Migrate THCTensor_copyIgnoringOverlaps to ATen

[peterbell10]

Stack from ghstack:

• -> Migrate THCTensor_copyIgnoringOverlaps to ATen #63505
• Migrate legacy lstsq from THC to ATen (CUDA) #63504
• Kill THCUNN #63429
• Migrate thnn_conv2d from THC to ATen #63428

This isn't a public operator, just a helper function used in CUDA_tensor_apply.

Differential Revision: D30441305

[facebook-github-bot]

🔗 Helpful links

• 🧪 See artifacts and rendered test results at hud.pytorch.org/pr/63505
• 📄 Preview docs built from this PR

💊 CI failures summary and remediations

As of commit 6779485 (more details on the Dr. CI page):

---

• 3/4 failures possibly* introduced in this PR
  • 1/3 non-scanned failure(s)
• 1/4 tentatively recognized as flaky ❄️
  • Click here to rerun these jobs

---

🕵️ 2 new failures recognized by patterns

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

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

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

2021-08-20T05:50:03.7251126Z Build left local git repository checkout dirty

2021-08-20T05:50:03.1430199Z real	132m1.600s
2021-08-20T05:50:03.1430635Z user	199m22.185s
2021-08-20T05:50:03.1431034Z sys	97m59.129s
2021-08-20T05:50:03.1431489Z + assert_git_not_dirty
2021-08-20T05:50:03.1432816Z + [[ linux-xenial-cuda11.1-py3.6-gcc7-default != *rocm* ]]
2021-08-20T05:50:03.1434147Z + [[ linux-xenial-cuda11.1-py3.6-gcc7-default != *xla* ]]
2021-08-20T05:50:03.1435868Z ++ git status --porcelain
2021-08-20T05:50:03.7248724Z + git_status='?? third_party/breakpad/'
2021-08-20T05:50:03.7249557Z + [[ -n ?? third_party/breakpad/ ]]
2021-08-20T05:50:03.7250358Z + echo 'Build left local git repository checkout dirty'
2021-08-20T05:50:03.7251126Z Build left local git repository checkout dirty
2021-08-20T05:50:03.7251887Z + echo 'git status --porcelain:'
2021-08-20T05:50:03.7252560Z git status --porcelain:
2021-08-20T05:50:03.7253397Z + echo '?? third_party/breakpad/'
2021-08-20T05:50:03.7253980Z ?? third_party/breakpad/
2021-08-20T05:50:03.7254429Z + exit 1
2021-08-20T05:50:03.7254864Z + cleanup
2021-08-20T05:50:03.7255272Z + retcode=1
2021-08-20T05:50:03.7255681Z + set +x
2021-08-20T05:50:03.7256170Z =================== sccache compilation log ===================
2021-08-20T05:50:03.7517499Z =========== If your build fails, please take a look at the log above for possible reasons ===========

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

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

2021-08-20T05:11:54.3992038Z Build left local git repository checkout dirty

2021-08-20T05:11:53.8242028Z real	91m17.676s
2021-08-20T05:11:53.8242529Z user	148m37.241s
2021-08-20T05:11:53.8242909Z sys	34m43.307s
2021-08-20T05:11:53.8243298Z + assert_git_not_dirty
2021-08-20T05:11:53.8244652Z + [[ linux-xenial-cuda11.1-py3.6-gcc7-default != *rocm* ]]
2021-08-20T05:11:53.8245933Z + [[ linux-xenial-cuda11.1-py3.6-gcc7-default != *xla* ]]
2021-08-20T05:11:53.8246884Z ++ git status --porcelain
2021-08-20T05:11:54.3988721Z + git_status='?? third_party/breakpad/'
2021-08-20T05:11:54.3990265Z + [[ -n ?? third_party/breakpad/ ]]
2021-08-20T05:11:54.3991074Z + echo 'Build left local git repository checkout dirty'
2021-08-20T05:11:54.3992038Z Build left local git repository checkout dirty
2021-08-20T05:11:54.3993442Z + echo 'git status --porcelain:'
2021-08-20T05:11:54.3994696Z git status --porcelain:
2021-08-20T05:11:54.3995334Z + echo '?? third_party/breakpad/'
2021-08-20T05:11:54.3996812Z ?? third_party/breakpad/
2021-08-20T05:11:54.3997260Z + exit 1
2021-08-20T05:11:54.3997670Z + cleanup
2021-08-20T05:11:54.3998036Z + retcode=1
2021-08-20T05:11:54.3998404Z + set +x
2021-08-20T05:11:54.3998852Z =================== sccache compilation log ===================
2021-08-20T05:11:54.4260657Z =========== If your build fails, please take a look at the log above for possible reasons ===========

---

❄️ 1 failure tentatively classified as flaky

but reruns have not yet been triggered to confirm:

linux-bionic-py3.8-gcc9-coverage / build (1/1)

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

2021-08-19T22:56:04.2588528Z Failed to recurse into submodule path 'third_party/ideep'

2021-08-19T22:56:03.4815599Z Submodule path 'third_party/python-six': checked out '15e31431af97e5e64b80af0a3f598d382bcdd49a'
2021-08-19T22:56:03.5501668Z Submodule path 'third_party/sleef': checked out 'e0a003ee838b75d11763aa9c3ef17bf71a725bff'
2021-08-19T22:56:03.7212070Z Submodule path 'third_party/tbb': checked out 'a51a90bc609bb73db8ea13841b5cf7aa4344d4a9'
2021-08-19T22:56:03.8311601Z Submodule path 'third_party/tensorpipe': checked out 'e45b2338d0a31192a7e413f3fbbfa7fd90504a37'
2021-08-19T22:56:03.8989389Z Submodule path 'third_party/tensorpipe/third_party/googletest': checked out 'aee0f9d9b5b87796ee8a0ab26b7587ec30e8858e'
2021-08-19T22:56:03.9436710Z Submodule path 'third_party/tensorpipe/third_party/libnop': checked out 'aa95422ea8c409e3f078d2ee7708a5f59a8b9fa2'
2021-08-19T22:56:04.0313222Z Submodule path 'third_party/tensorpipe/third_party/libuv': checked out '1dff88e5161cba5c59276d2070d2e304e4dcb242'
2021-08-19T22:56:04.0938127Z Submodule path 'third_party/tensorpipe/third_party/pybind11': checked out 'a23996fce38ff6ccfbcdc09f1e63f2c4be5ea2ef'
2021-08-19T22:56:04.1241188Z Submodule path 'third_party/tensorpipe/third_party/pybind11/tools/clang': checked out '6a00cbc4a9b8e68b71caf7f774b3f9c753ae84d5'
2021-08-19T22:56:04.2527585Z Submodule path 'third_party/zstd': checked out 'aec56a52fbab207fc639a1937d1e708a282edca8'
2021-08-19T22:56:04.2588528Z Failed to recurse into submodule path 'third_party/ideep'
2021-08-19T22:56:04.2634150Z ##[error]The process '/usr/bin/git' failed with exit code 1
2021-08-19T22:56:04.2783133Z ##[group]Run .github/scripts/wait_for_ssh_to_drain.sh
2021-08-19T22:56:04.2783699Z �[36;1m.github/scripts/wait_for_ssh_to_drain.sh�[0m
2021-08-19T22:56:04.2795155Z shell: /usr/bin/bash -e {0}
2021-08-19T22:56:04.2795474Z env:
2021-08-19T22:56:04.2795994Z   BUILD_ENVIRONMENT: linux-bionic-py3.8-gcc9-coverage
2021-08-19T22:56:04.2797093Z   DOCKER_IMAGE_BASE: 308535385114.dkr.ecr.us-east-1.amazonaws.com/pytorch/pytorch-linux-bionic-py3.8-gcc9
2021-08-19T22:56:04.2798103Z   SCCACHE_BUCKET: ossci-compiler-cache-circleci-v2
2021-08-19T22:56:04.2798651Z   TORCH_CUDA_ARCH_LIST: 5.2
2021-08-19T22:56:04.2798953Z   IN_CI: 1

---

ci.pytorch.org: 1 failed

• Failed: pr/pytorch-linux-bionic-rocm4.2-py3.6

---

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.

[ngimel]

One day we'll get rid of CUDA_tensor_Apply2, and yeah, we probably should prohibit overlapping writes even before that.
As far as I can see, it's used only for distributions (poisson, gamma, bernoulli), and those shouldn't accept self-overlapping tensor.

[ngimel]

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

[facebook-github-bot]

@ngimel merged this pull request in d454c9e.