Skip to content

Conversation

@vacu9708
Copy link
Contributor

@vacu9708 vacu9708 commented Jun 23, 2025

Summary (Fix issue #18030)

This PR follows up on #18072 by updating the remaining Reduce operators to accept axes as an input instead of an attribute.

Changes

  • Support axes as an input for Reduce ops (e.g., ReduceL2, ReduceMax, …)
  • Add corresponding test cases
Axes provided? noop_with_empty_axes Behavior
No 0 Reduce over all dimensions
No 1 Return the input unchanged
Yes any Reduce over the specified axes

To do

  • The ONNX spec allows runtime axes but TVM currently supports only constant axes
    • Modify TVM's reduce operators to support runtime axes

- Support axes as an input for Reduce ops (e.g., ReduceL2, ReduceMax, …)
- Add corresponding test cases
@Hzfengsy Hzfengsy merged commit 437d00a into apache:main Jun 24, 2025
11 checks passed
@vacu9708 vacu9708 deleted the reduce_ops branch July 9, 2025 23:52
ShiboXing pushed a commit to ShiboXing/tvm that referenced this pull request Aug 10, 2025
- Support axes as an input for Reduce ops (e.g., ReduceL2, ReduceMax, …)
- Add corresponding test cases
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants