OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-07 12:21:27 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity
26,155 Commits 8,304 Branches 150 Tags 1.5 GiB
c516f84525
Commit Graph

74 Commits

Author SHA1 Message Date
Elias Ellison
c516f84525 [JIT] Add Lower Tuples Call & Run remove mutation after list unrolling (#36829) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36829

This changes the IR complexity from the previous PR for the following tests:
```
('Name', 'Ifs/Loops', 'non-tensor ops')
Before:  ('max_unpool1d', 0, 3)
After:  ('max_unpool1d', 0, 0)
Before:  ('max_unpool2d', 0, 3)
After:  ('max_unpool2d', 0, 0)
Before:  ('max_unpool3d', 0, 4)
After:  ('max_unpool3d', 0, 0)
Before:  ('adaptive_max_pool2d', 0, 3)
After:  ('adaptive_max_pool2d', 0, 0)
Before:  ('adaptive_max_pool3d', 0, 4)
After:  ('adaptive_max_pool3d', 0, 0)
Before:  ('adaptive_avg_pool2d', 0, 3)
After:  ('adaptive_avg_pool2d', 0, 0)
Before:  ('adaptive_avg_pool3d', 0, 4)
After:  ('adaptive_avg_pool3d', 0, 0)
Before:  ('upsample', 13, 68)
After:  ('upsample', 4, 28)
Before:  ('upsample', 13, 68)
After:  ('upsample', 0, 5)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 13, 67)
After:  ('interpolate', 4, 27)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 13, 67)
After:  ('interpolate', 4, 27)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 13, 67)
After:  ('interpolate', 4, 27)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 13, 67)
After:  ('interpolate', 4, 27)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 14, 59)
After:  ('interpolate', 0, 3)
Before:  ('interpolate', 13, 57)
After:  ('interpolate', 4, 21)
Before:  ('interpolate', 14, 59)
After:  ('interpolate', 0, 3)
Before:  ('interpolate', 14, 59)
After:  ('interpolate', 0, 3)
Before:  ('interpolate', 13, 57)
After:  ('interpolate', 4, 21)
Before:  ('interpolate', 14, 59)
After:  ('interpolate', 0, 3)
Before:  ('interpolate', 14, 59)
After:  ('interpolate', 0, 3)
Before:  ('interpolate', 13, 57)
After:  ('interpolate', 4, 21)
Before:  ('interpolate', 14, 59)
After:  ('interpolate', 0, 3)
Before:  ('interpolate', 13, 77)
After:  ('interpolate', 4, 33)
Before:  ('interpolate', 14, 77)
After:  ('interpolate', 0, 5)
Before:  ('interpolate', 14, 77)
After:  ('interpolate', 0, 5)
Before:  ('interpolate', 13, 77)
After:  ('interpolate', 4, 33)
Before:  ('interpolate', 14, 77)
After:  ('interpolate', 0, 5)
Before:  ('interpolate', 14, 77)
After:  ('interpolate', 0, 5)
Before:  ('interpolate', 13, 77)
After:  ('interpolate', 4, 33)
Before:  ('interpolate', 14, 77)
After:  ('interpolate', 0, 5)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 14, 68)
After:  ('interpolate', 0, 4)
Before:  ('interpolate', 15, 103)
After:  ('interpolate', 1, 23)
Before:  ('interpolate', 14, 70)
After:  ('interpolate', 0, 6)
Before:  ('interpolate', 15, 103)
After:  ('interpolate', 1, 21)
Before:  ('interpolate', 14, 70)
After:  ('interpolate', 0, 6)
Before:  ('interpolate', 15, 91)
After:  ('interpolate', 1, 13)
Before:  ('interpolate', 14, 59)
After:  ('interpolate', 0, 3)
Before:  ('interpolate', 15, 93)
After:  ('interpolate', 1, 16)
Before:  ('interpolate', 14, 61)
After:  ('interpolate', 0, 5)
Before:  ('interpolate', 15, 111)
After:  ('interpolate', 1, 28)
Before:  ('interpolate', 14, 77)
After:  ('interpolate', 0, 5)
Before:  ('interpolate', 15, 113)
After:  ('interpolate', 1, 27)
Before:  ('interpolate', 14, 79)
After:  ('interpolate', 0, 7)
Before:  ('test_nn_AdaptiveMaxPool2d_single', 0, 3)
After:  ('test_nn_AdaptiveMaxPool2d_single', 0, 0)
Before:  ('test_nn_AdaptiveMaxPool2d_tuple', 0, 3)
After:  ('test_nn_AdaptiveMaxPool2d_tuple', 0, 0)
Before:  ('test_nn_AdaptiveMaxPool3d_single', 0, 4)
After:  ('test_nn_AdaptiveMaxPool3d_single', 0, 0)
Before:  ('test_nn_AdaptiveMaxPool3d_tuple', 0, 4)
After:  ('test_nn_AdaptiveMaxPool3d_tuple', 0, 0)
Before:  ('test_nn_AdaptiveMaxPool3d_single_nonatomic', 0, 4)
After:  ('test_nn_AdaptiveMaxPool3d_single_nonatomic', 0, 0)
Before:  ('test_nn_AdaptiveMaxPool3d_tuple_nonatomic', 0, 4)
After:  ('test_nn_AdaptiveMaxPool3d_tuple_nonatomic', 0, 0)
Before:  ('test_nn_AdaptiveAvgPool2d_single', 0, 3)
After:  ('test_nn_AdaptiveAvgPool2d_single', 0, 0)
Before:  ('test_nn_AdaptiveAvgPool2d_single_1x1output', 0, 3)
After:  ('test_nn_AdaptiveAvgPool2d_single_1x1output', 0, 0)
Before:  ('test_nn_AdaptiveAvgPool2d_tuple', 0, 3)
After:  ('test_nn_AdaptiveAvgPool2d_tuple', 0, 0)
Before:  ('test_nn_AdaptiveAvgPool3d_single', 0, 4)
After:  ('test_nn_AdaptiveAvgPool3d_single', 0, 0)
Before:  ('test_nn_AdaptiveAvgPool3d_tuple', 0, 4)
After:  ('test_nn_AdaptiveAvgPool3d_tuple', 0, 0)
```

Test Plan: Imported from OSS

Differential Revision: D21160758

Pulled By: eellison

fbshipit-source-id: 68ccbf3af74398e8dbad7e6bedb639635dafdb2e
 2020-04-28 23:28:02 -07:00
Jerry Zhang
6fa76b8a0c [jit] __deepcopy__ for RecursiveScriptModule (#32684) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/32684

Previously we have `clone` and `clone_instance`, where `clone` will clone both type
and value, and `clone_instance` only clone the value, both of them are shallow copies.
We need to re-evaluate whether we should expose them as a user facing API.
I think we should hide `clone`, but `clone_instance` might be useful as well, especially
when we are copying a model with very large weights, people might just want to do shallow copy.

This PR adds a `deepcopy` that might be useful as a user API, which deep copies the values, including
Tensor, but we didn't deepcopy `Blob`, `Capsule`, `Future` or `PyObject`.
For more discussions please see the following issue.

fixes: https://github.com/pytorch/pytorch/issues/32519

Test Plan: Imported from OSS

Differential Revision: D21220756

fbshipit-source-id: 476bf11fe82c08fac36e7457879a09f545ffdc5e
 2020-04-28 18:47:11 -07:00
Nikolay Korovaiko
a80a438e37 correctly set and restore states in te tests (#37210) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/37210

Differential Revision: D21238634

Pulled By: Krovatkin

fbshipit-source-id: 6462239753399c10c871baa5d5fdff5465cf2544
 2020-04-24 20:16:51 -07:00
Xiang Gao
3880f14b64 Canonicalize includes in torch, and add tests for it (#36303) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/36303

Test Plan: Imported from OSS

Differential Revision: D20943003

Pulled By: ezyang

fbshipit-source-id: 81fcbaccc1a7eec422bd8347d196bb66a5467884
 2020-04-23 08:09:21 -07:00
David Reiss
63e5058c88 Fix naming of "strides" method in TensorType (#36727) 
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
 2020-04-16 17:07:27 -07:00
Elias Ellison
9cbeb0faed [JIT] Dont optimize shape peepholes on inline (#36404) 
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
 2020-04-15 17:49:48 -07:00
Negin Raoof
f99a28f515 [ONNX] Adding a pass to replace interpolate function with aten::__interpolate (#35744) 
Summary:
Since aten;:__interpolate is removed in https://github.com/pytorch/pytorch/pull/34514, we need a pass replace interpolate function with aten::__interpolate for ONNX export.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35744

Reviewed By: hl475

Differential Revision: D20907041

Pulled By: houseroad

fbshipit-source-id: f2d2cdfec47389245c50f538267124eedf682adf
 2020-04-14 23:16:22 -07:00
Wanchao Liang
999d7f6ab2 [jit] tracer flag to guard risky behaivors (#36277) 
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
 2020-04-13 22:35:03 -07:00
Mikhail Zolotukhin
765bf8f03d Remove duplicate bindings from torch/csrc/jit/python/init.cpp. (#36492) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/36492

Test Plan: Imported from OSS

Differential Revision: D20995235

Pulled By: ZolotukhinM

fbshipit-source-id: 6afa3a956e57c2fb94bb29d332177be73a2bac2a
 2020-04-13 12:28:32 -07:00
Mike Ruberry
62f9312abd Revert D20783298: Fix naming of "strides" method in TensorType 
Test Plan: revert-hammer

Differential Revision:
D20783298

Original commit changeset: 8fcc146284af

fbshipit-source-id: 30e3cb6d7a30d82048534d4d2e794b7e08ae01bb
 2020-04-09 04:24:43 -07:00
David Reiss
16980e455f Fix naming of "strides" method in TensorType (#35170) 
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
 2020-04-08 15:59:28 -07:00
David Reiss
645d57ea01 Expose JIT Module's "register_attribute" to Python (#35630) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35630

Prefix underscored for now because the semantics of this method can be
confusing.  It adds a new attribute to the *type*, which can be shared
by several objects.

Test Plan:
Next diff in stack uses it, and has unit tests.

Imported from OSS

Differential Revision: D20904253

fbshipit-source-id: dcbf60eacf0e0e075c19238165aa33954aa73b5f
 2020-04-08 13:09:28 -07:00
Kimish Patel
d559a47933 Enable relu fusion with prepacked linear/conv. (#35705) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35705

Introduces a pass for relu fusion.

Test Plan:
python test/test_xnnpack_integration.py

Imported from OSS

Differential Revision: D20746592

fbshipit-source-id: 6c22f60a20e9121618c85077b9b58fb8d4082b3b
 2020-04-03 15:38:45 -07:00
Mikhail Zolotukhin
af5121f62a Invoke TensorExpr fuser pass from a graph executor. (#35913) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35913

The pass itself is still disabled by default, but with this change we
don't need to register it as a custom pass anymore. It allows us to
control its behavior with env variables more easily.

Test Plan: Imported from OSS

Reviewed By: suo

Differential Revision: D20827189

Pulled By: ZolotukhinM

fbshipit-source-id: e74d90b5e46422e7ab7bc40974a805220da50fbc
 2020-04-03 12:20:26 -07:00
davidriazati
6e13a7787b [jit] Fix type comparisons segfault (#35929) 
Summary:
Pybind will convert `None`s to `nullptr`s, so this adds a check to make
sure those don't get into the actual type comparison logic. Fixes #35778
](https://our.intern.facebook.com/intern/diff/20831278/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35929

Pulled By: driazati

Differential Revision: D20831278

fbshipit-source-id: 5800050e5eec280072afde58141ad00c1e8db8e2
 2020-04-03 11:33:48 -07:00
Elias Ellison
2595c62208 [JIT] Better error on default params error (#35888) 
Summary:
Someone messaged me abt this when a better error msg would have solved their problem
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35888

Differential Revision: D20819538

Pulled By: eellison

fbshipit-source-id: 95d124bfd162e1747dcdf7a981703a279a5dfaa6
 2020-04-02 15:31:22 -07:00
Christian Sarofeen
6d24f8fe21 Infrastructure for a new CUDA Fuser (#34785) 
Summary:
**Summary:** This PR contains the infrastructure of a new CUDA fuser. This CUDA fuser is based on many of the same principles of TensorExpressions and Halide, however the implementation is ground up. The fusion pass itself is similar to the default CUDA fuser, however, it has undergone some refactoring and is using the new code generation infrastructure. For those who are interested in how the code generation in this PR works, I would recommend reviewing _test/cpp/jit/test_gpu_fusion.cpp_ as well as the long comment section at the beginning of _torch/csrc/jit/codegen/cuda/transform_replay.h_  One of the largest differences between our approach and that of TVM/Halide, is the concept of "TensorView". TensorView from a high level should be thought of similarly to how we think of working with Tensors in PyTorch. It's an N-D object which can undergo transformations that change its dimensionality. Dimensionality changes are done through the operations split/merge/reorder/computeAt. These transformations are similar to split/fuse/reorder/compute_at of TVM, they modify how a tensor is iterated over to generate GPU code. Interestingly, in our scheme these transformations are applied to tensors and only impact how that tensor is generated.

**Warning:** This PR is purposefully not feature complete with the current fuser. We wanted to separate out the infrastructure from the fusion capabilities. Once in, smaller incremental PRs will be submitted to expand capabilities of the fuser.

**Short term goals:**

Parity with current CUDA fuser (including performance):
- Dynamic shapes (no recompilation)
- Implicit handling of braodcast (broadcasted tensors are treated as tensors of the braodcasted size in the generated code)
- Dropout

**Mid-term goals:**

- Transposes fused with pointwise operations where transpose involves only 2 axes (across the fused operation).
- 1-D reductions fused with pointwise operations
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34785

Reviewed By: ZolotukhinM

Differential Revision: D20650977

Pulled By: soumith

fbshipit-source-id: ee39c95a880e1b9822e874ed4cc180971572bf63
 2020-04-02 09:22:42 -07:00
Michael Suo
866d9d4e6a [jit] Fix name collision on load (#35720) 
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
 2020-04-01 00:02:38 -07:00
Michael Suo
06dcb70905 [jit] Fix Type equality in some cases (#35719) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/35719

Test Plan: Imported from OSS

Differential Revision: D20749422

Pulled By: suo

fbshipit-source-id: 09b697766c1eb3e56f4cf8acc7e854b0981d7991
 2020-03-31 22:29:12 -07:00
Supriya Rao
a090de380c [quant][graph] Add quant fusion for dynamic quantization (#35586) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35586

This pass fuses the choose_qparams-quant-dequant sequence
Fusion for weight tensor is the same as static quant.

Test Plan:
python test/test_quantize_script.py

Imported from OSS

Differential Revision: D20755680

fbshipit-source-id: b7443770642b6e6fa0fa9da8a44637e9b2d4df70
 2020-03-30 23:34:56 -07:00
Supriya Rao
1f7ee7b6b7 [quant][graph] Add pass to insert quant dequant for dynamic quantization (#35448) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35448

Add _choose_qparams_per_tensor which returns scale and zero_point similar to the dynamic quantization in the operator

Test Plan:
python test/test_quantize_script.py

Imported from OSS

Differential Revision: D20755679

fbshipit-source-id: c9066d8f1bb3e331809be26c4be806faafc9b981
 2020-03-30 23:33:32 -07:00
Jerry Zhang
6fc2403951 [quant][graphmode] qconfig_dict support None (#35336) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/35336

Test Plan:
python test/test_quantization.py

Imported from OSS

Differential Revision: D20655302

fbshipit-source-id: b453f3240ac487aa29629953b4d71274dbbc25fc
 2020-03-29 12:47:47 -07:00
Mikhail Zolotukhin
cd00bbc23f clang-format. (#35605) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/35605

Test Plan: Imported from OSS

Reviewed By: orionr

Differential Revision: D20720486

Pulled By: ZolotukhinM

fbshipit-source-id: f081a9fb6ef84fdce3b8f071d5e251e267854a18
 2020-03-28 11:45:06 -07:00
davidriazati
f27403d761 [jit] Fix named tuple resolution (#35409) 
Summary:
Fixes #29035

Previously we were missing a case for namedtuples in our Python value resolution logic, so they were just getting resolved as regular Python values, hence the `OSError`s in the linked issue
](https://our.intern.facebook.com/intern/diff/20653496/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35409

Pulled By: driazati

Differential Revision: D20653496

fbshipit-source-id: b5db1a11e918175aa02fda92993d233695417c56
 2020-03-27 17:07:26 -07:00
Nikolay Korovaiko
9e22d15f14 Enable tensorexpr cpp tests in CI. try #2 (#35454) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/35454

Differential Revision: D20665160

Pulled By: Krovatkin

fbshipit-source-id: e04cbe92b2ee5a3288f3c4e5c83533bfea85bf85
 2020-03-27 12:09:55 -07:00
Bram Wasti
a3e10d2a17 Expose enablement of TensorExpr fuser as env variable (#35341) 
Summary:
This commit allows one to use an environment variable to enable the fuser in torch/csrc/jit/tensorexpr/

```
PYTORCH_TENSOREXPR=1 python benchmark.py
```

This commit also changes the registration to happen by default, removing the requirement for the python exposed "_jit_register_tensorexpr_fuser"
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35341

Reviewed By: ZolotukhinM

Differential Revision: D20676348

Pulled By: bwasti

fbshipit-source-id: 4c997cdc310e7567c03905ebff72b3e8a4c2f464
 2020-03-26 14:31:57 -07:00
Meghan Lele
6384c2d81b [JIT] clang-format JIT code (#35115) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35115

This commit runs the newly added tools/clang_format.py on the JIT
codebase and includes all of the formatting changes thus produced.

Testing:
Ran the script, CI.

Test Plan: Imported from OSS

Reviewed By: eellison

Differential Revision: D20568523

Pulled By: SplitInfinity

fbshipit-source-id: e09bdb982ccf090eecfb7c7b461b8d0681eef82b
 2020-03-26 11:24:51 -07:00
Suraj Menon
aa01a95c6d Revert D20630760: [pytorch][PR] Enable NNC tests vol. i. add test_tensorexpr.py tests [WIP] 
Test Plan: revert-hammer

Differential Revision:
D20630760

Original commit changeset: 7d2f27aca6b1

fbshipit-source-id: 28ac92b3390651a4a67061d6ebf208515b9b9463
 2020-03-25 20:34:46 -07:00
Kimish Patel
dc2c4d02f9 Add a wrapper to wrap all optimization for mobile. (#35227) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35227

This wraps.
1. Conv BN folding (not mobile specific)
2. insert XNNPACK conv2d/Linear ops
3. Remove prepacking ops.

Test Plan: Imported from OSS

Differential Revision: D20603562

fbshipit-source-id: ff373af7112c070ec6198bac51845282e09ff1f8
 2020-03-25 19:21:14 -07:00
Nikolay Korovaiko
f3a5081bd4 Enable NNC tests vol. i. add test_tensorexpr.py tests [WIP] (#34897) 
Summary:
This  PR add tensorexpr cpp tests to test_jit.py
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34897

Differential Revision: D20630760

Pulled By: Krovatkin

fbshipit-source-id: 7d2f27aca6b1e23e3ffed1c765d8f590688118e3
 2020-03-25 17:23:48 -07:00
Elias Ellison
aab4beb87f [JIT] Pass To Safely Remove Aten Inplace Ops (#33186) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33186

This helps create larger functional graphs. It has the potential to increase memory use, so in order to land this on by default we would probably also do a reuse of buffers pass.

This is currently O(n * | Removed Nodes | ) because we have to rebuild the alias Db each time we make a change. This pass is critical to creating functional graphs, so this might be a compelling use case to build incremental updates to alias Db.

Test Plan: Imported from OSS

Differential Revision: D20603189

Pulled By: eellison

fbshipit-source-id: 105db52bf38e02188ca6df6d36294466d3309a0a
 2020-03-24 23:45:58 -07:00
Elias Ellison
5b2f8cef08 [JIT] Functional Graph Pass (#33020) 
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
 2020-03-24 23:44:18 -07:00
Wanchao Liang
4a84ac5f5d [jit] make Future type annotation available in Python (#27637) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27637

Fixes https://github.com/pytorch/pytorch/issues/26578

Test Plan: Imported from OSS

Differential Revision: D20626866

fbshipit-source-id: 20d6a3a719fddcb33e0e17a56d7123535fa20d65
 2020-03-24 14:36:05 -07:00
davidriazati
44622bbda9 [jit] Add lazy script decorator (#34935) 
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
 2020-03-24 13:43:18 -07:00
Supriya Rao
55019d357e [quant][graphmode] Add observers for dynamic quant (#35121) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35121

For dynamic quantization we insert observers at the input to mimic the quatization of activations that happens in the operator
Observer for weight is inserted similar to static quant

Test Plan:
python test/test_quantize_script.py

Sample output for single layer FC

.graph(%self : __torch__.___torch_mangle_4.M,
      %x.2 : Tensor):
  %_observer_1 : __torch__.torch.quantization.observer.MinMaxObserver = prim::GetAttr[name="_observer_1"](%self)
  %x.1 : Tensor = prim::CallMethod[name="forward"](%_observer_1, %x.2)
  %2 : __torch__.torch.nn.modules.linear.___torch_mangle_5.Linear = prim::GetAttr[name="fc"](%self)
  %3 : Tensor = prim::CallMethod[name="forward"](%2, %x.1) # test/test_quantize_script.py:19:23
  return (%3)

graph(%self : __torch__.torch.nn.modules.linear.___torch_mangle_5.Linear,
      %input.1 : Tensor):
 %2 : Function = prim::Constant[name="linear"]()
 %3 : Tensor = prim::GetAttr[name="weight"](%self)
 %_observer_0 : __torch__.torch.quantization.observer.MinMaxObserver = prim::GetAttr[name="_observer_0"](%self)
 %7 : Tensor = prim::CallMethod[name="forward"](%_observer_0, %3)
 %4 : Tensor = prim::GetAttr[name="bias"](%self)
 %5 : Tensor = prim::CallFunction(%2, %input.1, %7, %4) # /home/supriyar/miniconda3/envs/pytorch_py3/lib/python3.7/site-packages/torch/nn/modules/linear.py:87:15
 return (%5)

Imported from OSS

Differential Revision: D20599144

fbshipit-source-id: 9a8fa0e8655b9908826b981dce8a11d86efce5df
 2020-03-24 10:54:16 -07:00
James Reed
618c6214aa [reapply][JIT] Namespaces for TorchBind (#35254) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35254

Reapply D20541090 with some BC fixes
ghstack-source-id: 100733987

Test Plan: buck test mode/dev-nosan //caffe2/torch/fb/predictor/model_repo/tests:ai_infra_representative_model_shard_6_test -- 'RepresentativeModelTest\/ShardedRepresentativeModelTest\.RunModel\/0'

Reviewed By: zdevito

Differential Revision: D20607111

fbshipit-source-id: 80f148d860571208c93e9308128cd480ff089f74
 2020-03-24 00:39:48 -07:00
Elias Ellison
7ab25b2e6b [JIT] add id function (#34975) 
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
 2020-03-23 17:10:13 -07:00
albanD
0e0386b434 Revert "[JIT] add id function (#34975)" (#35209) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35209

This reverts commit 62f11f0a35.

Test Plan: Imported from OSS

Differential Revision: D20596847

Pulled By: albanD

fbshipit-source-id: e6777e42356aac772e59f0466a92cc13258218c1
 2020-03-23 08:42:09 -07:00
Kimish Patel
e1c092fe3a Changes to transition to generic API for ops with weight prepacking (#35010) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35010

semantics.

This PR moves all the xnnpack specific interfces to a generic interface.
Accordingly removes xnnpac specific reference from API and some variable
names.
What has not yet changed:

TODO:
USE_XNNPACK is still used. This can be removed where no XNNPACK
specific things are done. e.g., RegisterOpContext.cpp and
xnnpack_rewrite.cpp.
Also the filename and structure also remains. Some of the generic class
definition can be moved non-XNNPACK specific folder.

Test Plan:
python test/test_xnnpack_integration.py

Imported from OSS

Differential Revision: D20526416

fbshipit-source-id: 2e1725345c44bbb26bdc448097a7384eca121387
 2020-03-22 08:31:53 -07:00
Lu Fang
a100cf5146 Revert D20541090: [JIT][torchbind] Namespaces for torchbind classes 
Test Plan: revert-hammer

Differential Revision:
D20541090

Original commit changeset: ce3d9391dd3c

fbshipit-source-id: acc1d660fbda611941381315507dfe594c385db1
 2020-03-21 12:20:44 -07:00
Wanchao Liang
c21fde6421 [jit] make jit/rpc share the same PythonFutureWrapper (#35039) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35039

This is the initial step towards merging ivalue future and rpc future

Test Plan: Imported from OSS

Differential Revision: D20537164

Pulled By: wanchaol

fbshipit-source-id: d4f148c88e49ed6b0881ca4b4dd945ea24166183
 2020-03-20 22:35:34 -07:00
James Reed
e0496a70fc [JIT][torchbind] Namespaces for torchbind classes (#35054) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/35054

Test Plan: Imported from OSS

Differential Revision: D20541090

Pulled By: jamesr66a

fbshipit-source-id: ce3d9391dd3cdf619042b8f6ba2645f4c1fc875c
 2020-03-20 20:07:02 -07:00
Elias Ellison
62f11f0a35 [JIT] add id function (#34975) 
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
 2020-03-20 20:03:10 -07:00
Mikhail Zolotukhin
12f0052eee Add TensorExpr Fuser tests (resubmit). (#35085) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/35085

Test Plan: Imported from OSS

Differential Revision: D20552334

Pulled By: ZolotukhinM

fbshipit-source-id: 628fcf4719a879f18978ff8a0a64afbb045df645
 2020-03-20 13:19:31 -07:00
Natalia Gimelshein
3c90a90730 Revert D20540599: Add TensorExpr Fuser tests. 
Test Plan: revert-hammer

Differential Revision:
D20540599

Original commit changeset: ced9b6657fe7

fbshipit-source-id: e8fa11f20207c35f39b3fbe6f45fc627715377c1
 2020-03-19 18:37:32 -07:00
Mikhail Zolotukhin
7b59f41009 Add TensorExpr Fuser tests. (#35052) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/35052

Differential Revision: D20540599

Test Plan: Imported from OSS

Pulled By: ZolotukhinM

fbshipit-source-id: ced9b6657fe72bca61833ab5d59bdaddcacd114b
 2020-03-19 14:31:54 -07:00
davidriazati
e35dd4f603 [jit] Include call stack in OSError message (#34669) 
Summary:
Previously there was no indication of why you would get an `OSError` for something (such as the generated methods of a `dataclass`).
](https://our.intern.facebook.com/intern/diff/20426570/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34669

Pulled By: driazati

Differential Revision: D20426570

fbshipit-source-id: 45d63631984fa26a87c03de5523fb10d8abbc6db
 2020-03-18 15:10:23 -07:00
James Reed
09a7788a2f [torchbind] Improve IValue custom class API and remove most Capsule stuff (#34848) 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/34848

Test Plan: Imported from OSS

Differential Revision: D20480514

Pulled By: jamesr66a

fbshipit-source-id: 1c595faf34e00aab0a6202a8902426bd310551c3
 2020-03-17 20:39:34 -07:00
Mikhail Zolotukhin
95833a49e6 [TensorExpr] Pull changes from bertmaher/pytorch_fusion. (#34842) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34842

This PR (hopefully the last one of such kind) is merging changes from a
side branch where tensor expessions based fuser work has been done so
far. This PR is is a squashed version of changes in the side branch,
which is available here: https://github.com/bertmaher/pytorch

Differential Revision: D20478208

Test Plan: Imported from OSS

Pulled By: ZolotukhinM

fbshipit-source-id: 21556e009f1fd88099944732edba72ac40e9b9c0
 2020-03-17 11:02:48 -07:00
Mikhail Zolotukhin
35e7efeb9a [TensorExpr] Add CUDA codegen. (#34227) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34227

This PR adds a CUDA support to tensor expressions.

Differential Revision: D20251836

Test Plan: Imported from OSS

Pulled By: ZolotukhinM

fbshipit-source-id: ab36a55834cceff30c8371fef6cca1054a32f017
 2020-03-16 11:49:29 -07:00