Skip to content

Conversation

@jianyizh
Copy link
Contributor

@jianyizh jianyizh commented Sep 4, 2025

follows #1883, shape [4096,256,6,6] channel last with output shape [6,6] in torchbench alexnet can get ~4x improvement on bmg

Copilot AI review requested due to automatic review settings September 4, 2025 03:09
Copy link
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull Request Overview

This PR optimizes the adaptive average pool operation by introducing a vectorized implementation for channel-last memory format. The changes add vectorization support to improve memory access patterns and performance for 2D adaptive average pooling operations.

Key changes:

  • Add vectorized kernel implementation for adaptive average pooling in channel-last format
  • Replace the original channel-last kernel with the new optimized vectorized version
  • Add necessary memory access utilities for vectorization support

Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.

@jianyizh jianyizh force-pushed the jianyi/adptive_avg_pool branch from dc988db to e875db2 Compare September 8, 2025 04:31
@weishi-deng weishi-deng self-requested a review September 12, 2025 02:40
@chuanqi129 chuanqi129 added this pull request to the merge queue Sep 12, 2025
Merged via the queue into main with commit a5671d2 Sep 12, 2025
24 checks passed
@chuanqi129 chuanqi129 deleted the jianyi/adptive_avg_pool branch September 12, 2025 02:41
zhangxiaoli73 pushed a commit that referenced this pull request Sep 22, 2025
follows #1883, shape [4096,256,6,6] channel last with output shape [6,6]
in torchbench alexnet can get ~4x improvement on bmg

---------

Co-authored-by: Copilot <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants