add vectorization path on maxpool forward channel last #1883
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR adds a vectorized code path for the max pooling forward operation when using channel-last memory layout, providing significant performance improvements on Intel GPU architectures. The optimization uses vectorized memory operations and SYCL kernels to improve throughput.
Key changes:
- Introduces a new vectorized kernel
MaxPool2dChannelLastVecthat processes multiple channels simultaneously - Adds automatic vector size selection (8, 4, 2, or 1) based on data alignment and hardware capabilities
- Implements dynamic work group sizing based on hardware thread availability
Co-authored-by: Copilot <[email protected]>
liangan1
approved these changes
Aug 19, 2025
chunhuanMeng
approved these changes
Aug 19, 2025
github-merge-queue bot
pushed a commit
that referenced
this pull request
Sep 12, 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]>
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
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Part 1 of #1861
tested on shapes from alexnet training
on BMG, 831719 Scoreboard stalls decrease to 497,098. instruction fetch and distance stall also get better.