[CUDA] Workaround shmem limit for certain input sizes in AdaptiveAvgPool1D

[eqy]

Reference issue #68248

CC @ptrblck @malfet @xwang233

cc @albanD @mruberry @jbschlosser @walterddr @mikaylagawarecki @ptrblck

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/115231

• 📄 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 the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (3 Unrelated Failures)

As of commit 23323f3 with merge base 87ea6fb ():

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• periodic / macos-12-py3-x86-64 / test (default, 1, 4, macos-12) (gh)
• periodic / macos-12-py3-x86-64 / test (default, 3, 4, macos-12) (gh)
• periodic / macos-12-py3-x86-64 / test (default, 4, 4, macos-12) (gh)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[eqy]

Also related: #70701

[mikaylagawarecki]

Looks ok to me and failures look unrelated let's maybe hold off on merging till @marcmk6 verifies that it fixes the issue

[eqy]

@pytorchmergebot rebase

[pytorchmergebot]

@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here

[pytorchmergebot]

Successfully rebased adaptiveavgpool1dshmem onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout adaptiveavgpool1dshmem && git pull --rebase)

[marcmk6]

@eqy @mikaylagawarecki
Hi, I couldn't figure out how to build pytorch from source code. I did try to follow the instructions but got cpp compilation errors that I don't understand. So I'm sorry I may not be able to verify the fix, assuming I have to build it by myself.

[eqy]

@marcmk6 I was able to reproduce the error and the test (uses the same repro) passes. However we could check that it suffices for your particular hardware: what model(s) of GPU are you using?

[marcmk6]

@eqy I got RTX A5000 and an old Tesla P4

[eqy]

@pytorchmergebot rebase

[eqy]

@pytorchmergebot rebase

[pytorchmergebot]

@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here

[pytorchmergebot]

Successfully rebased adaptiveavgpool1dshmem onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via git checkout adaptiveavgpool1dshmem && git pull --rebase)

[eqy]

I've verified that this is fixed on P100 and A40 (the closest GPUs I have access to compared to P4 and A5000), merging.

@pytorchmergebot merge

[pytorchmergebot]

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here