[PyTorch] Stack-allocate boxed args for RecordFunction

[swolchok]

Stack from ghstack (oldest at bottom):

• -> [PyTorch] Stack-allocate boxed args for RecordFunction #76266

Saving a heap allocation in this path improves performance.

Differential Revision: D34090699

[facebook-github-bot]

🔗 Helpful links

• 🧪  See artifacts and rendered test results at hud.pytorch.org/pr/76266
• 📄  Preview Python docs built from this PR
• 📄  Preview C++ docs built from this PR
• ❓Need help or want to give feedback on the CI? Visit our office hours
• ↩️  [fb-only] Re-run with SSH instructions

❌ 2 New Failures

As of commit 48895d1 (more details on the Dr. CI page):

Expand to see more

---

• 2/2 failures introduced in this PR

---

🕵️ 2 new failures recognized by patterns

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

trunk / linux-bionic-cuda10.2-py3.9-gcc7 / test (default, 2, 2, linux.4xlarge.nvidia.gpu) (1/2)

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

2022-05-20T17:48:12.2313759Z RuntimeError: test_sparse_csr failed!

2022-05-20T17:48:11.5246675Z 
2022-05-20T17:48:11.5246809Z Generating XML reports...
2022-05-20T17:48:11.6841472Z Generated XML report: test-reports/python-unittest/test_sparse_csr/TEST-TestSparseCSRCUDA-20220520174739.xml
2022-05-20T17:48:11.6843227Z Generated XML report: test-reports/python-unittest/test_sparse_csr/TEST-TestSparseCSRSampler-20220520174739.xml
2022-05-20T17:48:11.7354585Z Generated XML report: test-reports/python-unittest/test_sparse_csr/TEST-TestSparseCompressedCUDA-20220520174739.xml
2022-05-20T17:48:12.2308554Z Traceback (most recent call last):
2022-05-20T17:48:12.2311054Z   File "/var/lib/jenkins/workspace/test/run_test.py", line 1074, in <module>
2022-05-20T17:48:12.2311819Z     main()
2022-05-20T17:48:12.2312281Z   File "/var/lib/jenkins/workspace/test/run_test.py", line 1052, in main
2022-05-20T17:48:12.2312995Z     raise RuntimeError(err_message)
2022-05-20T17:48:12.2313759Z RuntimeError: test_sparse_csr failed!
2022-05-20T17:48:12.7866072Z 
2022-05-20T17:48:12.7866712Z real	6m31.042s
2022-05-20T17:48:12.7867040Z user	9m50.392s
2022-05-20T17:48:12.7867309Z sys	1m17.978s
2022-05-20T17:48:12.7867742Z + cleanup
2022-05-20T17:48:12.7868112Z + retcode=1
2022-05-20T17:48:12.7868488Z + set +x
2022-05-20T17:48:12.7918914Z ##[error]Process completed with exit code 1.
2022-05-20T17:48:12.7967158Z ##[group]Run pytorch/pytorch/.github/actions/get-workflow-job-id@master
2022-05-20T17:48:12.7967514Z with:

pull / linux-bionic-rocm5.1-py3.7 / test (default, 1, 2, linux.rocm.gpu) (2/2)

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

2022-05-20T18:40:52.8636762Z RuntimeError: test_sparse_csr failed!

2022-05-20T18:40:49.3638896Z 
2022-05-20T18:40:49.3639226Z Generating XML reports...
2022-05-20T18:40:49.6296118Z Generated XML report: test-reports/python-unittest/test_sparse_csr/TEST-TestSparseCSRCUDA-20220520184015.xml
2022-05-20T18:40:49.6320853Z Generated XML report: test-reports/python-unittest/test_sparse_csr/TEST-TestSparseCSRSampler-20220520184015.xml
2022-05-20T18:40:49.6881829Z Generated XML report: test-reports/python-unittest/test_sparse_csr/TEST-TestSparseCompressedCUDA-20220520184015.xml
2022-05-20T18:40:52.8617632Z Traceback (most recent call last):
2022-05-20T18:40:52.8618978Z   File "test/run_test.py", line 1074, in <module>
2022-05-20T18:40:52.8628622Z     main()
2022-05-20T18:40:52.8629753Z   File "test/run_test.py", line 1052, in main
2022-05-20T18:40:52.8635423Z     raise RuntimeError(err_message)
2022-05-20T18:40:52.8636762Z RuntimeError: test_sparse_csr failed!
2022-05-20T18:40:54.8404957Z 
2022-05-20T18:40:54.8405731Z real	6m34.281s
2022-05-20T18:40:54.8406521Z user	22m34.657s
2022-05-20T18:40:54.8407229Z sys	3m6.947s
2022-05-20T18:40:54.8408388Z + cleanup
2022-05-20T18:40:54.8409083Z + retcode=1
2022-05-20T18:40:54.8409736Z + set +x
2022-05-20T18:40:54.8541425Z ##[error]Process completed with exit code 1.
2022-05-20T18:40:54.8629772Z ##[group]Run # copy test results back to the mounted workspace, needed sudo, resulting permissions were correct
2022-05-20T18:40:54.8630305Z �[36;1m# copy test results back to the mounted workspace, needed sudo, resulting permissions were correct�[0m

---

This comment was automatically generated by Dr. CI (expand for details).

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

Click here to manually regenerate this comment.

[swolchok]

An interesting follow-up would be to make it possible to avoid heap-allocating a Stack for calls to boxed KernelFunctions made via make_boxed_from_unboxed_functor. I have no idea how common this is, though.

[ezyang]

This needs to be kept consistent with our other boxing logic, is that right?

[ezyang]

I'd suggest a cross-ref with aten/src/ATen/core/boxing/impl/boxing.h (or perhaps moving it to this file entirely)

[ezyang]

The name here is a bit ambiguous; what we're doing is placement new boxing; would be nice if the name had this so people don't try to use this invalidly.

[swolchok]

it's in detail and takes a weird typedef; they'll have trouble using it without writing obvious "here be dragons" stuff like aligned_storage

[ezyang]

mechanically looks all fine, just some naming / file placement stuff

[swolchok]

@pytorchbot merge

[github-actions]

Hey @swolchok.
You've committed this PR, but it does not have both a 'release notes: ...' and 'topics: ...' label. Please add one of each to the PR. The 'release notes: ...' label should represent the part of PyTorch that this PR changes (fx, autograd, distributed, etc) and the 'topics: ...' label should represent the kind of PR it is (not user facing, new feature, bug fix, perf improvement, etc). The list of valid labels can be found here for the 'release notes: ...' and here for the 'topics: ...'.
For changes that are 'topic: not user facing' there is no need for a release notes label.