Summary:
Unrolling support has been added in a way that we get good performing code on GPUs. Not sure how long this link will last but an example of a generated unrolled kernel is:
https://godbolt.org/z/i0uAv3
What can be seen from there is multiple calls of "ld.global.f32" without "ld.store.f32" in between them (and vice versa). This means that we are launching multiple loads that can be run in parallel, as well as multiple stores that can be run in parallel. This can be a crucial optimization for memory bound kernels. This was generally a point of concern in TVM as an attempt of a similar kernel from TVM produces: https://godbolt.org/z/Vu97vG which surrounds load - store pairs in conditional branches preventing the benefits of unrolling.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36435
Reviewed By: ZolotukhinM
Differential Revision: D21024011
Pulled By: soumith
fbshipit-source-id: e852e282fa7a304aba962e1926f756098c011fe0
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:
This PR completely refactors the code lowering process from our IR to CUDA. Before we had one giant step that would go from a relatively high level IR straight to CUDA, now we're lowering this first into concepts like ForLoop, IfThenElse, TensorIndex, Allocate. This lowering will allow us to do more complex code lowering like reductions and unrolling. Unrolling will quickly follow this PR.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/36199
Reviewed By: dzhulgakov
Differential Revision: D20925220
Pulled By: soumith
fbshipit-source-id: 8f621c694c68a1aad8653e625d7287fe2d8b35dc
Summary:
Just run `./tools/clang_format.py --verbose` and `git commit --all`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35969
Test Plan: CI
Differential Revision: D20845626
Pulled By: malfet
fbshipit-source-id: 0ae9a91dfa33417a021e7e9d233baba4188daf81
Summary:
This enables the serialization part of this change (the deserialization stuff is already landed #33255)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35741
Pulled By: driazati
Differential Revision: D20758124
fbshipit-source-id: e2cdefa99c3bec991491e5e967e7f1661ca7ffd9
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
Summary: This diff fixes the issues with current handling of debug information passed along the execution of the model. (For example, it is possible that multiple calls to the debug guard may override each other)
Test Plan: CI test/cpp/jit
Reviewed By: dzhulgakov
Differential Revision: D20602775
fbshipit-source-id: 4683957954028af81a1a0f1f12b243650230c9bb
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34710
Extending RecordFunction API to support new recording scopes (such as TorchScript functions), as well as giving more flexibility to set sampling rate.
Test Plan: unit test (test_misc.cpp/testRecordFunction)
Reviewed By: gdankel, dzhulgakov
Differential Revision: D20158523
fbshipit-source-id: a9e0819d21cc06f4952d92d43246587c36137582
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
Summary:
Same to `else`, `endif` and `elseif`.
Also prefer lowercase over uppercase ones
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35343
Test Plan: None at all
Differential Revision: D20638789
Pulled By: malfet
fbshipit-source-id: 8058075693185e66f5dda7b825b725e139d0d000
Summary:
Clamp input tensor values to [3, 3] to limit how small `tanh` gradint can get
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35196
Test Plan: CI + `bin/test_jit --gtest_filter=JitTest.ADFormulas --gtest_repeat=60000 --gtest_break_on_failure`
Differential Revision: D20611256
Pulled By: malfet
fbshipit-source-id: 8640faa5d8567d6c6df8cc5df80c2e65407116eb
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34985
IValue is part of the overall runtime system, not just the JIT. So it
should be tested in the ATen tests.
The real motivation though is so that I can use gtest directly, not the
hacked-up version the JIT uses.
Test Plan: Imported from OSS
Differential Revision: D20537902
Pulled By: suo
fbshipit-source-id: 09897e015ecde24aa8996babeaa08d98db90ef0d
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34280
To have prim ops searchable for lite interpreter, overloaded names need to be added for the operators with the same name but different schema. For example, aten::add in register_prim_ops.cpp. The difference is a combination of args and output type.
`"aten::add(str a, str b) ->str"`
`"aten::add(int a, int b) ->int"`
`"aten::add(float a, float b) ->float"`
`"aten::add(int a, float b) ->float"`
`"aten::add(float a, int b) ->float"`
`"aten::add(Scalar a, Scalar b) ->Scalar"`
Solution:
Use the argument type and/or output type (the same to the existing overloaded names). The overloaded name should be minimum as long as the operators can be differentiated. For other operators please look into the source code change for details.
`"aten::add.str(str a, str b) ->str"`
`"aten::add.int(int a, int b) ->int"`
`"aten::add.float(float a, float b) ->float"`
`"aten::add.int_float(int a, float b) ->float"`
`"aten::add.float_int(float a, int b) ->float"`
`"aten::add.Scalar_Scalar(Scalar a, Scalar b) ->Scalar"`
Test Plan: Imported from OSS
Differential Revision: D20456997
Pulled By: iseeyuan
fbshipit-source-id: 2c3dc324b4a4e045559f62c6cc2a10fbb9a72dcf
Summary:
Now that lists are no longer specialized, we can register only one operator for list ops that are generic to their element type.
This PR reorgs lists into three sets of ops:
- CREATE_GENERIC_LIST_OPS
- CREATE_SPECIALIZED_LIST_OPS
- CREATE_COMPARATOR_LIST_OPS_SPECIALIZED (we didn't bind certain specialized ops to Tensor)
This is important to land quickly because mobile is finalizing its bytecode soon, after which we could not remove these ops.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34520
Reviewed By: iseeyuan
Differential Revision: D20429775
Pulled By: eellison
fbshipit-source-id: ae6519f9b0f731eaa2bf4ac20736317d0a66b8a0
Summary:
Now that lists are no longer specialized, we can register only one operator for list ops that are generic to their element type.
This PR reorgs lists into three sets of ops:
- CREATE_GENERIC_LIST_OPS
- CREATE_SPECIALIZED_LIST_OPS
- CREATE_COMPARATOR_LIST_OPS_SPECIALIZED (we didn't bind certain specialized ops to Tensor)
This is important to land quickly because mobile is finalizing its bytecode soon, after which we could not remove these ops.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34520
Differential Revision: D20368543
Pulled By: eellison
fbshipit-source-id: ad0c6d70d2a6be6ff0e948d6786052167fc43e27
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34515
Once upon a time we thought this was necessary. In reality it is not, so
removing it.
For backcompat, our public interface (defined in `api/`) still has
typedefs to the old `script::` names.
There was only one collision: `Pass` as a `Stmt` and `Pass` as a graph
transform. I renamed one of them.
Test Plan: Imported from OSS
Differential Revision: D20353503
Pulled By: suo
fbshipit-source-id: 48bb911ce75120a8c9e0c6fb65262ef775dfba93
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34588
I constructed the patch by deleting OperatorOptions and then rerouting
all queries for AliasAnalysisKind to FunctionSchema. Some of the
behavior is kind of bogus: we really shouldn't be mutating FunctionSchema
after the fact, but that won't get fixed until we actually switch to
true schema merging.
Reland of https://github.com/pytorch/pytorch/pull/34160
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Differential Revision: D20387079
Pulled By: ezyang
fbshipit-source-id: d189f7a6ad8cd186b88b6fbfa3f189994eea14e8
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34160
I constructed the patch by deleting OperatorOptions and then rerouting
all queries for AliasAnalysisKind to FunctionSchema. Some of the
behavior is kind of bogus: we really shouldn't be mutating FunctionSchema
after the fact, but that won't get fixed until we actually switch to
true schema merging.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Differential Revision: D20282846
Pulled By: ezyang
fbshipit-source-id: ba7bca6e8adc3365789639b88e54c4e881b1692e
Summary:
Stacked PRs
* #33474 - [jit] Remove list specializations from pickler
* **#33255 - [jit] Add type tags to lists/dicts in pickle**
This adds a global call to `torch.jit._pickle.restore_type_tags` for
lists and dicts so that we can preserve their types after serialization.
](https://our.intern.facebook.com/intern/diff/20346780/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33255
Pulled By: driazati
Differential Revision: D20346780
fbshipit-source-id: c8534954ef4adb2e3c880401acbee30cd284f3db
Summary:
**Summary**
There is often a need to create a Tensor when writing IR by hand for JIT
optimisation pass unit tests. The only options for this today are real
Tensor creation functions like `aten::ones`. Any test that uses these functions
must also use the same default arguments as the Python/C++ API, which means
that all of the tests have to be updated when the API is updated. This commit
introduces a new primitive, `prim::MakeTestTensor` with schema `() -> Tensor` that
should be used in unit tests instead of real Tensor creation functions. This new
primitive has no public-facing API, so the maintenance burden is much lower.
**Testing**
This commit updates the alias analysis and DCE tests to use `prim::MakeTestTensor` instead of
`aten::rand`, `aten::ones`, and `aten::zeros`.
```
$ ./bin/test_jit
CUDA not available. Disabling CUDA and MultiCUDA tests
Note: Google Test filter = *-*_CUDA:*_MultiCUDA
[==========] Running 75 tests from 1 test case.
[----------] Global test environment set-up.
[----------] 75 tests from JitTest
[ RUN ] JitTest.ADFormulas
[ OK ] JitTest.ADFormulas (82 ms)
[ RUN ] JitTest.Attributes
[ OK ] JitTest.Attributes (0 ms)
...
...
...
[ RUN ] JitTest.LiteInterpreterPrim
[ OK ] JitTest.LiteInterpreterPrim (0 ms)
[ RUN ] JitTest.LiteInterpreterLoadOrigJit
[ OK ] JitTest.LiteInterpreterLoadOrigJit (2 ms)
[----------] 75 tests from JitTest (150 ms total)
[----------] Global test environment tear-down
[==========] 75 tests from 1 test case ran. (150 ms total)
[ PASSED ] 75 tests.
```
**Fixes**
This pull request fixes https://github.com/pytorch/pytorch/issues/33500.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34334
Differential Revision: D20296437
Pulled By: SplitInfinity
fbshipit-source-id: df4e7b0881ae4913424e5a409bfa171a61c3e568
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33807
afaik this is unused, so removing it from the source tree. RIP :(
Test Plan: Imported from OSS
Differential Revision: D20122118
Pulled By: suo
fbshipit-source-id: cb45943f5b9f969482301a2f9fe540326dbc78f2
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34099
This change effectively applies into IValue's future impl a few fixes
we discovered when using the torch::utils::Future<T> impl.
The parallel impls should probably eventually be merged, but until then:
- Don't hold the lock when invoking the callbacks. This makes
it effectively impossible (deadlocks) to call value() to get
the value from inside the callback.
- We discovered that it was slightly cleaner in practice to
notify condition variables prior to invoking callbacks
(best to unblock paused threads ASAP, before spawning new work).
- Fix some var naming inconsistency.
- Add a some caffe2 cpp test coverage.
ghstack-source-id: 99336569
Test Plan:
```
buck test mode/dev //caffe2/test/cpp/jit:jit -- 'JitTest\.IValueFuture'
```
Differential Revision: D20203278
fbshipit-source-id: 6e805ba547899dab9aab458e4b23049db31f930e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34122
Earlier work added support for async rpc cases when RecordFunction's
end callbacks might be called in a different thread; in addition some
extra care was needed to handle pointer to parent function;
This PR makes RecordFunction aware of potentially multiple threads in
use, as well as removes unused parent() call and restricts current()
RecordFunction to scope-based record functions (RECORD_FUNCTION macro)
Test Plan: unit tests
Differential Revision: D20297709
Pulled By: ilia-cher
fbshipit-source-id: 46a59e1b2eea0bbd8a59630385e193b38d30f9d1
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33294
1. Serialize bytecode of __setstate__ and run it when loading the model.
2. One use case is quantization. To test this use case a few operators are registered temporarily for lite interpreter. The "_" prefix registration will be removed when the operators are all migrated to mobile.
Test Plan: Imported from OSS
Differential Revision: D20162898
Pulled By: iseeyuan
fbshipit-source-id: 7a3180807bf38fbce594d86993896861f12bb58c
Summary:
**Summary**
There is often a need to create a Tensor when writing IR by hand for JIT
optimisation pass unit tests. The only options for this today are real
Tensor creation functions like `aten::ones`. Any test that uses these functions
must also use the same default arguments as the Python/C++ API, which means
that all of the tests have to be updated when the API is updated. This commit
introduces a new primitive, `prim::MakeTestTensor` with schema `() -> Tensor` that
should be used in unit tests instead of real Tensor creation functions. This new
primitive has no public-facing API, so the maintenance burden is much lower.
**Testing**
This commit updates the alias analysis and DCE tests to use `prim::MakeTestTensor` instead of
`aten::rand`, `aten::ones`, and `aten::zeros`.
```
$ ./bin/test_jit
CUDA not available. Disabling CUDA and MultiCUDA tests
Note: Google Test filter = *-*_CUDA:*_MultiCUDA
[==========] Running 75 tests from 1 test case.
[----------] Global test environment set-up.
[----------] 75 tests from JitTest
[ RUN ] JitTest.ADFormulas
[ OK ] JitTest.ADFormulas (82 ms)
[ RUN ] JitTest.Attributes
[ OK ] JitTest.Attributes (0 ms)
...
...
...
[ RUN ] JitTest.LiteInterpreterPrim
[ OK ] JitTest.LiteInterpreterPrim (0 ms)
[ RUN ] JitTest.LiteInterpreterLoadOrigJit
[ OK ] JitTest.LiteInterpreterLoadOrigJit (2 ms)
[----------] 75 tests from JitTest (150 ms total)
[----------] Global test environment tear-down
[==========] 75 tests from 1 test case ran. (150 ms total)
[ PASSED ] 75 tests.
```
**Fixes**
This pull request fixes https://github.com/pytorch/pytorch/issues/33500.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33914
Differential Revision: D20150304
Pulled By: SplitInfinity
fbshipit-source-id: c88f5289055a02dc20b7a5dcdf87469f9816d020
Summary:
Stacked PRs
* #33474 - [jit] Remove list specializations from pickler
* **#33255 - [jit] Add type tags to lists/dicts in pickle**
This adds a global call to `torch.jit._pickle.restore_type_tags` for
lists and dicts so that we can preserve their types after serialization.
](https://our.intern.facebook.com/intern/diff/19868637/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/33255
Pulled By: driazati
Reviewed By: xman1979, Tianshu-Bao
Differential Revision: D19868637
fbshipit-source-id: 2f1826e6679a786ca209198690269f399a542c04
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
