Skip to content

Conversation

@eqy
Copy link
Collaborator

@eqy eqy commented May 13, 2021

#55070

square really just calls pow so it seems a separate TensorIterator doesn't need to be built for it

@facebook-github-bot
Copy link
Contributor

facebook-github-bot commented May 13, 2021

🔗 Helpful links

💊 CI failures summary and remediations

As of commit b93b02c (more details on the Dr. CI page):


  • 31/31 failures possibly* introduced in this PR
    • 1/31 non-scanned failure(s)

🕵️ 29 new failures recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

See GitHub Actions build Lint / mypy (1/29)

Step: "Run autogen" (full log | diagnosis details | 🔁 rerun)

2021-05-17T17:16:01.2939137Z AssertionError: un...is function; just delete the dispatch key entirely
2021-05-17T17:16:01.2928763Z   File "/opt/hostedtoolcache/Python/3.8.10/x64/lib/python3.8/runpy.py", line 87, in _run_code
2021-05-17T17:16:01.2929675Z     exec(code, run_globals)
2021-05-17T17:16:01.2930601Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 1008, in <module>
2021-05-17T17:16:01.2931409Z     main()
2021-05-17T17:16:01.2932221Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 836, in main
2021-05-17T17:16:01.2933192Z     native_functions = parse_native_yaml(native_yaml_path)
2021-05-17T17:16:01.2934272Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 89, in parse_native_yaml
2021-05-17T17:16:01.2935315Z     rs.append(NativeFunction.from_yaml(e, loc))
2021-05-17T17:16:01.2936308Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/model.py", line 344, in from_yaml
2021-05-17T17:16:01.2937733Z     assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
2021-05-17T17:16:01.2939137Z AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
2021-05-17T17:16:01.2940265Z   in aten/src/ATen/native/native_functions.yaml:3893:
2021-05-17T17:16:01.2941475Z     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
2021-05-17T17:16:01.2942051Z 
2021-05-17T17:16:01.2942633Z real	0m0.601s
2021-05-17T17:16:01.2943130Z user	0m0.525s
2021-05-17T17:16:01.2943902Z sys	0m0.040s
2021-05-17T17:16:01.2951315Z ##[error]Process completed with exit code 1.
2021-05-17T17:16:01.3063113Z Post job cleanup.
2021-05-17T17:16:01.5402098Z [command]/usr/bin/git version
2021-05-17T17:16:01.5402587Z git version 2.31.1

See GitHub Actions build Lint / quick-checks (2/29)

Step: "C++ docs check" (full log | diagnosis details | 🔁 rerun)

2021-05-17T17:16:15.6957643Z AssertionError: un...is function; just delete the dispatch key entirely
2021-05-17T17:16:15.6939350Z   File "/opt/hostedtoolcache/Python/3.9.5/x64/lib/python3.9/runpy.py", line 87, in _run_code
2021-05-17T17:16:15.6940644Z     exec(code, run_globals)
2021-05-17T17:16:15.6941935Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 1008, in <module>
2021-05-17T17:16:15.6944190Z     main()
2021-05-17T17:16:15.6945727Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 836, in main
2021-05-17T17:16:15.6948349Z     native_functions = parse_native_yaml(native_yaml_path)
2021-05-17T17:16:15.6949883Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 89, in parse_native_yaml
2021-05-17T17:16:15.6951387Z     rs.append(NativeFunction.from_yaml(e, loc))
2021-05-17T17:16:15.6953303Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/model.py", line 344, in from_yaml
2021-05-17T17:16:15.6955378Z     assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
2021-05-17T17:16:15.6957643Z AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
2021-05-17T17:16:15.6959271Z   in aten/src/ATen/native/native_functions.yaml:3893:
2021-05-17T17:16:15.6961290Z     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
2021-05-17T17:16:15.7198933Z ##[error]Process completed with exit code 1.
2021-05-17T17:16:15.7263669Z ##[group]Run set -eux
2021-05-17T17:16:15.7264175Z �[36;1mset -eux�[0m
2021-05-17T17:16:15.7264999Z �[36;1mpython torch/testing/check_kernel_launches.py |& tee "${GITHUB_WORKSPACE}"/cuda_kernel_launch_checks.txt�[0m
2021-05-17T17:16:15.7303380Z shell: /bin/bash -e {0}
2021-05-17T17:16:15.7303807Z env:
2021-05-17T17:16:15.7304404Z   pythonLocation: /opt/hostedtoolcache/Python/3.9.5/x64
2021-05-17T17:16:15.7305232Z   LD_LIBRARY_PATH: /opt/hostedtoolcache/Python/3.9.5/x64/lib

See GitHub Actions build Lint / clang-tidy (3/29)

Step: "Generate build files" (full log | diagnosis details | 🔁 rerun)

2021-05-17T17:21:10.9357383Z AssertionError: un...is function; just delete the dispatch key entirely
2021-05-17T17:21:10.9340966Z   File "/opt/hostedtoolcache/Python/3.9.5/x64/lib/python3.9/runpy.py", line 87, in _run_code
2021-05-17T17:21:10.9342173Z     exec(code, run_globals)
2021-05-17T17:21:10.9343166Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 1008, in <module>
2021-05-17T17:21:10.9344400Z     main()
2021-05-17T17:21:10.9345208Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 836, in main
2021-05-17T17:21:10.9346821Z     native_functions = parse_native_yaml(native_yaml_path)
2021-05-17T17:21:10.9348005Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/gen.py", line 89, in parse_native_yaml
2021-05-17T17:21:10.9353502Z     rs.append(NativeFunction.from_yaml(e, loc))
2021-05-17T17:21:10.9354485Z   File "/home/runner/work/pytorch/pytorch/tools/codegen/model.py", line 344, in from_yaml
2021-05-17T17:21:10.9355800Z     assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
2021-05-17T17:21:10.9357383Z AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
2021-05-17T17:21:10.9358611Z   in /home/runner/work/pytorch/pytorch/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
2021-05-17T17:21:10.9360227Z     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
2021-05-17T17:21:10.9572369Z -- 
2021-05-17T17:21:10.9574445Z CMake Error at cmake/Codegen.cmake:213 (message):
2021-05-17T17:21:10.9575165Z   Failed to get generated_cpp list
2021-05-17T17:21:10.9576048Z Call Stack (most recent call first):
2021-05-17T17:21:10.9576684Z   caffe2/CMakeLists.txt:2 (include)
2021-05-17T17:21:10.9577108Z 
2021-05-17T17:21:10.9577407Z 
2021-05-17T17:21:10.9618767Z -- Configuring incomplete, errors occurred!

See GitHub Actions build Test tools / test (4/29)

Step: "Run tests" (full log | diagnosis details | 🔁 rerun)

2021-05-17T17:19:36.0117386Z AssertionError: 'x.../parallel/common_types.pyi:2:1: e[1325 chars]d]\n'
2021-05-17T17:19:36.0107831Z   File "/opt/hostedtoolcache/Python/3.9.5/x64/lib/python3.9/unittest/async_case.py", line 65, in _callTestMethod
2021-05-17T17:19:36.0108780Z     self._callMaybeAsync(method)
2021-05-17T17:19:36.0109780Z   File "/opt/hostedtoolcache/Python/3.9.5/x64/lib/python3.9/unittest/async_case.py", line 88, in _callMaybeAsync
2021-05-17T17:19:36.0110948Z     return self._asyncioTestLoop.run_until_complete(fut)
2021-05-17T17:19:36.0111979Z   File "/opt/hostedtoolcache/Python/3.9.5/x64/lib/python3.9/asyncio/base_events.py", line 642, in run_until_complete
2021-05-17T17:19:36.0112810Z     return future.result()
2021-05-17T17:19:36.0113733Z   File "/opt/hostedtoolcache/Python/3.9.5/x64/lib/python3.9/unittest/async_case.py", line 102, in _asyncioLoopRunner
2021-05-17T17:19:36.0114559Z     ret = await awaitable
2021-05-17T17:19:36.0115340Z   File "/home/runner/work/pytorch/pytorch/tools/test/test_actions_local_runner.py", line 179, in test_mypy
2021-05-17T17:19:36.0116213Z     self.assertEqual(expected, f.getvalue())
2021-05-17T17:19:36.0117386Z AssertionError: 'x my[862 chars]nt]\n' != 'x my[862 chars]nt]\ntorch/nn/parallel/common_types.pyi:2:1: e[1325 chars]d]\n'
2021-05-17T17:19:36.0118222Z   x mypy (skipped typestub generation)
2021-05-17T17:19:36.0119168Z   torch/some_stubs.pyi:3:17: error: Incompatible types in assignment (expression has type "None", variable has type "str")  [assignment]
2021-05-17T17:19:36.0120535Z   torch/some_stubs.pyi:4:17: error: Incompatible types in assignment (expression has type "float", variable has type "str")  [assignment]
2021-05-17T17:19:36.0121815Z   torch/some_cool_file.py:3:17: error: Incompatible types in assignment (expression has type "None", variable has type "str")  [assignment]
2021-05-17T17:19:36.0123116Z   torch/some_cool_file.py:4:17: error: Incompatible types in assignment (expression has type "float", variable has type "str")  [assignment]
2021-05-17T17:19:36.0124420Z   caffe2/some_cool_file.py:3:17: error: Incompatible types in assignment (expression has type "None", variable has type "str")  [assignment]
2021-05-17T17:19:36.0125763Z   caffe2/some_cool_file.py:4:17: error: Incompatible types in assignment (expression has type "float", variable has type "str")  [assignment]
2021-05-17T17:19:36.0127002Z + torch/nn/parallel/common_types.pyi:2:1: error: Module 'torch' has no attribute 'device'  [attr-defined]
2021-05-17T17:19:36.0128098Z + torch/nn/utils/rnn.pyi:42:28: error: Variable "torch.types._dtype" is not valid as a type  [valid-type]
2021-05-17T17:19:36.0129872Z + torch/nn/utils/rnn.pyi:42:28: note: See https://mypy.readthedocs.io/en/latest/common_issues.html#variables-vs-type-aliases

See GitHub Actions build Linux CI (pytorch-linux-xenial-py3.6-gcc5.4) / build (5/29)

Step: "Build PyTorch" (full log | diagnosis details | 🔁 rerun)

2021-05-17T17:17:15.1110424Z AssertionError: un...is function; just delete the dispatch key entirely
2021-05-17T17:17:15.1104605Z   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
2021-05-17T17:17:15.1105031Z     exec(code, run_globals)
2021-05-17T17:17:15.1105489Z   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
2021-05-17T17:17:15.1106163Z     main()
2021-05-17T17:17:15.1106843Z   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
2021-05-17T17:17:15.1107404Z     native_functions = parse_native_yaml(native_yaml_path)
2021-05-17T17:17:15.1107972Z   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
2021-05-17T17:17:15.1108544Z     rs.append(NativeFunction.from_yaml(e, loc))
2021-05-17T17:17:15.1109112Z   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
2021-05-17T17:17:15.1109711Z     "unnecessary dispatch table for this function; just delete the dispatch " \
2021-05-17T17:17:15.1110424Z AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
2021-05-17T17:17:15.1111305Z   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
2021-05-17T17:17:15.1112270Z     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
2021-05-17T17:17:15.1248972Z -- 
2021-05-17T17:17:15.1249639Z CMake Error at cmake/Codegen.cmake:213 (message):
2021-05-17T17:17:15.1250221Z   Failed to get generated_cpp list
2021-05-17T17:17:15.1250600Z Call Stack (most recent call first):
2021-05-17T17:17:15.1251179Z   caffe2/CMakeLists.txt:2 (include)
2021-05-17T17:17:15.1251450Z 
2021-05-17T17:17:15.1251597Z 
2021-05-17T17:17:15.1409549Z -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_bionic_py3_8_gcc9_coverage_build (6/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:25:31 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:25:31   File "/opt/conda/lib/python3.8/runpy.py", line 87, in _run_code
May 17 17:25:31     exec(code, run_globals)
May 17 17:25:31   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:25:31     main()
May 17 17:25:31   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:25:31     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:25:31   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:25:31     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:25:31   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 344, in from_yaml
May 17 17:25:31     assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
May 17 17:25:31 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:25:31   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:25:31     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:25:31 -- 
May 17 17:25:31 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:25:31   Failed to get generated_cpp list
May 17 17:25:31 Call Stack (most recent call first):
May 17 17:25:31   caffe2/CMakeLists.txt:2 (include)
May 17 17:25:31 
May 17 17:25:31 
May 17 17:25:31 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_bionic_py3_6_clang9_noarch_build (7/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:26:38 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:26:38   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:26:38     exec(code, run_globals)
May 17 17:26:38   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:26:38     main()
May 17 17:26:38   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:26:38     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:26:38   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:26:38     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:26:38   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:26:38     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:26:38 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:26:38   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:26:38     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:26:39 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:26:39   Failed to get generated_cpp list
May 17 17:26:39 Call Stack (most recent call first):
May 17 17:26:39   caffe2/CMakeLists.txt:2 (include)
May 17 17:26:39 
May 17 17:26:39 
May 17 17:26:39 -- 
May 17 17:26:39 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch-linux-xenial-py3-clang5-android-ndk-r19c-gradle-custom-build-single (8/29)

Step: "pytorch android gradle custom build single architecture (for PR)" (full log | diagnosis details | 🔁 rerun)

May 17 17:28:08 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:28:08   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:28:08     exec(code, run_globals)
May 17 17:28:08   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:28:08     main()
May 17 17:28:08   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:28:08     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:28:08   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:28:08     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:28:08   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:28:08     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:28:08 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:28:08   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:28:08     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:28:08 -- 
May 17 17:28:08 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:28:08   Failed to get generated_cpp list
May 17 17:28:08 Call Stack (most recent call first):
May 17 17:28:08   caffe2/CMakeLists.txt:2 (include)
May 17 17:28:08 
May 17 17:28:08 
May 17 17:28:08 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_xenial_py3_6_gcc5_4_build (9/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:23:55 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:23:55   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:23:55     exec(code, run_globals)
May 17 17:23:55   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:23:55     main()
May 17 17:23:55   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:23:55     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:23:55   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:23:55     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:23:55   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:23:55     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:23:55 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:23:55   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:23:55     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:23:55 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:23:55   Failed to get generated_cpp list
May 17 17:23:55 Call Stack (most recent call first):
May 17 17:23:55   caffe2/CMakeLists.txt:2 (include)
May 17 17:23:55 
May 17 17:23:55 
May 17 17:23:55 -- 
May 17 17:23:55 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_ios_12_0_0_x86_64_build (10/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:24:52 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:24:52   File "/Users/distiller/anaconda/lib/python3.8/runpy.py", line 87, in _run_code
May 17 17:24:52     exec(code, run_globals)
May 17 17:24:52   File "/Users/distiller/project/tools/codegen/gen.py", line 1008, in <module>
May 17 17:24:52     main()
May 17 17:24:52   File "/Users/distiller/project/tools/codegen/gen.py", line 836, in main
May 17 17:24:52     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:24:52   File "/Users/distiller/project/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:24:52     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:24:52   File "/Users/distiller/project/tools/codegen/model.py", line 344, in from_yaml
May 17 17:24:52     assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
May 17 17:24:52 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:24:52   in /Users/distiller/project/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:24:52     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:24:52 -- 
May 17 17:24:52 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:24:52   Failed to get generated_cpp list
May 17 17:24:52 Call Stack (most recent call first):
May 17 17:24:52   caffe2/CMakeLists.txt:2 (include)
May 17 17:24:52 
May 17 17:24:52 
May 17 17:24:52 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_ios_12_0_0_x86_64_lite_interpreter_build (11/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:23:38 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:23:38   File "/Users/distiller/anaconda/lib/python3.8/runpy.py", line 87, in _run_code
May 17 17:23:38     exec(code, run_globals)
May 17 17:23:38   File "/Users/distiller/project/tools/codegen/gen.py", line 1008, in <module>
May 17 17:23:38     main()
May 17 17:23:38   File "/Users/distiller/project/tools/codegen/gen.py", line 836, in main
May 17 17:23:38     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:23:38   File "/Users/distiller/project/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:23:38     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:23:38   File "/Users/distiller/project/tools/codegen/model.py", line 344, in from_yaml
May 17 17:23:38     assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
May 17 17:23:38 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:23:38   in /Users/distiller/project/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:23:38     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:23:38 -- 
May 17 17:23:38 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:23:38   Failed to get generated_cpp list
May 17 17:23:38 Call Stack (most recent call first):
May 17 17:23:38   caffe2/CMakeLists.txt:2 (include)
May 17 17:23:38 
May 17 17:23:38 
May 17 17:23:38 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_xenial_py3_clang5_android_ndk_r19c_x86_32_build (12/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:26:35 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:26:35   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:26:35     exec(code, run_globals)
May 17 17:26:35   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:26:35     main()
May 17 17:26:35   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:26:35     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:26:35   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:26:35     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:26:35   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:26:35     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:26:35 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:26:35   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:26:35     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:26:35 -- 
May 17 17:26:35 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:26:35   Failed to get generated_cpp list
May 17 17:26:35 Call Stack (most recent call first):
May 17 17:26:35   caffe2/CMakeLists.txt:2 (include)
May 17 17:26:35 
May 17 17:26:35 
May 17 17:26:35 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_bionic_cuda10_2_cudnn7_py3_9_gcc7_build (13/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:27:14 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:27:14   File "/opt/conda/lib/python3.9/runpy.py", line 87, in _run_code
May 17 17:27:14     exec(code, run_globals)
May 17 17:27:14   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:27:14     main()
May 17 17:27:14   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:27:14     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:27:14   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:27:14     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:27:14   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 344, in from_yaml
May 17 17:27:14     assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
May 17 17:27:14 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:27:14   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:27:14     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:27:14 -- 
May 17 17:27:14 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:27:14   Failed to get generated_cpp list
May 17 17:27:14 Call Stack (most recent call first):
May 17 17:27:14   caffe2/CMakeLists.txt:2 (include)
May 17 17:27:14 
May 17 17:27:14 
May 17 17:27:14 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_windows_vs2019_py36_cuda11.1_build (14/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
  File "C:\Jenkins\Miniconda3\lib\runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "C:\Users\circleci\project\tools\codegen\gen.py", line 1008, in <module>
    main()
  File "C:\Users\circleci\project\tools\codegen\gen.py", line 836, in main
    native_functions = parse_native_yaml(native_yaml_path)
  File "C:\Users\circleci\project\tools\codegen\gen.py", line 89, in parse_native_yaml
    rs.append(NativeFunction.from_yaml(e, loc))
  File "C:\Users\circleci\project\tools\codegen\model.py", line 345, in from_yaml
    "unnecessary dispatch table for this function; just delete the dispatch " \
AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
  in C:/Users/circleci/project/cmake/../aten/src/ATen\native/native_functions.yaml:3893:
    square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
-- 
CMake Error at cmake/Codegen.cmake:213 (message):
  Failed to get generated_cpp list
Call Stack (most recent call first):
  caffe2/CMakeLists.txt:2 (include)


-- Configuring incomplete, errors occurred!

See CircleCI build pytorch_macos_10_13_py3_build (15/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:24:29 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:24:29   File "/Users/distiller/workspace/miniconda3/lib/python3.7/runpy.py", line 85, in _run_code
May 17 17:24:29     exec(code, run_globals)
May 17 17:24:29   File "/Users/distiller/project/tools/codegen/gen.py", line 1008, in <module>
May 17 17:24:29     main()
May 17 17:24:29   File "/Users/distiller/project/tools/codegen/gen.py", line 836, in main
May 17 17:24:29     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:24:29   File "/Users/distiller/project/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:24:29     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:24:29   File "/Users/distiller/project/tools/codegen/model.py", line 345, in from_yaml
May 17 17:24:29     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:24:29 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:24:29   in /Users/distiller/project/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:24:29     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:24:29 -- 
May 17 17:24:29 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:24:29   Failed to get generated_cpp list
May 17 17:24:29 Call Stack (most recent call first):
May 17 17:24:29   caffe2/CMakeLists.txt:2 (include)
May 17 17:24:29 
May 17 17:24:29 
May 17 17:24:29 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_xenial_py3_clang7_onnx_build (16/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:23:39 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:23:39   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:23:39     exec(code, run_globals)
May 17 17:23:39   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:23:39     main()
May 17 17:23:39   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:23:39     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:23:39   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:23:39     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:23:39   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:23:39     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:23:39 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:23:39   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:23:39     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:23:39 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:23:39   Failed to get generated_cpp list
May 17 17:23:39 Call Stack (most recent call first):
May 17 17:23:39   caffe2/CMakeLists.txt:2 (include)
May 17 17:23:39 
May 17 17:23:39 
May 17 17:23:39 -- 
May 17 17:23:39 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_xenial_py3_clang5_asan_build (17/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:24:21 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:24:21   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:24:21     exec(code, run_globals)
May 17 17:24:21   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:24:21     main()
May 17 17:24:21   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:24:21     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:24:21   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:24:21     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:24:21   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:24:21     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:24:21 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:24:21   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:24:21     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:24:21 -- 
May 17 17:24:21 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:24:21   Failed to get generated_cpp list
May 17 17:24:21 Call Stack (most recent call first):
May 17 17:24:21   caffe2/CMakeLists.txt:2 (include)
May 17 17:24:21 
May 17 17:24:21 
May 17 17:24:21 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_xenial_cuda11_1_cudnn8_py3_gcc7_build (18/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:29:17 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:29:17   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:29:17     exec(code, run_globals)
May 17 17:29:17   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:29:17     main()
May 17 17:29:17   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:29:17     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:29:17   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:29:17     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:29:17   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:29:17     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:29:17 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:29:17   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:29:17     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:29:17 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:29:17   Failed to get generated_cpp list
May 17 17:29:17 Call Stack (most recent call first):
May 17 17:29:17   caffe2/CMakeLists.txt:2 (include)
May 17 17:29:17 
May 17 17:29:17 
May 17 17:29:17 -- 
May 17 17:29:17 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch-linux-xenial-py3-clang5-android-ndk-r19c-gradle-custom-build-single_lite_interpreter (19/29)

Step: "pytorch android gradle custom build single architecture (for PR)" (full log | diagnosis details | 🔁 rerun)

May 17 17:26:56 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:26:56   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:26:56     exec(code, run_globals)
May 17 17:26:56   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:26:56     main()
May 17 17:26:56   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:26:56     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:26:56   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:26:56     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:26:56   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:26:56     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:26:56 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:26:56   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:26:56     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:26:56 -- 
May 17 17:26:56 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:26:56   Failed to get generated_cpp list
May 17 17:26:56 Call Stack (most recent call first):
May 17 17:26:56   caffe2/CMakeLists.txt:2 (include)
May 17 17:26:56 
May 17 17:26:56 
May 17 17:26:56 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_macos_10_15_py3_build (20/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:26:49 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:26:49   File "/Users/distiller/workspace/miniconda3/lib/python3.8/runpy.py", line 87, in _run_code
May 17 17:26:49     exec(code, run_globals)
May 17 17:26:49   File "/Users/distiller/project/tools/codegen/gen.py", line 1008, in <module>
May 17 17:26:49     main()
May 17 17:26:49   File "/Users/distiller/project/tools/codegen/gen.py", line 836, in main
May 17 17:26:49     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:26:49   File "/Users/distiller/project/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:26:49     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:26:49   File "/Users/distiller/project/tools/codegen/model.py", line 344, in from_yaml
May 17 17:26:49     assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
May 17 17:26:49 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:26:49   in /Users/distiller/project/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:26:49     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:26:49 -- 
May 17 17:26:49 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:26:49   Failed to get generated_cpp list
May 17 17:26:49 Call Stack (most recent call first):
May 17 17:26:49   caffe2/CMakeLists.txt:2 (include)
May 17 17:26:49 
May 17 17:26:49 
May 17 17:26:49 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_bazel_build (21/29)

Step: "Bazel Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:25:29 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:25:29 Use --sandbox_debug to see verbose messages from the sandbox
May 17 17:25:29 Traceback (most recent call last):
May 17 17:25:29   File "/var/lib/jenkins/.cache/bazel/_bazel_jenkins/fdf6d09bf4b4f04a71e2a7dfceb40620/sandbox/processwrapper-sandbox/196/execroot/pytorch/bazel-out/host/bin/gen.runfiles/pytorch/tools/setup_helpers/gen.py", line 11, in <module>
May 17 17:25:29     tools.codegen.gen.main()
May 17 17:25:29   File "/var/lib/jenkins/.cache/bazel/_bazel_jenkins/fdf6d09bf4b4f04a71e2a7dfceb40620/sandbox/processwrapper-sandbox/196/execroot/pytorch/bazel-out/host/bin/gen.runfiles/pytorch/tools/codegen/gen.py", line 836, in main
May 17 17:25:29     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:25:29   File "/var/lib/jenkins/.cache/bazel/_bazel_jenkins/fdf6d09bf4b4f04a71e2a7dfceb40620/sandbox/processwrapper-sandbox/196/execroot/pytorch/bazel-out/host/bin/gen.runfiles/pytorch/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:25:29     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:25:29   File "/var/lib/jenkins/.cache/bazel/_bazel_jenkins/fdf6d09bf4b4f04a71e2a7dfceb40620/sandbox/processwrapper-sandbox/196/execroot/pytorch/bazel-out/host/bin/gen.runfiles/pytorch/tools/codegen/model.py", line 345, in from_yaml
May 17 17:25:29     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:25:29 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:25:29   in aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:25:29     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:25:29 ----------------
May 17 17:25:29 Note: The failure of target //:gen (with exit code 1) may have been caused by the fact that it is running under Python 3 instead of Python 2. Examine the error to determine if that appears to be the problem. Since this target is built in the host configuration, the only way to change its version is to set --host_force_python=PY2, which affects the entire build.
May 17 17:25:29 
May 17 17:25:29 If this error started occurring in Bazel 0.27 and later, it may be because the Python toolchain now enforces that targets analyzed as PY2 and PY3 run under a Python 2 and Python 3 interpreter, respectively. See https://github.com/bazelbuild/bazel/issues/7899 for more information.
May 17 17:25:29 ----------------
May 17 17:25:29 Target //:torch failed to build
May 17 17:25:29 Use --verbose_failures to see the command lines of failed build steps.
May 17 17:25:29 ERROR: /var/lib/jenkins/workspace/BUILD.bazel:646:1 Executing genrule //:generated_cpp failed (Exit 1) bash failed: error executing command /bin/bash -c ... (remaining 1 argument(s) skipped)

See CircleCI build pytorch_linux_bionic_rocm3_9_py3_6_build (22/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:27:44 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:27:44   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:27:44     exec(code, run_globals)
May 17 17:27:44   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:27:44     main()
May 17 17:27:44   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:27:44     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:27:44   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:27:44     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:27:44   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:27:44     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:27:44 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:27:44   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:27:44     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:27:44 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:27:44   Failed to get generated_cpp list
May 17 17:27:44 Call Stack (most recent call first):
May 17 17:27:44   caffe2/CMakeLists.txt:2 (include)
May 17 17:27:44 
May 17 17:27:44 
May 17 17:27:44 -- 
May 17 17:27:44 -- Configuring incomplete, errors occurred!

See CircleCI build binary_linux_libtorch_3_7m_cpu_gcc5_4_cxx11-abi_shared-with-deps_build (23/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
  File "/opt/conda/lib/python3.8/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/pytorch/tools/codegen/gen.py", line 1008, in <module>
    main()
  File "/pytorch/tools/codegen/gen.py", line 836, in main
    native_functions = parse_native_yaml(native_yaml_path)
  File "/pytorch/tools/codegen/gen.py", line 89, in parse_native_yaml
    rs.append(NativeFunction.from_yaml(e, loc))
  File "/pytorch/tools/codegen/model.py", line 344, in from_yaml
    assert dispatch != {DispatchKey.CompositeImplicitAutograd: cpp.name(func)}, \
AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
  in /pytorch/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
    square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
-- 
CMake Error at cmake/Codegen.cmake:213 (message):
  Failed to get generated_cpp list
Call Stack (most recent call first):
  caffe2/CMakeLists.txt:2 (include)


-- Configuring incomplete, errors occurred!

See CircleCI build pytorch_vulkan_linux_bionic_py3_6_clang9_build (24/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:25:54 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:25:54   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:25:54     exec(code, run_globals)
May 17 17:25:54   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:25:54     main()
May 17 17:25:54   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:25:54     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:25:54   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:25:54     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:25:54   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:25:54     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:25:54 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:25:54   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:25:54     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:25:54 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:25:54   Failed to get generated_cpp list
May 17 17:25:54 Call Stack (most recent call first):
May 17 17:25:54   caffe2/CMakeLists.txt:2 (include)
May 17 17:25:54 
May 17 17:25:54 
May 17 17:25:54 -- 
May 17 17:25:54 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_xenial_py3_clang5_mobile_custom_build_static (25/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:26:24 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:26:24   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:26:24     exec(code, run_globals)
May 17 17:26:24   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:26:24     main()
May 17 17:26:24   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:26:24     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:26:24   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:26:24     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:26:24   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:26:24     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:26:24 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:26:24   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:26:24     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:26:24 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:26:24   Failed to get generated_cpp list
May 17 17:26:24 Call Stack (most recent call first):
May 17 17:26:24   caffe2/CMakeLists.txt:2 (include)
May 17 17:26:24 
May 17 17:26:24 
May 17 17:26:24 -- 
May 17 17:26:24 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_xla_linux_bionic_py3_6_clang9_build (26/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:27:09 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:27:09   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:27:09     exec(code, run_globals)
May 17 17:27:09   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:27:09     main()
May 17 17:27:09   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:27:09     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:27:09   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:27:09     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:27:09   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:27:09     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:27:09 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:27:09   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:27:09     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:27:09 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:27:09   Failed to get generated_cpp list
May 17 17:27:09 Call Stack (most recent call first):
May 17 17:27:09   caffe2/CMakeLists.txt:2 (include)
May 17 17:27:09 
May 17 17:27:09 
May 17 17:27:09 -- 
May 17 17:27:09 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_xenial_py3_clang5_mobile_custom_build_dynamic (27/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:26:31 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:26:31   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:26:31     exec(code, run_globals)
May 17 17:26:31   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:26:31     main()
May 17 17:26:31   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:26:31     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:26:31   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:26:31     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:26:31   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:26:31     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:26:31 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:26:31   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:26:31     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:26:31 -- 
May 17 17:26:31 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:26:31   Failed to get generated_cpp list
May 17 17:26:31 Call Stack (most recent call first):
May 17 17:26:31   caffe2/CMakeLists.txt:2 (include)
May 17 17:26:31 
May 17 17:26:31 
May 17 17:26:31 -- Configuring incomplete, errors occurred!

See CircleCI build pytorch_windows_vs2019_py36_cpu_build (28/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
  File "C:\Jenkins\Miniconda3\lib\runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "C:\Users\circleci\project\tools\codegen\gen.py", line 1008, in <module>
    main()
  File "C:\Users\circleci\project\tools\codegen\gen.py", line 836, in main
    native_functions = parse_native_yaml(native_yaml_path)
  File "C:\Users\circleci\project\tools\codegen\gen.py", line 89, in parse_native_yaml
    rs.append(NativeFunction.from_yaml(e, loc))
  File "C:\Users\circleci\project\tools\codegen\model.py", line 345, in from_yaml
    "unnecessary dispatch table for this function; just delete the dispatch " \
AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
  in C:/Users/circleci/project/cmake/../aten/src/ATen\native/native_functions.yaml:3893:
    square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
-- 
CMake Error at cmake/Codegen.cmake:213 (message):
  Failed to get generated_cpp list
Call Stack (most recent call first):
  caffe2/CMakeLists.txt:2 (include)


-- Configuring incomplete, errors occurred!

See CircleCI build pytorch_linux_xenial_py3_clang5_mobile_build (29/29)

Step: "Build" (full log | diagnosis details | 🔁 rerun)

May 17 17:23:52 AssertionError: unnecessary dis...is function; just delete the dispatch key entirely
May 17 17:23:52   File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code
May 17 17:23:52     exec(code, run_globals)
May 17 17:23:52   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 1008, in <module>
May 17 17:23:52     main()
May 17 17:23:52   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 836, in main
May 17 17:23:52     native_functions = parse_native_yaml(native_yaml_path)
May 17 17:23:52   File "/var/lib/jenkins/workspace/tools/codegen/gen.py", line 89, in parse_native_yaml
May 17 17:23:52     rs.append(NativeFunction.from_yaml(e, loc))
May 17 17:23:52   File "/var/lib/jenkins/workspace/tools/codegen/model.py", line 345, in from_yaml
May 17 17:23:52     "unnecessary dispatch table for this function; just delete the dispatch " \
May 17 17:23:52 AssertionError: unnecessary dispatch table for this function; just delete the dispatch key entirely
May 17 17:23:52   in /var/lib/jenkins/workspace/cmake/../aten/src/ATen/native/native_functions.yaml:3893:
May 17 17:23:52     square.out(Tensor self, *, Tensor(a!) out) -> Tensor(a!)
May 17 17:23:52 -- 
May 17 17:23:52 CMake Error at cmake/Codegen.cmake:213 (message):
May 17 17:23:52   Failed to get generated_cpp list
May 17 17:23:52 Call Stack (most recent call first):
May 17 17:23:52   caffe2/CMakeLists.txt:2 (include)
May 17 17:23:52 
May 17 17:23:52 
May 17 17:23:52 -- Configuring incomplete, errors occurred!

1 failure not recognized by patterns:

Job Step Action
GitHub Actions Linux CI (pytorch-linux-xenial-py3.6-gcc5.4) / render_test_results Download PyTorch Test Reports 🔁 rerun

ci.pytorch.org: 1 failed


This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

@eqy
Copy link
Collaborator Author

eqy commented May 14, 2021

Looks like I broke derivatives :/

@eqy eqy requested review from albanD and soulitzer as code owners May 14, 2021 00:30
@bdhirsh bdhirsh added the triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module label May 14, 2021
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, pow is already structured. The const_cast is needed because the C++ API signatures generated in build/aten/src/ATen/Functions.h all return mutable Tensor& instead of const Tensor&. We have an open issue for it here: #54001

@bdhirsh
Copy link
Contributor

bdhirsh commented May 14, 2021

Looks like I broke derivatives :/

Oh, this one's interesting. square LOOKS like it has backend specific implementations, since it has a CPU/CUDA dispatch entry for the out variant. But it looks like that entry is wrong; square_out is actually a mathematically composite kernel, since its implementation just calls into at::pow_out. That's why it was able to get away without having any derivatives.yaml entries. By trying to make it structured, the codegen will try to register cpu/cuda-specific implementations of square and square_, which assume that there's an actual autograd formula.

I think that you can just delete the dispatch table entry of square_out in native_functions.yaml, which will tell the codegen that all 3 ops are actually composite kernels:

  dispatch:
    CPU, CUDA: square_out

@eqy
Copy link
Collaborator Author

eqy commented May 14, 2021

Thanks! I was confused as to how moving to structured could have broke derivatives if it didn't need one registered previously. Interestingly adding a square entry in derivatives.yaml didn't seem to fix it but I'm not familiar with the intricacies of registering a function there.

@eqy
Copy link
Collaborator Author

eqy commented May 14, 2021

It looks like without the dispatch table codegen doesn't actually generate a function definition. This seems tricky as I can't find any examples of structured functions without dispatch table either.

@bdhirsh
Copy link
Contributor

bdhirsh commented May 14, 2021

It looks like without the dispatch table codegen doesn't actually generate a function definition. This seems tricky as I can't find any examples of structured functions without dispatch table either.

Oh, you can fix that by manually specifying that the operator is composite. So you can update the dispatch table to:

 dispatch:
    CompositeImplicitAutograd: square_out

Right now the codegen assumes that your op is composite if you don't specify a dispatch entry, but it looks like it only does that for unstructured operators (here). @ezyang is there any reason we don't allow defaulting the dispatch table for structured ops? I know we have some logic to auto-generate composite kernels for structured ops in some cases, but I don't think that would interfere with the defaulting here.

@eqy
Copy link
Collaborator Author

eqy commented May 14, 2021

codegen doesn't seem to like that on my end:
AssertionError: Do not explicitly specify CompositeExplicitAutograd dispatch key on structured functions, they will be automatically generated for you

@bdhirsh
Copy link
Contributor

bdhirsh commented May 14, 2021

codegen doesn't seem to like that on my end:
AssertionError: Do not explicitly specify CompositeExplicitAutograd dispatch key on structured functions, they will be automatically generated for you

EDIT: Nevermind, the codegen doesn't let you do that either. I think that's a bug, let me throw up a quick patch that should allow you do not specify a dispatch table for CompositeImplicitAutograd, structured ops.


Ah, you want CompositeImplicitAutograd, not CompositeExplicitAutograd. the Implicit is the part that means the op is pasing off the work of dealing with autograd to whatever ops it calls, which is what we want to do here (square should let pow figure out how to properly deal with the backward pass).

Yeah, the codegen does some fancy autogeneration stuff for CompositeExplicitAutograd in some cases, which is why it throws you an error when you write it yourself for structured ops.

@bdhirsh
Copy link
Contributor

bdhirsh commented May 17, 2021

@eqy The fix is a one line change (see here)- do you want to add it in to your existing PR? That'll test the fix more fully.

bdhirsh added a commit that referenced this pull request Jun 2, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 2, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 2, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 2, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 3, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 3, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 10, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 10, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 11, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 11, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 11, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 11, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 11, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 11, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 11, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 11, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 14, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 14, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 14, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 14, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 15, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 15, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 15, 2021
…at::meta::"

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
bdhirsh added a commit that referenced this pull request Jun 15, 2021
**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.


Differential Revision: [D28711299](https://our.internmc.facebook.com/intern/diff/D28711299)

[ghstack-poisoned]
facebook-github-bot pushed a commit that referenced this pull request Jun 15, 2021
Summary:
Pull Request resolved: #58570

**What the PR does**
Generate a fast-path `at::meta::{op}` API for calling meta functions without having to go through the dispatcher. This will be important for perf for external backends that want to use meta functions for shape checking (which seems likely to be what we end up doing for LazyTensorCore).

**Details**
In order to avoid naming collisions I had to make two small changes:
- rename `MetaFunctions.h` template -> `NativeMetaFunctions.h` (this is the file that declares the impl() function for every structured operator).
- rename the meta class: `at::meta::{op}::meta()` -> `at::meta::structured_{op}::meta()`

I also deleted a few unnecessary includes, since any file that includes NativeFunctions.h will automatically include NativeMetaFunctions.h.

**Why I made the change**
This change isn't actually immediately used anywhere; I already started writing it because I thought it would be useful for structured composite ops, but that isn't actually true (see [comment](#58266 (comment))). The change feels useful and unambiguous though so I think it's safe to add. I added explicit tests for C++ meta function calls just to ensure that I wrote it correctly - which is actually how I hit the internal linkage issue in the PR below this in the stack.

Test Plan: Imported from OSS

Reviewed By: pbelevich

Differential Revision: D28711299

Pulled By: bdhirsh

fbshipit-source-id: d410d17358c2b406f0191398093f17308b3c6b9e
@ezyang ezyang requested a review from SplitInfinity July 6, 2021 16:40
@pytorchbot
Copy link
Collaborator

Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as Stale.
Feel free to remove the Stale label if you feel this was a mistake.
If you are unable to remove the Stale label please contact a maintainer in order to do so.
Stale pull requests will automatically be closed 30 days after being marked Stale

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

cla signed open source Stale triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants