Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63776
I reverted this out of an abundance of caution because some test
failures occurred, but they were all due to precision issues fixed lower in
this stack. Let's try again.
I've rolled the elimination of the allow-parallelism-in-fusions toggle into
this diff since they're pretty tightly coupled.
ghstack-source-id: 136529847
Test Plan: CI
Reviewed By: huiguoo
Differential Revision: D30484555
fbshipit-source-id: 38fd33520f710585d1130c365a8c60c9ce794a59
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62763
This PR is to fix the issue that the graph inputs might be updated when we export the model in inference mode.
When a model is export in inference mode, some optimizations will be made. One side effect of these optimizations is: the inputs of graph might be adjusted. Such optimizatiosn include:
1. Conv and BatchNorm op fusion.
2. Do constant folding.
If the user sets export_params=False, or set keep_initializers_as_inputs=True, it's highly possible that the user wants to provide the corresponding parameters or initiliazers as the inputs of the graph.
In such situation, no matter the model is export in inference mode or training mode, exporter needs to prevent above optimizations from adjusting the graph inputs. By this, the inputs of graph could match inputs that users provided.
The changes in this PR, add an additional common judgement to see if the above optimizations needs to be done or not. From the value of export_params and keep_initializers_as_inputs arguments, infer if the graph inputs are allowed to be adjusted.
If no, these optimizations will be ignored, even other requirements are matched.
Besides these code changes, the comments of some parameters below have been updated so that users have more thoughts when they consider how to leverage these parameters for different purposes:
1. export_params
2. training
3. do_constant_folding
4. keep_initializers_as_inputs
Test Plan: Imported from OSS
Reviewed By: SplitInfinity
Differential Revision: D30375183
Pulled By: msaroufim
fbshipit-source-id: 4db8b9695649eb32a3a0fefa950ee2e5651bdba0
Co-authored-by: fatcat-z <jiz@microsoft.com>
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/59492
Adding code to find common expressions from the two subblocks of an if
operation and hoist them before the if block.
This also allows Dead Code Elimination to
then eliminate some if blocks.
Also eliminated some dead code in the codebase.
Test Plan:
python test_jit.py TestIfHoisting
Imported from OSS
Reviewed By: ngimel
Differential Revision: D29399533
fbshipit-source-id: 9336b9dc48c02c38862f98f98cd72fc1767a1802
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/59814
Using opinfos to test shape analysis. By default, we just check that we don't give incorrect answers, and then if `assert_jit_shape_analysis` is true, tests that we correctly propagates the full shape. and it found a couple bugs {emoji:1f603}
Test Plan: Imported from OSS
Reviewed By: Krovatkin
Differential Revision: D30200058
Pulled By: eellison
fbshipit-source-id: 6226be87f5390277cfa5a1fffaa1b072d4bc8803
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62200
This commit brings back the `RemoveInplaceOps` pass removed in D29523283 (dec5aa2260) that apparently had a bunch of internal users.
Test Plan: danthe3rd
Reviewed By: danthe3rd
Differential Revision: D29833316
fbshipit-source-id: 6cf13d463ab0a5e50ba3eb3243f79a9c51623809
Summary:
* Minor: spelling, grammar.
* Add calls to `GRAPH_DUMP()` where they were missing.
* Add or expand a few comments.
* Move a few comments to seemingly more appropriate spots.
* In canonicalize_graph_fuser_ops.cpp inline `runnableInputs()` since it
was only called in one place and had a misleading comment and
confusing name.
* In `PeepholeOptimizeImpl::optimizeBlock()`, set `changed = true;` when
removing `aten::is_complex`. Pretty sure its absence was a bug.
* Delete unused `_jit_pass_remove_inplace_ops` and and its
implementation `RemoveInplaceOps()`.
* In `preprocessCaffe2Ops()`, remove redundant check for nested optional
types. It was already checked in `checkONNXCompatibility()`.
* In `EncoderBase::AddAttribute`, log the unexpected attribute kind.
I don't remember the repro case now but I did hit this error at some
point and this additional logging made it easier to understand.
* In `fuseConvBatchNorm()` in eval_peephole.cpp, consistently use
camelCase instead of snake_case for local variables.
* Add curly braces around the bodies of if and loops.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/60390
Reviewed By: Krovatkin
Differential Revision: D29523283
Pulled By: SplitInfinity
fbshipit-source-id: 4e16c5648616f53da07d68dab7fdf252e06a0752
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/57334
Here's a possibly controversial PR. These counters got in the way of
generalizing the fuser tests to handle arbitrary devices, and I guess I'm just
generally skeptical that they provide much value. While true that they let us
observe whether fusion groups were created, we already have assertions based on
the shape of the graph, and I'm not sure that I trust those any less than these
counters.
Test Plan: Imported from OSS
Reviewed By: ZolotukhinM
Differential Revision: D29471484
Pulled By: bertmaher
fbshipit-source-id: f6d76f6e72dbfb581acff1d834b0c74500941b57
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/59735
1. Fixes ABA storage identity problem during serialization for `torch.package` by keeping reference of serialized storages through lifetime of `PackageExporter` to prevent reuse of memory address. Achieved by extending logic used in solution to mobile's same issue.
2. Adds determinism to naming scheme of serialized storages in export code paths which utilize `tensor_cdata_naming_scheme`(introduced 2nd mapping in `StorageContext`, now maps `storage cdata ptr` -> `unique id`, `unique id` -> `c10::Storage`)
3. Additionally uses presence of a storage in the `StorageContext` instance as marker for if a storage has been serialized or not, removing the need to scan the `PythonStreamWriter` for presence of the storage's serialization file
Test Plan: Imported from OSS
Reviewed By: suo
Differential Revision: D29075276
Pulled By: Lilyjjo
fbshipit-source-id: 15a5c30b1de99c5bd7079388f2db9b6ece2eca12
Summary:
Description:
- Before this, logging level could only be changed by changing the env
variable "PYTORCH_JIT_LOG_LEVEL"
- Can change the level from python now
- Have not added stream configuration for now
- Configuration is stored in a singleton class managing the options
Issue Link: https://github.com/pytorch/pytorch/issues/54188
Gotchas:
- Created separate functions
`::torch::jit::get_jit_logging_levels/set_jit_logging_levels` instead of
using the singleton class's method directly
- This is because when running test cases, two different instances
of the singleton are created for the test suite and the actual code
(`jit_log.cpp`)
- On using these methods directly, `is_enabled` calls the singleton
in `jit_log.cpp` while we are setting the config using another
singleton
- See: https://stackoverflow.com/questions/55467246/my-singleton-can-be-called-multiple-times
API:
- To set the level: `torch._C._jit_set_logging_option("level")`
- To get the level: `torch._C._jit_get_logging_option()`
Testing:
- UTs were added for C++
- A very simple UT was added for python to just check if the API is
being called correctly
- The API was checked by running trace in a sample python file
- Set env variable to "" and used `_jit_set_logging_option` in python to set the variable to `>dead_code_elimination`
- The error output had logs of form [DUMP..] [UPDATE...] etc
Fixes https://github.com/pytorch/pytorch/issues/54188
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58821
Reviewed By: soulitzer
Differential Revision: D29116712
Pulled By: ZolotukhinM
fbshipit-source-id: 8f2861ee2bd567fb63b405953d035ca657a3200f
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56966
This PR adds a toggle to shape analysis which won't inline complete tensor shapes as constants into the shape compute graph, which is a good stress test on the partial evaluation pipeline.
Test Plan: Imported from OSS
Reviewed By: bdhirsh
Differential Revision: D28444664
Pulled By: eellison
fbshipit-source-id: a62e424515a8837a4b596546efa93af5e8e61f10
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58300
Current state: During graph rewriting that can fuse nodes or add nodes
result in new nodes without debug information that was available in
original node. Thus we lose this information during graph rewrite.
This PR changes graph rewriting API to let user specify how the values
in the replacement pattern map to values in the pattern to be matched.
Then the graph rewriting will copy source range and inlined callstack
from the matched nodes onto the nodes being inserted.
(Note: this ignores all push blocking failures!)
Test Plan:
python test/test_jit.py
TestJit.test_pattern_based_rewrite_with_source_range_preserved
Imported from OSS
Reviewed By: malfet
Differential Revision: D28512465
fbshipit-source-id: 863173c29de726be85b3acbd3ddf3257eea36d13
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55926
This is necessary for code like conv2d where we wish to share a generic convolution shape function logic with that of conv2d but for conv2d always infer the output is dimension 4. I'm also hoping the refinement algorithm here could be refactored out and used to support refining tensor types from user annotations. i have a length comment explaining how this works, and the logic outside of data structures is pretty small and contained. Additionally, you might check out https://fb.quip.com/X7EVAdQ99Zzm for a very similar description of how to refine values based on comparison operators.
Test Plan: Imported from OSS
Reviewed By: ZolotukhinM
Differential Revision: D27750997
Pulled By: eellison
fbshipit-source-id: d962415af519ac37ebc9de88f2e1ea60a1374f7c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55925
This sets up the initial handling of symbolic shapes. As in the test, it doesn't work perfectly yet because it needs a couple other optimization passes. The basic description is pretty simple: we resolve tensor dimension indices to the same Value *, and before extracting out the output Tensor shape we substitute in symbolic shapes. We don't substitute during optimization because they are represented as negative numbers so we don't want them inadvertently used in Constant prop or something else.
Test Plan: Imported from OSS
Reviewed By: ZolotukhinM
Differential Revision: D27750996
Pulled By: eellison
fbshipit-source-id: 6984e7276b578f96b00fc2025cef0e13f594b6e6
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/54809
I'm going to post on dev-discuss soon with a more thorough explanation of the design and advantages of this shape analysis, so I'm leaving out that for now.
There is still a ton left to do, I'm posting this initial version so we can get something on master multiple can work on. List of many remaining steps to do:
- [ ] Add symbolic shapes support
- [ ] Bind shape functions for operators in C++
- [ ] Make classes of operators share the same shape function (e.g. pointwise, broadcast two inputs)
- [ ] Refactor APIs
- [ ] Only iteratively optimize shape function while a change has been made
- [ ] Expand coverage of coverage to common ops
- [ ] Add shape analysis pass on Graph that handles Ifs and Loops
- [ ] Allow concurrent reads to the operator map
- [ ] Successive applications of same inputs to same shape function (e.g. series of pointwise ops)
For this review, I am mostly looking for comments related to the implementation of symolic_shape_analysis.cpp, with the caveats listed above. I am not really looking for comments related to api/registration/graph level analysis as those are all planned to be changed. I am fine landing this as is or waiting until necessary components of the TODOs above are finished.
Test Plan: Imported from OSS
Reviewed By: pbelevich
Differential Revision: D27750998
Pulled By: eellison
fbshipit-source-id: 4338b99e8651df076291c6b781c0e36a1bcbec03
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58382
Calling markCompleted on a Future now first acquires the Future's mutex (as usual) but then sometimes tries to acquire the GIL during the DataPtr extraction while still holding the Future's mutex. (This happens when the value passed to markCompleted is a Python object). This can cause a deadlock if someone else calls any of the other methods of Future while holding the GIL.
There are two solutions to this: avoid holding the Future's mutex when extracting DataPtrs, and avoid holding the GIL while invoking the Future's method. In this PR I'm going for the latter, because it's a very simple immediate fix, but I believe this is brittle and that we should probably also consider the former fix.
ghstack-source-id: 129105358
Test Plan: The repro in https://github.com/pytorch/pytorch/issues/58239 now doesn't deadlock.
Reviewed By: mrshenli
Differential Revision: D28472816
fbshipit-source-id: 1bc9bca426dd004f9eb2568db1ffd38f014450e2
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56152
Currently, the Bundled Inputs API mutates the module in-place. It adds class methods and not instance methods. This results in a small problem that one can't re-run an already executed cell in Bento if the class has already been subject to bundled inputs.
In addition, there is no way to add bundled inputs to a module that has bundled inputs added already. This API provides a way to solve this problem as well by adding an `ignored_methods` to the call to `clone()` by allowing the implementation of bundled inputs to pass in the methods that it will add as `ignored_methods` so that when it does try to add those methods, it will be able to do so successfully.
We'll have to be careful when ignoring those methods during the call to `torch.jit._clone_module_with_class` since any bundled input that relies on a user-provided method will need to be preserved and not ignored during the clone.
Looking for feedback on whether this is an acceptable direction.
ghstack-source-id: 128908360
Test Plan:
Added unit test and ran it as `buck test //caffe2/test:mobile`
Also see this Bento Notebook: https://www.internalfb.com/intern/anp/view/?id=550829
Reviewed By: gmagogsfm
Differential Revision: D27788394
fbshipit-source-id: 48109cd4583506d4efdb345e4ba31385db23a273
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/57292
In Future (and soon in other places too) we need to receive a list of devices from Python-land. We don't want to just take their indices because we need full devices in order to infer the type from them. torch.device is not defined through pybind, it's defined through a plain `PyModule_AddObject` call with CPython, thus pybind isn't naturally able to understand and convert it. However we can provide a custom type caster which fixes that. We have this already for at::Tensor, at::Generator, ...
ghstack-source-id: 127916268
Test Plan: CI
Reviewed By: mrshenli
Differential Revision: D28092732
fbshipit-source-id: 1c31d0b85a4d5c9e7bde8161efbb7574d505157c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/57052
This PR caps a stack whose goal was to merge CUDAFuture into ivalue::Future. CUDAFuture used to be a subclass of ivalue::Future, which was already pretty good, but it meant that in several places we needed `#ifdef`s or registries in order to create the right type of class, which was annoying. We've made CUDAFuture device-agnostic, by using generic helpers, so that it doesn't depend on CUDA. Now all its code can be inserted into ivalue::Future.
This PR does this very naively, by copy-pasting CUDAFuture's code into the (previously empty) virtual methods of ivalue::Future. This helps ensure the correctness of this PR, as it's straightforward to see it behaves exactly like before. However we probably want to polish it a bit later to iron out so wrinkles.
ghstack-source-id: 127713138
(Note: this ignores all push blocking failures!)
Test Plan: CI
Reviewed By: mrshenli
Differential Revision: D28036829
fbshipit-source-id: 3e5b16402f5dc245c1fcb9d7bf06db64dcb0d2a3
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/57051
Make CUDAFuture autodetect the devicetype from its arguments (which thus change from DeviceIndices to full Devices). This in fact transforms CUDAFuture into a AnythingFuture, since it's not tied to CUDA in any way anymore. Having made it fully device-agnostic, we'll merge it into ivalue::Future in the next PR.
ghstack-source-id: 127713134
(Note: this ignores all push blocking failures!)
Test Plan: CI
Reviewed By: mrshenli
Differential Revision: D28032711
fbshipit-source-id: 8ba23b1b0d97f61db8693cd5f3c7bae7989a9bcd
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/57045
Went back and adjusted the previous optimizations to just be applied to every function.
Cleaned up api to match.
ghstack-source-id: 127214412
ghstack-source-id: 127536155
Test Plan: unit test
Reviewed By: kimishpatel
Differential Revision: D27950859
fbshipit-source-id: 214e83d5a19b452747fe223615815c10fa4aee58
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56717
The signal_handler was under the caffe2 namespacee but was being used
by PyTorch as well.
I've fixed this my moving it to the c10 namespace where now both C2 and PyTorch
can use it.
The signal_handler interface in caffe2/utils/signal_handler.h is kept the same
for backward compatiblity for C2, but most of the commmon code is moved to c10.
ghstack-source-id: 127446929
Test Plan: waitforbuildbot
Reviewed By: ezyang
Differential Revision: D27946738
fbshipit-source-id: d6228d1a0108f4c807d405e7a0bb799c5375388f
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56517
Currently a torch.futures.Future could wrap a CUDAFuture, but it could not create one from scratch. This prevented users from using CUDAFutures in some occasions, for example when using `rpc.functions.async_execution`, or in their own code. I don't see any reason for such a limitation, hence here I add support for this.
ghstack-source-id: 127261554
Test Plan: Added a test later in the stack
Reviewed By: mrshenli
Differential Revision: D27887190
fbshipit-source-id: ecbb39c1ad7cd189d478ded9c361448f05a270ad
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56172
Enable the standardOps include **Add\Sub\Mul\Div\Gemm\Pow\Mod** with low precision input in ORT
Test Plan: Imported from OSS
Reviewed By: pbelevich
Differential Revision: D27866136
Pulled By: SplitInfinity
fbshipit-source-id: f2cf5649fffefd68c0cc7b6dce94198751636727
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56120
This reverts commit ad17fadbfc (D27786457).
The big annoyance here is that depending on the threading mode you may not be
able to toggle num_threads at will, so the fusion tests won't fail.
I hate this solution, but I'm adding a secondary override for the TE fuser.
Now you need to both turn on fusion (_jit_override_can_fuse_on_cpu), and you're
OK if you're running with 1 thread, or you can add
`_jit_set_texpr_parallel_cpu_enabled` to enable it anyways.
This is (a) mainly for tests, since a real user probably won't fiddle aimlessly
with the thread count, and (b) will go away once NNC's threading support is
fully baked.
Test Plan: Imported from OSS
Reviewed By: Krovatkin
Differential Revision: D27788199
Pulled By: bertmaher
fbshipit-source-id: 070d04474f15e9689dbdf8cc1fde43050c6506b1
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/54619
Minor refactor to conv batchnorm folding to work on other functions besides forward
ghstack-source-id: 125767010
Test Plan: unit test and {P339453712}
Reviewed By: kimishpatel
Differential Revision: D27301452
fbshipit-source-id: 4e0cc544a171a970583979a496b2908935124497
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/53314
Introduction of api for optimizing non forward functions for mobile. As of this diff, all functions that you say to optimize will be preserved, and those functions will be run through canonical optimization. The intention is to stack each further optimization onto separate diffs since they touch multiple files, and it seems like it'd be a nightmare to review.
ghstack-source-id: 123909414
Test Plan:
torch.utils.mobile_optimizer.optimize_for_mobile(net, methods_to_optimize=["forward", "foo"]) runs fine
torch.utils.mobile_optimizer.optimize_for_mobile(net, methods_to_optimize={"foo"}) optimizes just foo if the model doesnt define forward otherwise optimizes foo and forward
torch.utils.mobile_optimizer.optimize_for_mobile(net, methods_to_optimize=["forward"]) runs fine
torch.utils.mobile_optimizer.optimize_for_mobile(net) runs fine if the model defines forward, Throws otherwise
Reviewed By: kimishpatel
Differential Revision: D26618689
fbshipit-source-id: 5bff1fb3f3f6085c4a649a8128af9c10f0fa9400
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/53306
* [ONNX] Fix for sequence of mutations in blocks (#51577)
Fixes consecutive mutations in a tensor inside blocks.
Also, support append and pop in blocks.
* Support inplace operations + indexing
* Clean up old pass for remove mutations
* Add loop test
* Fixes for set attr in loops
* Removing the new jit API flag
* [ONNX] Redesign onnx pass to enable shape type dependent pattern conversion - cont (#51795)
With the introduction of ONNX shape inference, shape and type are inferred on the fly as operators get converted from ATen to ONNX when running symbolic function. This resolves the shape/type requirement for the symbolic functions. The pre-onnx passes however, can not be supported by shape inference, since at that stage the operators in the graph are still ATen operators.
This PR is to update the design of ONNX pass, to enable a mechanism of capturing subgraphs of ATen operators of certain patterns, and convert them later, when shape/type information of upstream operators are available.
The new design will require pre-onnx passes that need shape/type to be written in two parts, encapsulation and conversion.
The encapsulation part will find the nodes of patterns, like how pre-onnx passes were written previously. But instead of converting the nodes, it will encapsulate them into a sub-block of a new placeholder node. This part is called before onnx pass, so it runs before calling symbolic functions.
The conversion part will be called inside the onnx pass. In onnx pass, run_symbolic_func will be called for each node in topological order. When it reaches the placeholder node, the conversion part will be invoked. It will convert the nodes inside the sub-block based on pattern. By that time, it will have shape/type of upstream operators available. After the conversion is complete, the placeholder node will be removed, and nodes inside its sub-block converted. Run_symbolic_func will be called for these nodes, and they will be converted from ATen operator to ONNX operator.
This PR includes several other fixes, listed below.
* ~~replace helper.cpp with onnx_utils.cpp for holding utility functions.~~
* fix EraseNumberTypes on Bool type, the code was outdated that back then Bool type doesn't exist.
* ~~enable onnx shape inference in export with parameter/initializer data.~~
* other code clean ups.
* fix insertion of identity nodes for loop opset 13 sequence output.
~~PR depends on #51603~~
* Fix after merge
* clang
* Fix clang
* Fix clang
* Fix warning message.
* Fixes for non-model param attributes
* Fix for caffe2
* Additional test
* clang
* Skip test for lower opsets
* fix clang-tidy
* Update init.cpp
* Update remove_inplace_ops_for_onnx.cpp
* Update remove_inplace_ops_for_onnx.cpp
* Update remove_inplace_ops_for_onnx.cpp
* Fix for clang formatting
Test Plan: Imported from OSS
Reviewed By: pbelevich, malfet
Differential Revision: D26922416
Pulled By: SplitInfinity
fbshipit-source-id: e7108620b39b6404c594910786c4d275fee59d84
Co-authored-by: Bowen Bao <bowbao@microsoft.com>
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/53304
With the introduction of ONNX shape inference, shape and type are inferred on the fly as operators get converted from ATen to ONNX when running symbolic function. This resolves the shape/type requirement for the symbolic functions. The pre-onnx passes however, can not be supported by shape inference, since at that stage the operators in the graph are still ATen operators.
This PR is to update the design of ONNX pass, to enable a mechanism of capturing subgraphs of ATen operators of certain patterns, and convert them later, when shape/type information of upstream operators are available.
The new design will require pre-onnx passes that need shape/type to be written in two parts, encapsulation and conversion.
The encapsulation part will find the nodes of patterns, like how pre-onnx passes were written previously. But instead of converting the nodes, it will encapsulate them into a sub-block of a new placeholder node. This part is called before onnx pass, so it runs before calling symbolic functions.
The conversion part will be called inside the onnx pass. In onnx pass, run_symbolic_func will be called for each node in topological order. When it reaches the placeholder node, the conversion part will be invoked. It will convert the nodes inside the sub-block based on pattern. By that time, it will have shape/type of upstream operators available. After the conversion is complete, the placeholder node will be removed, and nodes inside its sub-block converted. Run_symbolic_func will be called for these nodes, and they will be converted from ATen operator to ONNX operator.
This PR includes several other fixes, listed below.
* ~~replace helper.cpp with onnx_utils.cpp for holding utility functions.~~
* fix EraseNumberTypes on Bool type, the code was outdated that back then Bool type doesn't exist.
* ~~enable onnx shape inference in export with parameter/initializer data.~~
* other code clean ups.
* fix insertion of identity nodes for loop opset 13 sequence output.
~~PR depends on #51603~~
Test Plan: Imported from OSS
Reviewed By: SplitInfinity
Differential Revision: D26922417
Pulled By: malfet
fbshipit-source-id: 14ed06158d539e2451c2e5e63ba1b32fb0f75095
Summary:
This PR adds an implementation for `aten::cat` in NNC without any conditionals. This version is not enabled by default.
Here is the performance of some micro benchmarks with and without conditionals. There is up to 50% improvement in performance without conditionals for some of the shapes.
aten::cat implementation in NNC **with** conditionals
```
$ python -m benchmarks.tensorexpr --device cpu --mode fwd --jit_mode trace --cpu_fusion concat
pt: concat2d2input_fwd_cpu_1_160_1_14_1: 5.44 us, SOL 0.26 GB/s, algorithmic 0.51 GB/s
pt: concat2d2input_fwd_cpu_1_580_1_174_1: 5.75 us, SOL 1.05 GB/s, algorithmic 2.10 GB/s
pt: concat2d2input_fwd_cpu_20_160_20_14_1: 6.87 us, SOL 4.05 GB/s, algorithmic 8.11 GB/s
pt: concat2d2input_fwd_cpu_20_580_20_174_1: 14.52 us, SOL 8.31 GB/s, algorithmic 16.62 GB/s
pt: concat2d2input_fwd_cpu_8_512_8_512_1: 9.58 us, SOL 6.84 GB/s, algorithmic 13.68 GB/s
```
aten::cat implementation in NNC **without** conditionals
```
$ python -m benchmarks.tensorexpr --device cpu --mode fwd --jit_mode trace --cpu_fusion --cat_wo_conditionals concat
pt: concat2d2input_fwd_cpu_1_160_1_14_1: 4.67 us, SOL 0.30 GB/s, algorithmic 0.60 GB/s
pt: concat2d2input_fwd_cpu_1_580_1_174_1: 5.65 us, SOL 1.07 GB/s, algorithmic 2.14 GB/s
pt: concat2d2input_fwd_cpu_20_160_20_14_1: 6.10 us, SOL 4.56 GB/s, algorithmic 9.12 GB/s
pt: concat2d2input_fwd_cpu_20_580_20_174_1: 7.44 us, SOL 16.22 GB/s, algorithmic 32.44 GB/s
pt: concat2d2input_fwd_cpu_8_512_8_512_1: 6.46 us, SOL 10.14 GB/s, algorithmic 20.29 GB/s
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/53128
Reviewed By: bertmaher
Differential Revision: D26758613
Pulled By: navahgar
fbshipit-source-id: 00f56b7da630b42bc6e7ddd4444bae0cf3a5780a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/51564
Constructor logic was spread throughout InferenceModule and StaticRuntime. This diff unifies the two. After a lot of discussion on this diff D25961626 it became apparent that `clone` is uglier than a cheap StaticRuntime.
This means StaticRuntime is effectively StaticModule and the only code in the new StaticRuntime is the `run` functions.
```
graph, schema = PrepareForStaticModule(torchscript_module)
sm = StaticModule(graph, schema, options)
sm(inputs)
// or create many cheap runtimes with the module
sr = StaticRuntime(sm)
sr(inputs)
```
Changelist:
- Rename InferenceModule StaticModule
- Move all logic for construction into StaticModule
- Create a new StaticRuntime that only has a unique memory planner (everything else is in StaticModule)
- Update comments with explanation
- Propagate all changes to predictor integration
- Propagate all changes to python integration
- Change semantics to be a bit more PyTorch-standard (no "run" calls, no "get_" getters).
Test Plan:
buck test //caffe2/test:static_runtime
buck test caffe2/benchmarks/static_runtime:static_runtime_cpptest
Reviewed By: hlu1
Differential Revision: D25592967
fbshipit-source-id: 8233bed03137ce129137af2d44bce0095033ef0f
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/51483
This PR moves the conv weights of a frozen model to MKLDNN, and AOT reorders the weights. When the weights are already in MKLDNN, just computing a single conv by converting the input and output from/to mkldnn provides large speedups. I benchmark'd the results of the top 200 shapes in predictor [here](https://www.internalfb.com/phabricator/paste/view/P171537938), as well as verified that it sped up popular models in torchvision.
Test Plan: Imported from OSS
Reviewed By: navahgar
Differential Revision: D26696703
Pulled By: eellison
fbshipit-source-id: 0b4441bee4f6e0890a4540fbca3bb5e58b8c5adf
Summary:
This is a second attempt to use graph executor to run forward on a gradient. This allows a secondary chance to profile intermediate tensor introduced by autodiff.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/52136
Reviewed By: pbelevich
Differential Revision: D26693978
Pulled By: Krovatkin
fbshipit-source-id: 91dde8009a210950af8e5173668ada241e16dd52
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/52910
**Summary**
PR #52158 tried to move all JIT bindings from `torch._C` to a new
submodule `torch._C._jit`, but that...did not go well. This pull request
adds the new `torch._C._jit` submodule, but does not migrate the
existing bindings. Instead, it adds a unit test that fails if any new
bindings are added to `torch._C`. A comment in the test instructs
developers to add their new binding to the allowlist if it really should
be in `torch._C`, or to add it to the appropriate submodule (e.g
`torch._C._jit`, for example). The idea is to prevent the issue
described in #51691 from getting *worse* if it cannot be fixed.
**Test Plan**
Continuous integration.
**Fixes**
This commit fixes#51691.
Test Plan: Imported from OSS
Reviewed By: albanD
Differential Revision: D26698373
Pulled By: SplitInfinity
fbshipit-source-id: ec9f5426051227a513d4fd09512b624420e0100b
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/51754
This API allows you to manage multiple python interpreters in a single
process to deploy PyTorch models packaged with torch.package.
torch/csrc/deploy/deploy.h contains the API definition
torch/csrc/deploy/test_deploy.cpp has some examples.
Notes:
* mutex is added to PyTorchStreamReader to make it safe to use from multiple threads at once.
* USE_DEPLOY is only true for the special libtorch_deployinterpreter.so library, when enabled
we use a hash table to maintain PyObject <> at::Tensor mappping rather than the internal pointer
in Tensor since >1 interpreter may have a reference to the tensor.
* serialization.py has some additional functions for creating pickle objects
but keeping storages in memory for use transfering tensors between interpreters
Test Plan: Imported from OSS
Reviewed By: wconstab
Differential Revision: D26329468
Pulled By: zdevito
fbshipit-source-id: d75f4ebb9a27f1d911179d9996041bcb3ca04a07
