by PyTorch instead of explicitly annotated by user makes many error messages more user-friendly
Currently Type has two string conversion methods. str() for IR printing and python_str() for serialization and error message generation. If we want to include more information in type printing while maintaining serialization/deserialization correctness, we need to split python_str() into annotation_str() and repr_str().
annotation_str is solely responsible for serialization, it strictly matches format of python type annotation. repr_str() is responsible for generating a human-readable error message that includes information like "this type is inferred, not explicitly annotated"
Closes https://github.com/pytorch/pytorch/issues/39449
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39544
Differential Revision: D21978759
Pulled By: gmagogsfm
fbshipit-source-id: 733566f5a62e748b5ca4bb3c5943ebb6d5b664d0
Summary:
Currently torch.Tensor subclasses (like torch.nn.Parameter) isn't a supported type annotation to torch script inputs. This PR allows it to be treated like torch.Tensor for compilation.
Closes https://github.com/pytorch/pytorch/issues/38235
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39487
Differential Revision: D21885827
Pulled By: gmagogsfm
fbshipit-source-id: 1ec51829b132b7b0293a6c526d73497b23dae113
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39483
I fixed all of the new errors that occurred because of the upgrade.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Differential Revision: D21884575
Pulled By: ezyang
fbshipit-source-id: 45c8e1f1ecb410c8d7c46dd3922ad70e982a0685
Summary:
Previously, on conversion from python -> c++ it was casted to double list through bad copy pasta. It's pretty unusual for someone to script a broadcasting list function directly since it's an internal api, so it was unlikely to affect anyone.
Fix for https://github.com/pytorch/pytorch/issues/39450
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39481
Reviewed By: jamesr66a
Differential Revision: D21870557
Pulled By: eellison
fbshipit-source-id: e704e5e87d2702a270b7d65c4df444246a134480
Summary:
**Summary**
This commit adds support for seralization and deserialization of
`ScriptModules` that have been lowered to a specific backend. Nothing
special was required to accomplish this, other than removing some code
in `unpickler.cpp` that guarded against the deserialization of `Any`
type objects. Now that lists and dicts are tagged with their types
during serialization, this check is no longer necessary.
**Test Plan**
This commit adds a unit test for testing that a lowered module still
produces the same results as Python and regular JIT after saving and
loading.
**Fixes**
This pull request fixes part of https://github.com/pytorch/pytorch/issues/37841.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38893
Differential Revision: D21825813
Pulled By: SplitInfinity
fbshipit-source-id: 77a7b84504e0dddf14c89b3ed5dd6b438c086f66
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38527
This PR solves issue (#37200).
Error is encountered during IR generation while trying to resolve the call to sum.
Should let user know it inferred the value for argument 'dim' to be of type 'Tensor'
because it was not annotated with an explicit type.
Test Plan:
Add code to reprodue the issue (#37200)
`python test/test_jit.py TestJit.test_inferred_as_tensor`
Differential Revision: D21743876
Pulled By: superwizard2019
fbshipit-source-id: 370ca32afea4d53b44d454f650f7d3006f86bcc6
Summary:
This updates assertEqual and assertEqual-like functions to either require both or neither of atol and rtol be specified. This should improve clarity around handling precision in the test suite, and it allows us to remove the legacy positional atol argument from assertEqual. In addition, the "message" kwarg is replace with a kwarg-only "msg" argument whose name is consistent with unittest's assertEqual argument.
In the future we could make "msg" an optional third positional argument to be more consistent with unittest's assertEqual, but requiring it be specified should be clear, and we can easily update the signature to make "msg" an optional positional argument in the future, too.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38872
Differential Revision: D21740237
Pulled By: mruberry
fbshipit-source-id: acbc027aa1d7877a49664d94db9a5fff91a07042
Summary:
This updates assertEqual and assertEqual-like functions to either require both or neither of atol and rtol be specified. This should improve clarity around handling precision in the test suite, and it allows us to remove the legacy positional atol argument from assertEqual. In addition, the "message" kwarg is replace with a kwarg-only "msg" argument whose name is consistent with unittest's assertEqual argument.
In the future we could make "msg" an optional third positional argument to be more consistent with unittest's assertEqual, but requiring it be specified should be clear, and we can easily update the signature to make "msg" an optional positional argument in the future, too.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38872
Differential Revision: D21717199
Pulled By: mruberry
fbshipit-source-id: 9feb856f94eee911b44f6c7140a1d07c1b026d3a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38746
Factors out testing of op alias normalization so that there is a registry used for tests.
Test Plan: Imported from OSS
Differential Revision: D21673107
Pulled By: eellison
fbshipit-source-id: e06653cdf24f14a4253dd054e4d402d171d16a11
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38735
Follow up to my comment https://github.com/pytorch/pytorch/pull/36597/#issuecomment-613674329
This adds a pass to convert op aliases into a normalized form. Having two ops generated in our IR that do the same thing makes the IR harder for downstream consumers of the IR, such as TorchScript passes but also ONNX, glow, etc.
Another solution would have been to fix our code generation to only emit `aten::abs` from the start. This seems trickier, and doesn't really buy us much if we still have to expose `aten::absolute` in C++, as glaringlee of the C++ API thinks we should.
Bike shedding: maybe this should be `CanonicalizeOps` instead
Test Plan: Imported from OSS
Differential Revision: D21673108
Pulled By: eellison
fbshipit-source-id: c328618907de1af22e07f57fd27fa619978c2817
Summary:
Updates our tests in preparation of integer division using torch.div and torch.addcdiv throwing a runtime error by avoiding integer division using torch.div. This creates a brief period where integer division using torch.div is untested, but that should be OK (since it will soon throw a runtime error).
These callsites were identified using https://github.com/pytorch/pytorch/issues/36897.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38621
Differential Revision: D21612823
Pulled By: mruberry
fbshipit-source-id: 749c03a69feae02590b4395335163d9bf047e162
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38453
Two fixes:
- RecordFunction in JIT interpreter should exist during the execution
of the frame, and not just when we enter the frame
- When creating a JIT continuation in wait instruction, we'd want to
preserve the original thread local context, right now when we resume
execution in continuation we preserve the thread local state of the
thread that set future value (i.e. executed a forked task)
Test Plan: unittest, CI
Reviewed By: ngimel
Differential Revision: D21565959
Pulled By: ilia-cher
fbshipit-source-id: 206b98e3bfb0052fc8e4031da778e372cc71afc1
Summary:
After an early return, we conditionalize all further execution. This means that currently the pattern of
`if return elif return elif return` generates better code than `if return if return if return`. It's obviously not good to have semantically equivalent code generate worse IR, so we should rewrite the graph to handle this case. This came up in https://github.com/pytorch/pytorch/pull/37171
```
torch.jit.script
def test_foo(x: bool, y: bool):
if x:
return 1
return 2
print(test_foo.code)
```
generates:
```
def test_foo(x: bool,
y: bool) -> int:
_0 = uninitialized(int)
if x:
_1, _2 = True, 1
else:
_1, _2 = False, _0
if _1:
_3 = _2
else:
_3 = 2
return _3
```
while
```
torch.jit.script
def test_foo(x: bool, y: bool):
if x:
return 1
else:
return 2
print(test_foo.code)
```
generates:
```
def test_foo(x: bool,
y: bool) -> int:
if x:
_0 = 1
else:
_0 = 2
return _0
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38282
Differential Revision: D21576733
Pulled By: eellison
fbshipit-source-id: 80cf1ad7fbda6d8d58557abbfb21c90eafae7488
Summary:
TorchScript currently doesn’t support `*args, **kwargs` in method signature, which is extensively used in DPER3 low-level modules’ forward method. In order to make DPER3 low-level modules scriptable, I was thinking about a solution of having a forward method *only* for TorchScript, and replace the forward method when we are not in scripting mode.
This solution works today, and I would like to add a test to make sure it will always work in the future.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38158
Differential Revision: D21485657
Pulled By: yf225
fbshipit-source-id: df7368e8a5265418be7c305e6666ffd76e595466
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35623
Python 2 has reached end-of-life and is no longer supported by PyTorch.
This test case is valid syntax in Python 3.
Test Plan: CI
Differential Revision: D20842874
Pulled By: dreiss
fbshipit-source-id: 9f12e046f827d4f9d5eca99b0b0b46f73e06ff51
Summary:
Previously, we weren't adding the location to implicit conversions, so the error message wouldn't show location when these ops failed.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38442
Differential Revision: D21563500
Pulled By: eellison
fbshipit-source-id: 19dd786ab8580f11ed919aac669efeed0ef52dcb
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37994
Before, reassigning a method in a module (like `forward = _forward`)
didn't work, because we look at the function object's name for our def
name when building AST. Mkae that overrideable to handle cases like
reassignment
Test Plan: Imported from OSS
Differential Revision: D21444535
Pulled By: suo
fbshipit-source-id: 4f045f18b5a146edc8005689af525d7d7ed8dd5f
Summary:
Fix for https://github.com/pytorch/pytorch/issues/37986
Follows the stack in https://github.com/pytorch/pytorch/pull/33783 stack to make functions in `torch/functional.py` resolve to their python implementations. Because the return type of `torch.unique` depends on `return_inverse` and `return_counts` I had to refactor the implementation to use our boolean_dispatch mechanism.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38156
Differential Revision: D21504449
Pulled By: eellison
fbshipit-source-id: 7efb1dff3b5c00655da10168403ac4817286ff59
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38253
This pass removes dropout and dropout_ nodes when training is false. It
requires to have run freeze_module pass which does both inlining and constant
propagation, without which training variable remains as attribute instead of
constant.
ghstack-source-id: 103939141
Test Plan: python test/test_jit.py TestScript.test_remove_dropout
Reviewed By: dreiss
Differential Revision: D21505863
fbshipit-source-id: 42ea45804e4653b625b6a254c8d8480757264aa8
Summary:
Followup of https://github.com/pytorch/pytorch/issues/37848 I realized that it's better to condition on `Value` type instead of token type. So now it also support indexing through list variables (used to be list literal only).
Also apparently our eager frontend accept indexing with float list as well, so matched this edge case behavior as well.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37966
Reviewed By: suo
Differential Revision: D21439642
Pulled By: ailzhang
fbshipit-source-id: cedb8431ef38747d4aa9909a6bbf8e954dbe0e25
Summary:
**Summary**
This commit adds `torch::jit::RegisterBackend`, an API that allows
external backends to be registered for the execution of JIT subgraphs
outside the JIT interpreter. In order to register an external backend,
one must extend the provided abstract class `PyTorchBackendInterface` and provide
two additional functions: one that creates an instance of the aforementioned subclass
of `PyTorchBackendInterface`, and another that preprocesses a `ScriptModule` so that
it can run on the backend. Then, a `ScriptModule` that can compile and execute a given
JIT subgraph using the functions provided at registration time is generated
for each registered backend.
**Testing**
This commit adds a unit test that uses a minimal test backend
to make sure that the registration endpoint and generated
`ScriptModule` work.
```
$ python test/test_jit.py TestBackends
Fail to import hypothesis in common_utils, tests are not derandomized
.
----------------------------------------------------------------------
Ran 1 test in 0.183s
OK
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35833
Differential Revision: D21231955
Pulled By: SplitInfinity
fbshipit-source-id: 452db1123d0e5d83f97fe5da8a00fdfdb50dbef9
Summary:
this is failing in the profiling_executor job
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37961
Differential Revision: D21434341
Pulled By: eellison
fbshipit-source-id: b34f94b1595ef6f6edee76cd200f951a2ef21f22
Summary:
The existing contextmanager only conditionally enabled_profiling_mode, which was counter intuitive. When we changed the default executor it broke internal benchmarking as a result.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37825
Differential Revision: D21404611
Pulled By: eellison
fbshipit-source-id: 306b3c333ef4eb44ab6a6e5ab4e0682e5ce312ce
Summary:
We used to only support indexing through
- numbers like `x[0, 1]`
- tuple like `x[(0, 1)]`
- tensor like `x[torch.tensor([0, 1])]`
This PR adds support for indexing through list which is equivalent to tensor.
- `x[[0, 1, 5]]`
- `x[[0, 1], [0, 1]]`
- `x[[[0, 1], [0, 1]], [[0, 1], [0, 1]]]`
Note for `x[[0, 1, 5]]` we had a bug in AST conversion code so we used to treat it like `x[0, 1, 5]` which means it might accidentally run and produce wrong result(fixes https://github.com/pytorch/pytorch/issues/37286 fixes https://github.com/pytorch/pytorch/issues/18616), now that it's fixed we probably want to mark it as BC breaking.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37848
Reviewed By: suo
Differential Revision: D21409840
Pulled By: ailzhang
fbshipit-source-id: 6f2d962885c6dc009cb384d98be1822f5ca7a189
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37842
Fixes https://github.com/pytorch/pytorch/issues/23993.
Previously our name lookup function for the tracer was looking in
f.globals for names. For example:
```
sample1 = torch.ones(1)
sample2 = torch.ones(1)
traced = torch.jit.trace(my_mod, ((sample1, sample2,),))
> produces a graph with something like:
> %sample1, %sample2 = prim::TupleUnpack(%input)
```
This is not great if you are, e.g. trace checking, because a non-local
bit of interpreter state is affected the graph produced:
```
traced = torch.jit.trace(my_mod, _clone_inputs((sample, sample,),))
> produces a graph with something like
> %0, %1 = prim::TupleUnpack(%input)
```
I have removed this functionality, as I don't think it provides huge
value. Things that look locally for names will still work, so e.g.
inputs, intermediate variables, and the like will be named correctly.
Test Plan: Imported from OSS
Differential Revision: D21406478
Pulled By: suo
fbshipit-source-id: 3c7066b95d4a6e9b528888309954b02dadbc1a07
Summary:
We were previously only looking at class attributes, so that didn't include methods etc, and would silently give wrong semantics. This makes hasAttr go through the same resolution as our other attribute lookups.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37424
Differential Revision: D21282633
Pulled By: eellison
fbshipit-source-id: 8e970f365c2740d137a02331739c2ed93747b918
Summary:
as a part of moving to the dynamic shapes we are now passing `frame_id` to each profiling callback. The implementation of that requires copying profiling callbacks into Interpreter, so `first`s are actually different for every run. The dynamic shapes merging algorithm won't be using `first`, but in the meantime, while we get there, this should be a good enough fix.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36806
Differential Revision: D21307173
Pulled By: Krovatkin
fbshipit-source-id: 7dade56ebcc72ebd40bb7f3d636c7b83c99b628f
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37464
Fixes https://github.com/pytorch/pytorch/issues/23993.
There are two fixes here:
1. Previously our name lookup function for the tracer was looking in
f.globals for names. For example:
```
sample = torch.ones(1)
traced = torch.jit.trace(my_mod, ((sample, sample,),))
# produces a graph with something like
# %sample, %sample = prim::TupleUnpack(%input)
```
This is not great if you are, e.g. trace checking, because a non-local
bit of interpreter state is affected the graph produced:
```
traced = torch.jit.trace(my_mod, _clone_inputs((sample, sample,),))
# produces a graph with something like
# %0, %1 = prim::TupleUnpack(%input)
```
I have removed this functionality, as I don't think it provides huge
value. Things that look locally for names will still work, so e.g.
inputs, intermediate variables, and the like will be named correctly.
2. Previously, our input cloning for trace checking didn't do a memoized
deep copy. So:
```
_clone_inputs((sample, sample, sample))
```
produces a tuple with three non-aliased tensors. That's wrong! Use
copy.deepcopy with a memoization argument to fix this.
Test Plan: Imported from OSS
Differential Revision: D21297549
Pulled By: suo
fbshipit-source-id: 981d5879a4a244520dd68489767129ff357f1497
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37012
Removes an if statement in `torch.nn.functional.affine_grid`
Test Plan: Imported from OSS
Differential Revision: D21160755
Pulled By: eellison
fbshipit-source-id: 8b030936c9fbdb05b44abc9f254805d102f2acc2
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36953
Add support for generic lists as a constant. generic dicts & tuples are already implemented. This is a pretty common pattern and cuts down on the number of non-tensor nodes executed in interpolate tests.
Test Plan: Imported from OSS
Differential Revision: D21160761
Pulled By: eellison
fbshipit-source-id: 1e6b7b25b7580f09067794772d44e615601c60c4
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/37088
For an inlined expression tree like `(e_0, (e_1, e_long))` the previous
algoritm only scanned the same statement as `e_long`, splitting the
inlined expressions across lines. Because it did not scan `e_0`, `e_0`
would still get emitted inline, causing it to reverse order with `e_1` and
`e_long`. The new algorithm scans starting at `e_long` and going all
the way back up the expression until it reaches the end of the inlined
statement. Caching of what has already been scanned has been added so that
if there was a second long long `e_long2` after `e_long`, it would not
rescan and re-inline the statements that were already split.
Test Plan: Imported from OSS
Differential Revision: D21180394
Pulled By: zdevito
fbshipit-source-id: 4d142c83a04c89a47d04282f67a513f82cf153c0
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35615
Python 2 has reached end-of-life and is no longer supported by PyTorch.
Now we can clean up a lot of cruft that we put in place to support it.
These changes were all done manually, and I skipped anything that seemed
like it would take more than a few seconds, so I think it makes sense to
review it manually as well (though using side-by-side view and ignoring
whitespace change might be helpful).
Test Plan: CI
Differential Revision: D20842886
Pulled By: dreiss
fbshipit-source-id: 8cad4e87c45895e7ce3938a88e61157a79504aed
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36188
* Need to remove n^2 behavior for scanning whether to split or not
otherwise long inline chains will take a long time re-scanning.
Test Plan: Imported from OSS
Differential Revision: D20907254
Pulled By: zdevito
fbshipit-source-id: ebfc1a4eefc26d5806381e7afd75b7a9cd4cde97
Summary:
Our test suite used to set double as its default scalar type, and when it was switched to not do so (to be more consistent with how users experience PyTorch), a few tests had to still set the default scalar type to double to function properly. Now that the jit no longer creates double tensors so frequently, it appears that test_jit no longer needs to set double as its default scalar type, too.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36982
Differential Revision: D21152120
Pulled By: mruberry
fbshipit-source-id: ea6d3c1ad55552dc5affa1fe1bd0e5189849e6d7
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34258
This PR allows both atol and rtol to be specified, uses defaults based on the prior analysis (spreadsheet attached to https://github.com/pytorch/pytorch/pull/32538), but retains the absolute tolerance behavior in cases where precision was previously specified explicitly.
Test Plan: Imported from OSS
Differential Revision: D21110255
Pulled By: nairbv
fbshipit-source-id: 57b3a004c7d5ac1be80ee765f03668b1b13f4a7e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36696
This PR add dictionary as a supported output of tracer under the strict
flag.
Test Plan: Imported from OSS
Reviewed By: houseroad
Differential Revision: D21056962
Pulled By: wanchaol
fbshipit-source-id: ace498182d636de853cf8a1efb3dc77f5d53db29
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36727
Looks like this was renamed by accident in 0cbd7fa46f
Test Plan:
Unit test.
Lint.
Differential Revision: D21076697
Pulled By: dreiss
fbshipit-source-id: dbd18cb41c7b26479984a7a7b12ad41a4c5b7658
Summary:
Previously we were always creating a double tensor from `torch.tensor(1.)`, whereas python eager uses the current default dtype. Fix for https://github.com/pytorch/pytorch/issues/36369
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36587
Differential Revision: D21043617
Pulled By: eellison
fbshipit-source-id: 38da303594f52e06941d86b6e57c4a06e7d36938
Summary:
With https://github.com/pytorch/pytorch/pull/35562, we are running peephole optimization on inlining to reduce the number of nodes that are copied.
The tracer encodes the sizes in the graph like:
```
graph(%0 : Double(7)):
%1 : Function = prim::Constant[name="tensor_size"]()
%2 : Tensor = prim::CallFunction(%1, %0)
return (%2)
```
however people would like to reuse the graph with different shapes so running size invalidations would invalidate that. long term it might be better for the tracer to not include shape information but there are downstream users of that.
Separates out FuseAddMM from peephole so that now there is a single `disable_size_optimizations` parameter, and onnx explicitly invokes fuseaddmm.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36404
Differential Revision: D20968974
Pulled By: eellison
fbshipit-source-id: 56f8f1699e3b0adeeccdfd5a67bb975fd41a2913
Summary:
Previously we were copying the bound method of the original class to the
new script module class, which causes `self` to be wrong. This PR
changes it so we fetch the unbound function, then bind it to the new
script module, then attach it to the module.
Fixes#28280
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36546
Pulled By: driazati
Differential Revision: D21023329
fbshipit-source-id: 6b3f8404700860151792f669a9c02fbd13365272
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36277
This PR introduce a flag to the tracer that guard the risky behaviors
like adding list/dict as output of the tracer. Currently to ensure not
BC breaking user, we throw warning if the tracer output is list, and
will throw error when the tracer output is dict to enforce using this
flag (next PR)
Test Plan: Imported from OSS
Differential Revision: D20998157
Pulled By: wanchaol
fbshipit-source-id: 0d2c55f1a263a48b1b92dd6ad54407815e0a6f72
Summary:
To make them compatible with python3.7 and python3.8
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36422
Test Plan: CI
Differential Revision: D21006399
Pulled By: malfet
fbshipit-source-id: 725df277ff3e4479fc2c39d16a30fbf301fde9e5
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36411
This PR remove pytorch specific defined assertwarns and use the unit
test one, also format some tests
Test Plan: Imported from OSS
Differential Revision: D20998159
Pulled By: wanchaol
fbshipit-source-id: 1280ecff2dd293b95a639d13cc7417fc819c2201
Summary:
AnyType wasn't listed as a mutable type, so the assertion triggered (yay!). Also update the `isMutableTypeInternal(from) != isMutableTypeInternal` logic to be more encompassing.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36178
Differential Revision: D20922356
Pulled By: eellison
fbshipit-source-id: 7060a62b18e98dc24b6004a66225c196aadb566e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35170
Looks like this was renamed by accident in 0cbd7fa46f
Test Plan:
Unit test.
Imported from OSS
Differential Revision: D20783298
fbshipit-source-id: 8fcc146284af022ec1afe8d651baf6721b190ad3
Summary:
In a comprehension like:
```
def f()->int:
i = 1
x = [i for i in range(7)]
return i
```
the variables inside the comprehension do not write to the function environment.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36105
Differential Revision: D20880699
Pulled By: eellison
fbshipit-source-id: 40af0f7470e0baeff7ef158cb461bf85c816d169
Summary:
Resubmit of https://github.com/pytorch/pytorch/pull/35424, only this time I run optimizations in the right order so the PR description is actually true.
This speeds up the inlining pass of FairSeq model from 180s -> 13s, and MaskRCNN model from 5s -> 1.5s.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35562
Differential Revision: D20738922
Pulled By: eellison
fbshipit-source-id: 1439cf9d1f0bc780e2d64a744694f8b3b7ba4b70
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35834
This handles the cases we did not handle before in AND and OR statements:
static_true || <unknown> -> static_true
static_false && <unknown> -> static_false
Test Plan: Imported from OSS
Differential Revision: D20801125
Pulled By: zdevito
fbshipit-source-id: 0ef94c3a14c7af91580fc5248a4ccfd9e8d6d481
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35720
When modules are saved, all relevant types are serialized according to
their qualified name with a compilation unit. Since qualified names are
guaranteed to be unique within a compilation unit, this normally works
fine.
On load, all types are registered in a compilation unit owned by the
script::Module. Type names are not unique across compilation units, so
if you load two modules with colliding type names, make them submodules
of yet another module, and save that module, there is the potential of a
name collision. See the added tests for examples if that description is
confusing.
The solution is to unique type names when serializing code by mangling
them if we detect a name collision.
Test Plan: Imported from OSS
Differential Revision: D20749423
Pulled By: suo
fbshipit-source-id: a8827ff1d4a89f3e7964dbbb49b4381863da3e6a
Summary:
This is a follow up from https://github.com/pytorch/pytorch/pull/34520, which removed specialized list ops. This removes templating from list ops.
it also has one minor other change, which is to move `aten::len(t[]) -> int` to `aten::len(Any[]) -> int` so that heterogenous tuples can be called with `len()`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35768
Differential Revision: D20772943
Pulled By: eellison
fbshipit-source-id: bc36a00920bc94ca8c5aa9eb7d5d7a640388ffbb
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34674
Two changes to make sure the op_names dumped in export_opnames() are consistent to what are actually used in bytecode.
* Inline graph before dumping the operator names.
* Use code of the graph (which is used in bytecode) instead of the nodes of graph.
Test Plan: Imported from OSS
Differential Revision: D20610715
Pulled By: iseeyuan
fbshipit-source-id: 53fa9c3b36f4f242b7f2b99b421f4adf20d4b1f6
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33020
This is a pass to create functional blocks. The other PRs in the stack help avoid some of the limitations that are are often found in graphs. It's possible that this would work well with a graph that is frozen. Follow up work items that will help this pass:
- We don't currently have any capacity in alias analysis to tell whether a Value that came from the wildcard set "re-escapes" back into the wildcard set.
- More comments on the semantics of the graph and correctness conditions
- We could consider using dynamic dag if the perf of this is a limitation.
- potential make Functional Graphs Functional Blocks instead, so that we do not repeatedly copy constants, also to make IR read easier.
Test Plan: Imported from OSS
Differential Revision: D20603188
Pulled By: eellison
fbshipit-source-id: 6822a6e65f4cc2676f8f6445fe8aa1cb858ebeeb
Summary:
Stacked PRs
* #34938 - [jit] Remove stray `script`
* **#34935 - [jit] Add lazy script decorator**
Some users maintain libraries of code that is largely trace-able but not
script-able. However, some functions may need to be `torch.jit.script`ed if
they contain control flow so the tracer will use the compiler version.
This however impacts library start up time as in #33418, so this PR adds
a workaround in the form of a `torch.jit._lazy_script_while_tracing`
that will only initialize the compiler if the function is called while
actually tracing.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34935
Pulled By: driazati
Differential Revision: D20569778
fbshipit-source-id: d87c88c02b1abc86b283729ab8db94285d7d4853
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34804
We want to replicate the quantize node for return values in blocks of prim::If
in order to create the quantization patterns.
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D20524453
fbshipit-source-id: 2268ac555f646158f4e1ffc98ccc8101d7504194
Summary:
add `id` function so to give uses a way of keeping a `seen` set of nn modules.
n practice, this is only used between values of `T` and `T` or `T` and `Optional[T]`, so in this implementation I made it so that None is the only value that can be zero. Python also only guarantees `id()` gives semantically meaningful results for pointer types.
EDIT: now only allowing id on class types
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34975
Reviewed By: driazati
Differential Revision: D20599564
Pulled By: eellison
fbshipit-source-id: 3c6666a9b9b0258198adc70969dd6332e3375e4f
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35221
When weight is reused, we only need to insert one observer for the weight
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D20602492
fbshipit-source-id: e003e6316f6615f3526f0d00fb7b722148b4749b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34411
To make sure dequantize and the node that uses the dequantized value reside in the
block, so that we can do quant fusion
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D20519603
fbshipit-source-id: 3e4c68d0a73142716e19ea6a64ae3a5d6d51fa41
Summary:
This PR adds a preprocessing step in Conv2dBatchNorm folding.
It traverses the module to check if the bias of Conv2d module is set to
None. If it is, it assume that this a traced module and insert
Optional[Tensor] type bias.
Furthermore it insert getAttr for bias in the forward graph and fixes
_convolution op to take values from getAttr.
It also fixes parametere extraction from BN module which may not
have weight and bias attributes if affine was set to False. In scripted
mode such a BN module will get weight and bias attributes set to None.
For the case of eps it gets const propagated in tracing. This is also
fixed.
Few tests cases are added.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34932
Test Plan:
python test/test_jit.py TestJit.test_foldbn_trivial
python test/test_jit.py TestJit.test_foldbn_trivial_nobias
python test/test_jit.py TestJit.test_foldbn_in_submodule
python test/test_jit.py TestJit.test_foldbn_shared_classtype
python test/test_jit.py TestJit.test_foldbn_complex_cases
python test/test_jit.py TestJit.test_nofoldbn_complex_cases
Differential Revision: D20536478
Pulled By: kimishpatel
fbshipit-source-id: 4e842976a380d0575a71001bb4481390c08c259e
Summary:
add `id` function so to give uses a way of keeping a `seen` set of nn modules.
n practice, this is only used between values of `T` and `T` or `T` and `Optional[T]`, so in this implementation I made it so that None is the only value that can be zero. Python also only guarantees `id()` gives semantically meaningful results for pointer types.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34975
Differential Revision: D20549677
Pulled By: eellison
fbshipit-source-id: cca5ed4ef013f7540f93abf49f91f9830dfdca14
Summary:
https://github.com/pytorch/pytorch/issues/34563 accidentally introduced a lint error due to an unused import. This PR removes this import.
Jit tests run as expected after this change:
```
> python test/test_jit.py
.....
Ran 2435 tests in 100.077s
OK (skipped=140, expected failures=1)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34778
Differential Revision: D20459708
Pulled By: tugrulince
fbshipit-source-id: bb742085fafc849ff3d9507d1557556e01fbeb4b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34671
Like the python arg parser, this tries to convert to the schema in order.
It introduces schema_match_exception which gets thrown when the schema doesn't match,
allowing the overload handler to try the next option.
Behavior will not 100% match the schema argument parser but should work for
simple cases using custom binding.
Test Plan: Imported from OSS
Differential Revision: D20432206
Pulled By: zdevito
fbshipit-source-id: 280839a2205ea3497db3a9b5741fccc1e2bff9a8
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34190
inplace modification of ClassType might affect other tests, so we want to do non-inplace modifications.
Actually the inplace argument will be removed soon.
Test Plan:
ci
Imported from OSS
Differential Revision: D20451765
fbshipit-source-id: e87ad528c4e7f84f5774b94a8e3e85568269682d
Summary:
With this PR, we can now support left and right shift operators in the JIT engine for <int, int> and <Tensor, int>.
Updated tests pass as expected:
```
> python test/test_jit.py
...
Ran 2427 tests in 84.861s
OK (skipped=139, expected failures=1)
```
Running the following code with Python results in the output below:
```
> cat ~/expressions.py
import torch
torch.jit.script
def fn(a, b):
# type: (int, int)
return (
a << b, # supported
b >> a, # supported
a & b,
a | b,
a ^ b
)
print(fn.graph)
```
```
> python ~/expressions.py
graph(%a.1 : int,
%b.1 : int):
%4 : int = aten::leftshift(%a.1, %b.1) # /home/ince/expressions.py:7:8
%7 : int = aten::rightshift(%b.1, %a.1) # /home/ince/expressions.py:8:8
%10 : int = aten::__and__(%a.1, %b.1) # /home/ince/expressions.py:9:8
%13 : int = aten::__or__(%a.1, %b.1) # /home/ince/expressions.py:10:8
%16 : int = aten::__xor__(%a.1, %b.1) # /home/ince/expressions.py:11:8
%17 : (int, int, int, int, int) = prim::TupleConstruct(%4, %7, %10, %13, %16)
return (%17)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34563
Differential Revision: D20434209
Pulled By: tugrulince
fbshipit-source-id: 886386c59755106e17b84778b8e495b80a6269cd
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/33927
Test Plan:
test will be added in later PRs
Imported from OSS
Differential Revision: D20354879
fbshipit-source-id: 03976f4b86c46dbdc4e45764a1e72f1a3855a404
Summary:
`torch.nn.functional.interpolate` was written as a builtin op when we scripted the standard library, because it has four possible overloads. As a result, whenever we make a change to `interpolate`, we need to make changes in two places, and it also makes it impossible to optimize the interpolate op. The builtin is tech debt.
I talked with ailzhang, and the symbolic script changes are good to remove (i guess that makes a third place we needed to re-implement interpolate).
I'm trying to get rid of unneccessary builtin operators because we're standardizing mobile bytecode soon, so we should try to get this landed as soon as possible.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34514
Differential Revision: D20391089
Pulled By: eellison
fbshipit-source-id: abc84cdecfac67332bcba6b308fca4db44303121
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33853
Quant fusion relies on inline, but inline will break the CallFunction("linaer", ...) into a if block
it will be hard to recognize this block and swap it with quantized::linear, in order to
preserve the op, we will swap all quantized functional linear into aten::linear.
They might produce different backward graph, but this is called in the step before we get quantized
model, so it shouldn't affect anything.
We'll integrate this with convert_script later in the new "finalize_quant" API
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D20343873
fbshipit-source-id: 423e03bf893b79267d2dc97bc997ee1bfe54ec0f
Summary:
Previously when emitting subscripts we only emitted actual values, but
now they may sometimes emit a `ModuleValue`, so it should stay as a
`SugaredValue`. This allows for the result of the subscript to be
treated as a real module (i.e. you can just do `self.modlist[1](inputs)`
instead of `self.modlist[1].forward(inputs)`)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34320
Pulled By: driazati
Differential Revision: D20345642
fbshipit-source-id: 2bedf9a454af747b704422f6bbb8370cbdf4bf61
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33481
We have to propagate observed property of values through ops like max_pool2d, flatten and
avoid inserting duplicated observers.
For example:
```
x1 = self.conv(x)
x2 = maxpool(x1)
x3 = self.conv(x2)
```
If x1 is observed, we should propagate this information through maxpool and
we should consider x2 as observed as well.
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D20261897
fbshipit-source-id: 7de354a3ccb2b6e1708f5c743d4d9f7272691a93
Summary:
`ConcreteModuleTypeBuilder` used to keep parameters together with all others attributes in an `unordered_map` often leading to reordering them while building up the type. Parameter order is semantically meaningful, so we need to preserve it.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34131
Differential Revision: D20331542
Pulled By: suo
fbshipit-source-id: 5b860025f7902654d6099751d3fb14b12f6f5a67
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32814
We skip quantization for the intermediate values for patterns like `Conv - ReLU`,
but currently we didn't skip quantizing the input/output of the graphs of matched modules,
since we now changed the way we add observers, this also needs to be updated.
Test Plan:
python test/test_jit.py -- 'TestJit.test_insert_observers_skip_values'
Imported from OSS
Differential Revision: D20208785
fbshipit-source-id: ce30f2c4c8ce737500d0b41357c80ec8b33aecf9
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33834
This changes how we report Tracebacks to make them more clear when
there are both serialized and non-serialized ranges. It now looks like:
```
Traceback (most recent call last):
File "foo.py", line 25, in <module>
s2(a, b)
File "/scratch/zdevito/pytorch/torch/nn/modules/module.py", line 550, in __call__
result = self.forward(*input, **kwargs)
RuntimeError: The following operation failed in the TorchScript interpreter.
Traceback of TorchScript, serialized code (most recent call last):
File "code/__torch__.py", line 7, in forward
x: Tensor,
y: Tensor) -> Tensor:
return (self).bar(x, y, )
~~~~~~~~~ <--- HERE
def bar(self: __torch__.Moo,
x: Tensor,
File "code/__torch__.py", line 11, in bar
x: Tensor,
y: Tensor) -> Tensor:
_0 = (self).baz(x, y, )
~~~~~~~~~ <--- HERE
_1 = torch.ones([3], dtype=None, layout=None, device=None, pin_memory=None)
return torch.add(_0, _1, alpha=1)
File "code/__torch__.py", line 17, in baz
x: Tensor,
y: Tensor) -> Tensor:
return torch.add(x, y, alpha=1)
~~~~~~~~~ <--- HERE
Traceback of TorchScript, original code (most recent call last):
File "foo.py", line 11, in forward
def forward(self, x, y):
return self.bar(x, y)
~~~~~~~~ <--- HERE
File "foo.py", line 9, in bar
def bar(self, x, y):
return self.baz(x, y) + torch.ones(3)
~~~~~~~~ <--- HERE
File "foo.py", line 7, in baz
def baz(self, x, y):
return x + y
~~~~~ <--- HERE
RuntimeError: The size of tensor a (4) must match the size of tensor b (5) at non-singleton dimension 1
```
It follows Python convension of having the most important information last
and reading from the bottom up.
Changes:
* Moved the error message to the end, to copy Python
* Report original traceback separate from serialized traceback
* Make sure root functions have names in the interpreter trace.
Test Plan: Imported from OSS
Differential Revision: D20126136
Pulled By: zdevito
fbshipit-source-id: fd01f9985e5d74e04c4d064c02e8bc320f4fac13
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33277
Currently we insert observer in the called graph, which is incorrect since graphs can be shared
and the decision of whether to insert observer or not might dependend on where the graph is called.
For example, for a call sequence `self.conv1(self.conv2(x))`, we can't inserting observer correctly
if `self.conv1` and `self.conv2` are sharing the same type in the current implementation, because we insert
observer in the graph of the forward method of Conv2d right now and this call sequence requires us to insert
only one observer for the output of self.conv1/input of self.conv2.
We'll need to insert observers for input/output values of the graph in call site instead.
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D20208787
fbshipit-source-id: 739e1d877639c0d0ed24e573bbd36211defa6836
Summary:
This patch enables folding GetAttr nodes with their corresponding
values. _jit_pass_freeze_module API returns a new TorchScipt module
where all function calls and get attributes are inlined.
Usage:
frozen_model = torch._C._freeze_module(scrited_model._c)
frozen_model.forward(...)
This API currently optimizes the forward method. We will follow up to
to preserve and optimize methods and attributes that are annotated as
torch.jit.interface.
Several future improvements to JIT optimizations are required to maximize
clean up/de-sugar the graph and eliminate redundancies.
Ideally, we want to produce a graph that can easily be lowered to
GLOW and other low-level backends.
__
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32178
Differential Revision: D19419640
Pulled By: bzinodev
fbshipit-source-id: 52baffaba9bca2cd60a8e747baa68d57711ad42b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33705
The fact that there were two overloads appears to be a historical
artifact that dates back to when goldsborough originally added these
bindings in the first place. If TensorOptions is made optional,
then you only need one overload, not two, as they are exactly redundant
with each other. When MemoryFormat was added, it was made a little
harder to do this, as the C++ syntax at::empty_like(t, memory_format) would
not work if you collapsed the overload; but now it works because TensorOptions
supports MemoryFormat.
The upshot is, I can get rid of all the overloads and just have one overload.
Amazingly, this change is backwards compatible, as the test attests. While
I was at it, I also deleted the overload name from the functions entirely.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Differential Revision: D20073355
Pulled By: bhosmer
fbshipit-source-id: c6a8908213b32ccf6737ea864d135e2cce34f56b
Summary:
This bug has been hit a couple times recently. We need to handle all bivariant types, not just optional, when asserting mutability/immutability of pointed-to elements in alias analysis.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33952
Differential Revision: D20166025
Pulled By: eellison
fbshipit-source-id: cf3df9897a639641ef8303a08ba2b13523d01ef1
Summary:
Fixes#30775
This adds TorchScript implementations (copied from `python_variable.cpp`) for the remainin `Tensor` properties that were missing from the jit, in addition to a test that ensures new properties will trigger a failure so we can decide whether we want to add them as well.
For `some_tensor`, adds:
* `some_tensor.T`
* `some_tensor.ndim`
* `some_tensor.is_leaf`
* `some_tensor.name`
](https://our.intern.facebook.com/intern/diff/20153288/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33906
Pulled By: driazati
Differential Revision: D20153288
fbshipit-source-id: 2ddc48a14267077bc176065267e5ce52181b3d6b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33173
How to deal with ops that’s defined for both floating point and quantized Tensor?
Category of ops: the ones that doesn’t require observers, which means the quantization parameters(scale/zero_point) of the output of this op can be inferred from the quantization parameters of inputs.
For example:
avg_pool, max_pool, flatten, transpose, upsample
Another related topic to previous one is how do we deal with things like adaptive_avg_pool2d that does not require to be observed and it works with quantized tensor as well? If we insert quant/dequant for them, even the quant fusion becomes a numerically changing operation because the scale/zero_point for input and output are different.
Proposal
We can swap the operator with dequantize whenever we see it. For example, for pattern
Let’s say aten::general_op is defined for both floating point and quantized
%r = aten::conv(...)
%q = quantize(%r)
%dq = dequantize(%q)
%f = aten::general_op(%dq)
...
We detect that all inputs of aten::general_op is produced by dequantize, we’ll first delete all the dequantize for the inputs and then insert dequantize for each use of the output of the aten::general_op, note that this should work generally for all the case we might encounter.
After transformation we’ll have:
%r = aten::conv(...)
%q = quantize(%r)
%x = aten::general_op(%q)
%f = dequantize(%x)
...
1. Multiple inputs
1. We need to make sure all inputs of the aten::general_op are produced by dequantize before we do this transformation
2. Input used by multiple operators
1. We already did this by inserting dequantize for each use of the value
3. Output used by multiple operators
1. We’ll reuse the code that inserts dequantize(might need some refactor)
Note that current concat does not belong to this category right now since it does not inherit quantization parameters from inputs.
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D20123590
fbshipit-source-id: de2febe1f37e4079457a23acaeccbc6d9c9e1f8a
Summary:
Stacked PRs
* **#33578 - [jit] Unify augmented assign handling**
* #32993 - [jit] Fix aug assign for non-tensor attributes
We handle augmented assignments to `Select` and `Var` statements differently, but the actual in place update is the same for both, so this PR factors it out into a method so we don't have 2 code paths doing the same thing.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33578
Pulled By: driazati
Differential Revision: D20127647
fbshipit-source-id: 94f37acbd2551498de9d2ca09a514508266f7d31
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33531
We already insert dequantize for each use of the value, but there might still be cases where we only
see the value is used multiple times after inline. This pass adds the support to replicate dequantize
after inline to ensure output of dequantize is only used by one node, which is necessary to preserve all
quantization patterns like `dequant - conv - quant`
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D20123591
fbshipit-source-id: 6edb10a4566538bcf9379d332233f870372b7a63
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33779
This should eliminate random warnings and print spew from test_jit.
It also fixes a bug where we weren't properly comparing captured outputs
(!)
Test Plan: Imported from OSS
Differential Revision: D20124224
Pulled By: suo
fbshipit-source-id: 9241d21fdf9470531b0437427b28e325cdf08d3a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33504
Fix resolution fo functions that are bound onto torch in torch/functional.py. This does not fix compilation of all of those functions, those will be done in follow ups. Does torch.stft as a start.
Fixes#21478
Test Plan: Imported from OSS
Differential Revision: D20014591
Pulled By: eellison
fbshipit-source-id: bb362f1b5479adbb890e72a54111ef716679d127
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29495
This PR adds support for `_modules`, making it so we no longer need to special case support for `nn.Sequential`. I was getting internal errors around the previous approach using `self.define()`, so i am adding this PR as part of the stack.
Fix for https://github.com/pytorch/pytorch/issues/28998
Test Plan: Imported from OSS
Differential Revision: D18412561
Pulled By: eellison
fbshipit-source-id: a8b24ebee39638fccf63b2701f65f8bb0de84faa
Summary:
Stacked PRs
* **#33578 - [jit] Unify augmented assign handling**
* #32993 - [jit] Fix aug assign for non-tensor attributes
We handle augmented assignments to `Select` and `Var` statements differently, but the actual in place update is the same for both, so this PR factors it out into a method so we don't have 2 code paths doing the same thing.
](https://our.intern.facebook.com/intern/diff/20010383/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33578
Pulled By: driazati
Differential Revision: D20010383
fbshipit-source-id: 52e559ce907e95e5c169ab9d9690d0d235db36f3
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32379
Folding Conv2d - BatchNorm2d modules means recalculate the weight and bias of Conv2d module by incorproating the parameters
of BatchNorm2d, and also change the method calls to calling only forward of Conv2d module, this involves change of both module
types and graph because the bias of Conv2d is a parameter when it has value and is an attribute when it is
None(since JIT code has assumption of prameter being Tensor in multiple places), therefore
we'll need to remove the bias attribute when it is None and add a bias attribute later. Since ClassType might be shared, we separate
remove and add in separate steps and also keep track of the processed graph to avoid modifying the graph and type multiple times.
However we'll have to record the slot index of bias as well so we can replay the slot removal on other instances of Conv2d module.
Test Plan:
tbd
Imported from OSS
Differential Revision: D20078719
fbshipit-source-id: cee5cf3764f3e0c0a4a2a167b78dbada2e3835cc
Summary:
this adds enough infrastructure to run bailout checks in `checkScript`. I'll need to figure out the best way to enable it for nightly builds now.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32802
Differential Revision: D19974718
Pulled By: Krovatkin
fbshipit-source-id: 40485503f6d3ae14edcce98e1eec1f0559f3ad08
Summary:
The `not inline_everything` check was causing the jitter check to be skipped whenever we emitted a function. thanks SplitInfinity for pointing this out.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33468
Differential Revision: D19975934
Pulled By: eellison
fbshipit-source-id: 03faf8d2fd93f148100d8cf49cb67b8e15cf1f04
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33508
Ever since we switched to not inlining by default, some users have
complained since they relied on inlining occuring to, e.g. process the
graph with some other tool. Add an inlined_graph for convenience in
those cases.
Test Plan: Imported from OSS
Differential Revision: D19977638
Pulled By: suo
fbshipit-source-id: fe1fa92ff888959203d5d1995930d488b5f9e24c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32804
Constants are interpreter primitives so the op was not actually used.
This cleans up some of the logic around it.
This also fixes constant prop such that failures to look up an op
do not silently stop constant propagation. Instead, only errors
inside the op implementation itself will do this.
Test Plan: Imported from OSS
Differential Revision: D19673156
Pulled By: zdevito
fbshipit-source-id: 7beee59a6a67a6c2f8261d86bd505280fefa999e
Summary:
If a value has the type None, we can always replace it with a None constant.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33264
Differential Revision: D19878695
Pulled By: eellison
fbshipit-source-id: 5d0e7ffb37c5747997df093fec3183039d8dff4d
Summary:
Stacked PRs
* #32955 - [jit] Fix flipped PackedSequence outputs in script
* **#32953 - [jit] Support properties on `Device`**
PyTorch devices have a `index` and `type` property. This PR adds support for both to TorchScript
](https://our.intern.facebook.com/intern/diff/19849320/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32953
Pulled By: driazati
Differential Revision: D19849320
fbshipit-source-id: ce845258c6110058dd9ea1f759ef74b7ed2e786e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32941
The Python grammar allows single-statement one-line functions. So we
should allow it in the string parser.
Test Plan: Imported from OSS
Differential Revision: D19704153
Pulled By: suo
fbshipit-source-id: 8c06cc9c600aa2a9567b484a1ecc0360aad443e3
Summary:
Add `torch.jit.is_scripting` to the list of CondValues, or values that if they are an input to a if statement we only compile one side of the if. I'm not sure if we actually want this PR.
Pros:
- Makes it easier to add features that are not yet supported in TorchScript (like has_torch_function)
- The current idiom of writing `torch.jit.is_scripting` and factoring out the block to a function annotated with `torch.jit.ignore` is functionally equivalent and much more cumbersome
Cons:
- Makes it easier to add features that are not yet supported in TorchScript
- Perhaps is confusing as a reader what is being compiled. Potentially could give all caps name or otherwise change name to make it more visually stand out.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32871
Differential Revision: D19670383
Pulled By: eellison
fbshipit-source-id: 5257b0bd23c66f199d59a7f2c911e948301e5588
Summary:
The Python document <https://www.python.org/dev/peps/pep-0263/> gives
all examples using lowercase letters. Although it doesn't say
straightly, the following paragraph seems to indicate that uppercase
letters aren't legitimate:
> If a source file uses both the UTF-8 BOM mark signature and a magic encoding comment, the only allowed encoding for the comment is 'utf-8'. Any other encoding will cause an error.
My Emacs also complains about the uppercase letters every time I save
the file.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32850
Differential Revision: D19663281
Pulled By: ezyang
fbshipit-source-id: 48127d3c2fd6e22dd732a2766913735136ec2ebc
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32745
Some parameters (like `bias` in conv) are optional. To achieve this
previously, you had to add `bias` as a constant, which would invoke some
pretty weird behavior in the frontend, summarized as:
```
if bias is not None:
add it as a parameter normally
else: # bias is None
add it as a constant with the value None
```
There are several things bad about this:
1. Bias is not a constant. Marking it `__constants__` is confusing.
2. It basically relies on an implementation detail (the frontend
processes parameters before constants) to work.
Okay, whatever. I don't even know why we did this originally, but
getting rid of it doesn't break anything, so I assume improved NoneType
refinement has made this a non-issue.
Note on perf: this will make no difference; if bias was `None` it's still
folded out today, if bias is a Tensor it would be added as a parameter
both before and after this change
Test Plan: Imported from OSS
Differential Revision: D19628634
Pulled By: suo
fbshipit-source-id: d9128a09c5d096b938fcf567b8c23b09ac9ab37f
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32642
Previously, if we defined `__setstate__` but not `__getstate__`, we
would segfault. This PR turns that into a comprehensible error message
(and improves another error message as well).
Fixes https://github.com/pytorch/pytorch/issues/25886
Test Plan: Imported from OSS
Differential Revision: D19596463
Pulled By: suo
fbshipit-source-id: dbe76bc36bc747d65fb0223184c009e0e9ba072c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32380
We'll clone the module first and then fold conv bn and return a new
module
Test Plan:
.
Imported from OSS
Differential Revision: D19508033
fbshipit-source-id: 328e91a2c9420761c904a7f2b62dab4cfaaa31ac
Summary:
This API seems to be quite useful to make sure all bailouts in a graph are triggered. I used it for testing torchvision models and I was wondering if this might be something we might actually want to have? zdevito
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32518
Differential Revision: D19553147
Pulled By: Krovatkin
fbshipit-source-id: 7542c99051588b622091aec6d041c70731ca5d26
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32260
This makes it so you can actually pass the custom class as an arg to ScriptFunctions
Test Plan: Imported from OSS
Differential Revision: D19424252
Pulled By: jamesr66a
fbshipit-source-id: c3530186619655781dedbea03c2ad321aaff1cb8
Summary:
There was a user who did this and it would seg fault.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32503
Differential Revision: D19538481
Pulled By: eellison
fbshipit-source-id: dc3752028b9eff6ac88c025e8a2b5f8fd44ce32f
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30445
Create distributed and rpc directories under caffe/test for better management
of unit tests.
Differential Revision: D18702786
fbshipit-source-id: e9daeed0cfb846ef68806f6decfcb57c0e0e3606
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32491
This PR enables IValue to be able to hold a pure PyObject by adding a
new enum tag, a new jit_type to denote PyObject existance in IValue and
the JIT type system. We don't and not plan to expose this to user.
This is the basic piece that enable ivalue to be adopted broader like
making RRef always hold IValue, it might also simplify some compiler
logic
ghstack-source-id: 97039980
Test Plan: Imported from OSS
Differential Revision: D19502234
fbshipit-source-id: 90be001706d707d376cfbea25980fd82980df84a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31841
Add Tuple Constants to JIT. The constraint here is that all elements of a tuple must themself be insertable as a a constant. Previously tuples were special cased in constant propagation, but now that there are more passes that are inserted constants, such as freezing, we should just have tuples be representable as constants.
Test Plan: Imported from OSS
Differential Revision: D19439514
Pulled By: eellison
fbshipit-source-id: 3810ba08ee349fa5598f4b53ea64525996637b1a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31408
We'll error out when a graph is quantized with different QSchemes.
This only occurs when we have two modules that have same types (e.g. two Conv2d modules initialized with
same arguments) and quantized with two configs that would produce different quantized graphs, for example
per tensor affine and per channel affine. This is a rare case, so it should be OK to skip for now.
Actual support will come later.
Test Plan:
test_jit.py, test_quantization.py
Imported from OSS
Differential Revision: D19162366
fbshipit-source-id: 798f06d0ddef0c8458237ce88b62159cc77eec8b
Summary:
Unchecked cast just refines the type of a value, the value stays the same, so the output should alias the input.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32309
Differential Revision: D19439037
Pulled By: eellison
fbshipit-source-id: fe6902d0d9a5a9ef5e9c13e1dbd056576d8c327e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32232
Previously, we were using `operator<<` as the default way of printing
IValue constants during serialization. The semantics of `operator<<`
were ill-defined; and this bit us in particular with strings and lack of
quoting.
This PR defines the role of `operator<<`: much like Python `str()`, it
is intended to produce a human-readable-ish representation for
debugging purposes.
This PR also defines a new `repr()` function on IValue that is intended
to produce a valid Python expression that can be used to recreate an
object with the same value. `repr()` is not defined on all IValue kinds
(notably tensors!) for this reason.
Test Plan: Imported from OSS
Differential Revision: D19417036
Pulled By: suo
fbshipit-source-id: c102d509eaf95a28b6a62280bc99ca6f09603de5
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32187Fixes#32058. Previously we would build documentation during the pytorch
linux cuda build. We don't actually need to do this because we have a
dedicated python_doc_build job that builds the docs. With this change,
the CUDA build should run ~10 minutes faster, giving devs faster signal.
Test Plan: - Check the CUDA (10.1) build on this PR, make sure it doesn't build the docs.
Differential Revision: D19400417
Pulled By: zou3519
fbshipit-source-id: e8fb2b818146f33330e06760377a9afbc18a71ed
Summary:
`test_init_ops` calls `orthogonal_` which fails without lapack (this test was just missing a skip condition)
The cpp tests would fail with a `undefined symbol` error if run with `BUILD_TESTS=0`, so this PR skips them if that flag is `0`
](https://our.intern.facebook.com/intern/diff/19320064/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31965
Pulled By: driazati
Differential Revision: D19320064
fbshipit-source-id: d1dcd36714107688ded25a414e8969abe026bd03
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31978
Currently we keep a `mangleIndex_` that's intenral to compilation unit and
just increment the index when we found the original name is mangled, this doesn't
guarantee the new name is not defined.
This PR fixes the problem by querying whether the new name is defined or not.
fixes: https://github.com/pytorch/pytorch/issues/31268
Test Plan:
fixes the issue
Imported from OSS
Differential Revision: D19350535
fbshipit-source-id: fe3262b2838d4208ab72e2cd4a5970b3a792ae86
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31800
If we know that two constants are the same object, we can ignore other constraints and pool them together. This fixes an issue introduced by the other PR where quantization relied on constant pooling happening for correctness.
Test Plan: Imported from OSS
Differential Revision: D19269499
Pulled By: eellison
fbshipit-source-id: 9d4396125aa6899cb081863d463d4f024135cbf4
Summary:
This hooks up `inspect` so that Python functions get their parameters
names attached instead of naming them `0, 1, 2, ...`. This also fixes
issue #28537 where `ignore` functions were improperly typing `self`.
](https://our.intern.facebook.com/intern/diff/19256434/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29300
Pulled By: driazati
Differential Revision: D19256434
fbshipit-source-id: 6a1fe7bd0afab708b8439517798955d0abfeb44c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31343
Fix an issue in TorchScript tracing for modules with `c10::List<at::Tensor>` as an output. TensorList was not supported properly.
Test Plan: unit tests
Reviewed By: wanchaol
Differential Revision: D18850722
fbshipit-source-id: 87a223104d1361fe754d55deceeb1e8bbcad629b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29220
Support for accessing constant is added in previous
PRs, this PR re-enables the foldbn tests
Test Plan:
test_jit.py
Imported from OSS
Differential Revision: D18846848
fbshipit-source-id: 90ceaf42539ffee80b984e0d8b2420da66c263c3
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29219
We added class constant in previous PRs, this PR allows access to
class constant in the object API
Test Plan:
build/bin/test_jit
python test/test_jit.py
Imported from OSS
Differential Revision: D18846851
fbshipit-source-id: 888a6517d5f747d1f8ced283c0c2c30b2f6c72c6
Summary:
7zip and cmake are part of base image, no need to re-install. Remove the install step can make build/test more stable.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30897
Differential Revision: D19232961
Pulled By: mingbowan
fbshipit-source-id: fa3bbd1325839a2a977bf13fdbd97fda43793b8d
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31517
This is going to be used by upsample (which currently uses magic values to represent optionals).
For now, we just introduce a fake function for testing (torch._test_optional_float(x)).
Test Plan: Imported from OSS
Differential Revision: D19198721
Pulled By: gchanan
fbshipit-source-id: 0a1382fde0927c5d277d02d62bfb31fb574b8c74
Summary:
This is the first stab at running profile-insensitive optimizations on pre-profiled graphs. Running those optimizations has a potential to simplify graphs greatly before GuardElimination and GuardElimination should be able to remove more guards.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31392
Differential Revision: D19173639
Pulled By: Krovatkin
fbshipit-source-id: 2485a2a598c10f9b5445efb30b16439ad4551b3f
Summary:
Previously we would only catch `py::cast_error` which led to incomprehensible error messages like: `TypeError: 'NoneType' object is not iterable`. We are running arbitrary pybind code here, and not doing anything with the error message, so we should be less restrictive with the types of errors we catch.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31398
Differential Revision: D19166655
Pulled By: eellison
fbshipit-source-id: 84db8b3714c718b475913f2f4bb6f19e62f2d9ec
Summary:
Fixes#27495
This adds builtins as another piece of a concrete type. They're separate from normal functions since they represent the `BuiltinFunction` sugared value (which is a direct call to a builtin op). It also moves the builtins related logic from `jit/__init__.py` to `jit/_builtins.py` so it can be used from `jit/_recursive.py` to look up functions in the builtins table.
](https://our.intern.facebook.com/intern/diff/19149779/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31269
Pulled By: driazati
Differential Revision: D19149779
fbshipit-source-id: d4e5e5d7d7d528b75a2f503e6004394251a4e82d
Summary:
Stacked PRs
* #29940 - [jit] Fix parsing of big float literals
* **#29935 - [jit] Fix hex literal parsing**
* #29931 - [jit] Throw a better error for int too big for int64_t
Previously these were all parsed as `0`
](https://our.intern.facebook.com/intern/diff/19124944/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29935
Pulled By: driazati
Differential Revision: D19124944
fbshipit-source-id: 1ee0c1dee589933363a5efba069a2cfaf94373c5
Summary:
Add a section for unsupported ops, and modules. Automatically generate the properties and attributes that aren't bound, and for ops that have semantic mismatches set up tests so the docs stay up to date.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31329
Differential Revision: D19164472
Pulled By: eellison
fbshipit-source-id: 46290bb8a64d9de928cfb1eda5ff4558c3799c88
Summary:
Remove most of the testing for `weak_script`, since we removed it. Refactor a few of the existing tests to use recursive scripting api.
Fix for https://github.com/pytorch/pytorch/issues/23965
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31193
Differential Revision: D18966291
Pulled By: eellison
fbshipit-source-id: 6b1e18c293f55017868a14610d87b69be42bde12
Summary:
Resubmit of https://github.com/pytorch/pytorch/pull/30356 and https://github.com/pytorch/pytorch/pull/31014 :'(
The last commit contains the fix. There was an internal FBcode error not able to compile the previous `impl_default->second.equal(default_val.second))` line. I tried various fixes in C++ internally but couldn't figure anything out. This is a good example of the programming costs of going from python -> c++ for different types of objects, because the conceptual overhead has expanded in scope from (python) -> (python, c++, pybind).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31123
Differential Revision: D18936128
Pulled By: eellison
fbshipit-source-id: 7d8fd66a6dd4a3e9838f3a0b68c219b6565a9462
Summary:
Previously list elements were only unified for tensor lists.
This improves error messages and expands the unification logic
to include all types.
](https://our.intern.facebook.com/intern/diff/18837726/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30777
Pulled By: driazati
Differential Revision: D18837726
fbshipit-source-id: c4d275562a8429700987569426d694faa8f6002e
Summary:
This makes `nn.Transformer` usable from TorchScript. It preserves backwards compatibility via `__setstate__` on the encoder/decoder.
Fixes https://github.com/pytorch/pytorch/issues/24173
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28561
Differential Revision: D18124753
Pulled By: driazati
fbshipit-source-id: 7314843e5aa9c9bf974c4672e4edb24ed8ef4a6f
Summary:
Peephole optimize out type refinements when they are no longer refining the type.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/31024
Differential Revision: D18920958
Pulled By: eellison
fbshipit-source-id: 6d05d9812b9f9dcf001de760a78a2042fb832773
Summary:
Adds `torch.floor_divide` following the numpy's `floor_divide` api. I only implemented the out-of-place version, I can add the inplace version if requested.
Also fixes https://github.com/pytorch/pytorch/issues/27512
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30493
Differential Revision: D18896211
Pulled By: eellison
fbshipit-source-id: ee401c96ab23a62fc114ed3bb9791b8ec150ecbd
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30356
This finishes up the `torch.jit.overload` api for free-functions.
- defaults now required on the implementation function itself
- fully follows [overload spec](https://mypy.readthedocs.io/en/latest/more_types.html#function-overloading) such that the following is supported
```
overload
def mouse_event(x1: int, y1: int) -> ClickEvent: ...
def mouse_event(x1: int,
y1: int,
x2: Optional[int] = None,
y2: Optional[int] = None): ...
```
Note: `jit.overload` isn't supported yet for UDT, but is support for modules. This PR doesn't make the same changes for modules, if reviewers think I should include them then I could do so in a follow up PR or wait to land this. Since that's still an internal api I think it's fine, and the changes here would allow us to expose `torch.jit.overload` on free functions.
Test Plan: Imported from OSS
Differential Revision: D18864774
Pulled By: eellison
fbshipit-source-id: 6c566738bd6f0551a000a9ea8d56e403636b7856
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30877
Previously, when the environment tried to reassign variables which had been assigned to "inf" or "nan" it would fail because they are not simple values. Constant prop exposed this, a test was failing internally because of it.
Test Plan: Imported from OSS
Reviewed By: Krovatkin
Differential Revision: D18861016
Pulled By: eellison
fbshipit-source-id: b9b72978a26a0b00b13bf8ea7685825551f5a541
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30544
Run Constant Propagation upon compilation only on ops with non-aliasing inputs and outputs. This speeds up the first run of `torchvision.models.resnet18` by over 50% and speeds up compilation by about 25% (although the effects didn't seem additive with with https://github.com/pytorch/pytorch/pull/30503, so I'm going to land this PR first and then see if caching still has a sizable impact).
Running constant prop only with non-aliasing types does a lot of graph cleanup by removing constant ifs and a bunch of other smaller ops. It also avoids all the jitter problems we had when we tried running full constant prop previously. Bc it is idempotent it doesn't jitter, and it doesn't jitter graphs constructed from tracing because tracing doesn't emit any ops that only involve non-aliasing inputs.
Full constant prop isn't idempotent because what ops are run depends on the state of mutation in alias db, which will often change upon successive iterations of constant propagation, and bc it affects graphs constructed from tracing.
Edit: if we were okay with running constant propagation on graphs constructed from tracing (potentially making them hard to debug), an alternative would be to run constant propagation until the graph reaches a fixed point.
Test Plan: Imported from OSS
Differential Revision: D18833607
Pulled By: eellison
fbshipit-source-id: 92a0adb4882d67ed5a0db5c279f5e122aeeba54a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30892
Fixes all outstanding lints and actually installs a properly configured
flake8
Test Plan: Imported from OSS
Differential Revision: D18862825
Pulled By: suo
fbshipit-source-id: 08e9083338a7309272e17bb803feaa42e348aa85
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30821
While investigating while our tests didn't catch #30704 I noticed that none
of our tests in method_tests() were being run on CUDA. This diff moves
those tests into the new device-generic test framework so that we also get
CUDA coverage. For expediency, I blacklisted all tests which didn't work
on CUDA (rather than fix them); that's something we can leave for future PRs.
This is done by way of a new expectedFailure gadget.
Note that all occurences of skipIfNoLapack needed to be replaced with
skipCPUIfNoLapack.
I punted for test_jit; it's possible those tests should also run CUDA but a JIT
expert should take a look here.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Differential Revision: D18840089
Pulled By: ezyang
fbshipit-source-id: 66b613b5024c91d3e391c456bb642be7e73d4785
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30551
To enable quantizing with shared types, we need to insert GetAttr nodes for
quantization parameters since the code might be shared by multiple module instances
and we'd like to make quantized module instance also share the same code but with
different values of attributes.
Test Plan:
test_jit.py, test_quantization.py
Imported from OSS
Differential Revision: D18818652
fbshipit-source-id: fc95623cac59dcedd9e3f95397524eae515e7a11
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30037
Support quantization for modules with reused submodules, e.g. relu (automatically make unique)
We first do a pass on the graph to find all duplicate uses of the same module, and record the `Value`s of the
module instance, for each of these values we create a new module and change the access to that module.
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D18821483
fbshipit-source-id: 1698b981e9e9f0c728d9f03fcbcfbd260151f679
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30548
ClassTypes can be shared among different module instances, but previously we assumed
they would be unique, this PR enables the insert_observers pass to work with shared class types
Test Plan:
python test/test_jit.py
python test/test_quantization.py
Imported from OSS
Differential Revision: D18802465
fbshipit-source-id: b782e71e44a043af45577ac2b5c83e695155bb8b
Summary:
This fixes the second issue reported in https://github.com/pytorch/pytorch/issues/29909 namely, a loop counter is assigned the wrong values after transitioning to a bailout graph.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30186
Differential Revision: D18646845
Pulled By: Krovatkin
fbshipit-source-id: 1f7c601dd9f35892979385ffa132fb0886a4f203
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30362
Right now the qat modules(qat.ConvBn2d, qat.ConvBnReLU2d, qat.Conv2d)
are not convinent to support other dimensions of Conv, this PR refactors
these modules so that we can support Conv1d/Conv3d better
Test Plan:
python test/test_quantization.py
Imported from OSS
Differential Revision: D18691152
fbshipit-source-id: 5b561e6b054eadd31b98cabdf1ac67a61ee9b805
Summary:
In this PR, we mainly handle the case there are multiple usage of a Value when inserting the quant-dequant pair. This change will add one dequant for each usage of the Value.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30145
Differential Revision: D18671600
Pulled By: lly-zero-one
fbshipit-source-id: 61324a98861da85b80dcf7e930381311118ae53b
Summary:
This PR looks for a `constants.pkl` file at the top level in a zip file
in `torch.load`. If found, it calls `torch.jit.load` instead and issues
a warning to call `torch.jit.load` directly
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29339
Differential Revision: D18611095
Pulled By: driazati
fbshipit-source-id: f070a02f6b5509054fc3876b3e8356bbbcc183e1
Summary:
A prim::BailOut also needs to capture max trip counts as for some graphs they aren't constants and they are used in continuation graphs to figure out the remaining number of iterations to run.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/30097
Differential Revision: D18624446
Pulled By: Krovatkin
fbshipit-source-id: 085d25981c6669f65848996cd2d50066cc252048
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29577
`torch.autograd.grad` can return none is one of the input is not in the
autograd graph or not requires_grad, this fix it so that it return a
list of optional tensor instead of list of tensor.
This might have BC issue unfortunately, but I think it's rare both
internal and external (only training use it, and most of the training
use backward, instead of autograd.grad), so whitelist it.
Test Plan: Imported from OSS
Differential Revision: D18491642
fbshipit-source-id: d32b2b3446cf9e8b9a98f6d203a21a75643d8991
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29494
`calculate_qparams` of per channel quantization should return the axis, this
PR added this and also added corresponding support in graph mode
Test Plan:
python test/test_jit.py
Imported from OSS
Differential Revision: D18580905
fbshipit-source-id: f9691c1f043f8bca39f81716a4d0b10f60a65396
Summary:
This uses newly added InlinedCallStack to print the original call stack
even if the real call stack is shallower because of inlining.
This change also makes torchscript stacktraces look like python ones.
Example:
```
torch.jit.script
def baz(c, b):
return c + b
torch.jit.script
def foo(c, b):
return baz(c, b)
torch.jit.script
def bar(c, b):
return foo(c, b)
bar(torch.rand(10), torch.rand(9))
```
Output before:
```
Traceback (most recent call last):
File "fail.py", line 25, in <module>
bar(torch.rand(10), torch.rand(9))
RuntimeError: The size of tensor a (10) must match the size of tensor b (9) at non-singleton dimension 0
The above operation failed in interpreter, with the following stack trace:
at fail.py:15:11
torch.jit.script
def baz(c, b):
return c + b
~~~~~ <--- HERE
```
Output after:
```
Traceback (most recent call last):
File "fail.py", line 41, in <module>
bar(torch.rand(10), torch.rand(9))
RuntimeError: The size of tensor a (10) must match the size of tensor b (9) at non-singleton dimension 0
The above operation failed in interpreter.
Traceback (most recent call last):
File "fail.py", line 33
torch.jit.script
def bar(c, b):
return foo(c, b)
~~~ <--- HERE
File "fail.py", line 29, in foo
torch.jit.script
def foo(c, b):
return baz(c, b)
~~~ <--- HERE
File "fail.py", line 25, in baz
torch.jit.script
def baz(c, b):
return c + b
~~~~~ <--- HERE
```
Output of non-scripted python code:
```
Traceback (most recent call last):
File "fail.py", line 36, in <module>
bar(torch.rand(10), torch.rand(9))
File "fail.py", line 21, in bar
return foo(c, b)
File "fail.py", line 18, in foo
return baz(c, b)
File "fail.py", line 15, in baz
return c + b
RuntimeError: The size of tensor a (10) must match the size of tensor b (9) at non-singleton dimension 0
```
Differential Revision: D18532812
Test Plan: Imported from OSS
Pulled By: ZolotukhinM
fbshipit-source-id: e7e5ba5e4a8f1c7086406271d0f1685d9db8541a
Summary:
Stacked PRs
* https://github.com/pytorch/pytorch/issues/29244 - Use custom CRC
* **https://github.com/pytorch/pytorch/issues/29232 - Add zipfile serialization**
This adds a serialization method that uses a zipfile (https://github.com/pytorch/pytorch/issues/26567). Right now it is
guarded behind a flag `_use_new_zipfile_serialization`. In release mode it seems to have performance about the same / slightly better than the current serialization in some simple benchmarks for large/small tensors.
Follow ups:
* Flip the `_use_new_zipfile_serialization` flag
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29232
Differential Revision: D18332036
Pulled By: driazati
fbshipit-source-id: 1bac0847c4d599612cba905f2cac8248783be2f4
Summary:
Fix for https://github.com/pytorch/pytorch/issues/21545
We we were silently giving wrong semantics previously:
Python behavior:
```
def test(x=[]):
x.append(1)
return len(x)
print(test()) # 1
print(test()) # 2
```
By checking at the python layer, we prevent any new models from serializing this behavior but do not break existing serialized models.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29833
Differential Revision: D18513168
Pulled By: eellison
fbshipit-source-id: 6fe73f28e1f9d39dedeaf67a04718089d14401a1
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28988
Make ModuleList, Sequential, ModuleDict go through the same pathway as other modules, cleaning up a bunch of code and allowing them to define custom forwards and other methods.
EDIT: Previously, we would ignore an nn.Sequential attribute if it was not in `__constants__` ("did you forget to add it to Constants"). This PR scripts it even if it is not in `__constants__`. Is that what we want?
Test Plan: Imported from OSS
Differential Revision: D18402821
Pulled By: eellison
fbshipit-source-id: dd4f28fb0df0d1ba4ad1b3bc34ba141959a433f7
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29529
Pull Request resolved: https://github.com/pytorch/glow/pull/3771
We would like to replace `conv_prepack` with `conv2d_prepack` and `conv_unpack` with `conv2d_unpack`.
This makes the naming consistent between 2D and 3D conv:
```
torch.ops.quantized.conv2d_prepack
torch.ops.quantized.conv2d_unpack
torch.ops.quantized.conv2d
torch.ops.quantized.conv3d_prepack
torch.ops.quantized.conv3d_unpack
torch.ops.quantized.conv3d
```
We should do this earlier rather than later when we have more users for the quantized conv2d ops, for better engineering.
The replacement bash command is as the follows:
```
find ./ -type f -exec sed -i -e 's/quantized::conv_prepack/quantized::conv2d_prepack/g' {} \;
find ./ -type f -exec sed -i -e 's/quantized::conv_unpack/quantized::conv2d_unpack/g' {} \;
find ./ -type f -exec sed -i -e 's/torch.ops.quantized.conv_prepack/torch.ops.quantized.conv2d_prepack/g' {} \;
find ./ -type f -exec sed -i -e 's/torch.ops.quantized.conv_unpack/torch.ops.quantized.conv2d_unpack/g' {} \;
```
ghstack-source-id: 93661879
Test Plan: CI
Reviewed By: jackm321
Differential Revision: D18421079
fbshipit-source-id: 17ae8b1ee79223bd2c5d4bbccd57af6580c4ab12
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28985
Remove the observer module in the quantized model
Test Plan:
python test/test_jit.py 'TestJit.test_insert_quant_dequant'
Imported from OSS
Differential Revision: D18253777
fbshipit-source-id: 26081c4c3fd3dc049cafa8c0383219bc4c233589
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29432
This removes a lot of the private methods on torch._C.ScriptModule,
and instead implements functionality in terms of slot_dict_impl views
to implement _parameter, _buffers, and _modules in nn.Module.
A followup PR should also remove the _register_attribute,
_register_module, and _register_parameter methods, but this requires
more refactoring of the way tracing creates modules and replication
for data parallel works.
Test Plan: Imported from OSS
Differential Revision: D18387963
Pulled By: zdevito
fbshipit-source-id: f10d47afeb30c1e05d704ae5ac4166830933125c
Summary:
Fixes https://github.com/pytorch/pytorch/issues/17662
I'm not sure if `arange` needs to be in python_arg_parser at all, given the schemas in native_functions.yaml. In any case this at least fixes the dytpe mismatch.
In follow up PRs I will try to handle some of the other ops that do type inference at the python level, like randint.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27629
Differential Revision: D17885939
Pulled By: eellison
fbshipit-source-id: f97a8bc722b7ab77de1c42a992e49a4a3175ad60
Summary:
For the same reason we don't allow iteration over heterogenous types (modulelists/tuples) with types that don't have a static length, we also can't break/continue within them - we need to statically know all types.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29474
Differential Revision: D18406097
Pulled By: eellison
fbshipit-source-id: 70ed3fc4947b6237cdd6703135a988a5c13ce786
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29332
Even though we're statically typed, this can be useful, e.g. as
shorthand when iterating through a module list.
Test Plan: Imported from OSS
Differential Revision: D18393097
Pulled By: suo
fbshipit-source-id: aa42e955f88d1b8a876d0727055eb596453b9839
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29269
Hit this bug when I have an attribute of type `Optional[Tensor]` which
is initialized to None and reassigned later to some tensor.
Test Plan:
.
Imported from OSS
Differential Revision: D18364338
fbshipit-source-id: d8e1277a84ab7d80331cba83f5639469d398632e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28828
This updates torch::script::Module to more closely match the behavior
of nn.Module. In particular, it implements the (optionally recurisive)
iterators that retrieve submodules, parameters, and buffers and makes
their names match the python versions.
This also removes the individual accessors for Parameter, Module, Buffer, etc.
and replaces them with a single `attr` function which is equivalent to
writing `a.foo` in Python (`setattr` emulates `a.foo = v`).
As we build out the user-facing API for TorchScript values this will end
up matching how an attribute is accessed on general objects.
This PR preservers the python bindings for script::Module by emulating the
old API at the binding level. A followup will clean up the usage to more
directly match the C++ API.
Test Plan: Imported from OSS
Differential Revision: D18197611
Pulled By: zdevito
fbshipit-source-id: 7ee4dcbb258605d1c988314b05d938423f1ccee5
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29249
This splits out all the tests that are "easy", leaving `TestJit`,
`TestScript`, the autogenerated tests, and a small docs test.
Splitting those into reasonable chunks is more effort which is less
mechanical.
Differential Revision: D18339007
Test Plan: Imported from OSS
Pulled By: suo
fbshipit-source-id: 69164b9f9a2c379fe8923a846c98dd3c37ccb70e
Summary:
Type objects in python have an attribute `__abstractmethods__` that throws when it is accessed, so we were failing with an AttributeError whenever a type was used in TorchScript.
This pr prevents that error from happening. We can't just throw when a type is used because it could be used to access a static method: https://github.com/pytorch/pytorch/pull/27163
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28053
Differential Revision: D18332347
Pulled By: eellison
fbshipit-source-id: 9c7f2220f92674ad4d903621d9762cecc566ab0d
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28620
All Tensors are Variables now, they just happen to have requires_grad=False. Tensors ALWAYS have `VariableTensorId` in their type set.
When constructing this patch, I had to make decisions about what I would fix in this patch, and what I would leave for follow up PRs. Here is the cleanup that happens in this patch:
- The `is_variable` property is removed from TensorOptions. I removed this immediately because unlike Tensor::is_variable, TensorOptions::is_variable doesn't respect our VariableTensorId thread-local state. This means that there were a bunch of places where TensorOptions::is_variable was false, which is obviously bogus in the world when tensor and variable are merged. Instead of keeping the method as a function that always returns true, I just opted to remove it entirely (it's not public API.) All places we set `is_variable` are deleted.
- Knock on effect: there is no longer a separate DeprecatedTypeProperties for the variable and non-variable versions of type.
- Knock on effect: instead of asserting on TensorOptions::is_variable, instead we just test `at::impl::variable_is_excluded()`
- There is now only one copy of the cuDNN RNN dropout cache, not two (I'm not sure why we had two to begin with)
Some cleanup that doesn't happen in this patch:
- Eliminating unnecessary uses of `make_variable`
- Eliminating `Tensor::is_variable`
The most subtle part of this patch is retaining tracing behavior: the fact that everything is a Variable means that more code gets routed to VariableType than before; this can change traces. I identified two places where we didn't appropriately turn off VariableType, mostly factory functions:
- `torch.tensor` must turn off VariableType before invoking `at::empty` to construct the tensor, as it subsequently does direct data access
- `tensor_slow` (invoked when you pass a Python scalar to a tensor argument) must turn off VariableType before calling `scalar_to_tensor` so the scalar gets traced as constant, rather than as a call to `scalar_to_tensor`.
Honestly, these are all giant hacks, and should be replaced with a more specialized guard that just toggles tracing.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Reviewed By: dreiss
Differential Revision: D18171156
Pulled By: ezyang
fbshipit-source-id: 5b6a045beba37492647e350190f495114e86504d
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29061
It looks like we are too close to the maximum library size on
Windows. Kill Caffe2 operators to get us lower again.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Reviewed By: smessmer
Differential Revision: D18281083
Pulled By: ezyang
fbshipit-source-id: 8a11f9059dbf330f659bd96cc0cc2abc947723a8
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28255
Add support for treating Sequentials, ModuleLists, and ModuleDicts as iterables.
As previously, when emitting a for loop over a Module Container we unroll the for loop over all elements. We require that any Sugared Value in an iterable with a Module Container have a statically - determinable length.
Otherwise, if you zipped over a list of varying length and an nn.Sequential that alternated between returning a Tensor and a Dictionary, the output type would change based on the length of the list.
Fix for #17179
And https://github.com/pytorch/pytorch/issues/27401
and https://github.com/pytorch/pytorch/issues/27506
Test Plan: Imported from OSS
Reviewed By: ZolotukhinM
Differential Revision: D18278124
Pulled By: eellison
fbshipit-source-id: aca336a5b8da89c756b1f0884883649510cbde3c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28407
Given that we do not have support for inheitance or any polymorphism
strategy yet, we should guard against user from using it until we get
the full support so that user won't confuse by the weird behaviors.
Test Plan: Imported from OSS
Differential Revision: D18284310
fbshipit-source-id: f55a224f4190d57926d91ed98f6168d787387eb8
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27403
In fold_convbn pass, we need to recompute the parameter(weight, bias) for
conv, update the attribute of conv and update the access of bias in conv
because if the original conv have no bias, the `self.bias` access will be
inline and replaced by Constant node `None = prim::Constant()`, we need to
update this to use `GetAttr[name="bias"]` to make this work. But there is
also some work going on the handle constants, so we'll fix this pass after
that is done.
Test Plan:
.
Imported from OSS
Differential Revision: D18182918
fbshipit-source-id: bba510bc41ab58e0eb76f7b77335b6e3ffe2862d
Summary:
This makes MultiheadedAttention TorchScript compatible
It also breaks BC-compatibility for old models that do not have `_qkv_same_embed_dim` as an attribute.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28555
Pulled By: driazati
Differential Revision: D18124746
fbshipit-source-id: 5c5042fc6fc0e557db859a8ae05174cba5fce6a9
Summary:
Fix Slice/Select trace arguments. This PR stashes arguments to functions in order to avoid tracing them as constants.
This PR depends on a fix for select op in PR:
https://github.com/pytorch/pytorch/pull/25273
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26549
Reviewed By: hl475
Differential Revision: D17623851
Pulled By: houseroad
fbshipit-source-id: ae314004266688d2c25c5bada2dcedbfc4f39c5b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26572
Combined with isinstance specialization this allows a degree of polymorphic
functions to work without needing to use our weirder overload hacks.
We do not define any operators on Any, so the only thing you can do with it
is to put it in containers or type refine it using an isinstance check.
Any is restricted from appearing in non-argument position because we
cannot restore type tags if it ends up as a field in a class.
Test Plan: Imported from OSS
Differential Revision: D17530643
Pulled By: zdevito
fbshipit-source-id: f06f78ce84819f7773953a492f3d4c49219ee94c
Summary:
This PR fixes https://github.com/pytorch/pytorch/issues/25801 (see there for my verbose analysis).
As an example, for the following code:
```
import torch
torch.jit.script
def f1(x):
# type: (int, int) -> None
pass
```
this PR will change error message from this:
```
RuntimeError:
Number of type annotations (2) did not match the number of function parameters (1):
# type: (int, int) -> None
```
to this:
```
RuntimeError:
Number of type annotations (2) did not match the number of function parameters (1):
at __scratch__/example.py:4:0
torch.jit.script
def f1(x):
~~~~~~~~ <--- HERE
# type: (int, int) -> None
pass
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27195
Differential Revision: D17910902
Pulled By: driazati
fbshipit-source-id: af5c6353069d005752d6c7f0bd6a0c6db8437e55
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27772
This replaces unchecked_unwrap_optional with unchecked_cast. This
enables the generalization of type refinement so that it works for
isinstance checks as well. This also removes unchecked_unwrap_optional from
code we generate, which is good because it is a hard op to serialize well
since it doesn't directly encode the Optional[T] being unwrapped. In contrast,
unchecked_cast always explicitly lists the type.
Test Plan: Imported from OSS
Differential Revision: D17885424
Pulled By: zdevito
fbshipit-source-id: ce81077d6fbeaf2a802a2e0b17349aca85670466
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27773
We've changed how these functions are used over time, so I cleaned up
the header file API to match. In particular:
* tryMatchSchemas was added since the overload logic got copy/pasted
into three separate locations.
* With this change, tryMatchSchema is no longer public, as it is not needed
outside of tryMatchSchemas
* emitBuiltinFunction no longer needs a requires argument (it was always true)
* Argument order for all the schema matching stuff now puts the 'self'
builtin override last. This is only rarely used and was inconsistent with
matchSchema
Test Plan: Imported from OSS
Differential Revision: D17885425
Pulled By: zdevito
fbshipit-source-id: 064bc9fa4bd57b2e5366fff9f3c6ab9b9945e08b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27819
The idea here is to preserve the fact that `test_jit.py` contains all the JIT tests. So we import `JitTestCase`s from `jit/` into `test_jit.py` so that the test loader will find and run them when you do `python test_jit.py`. This also means that things like `-k` will work as expected.
The individual test files in `jit/` will throw if run directly, to prevent cases where the CI accidentally runs multiple versions of the same test.
Differential Revision: D17898105
Test Plan: Imported from OSS
Pulled By: suo
fbshipit-source-id: 0cd6f8421c86c90a6e1bae33a3fdbe998f570e07
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26722
Put them in a directory under jit/ to prep for test splitting
Test Plan: Imported from OSS
Differential Revision: D17550582
Pulled By: suo
fbshipit-source-id: a592b671ffe808f02d0a597d441bd98a18c9109e
Summary:
One fewer legacy decorator cluttering the test suite.
Functions relying on this decorator were updated or, in the case of test_sparse, the test suite was put back on double by default.
Note: this PR is blocked on https://github.com/pytorch/pytorch/issues/27599.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27628
Differential Revision: D17896254
Pulled By: mruberry
fbshipit-source-id: 13d460301f50ef4af7a660372432108164c0de1f
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26666
Changes:
- Introduce a `ConcreteModuleType` concept. This acts both as the key into the type
cache, and as the source of truth for `ModuleValue::attr` queries. It needs
to do both jobs because that's how we ensure correctness (if the types are
different, it's because `ModuleValue::attr` would return different things).
- Now `recursive_script` will first construct a `ConcreteModuleType` and search for a
pre-existing type before starting compilation.
- All previous paths to creating a `ScriptModule` (including inheriting from
`ScriptModule`) are now rewritten to go through `create_script_module`, so
that we have only a single place where construction happens.
Behavioral changes:
- Big change to `torch.jit.ScriptModule` inheritance: all attributes are now
recursively scripted if possible, matching recursive scripting semantics.
This makes it hard to keep something from being scripted (for example, a
Python submodule). Possibly we'll need an `ignore()` type thing for
attributes. In particular, this adds `self.training` to *every* ScriptModule, since
it's present on every `nn.Module`.
- I believe this change to be transparent to existing users of the inheritance API, since if you had an attribute that is unscriptable that you never used, there is no error. In some cases, we will create new attributes (even if they are unused), which will increase serialized model size from before.
Test Plan: Imported from OSS
Differential Revision: D17551196
Pulled By: suo
fbshipit-source-id: b476d1c9feb3ddfd63406d90989aaf9dfe890591
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26665
This is actually useful. For example: in batchnorm.py, all the tracked
stats are either `nn.Parameter` or `None`. We should register them as
params if they are set, or attributes with type NoneType if they are
not.
Test Plan: Imported from OSS
Reviewed By: shannonzhu
Differential Revision: D17551197
Pulled By: suo
fbshipit-source-id: 8d6f6d76d4dab0d524c4ffdfe0c1dd465771cd00
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26499
We've changed how these functions are used over time, so I cleaned up
the header file API to match. In particular:
* tryMatchSchemas was added since the overload logic got copy/pasted
into three separate locations.
* With this change, tryMatchSchema is no longer public, as it is not needed
outside of tryMatchSchemas
* emitBuiltinFunction no longer needs a requires argument (it was always true)
* Argument order for all the schema matching stuff now puts the 'self'
builtin override last. This is only rarely used and was inconsistent with
matchSchema
Test Plan: Imported from OSS
Differential Revision: D17488297
Pulled By: zdevito
fbshipit-source-id: a32d838ce35544972fa8767557acc22149081b55
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26271
This replaces unchecked_unwrap_optional with unchecked_cast. This
enables the generalization of type refinement so that it works for
isinstance checks as well. This also removes unchecked_unwrap_optional from
code we generate, which is good because it is a hard op to serialize well
since it doesn't directly encode the Optional[T] being unwrapped. In contrast,
unchecked_cast always explicitly lists the type.
Test Plan: Imported from OSS
Differential Revision: D17412856
Pulled By: zdevito
fbshipit-source-id: ded47eb086c4610998ad92bb1174225af00220f7
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27515
Resoving variable names using the local activation frames does not work
when using recursive scripting, but our current code tries to do it
(incorrectly) anyway. The reason it works is only because the script
call is in the same local frame as the definition. This will not be
true in practice and makes it seem like the API works in more cases
than it really does. This forces us to always use closure-based annotations,
documents it, and it fixes the tests so that they still pass.
Test Plan: Imported from OSS
Differential Revision: D17803403
Pulled By: zdevito
fbshipit-source-id: e172559c655b05f0acf96c34f5bdc849f4e09ce2
Summary:
This PR stop common_utils.py from setting the default tensor type when it's imported. See issue https://github.com/pytorch/pytorch/issues/27355. This is a frequent source of confusion for test writers.
Many tests relied on this setting (whether they knew it or not), and this PR also updates the test suite to pass without common_utils.py setting the default tensor type. Some larger test files now set the default floating dtype themselves, however. These test files are:
- test_autograd.py
- test_distributions.py
- test_jit.py
- test_nn.py
This is still a significant improvement from today, however. First, these files set the default floating dtype much more clearly than importing it from common_utils. Second, the rest of the test suite no longer sets this globally. Third, this PR is a springboard to updating those tests, too. In particular, as tests are made generic they can be moved aways from relying on this global setting.
Notable technical changes in this PR are:
- Significant updates to test_torch.py to make it pass without setting the default floating dtype globally.
- The default_floating_dtype decorator is now defined in common_utils, a couple versions of this operator were defined in test files previously.
- test_torch-specific parts of common_utils were refactored into test_torch.
- tensor creation methods in common_utils were updated to accept an optional dtype and device.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27444
Differential Revision: D17795235
Pulled By: mruberry
fbshipit-source-id: 7f77271c0c836e69f183ad9057a2c4b29f09d2e1
Summary:
Most of this was old cruft left over from special handling of `training` before we had a `bool` type. This makes all modules have a `training` attribute that is true by default and removes all other special handling.
Fixes#26884
](https://our.intern.facebook.com/intern/diff/17728129/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27109
Pulled By: driazati
Differential Revision: D17728129
fbshipit-source-id: 8ddc9fbb07a953dd05529538bfdd01ed88b5cb57
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26770
This PR added the interface/object serialization as module attribute, to
allow initializing object as a interface type during python
initialization. Because interface type can be backed by any class object
that implements that interface, if we declare it in
python/module.__init__, we will need to collect the run time types of the
value and serialize them to ensure complete code information
Test Plan: Imported from OSS
Differential Revision: D17742707
fbshipit-source-id: 7f614ad4f982996d320a0e2dd3515bf47370e730
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27110
Previously missing methods on some types like tensors would talk about
'builtins' which are only a thing inside of the compiler. Furthermore,
the error would only occur when the builtin was applied and it was discovered
that no builtin existed. This changes the error message so that it
discovers that method on our builtin types does not exist on attribute lookup.
Test Plan: Imported from OSS
Differential Revision: D17677616
Pulled By: zdevito
fbshipit-source-id: 2f7cf6c6093a9c832569c44f4b1044a2e56fe205
