Add missing C10_API to dispatch key TLS setter/getters

[ezyang]

Stack from ghstack:

• Replace catchall kernels with CompoundOp dispatch key. #32782 Replace catchall kernels with CompoundOp dispatch key.
• Stop using dispatchTypeId to do checks for tensor list unwrap. #32787 Stop using dispatchTypeId to do checks for tensor list unwrap.
• Centralize addition of "always on" dispatch keys. #32734 Centralize addition of "always on" dispatch keys.
• Make DispatchKeyGuards accept DispatchKey::Undefined #32729 Make DispatchKeyGuards accept DispatchKey::Undefined
• Rename DispatchKey::UndefinedTensorId to Undefined #32728 Rename DispatchKey::UndefinedTensorId to Undefined
• Delete copy/move constructors on these RAII guards. #32727 Delete copy/move constructors on these RAII guards.
• Add missing C10_API to dispatch key TLS setter/getters #32557 Add missing C10_API to dispatch key TLS setter/getters
• Improve documentation in dispatcher; remove unnecessary optional #32533 Improve documentation in dispatcher; remove unnecessary optional

Signed-off-by: Edward Z. Yang ezyang@fb.com

Differential Revision: D19579853

[kostmo]

💊 CircleCI build failures summary and remediations

As of commit 67890aa:

None of the build failures appear to be your fault.

• 3/3 recognized as flaky ❄️
  • Re-run these jobs?

Detailed failure analysis

One may explore the probable reasons each build failed interactively on the Dr. CI website.

❄️ 3 failures recognized as flaky

The following build failures have been detected as flaky and may not be your fault:

pytorch_linux_xenial_py3_6_gcc5_4_test (1/3)

Step: "Test" (full log | pattern match details) ❄️

Jan 28 17:28:22 RuntimeError: quantization scale should be > 0

Jan 28 17:28:22   File "test_quantization.py", line 730, in test_quantized_rnn 
Jan 28 17:28:22     y, (h, c) = cell_dq(x, (h, c)) 
Jan 28 17:28:22   File "/opt/conda/lib/python3.6/site-packages/torch/nn/modules/module.py", line 540, in __call__ 
Jan 28 17:28:22     result = self.forward(*input, **kwargs) 
Jan 28 17:28:22   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 373, in forward 
Jan 28 17:28:22     return self.forward_tensor(input, hx) 
Jan 28 17:28:22   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 334, in forward_tensor 
Jan 28 17:28:22     input, hx, batch_sizes, max_batch_size, sorted_indices) 
Jan 28 17:28:22   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 315, in forward_impl 
Jan 28 17:28:22     self.batch_first, dtype=self.dtype, use_dynamic=True) 
Jan 28 17:28:22 RuntimeError: quantization scale should be > 0 
Jan 28 17:28:22  
Jan 28 17:28:22 ---------------------------------------------------------------------- 
Jan 28 17:28:22 Ran 36 tests in 57.585s 
Jan 28 17:28:22  
Jan 28 17:28:22 FAILED (errors=1, skipped=1) 
Jan 28 17:28:22  
Jan 28 17:28:22 Generating XML reports... 
Jan 28 17:28:22 Traceback (most recent call last): 
Jan 28 17:28:22   File "test/run_test.py", line 456, in <module> 
Jan 28 17:28:22     main() 

pytorch_linux_xenial_cuda10_1_cudnn7_py3_NO_AVX2_test (2/3)

Step: "Test" (full log | pattern match details) ❄️

Jan 28 19:45:42 RuntimeError: quantization scale should be > 0

Jan 28 19:45:42   File "test_quantization.py", line 730, in test_quantized_rnn 
Jan 28 19:45:42     y, (h, c) = cell_dq(x, (h, c)) 
Jan 28 19:45:42   File "/opt/conda/lib/python3.6/site-packages/torch/nn/modules/module.py", line 540, in __call__ 
Jan 28 19:45:42     result = self.forward(*input, **kwargs) 
Jan 28 19:45:42   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 373, in forward 
Jan 28 19:45:42     return self.forward_tensor(input, hx) 
Jan 28 19:45:42   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 334, in forward_tensor 
Jan 28 19:45:42     input, hx, batch_sizes, max_batch_size, sorted_indices) 
Jan 28 19:45:42   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 315, in forward_impl 
Jan 28 19:45:42     self.batch_first, dtype=self.dtype, use_dynamic=True) 
Jan 28 19:45:42 RuntimeError: quantization scale should be > 0 
Jan 28 19:45:42  
Jan 28 19:45:42 ---------------------------------------------------------------------- 
Jan 28 19:45:42 Ran 36 tests in 57.614s 
Jan 28 19:45:42  
Jan 28 19:45:42 FAILED (errors=1, skipped=1) 
Jan 28 19:45:42  
Jan 28 19:45:42 Generating XML reports... 
Jan 28 19:45:42 Traceback (most recent call last): 
Jan 28 19:45:42   File "test/run_test.py", line 456, in <module> 
Jan 28 19:45:42     main() 

pytorch_linux_xenial_cuda10_1_cudnn7_py3_gcc7_test (3/3)

Step: "Test" (full log | pattern match details) ❄️

Jan 28 19:46:36 RuntimeError: quantization scale should be > 0

Jan 28 19:46:36   File "test_quantization.py", line 730, in test_quantized_rnn 
Jan 28 19:46:36     y, (h, c) = cell_dq(x, (h, c)) 
Jan 28 19:46:36   File "/opt/conda/lib/python3.6/site-packages/torch/nn/modules/module.py", line 540, in __call__ 
Jan 28 19:46:36     result = self.forward(*input, **kwargs) 
Jan 28 19:46:36   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 373, in forward 
Jan 28 19:46:36     return self.forward_tensor(input, hx) 
Jan 28 19:46:36   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 334, in forward_tensor 
Jan 28 19:46:36     input, hx, batch_sizes, max_batch_size, sorted_indices) 
Jan 28 19:46:36   File "/opt/conda/lib/python3.6/site-packages/torch/nn/quantized/dynamic/modules/rnn.py", line 315, in forward_impl 
Jan 28 19:46:36     self.batch_first, dtype=self.dtype, use_dynamic=True) 
Jan 28 19:46:36 RuntimeError: quantization scale should be > 0 
Jan 28 19:46:36  
Jan 28 19:46:36 ---------------------------------------------------------------------- 
Jan 28 19:46:36 Ran 36 tests in 59.177s 
Jan 28 19:46:36  
Jan 28 19:46:36 FAILED (errors=1, skipped=1) 
Jan 28 19:46:36  
Jan 28 19:46:36 Generating XML reports... 
Jan 28 19:46:36 Traceback (most recent call last): 
Jan 28 19:46:36   File "test/run_test.py", line 456, in <module> 
Jan 28 19:46:36     main() 

---

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 on the GitHub issue tracker.

This comment has been revised 6 times.

[pbelevich]

It would be nice to have an explanation why we need this change

[ezyang]

@pbelevich Without the annotation I can't actually use these methods from an external library, because they get hidden visibility. C10_API makes the symbols visible so they can be used externally.

[pbelevich]

@ezyang I know that, I don't know who and why is going to use them :)

[ezyang]

Oh! I wanted to use them in the example code in #32558 to toggle dispatch keys manually (as I needed some non-standard behavior that the pre-canned guards couldn't do) but I couldn't.

[facebook-github-bot]

@ezyang merged this pull request in 5ffa1ef.