Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Workaround ROCm global pool #15481

Merged
merged 8 commits into from
Apr 23, 2023
Merged

Workaround ROCm global pool #15481

merged 8 commits into from
Apr 23, 2023

Conversation

cloudhan
Copy link
Member

@cloudhan cloudhan commented Apr 12, 2023

Fix #15482

This also improve the perf for it, comparison is meanless tho, the original impl is incorrect.

shape Original (numerically incorrect) New (numerically correct)
1,64,255,255 0.16797351837158203 0.0012824535369873047
32,64,255,255 1.3541345596313477 0.008842945098876953
32,64,14,14 0.00106048583984375 0.0008864402770996094

PeixuanZuo
PeixuanZuo previously approved these changes Apr 19, 2023
Copy link
Contributor

@PeixuanZuo PeixuanZuo left a comment

Choose a reason for hiding this comment

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

LGTM, Thanks.

@kailums
Copy link
Contributor

kailums commented Apr 19, 2023

LGTM

@kailums
Copy link
Contributor

kailums commented Apr 20, 2023

approve

kailums
kailums previously approved these changes Apr 20, 2023
Copy link
Contributor

@kailums kailums left a comment

Choose a reason for hiding this comment

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

approve

@cloudhan cloudhan merged commit 9e44248 into main Apr 23, 2023
@cloudhan cloudhan deleted the guangyunhan/workaround-global-pool branch April 23, 2023 03:48
ShukantPal pushed a commit to ShukantPal/onnxruntime that referenced this pull request May 7, 2023
Implement global avg/max pool with reduction
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.

[ROCm] Global (average) Pooling unusable.
3 participants