Port min kernel to structured kernels.

[ysiraichi]

Stack from ghstack:

• -> Port min kernel to structured kernels. #61450
• Port max kernel to structured kernels. #61449

Tracking issue: #55070

Differential Revision: D29741713

[facebook-github-bot]

🔗 Helpful links

• 🧪  See artifacts and rendered test results at hud.pytorch.org/pr/61450
• 📄  Preview docs built from this PR
• 📄  Preview C++ docs built from this PR
• 🔧  Opt-in to CIFlow to control what jobs run on your PRs

💊 CI failures summary and remediations

As of commit 91507a1 (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:

linux-xenial-cuda11.3-py3.6-gcc7 / test (default, 2, 2, linux.8xlarge.nvidia.gpu) (1/1)

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

2021-09-21T17:08:02.1986802Z RuntimeError: CUDA error: device-side assert triggered

2021-09-21T17:07:59.0109996Z   File "/opt/conda/lib/python3.6/site-packages/torch/cuda/__init__.py", line 493, in synchronize
2021-09-21T17:07:59.0110835Z     return torch._C._cuda_synchronize()
2021-09-21T17:07:59.0111704Z RuntimeError: CUDA error: device-side assert triggered
2021-09-21T17:07:59.0112763Z CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
2021-09-21T17:07:59.0113775Z For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
2021-09-21T17:08:02.1977308Z /var/lib/jenkins/workspace/aten/src/ATen/native/cuda/TensorCompare.cu:161: _assert_async_cuda_kernel: block: [0,0,0], thread: [0,0,0] Assertion `input[0] != c10::complex<float>(0, 0)` failed.
2021-09-21T17:08:02.1982070Z Traceback (most recent call last):
2021-09-21T17:08:02.1982666Z   File "<string>", line 4, in <module>
2021-09-21T17:08:02.1984296Z   File "/opt/conda/lib/python3.6/site-packages/torch/cuda/__init__.py", line 493, in synchronize
2021-09-21T17:08:02.1985877Z     return torch._C._cuda_synchronize()
2021-09-21T17:08:02.1986802Z RuntimeError: CUDA error: device-side assert triggered
2021-09-21T17:08:02.1987851Z CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
2021-09-21T17:08:02.1988845Z For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
2021-09-21T17:08:02.4127859Z ok (12.692s)
2021-09-21T17:08:02.4199281Z   test_gather_bool (__main__.TestCuda) ... ok (0.007s)
2021-09-21T17:08:02.4243492Z   test_get_device_index (__main__.TestCuda) ... ok (0.004s)
2021-09-21T17:08:02.4304689Z   test_get_set_rng_state_all (__main__.TestCuda) ... ok (0.006s)
2021-09-21T17:08:02.4544704Z   test_grad_scaling_accumulation (__main__.TestCuda) ... ok (0.024s)
2021-09-21T17:08:02.5017497Z   test_grad_scaling_autocast (__main__.TestCuda) ... ok (0.047s)
2021-09-21T17:08:02.5291978Z   test_grad_scaling_clipping (__main__.TestCuda) ... ok (0.027s)
2021-09-21T17:08:02.5560886Z   test_grad_scaling_clipping_separate_unscale (__main__.TestCuda) ... ok (0.027s)

---

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.

[ysiraichi]

The CI failure is not related to this PR.

[bdhirsh]

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

[bdhirsh]

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

[bdhirsh]

Actually, is that test failure unrelated? It's a test suite for meta (TestSparseMETA), looks meta tensor related. The failing check lives here:

pytorch/c10/core/TensorImpl.h

Line 1151 in 6c8cb9b

storage_initialized(),

- meta tensors have storage (so has_storage() returns true), but it's nullptr so storage_initialized() returns false).

This sounds like a meta coverage thing where enabling the meta API for min may have gotten some tests further along, so you might need to mark a test in test_sparse.py with skip_meta?

[ezyang]

Hum, we probably shouldn't be running any of the sparse tests under meta tensors. Oops!

[ysiraichi]

@bdhirsh Yeah. That does seem to be the case. I thought I had @skipMeta all of them, though.
@ezyang Does adding "meta" as the except_for parameter when instantiating tests for both TestSparse and TestSparseUnaryUfuncs sounds reasonable?

[ezyang]

Does adding "meta" as the except_for parameter when instantiating tests for both TestSparse and TestSparseUnaryUfuncs sounds reasonable?

sgtm

[ezyang]

@bdhirsh is on vacation this week; if you want this imported faster than that give a holler

[ysiraichi]

Don't worry. I still want to do some refactoring in this PR.

[codecov]

Codecov Report

Merging #61450 (91507a1) into gh/ysiraichi/16/base (1b02641) will increase coverage by 6.29%.
The diff coverage is n/a.

@@                   Coverage Diff                    @@
##           gh/ysiraichi/16/base   #61450      +/-   ##
========================================================
+ Coverage                 60.08%   66.38%   +6.29%     
========================================================
  Files                       657      738      +81     
  Lines                     85345    94175    +8830     
========================================================
+ Hits                      51277    62514   +11237     
+ Misses                    34068    31661    -2407     

[bdhirsh]

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

[bdhirsh]

@ysiraichi let me know if you're still doing any refactoring. Otherwise I'll go ahead and try to land!

[ysiraichi]

@bdhirsh Sorry for the delay. I'm done refactoring this PR.

[bdhirsh]

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