AvgPool: Ensure all cells are valid in ceil mode

[peterbell10]

Closes #36977

This avoid the division by zero that was causing NaNs to appear in the output. AvgPooling2d and AvgPooling3d both had this issue on CPU and CUDA.

[dr-ci]

💊 CI failures summary and remediations

As of commit 0b00129 (more details on the Dr. CI page):

---

• 1/1 failures introduced in this PR

---

🕵️ 1 new failure recognized by patterns

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

pytorch_macos_10_13_py3_test (1/1)

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

Jul 13 16:44:18 [E request_callback_impl.cpp:168] Received error while processing request type 2: PickleError: ScriptModules cannot be deepcopied using copy.deepcopy or saved using torch.save. Mixed serialization of script and non-script modules is not supported. For purely script modules use my_script_module.save() instead.

Jul 13 16:44:18   /Users/distiller/workspace/miniconda3/lib/python3.7/site-packages/torch/distributed/rpc/internal.py(93): serialize 
Jul 13 16:44:18   /Users/distiller/workspace/miniconda3/lib/python3.7/site-packages/torch/distributed/rpc/internal.py(145): serialize 
Jul 13 16:44:18  
Jul 13 16:44:18 [E request_callback_impl.cpp:168] Received error while processing request type 2: PickleError: ScriptModules cannot be deepcopied using copy.deepcopy or saved using torch.save. Mixed serialization of script and non-script modules is not supported. For purely script modules use my_script_module.save(<filename>) instead. 
Jul 13 16:44:18  
Jul 13 16:44:18 At: 
Jul 13 16:44:18   /Users/distiller/workspace/miniconda3/lib/python3.7/site-packages/torch/jit/_script.py(570): __getstate__ 
Jul 13 16:44:18   /Users/distiller/workspace/miniconda3/lib/python3.7/site-packages/torch/distributed/rpc/internal.py(93): serialize 
Jul 13 16:44:18   /Users/distiller/workspace/miniconda3/lib/python3.7/site-packages/torch/distributed/rpc/internal.py(145): serialize 
Jul 13 16:44:18  
Jul 13 16:44:18 [E request_callback_impl.cpp:168] Received error while processing request type 2: PickleError: ScriptModules cannot be deepcopied using copy.deepcopy or saved using torch.save. Mixed serialization of script and non-script modules is not supported. For purely script modules use my_script_module.save(<filename>) instead. 
Jul 13 16:44:18  
Jul 13 16:44:18 At: 
Jul 13 16:44:18   /Users/distiller/workspace/miniconda3/lib/python3.7/site-packages/torch/jit/_script.py(570): __getstate__ 
Jul 13 16:44:18   /Users/distiller/workspace/miniconda3/lib/python3.7/site-packages/torch/distributed/rpc/internal.py(93): serialize 
Jul 13 16:44:18   /Users/distiller/workspace/miniconda3/lib/python3.7/site-packages/torch/distributed/rpc/internal.py(145): serialize 
Jul 13 16:44:18  
Jul 13 16:44:19 ok (2.252s) 
Jul 13 16:44:21   test_unexepected_kwarg_is_specified (__main__.JitRpcTestWithSpawn) ... ok (2.494s) 
Jul 13 16:44:24   test_user_rrefs_confirmed (__main__.JitRpcTestWithSpawn) ... ok (2.123s) 
Jul 13 16:44:26   test_user_rrefs_confirmed_remote (__main__.JitRpcTestWithSpawn) ... ok (2.316s) 

---

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 or post in the (internal) Dr. CI Users group.

See how this bot performed.

This comment has been revised 1 time.

[ezyang]

Needs benchmarks before and after

[facebook-github-bot]

@ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[ezyang]

Logic looks pretty reasonable. Benchmarking should show a perf hit but hopefully it is not too bad.

[facebook-github-bot]

@ezyang merged this pull request in 87bf04f.

[peterbell10]

I've run the long benchmarks for AvgPool2d and AvgPool3d. CUDA benchmarks show no measurable difference, cpu benchmark shows up to a 1% slowdown for AvgPool2d and 2% for AvgPool3d.