🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/164517

• 📄 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

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• Driver update on H100 and A100 instances

✅ No Failures

As of commit 2f475da with merge base a707042 ():
💚 Looks good so far! There are no failures yet. 💚

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

@pytorchbot merge

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

Do we drop the support for cuda version before 12.3?

Since otherwise, something like:

+at::Tensor multimem_one_shot_reduce_out(
+    const at::Tensor& input,
+    std::string reduce_op,
+    int64_t root,
+    std::string group_name,
+    at::Tensor out) {
+  TORCH_CHECK(false, "multimem_one_shot_reduce_out: requires CUDA 12.3+.");
+  return input;
+}

need to be added after #elif defined(CUDART_VERSION) && CUDART_VERSION < 12030
to be able to build

@pytorchbot revert -m="Diff reverted internally" -c="ghfirst"

This Pull Request has been reverted by a revert inside Meta. To re-land this change, please open another pull request, assign the same reviewers, fix the CI failures that caused the revert and make sure that the failing CI runs on the PR by applying the proper ciflow label (e.g., ciflow/trunk).)

@pytorchbot successfully started a revert job. Check the current status here.
Questions? Feedback? Please reach out to the PyTorch DevX Team

@kwen2501 your PR has been successfully reverted.

@shunting314 The code you cited is merely host-side code, without fancy CUDA APIs. Which part is not buildable? Can you share your error log?

I don't have the log available. But it says multimem_one_shot_reduce_out is not defined since the definition is in a conditional block guarded by the CUDA 12.3 check.

I see, thank you @shunting314

Added the temporary workaround as suggested.
Longer term, we should remove all those false impl's.

@pytorchbot merge

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