Skip to content

Skip extent-one dimensions in a device copy#8957

Merged
abadams merged 3 commits into
mainfrom
abadams/skip_extent_one_dimensions_in_device_copy
Feb 21, 2026
Merged

Skip extent-one dimensions in a device copy#8957
abadams merged 3 commits into
mainfrom
abadams/skip_extent_one_dimensions_in_device_copy

Conversation

@abadams
Copy link
Copy Markdown
Member

@abadams abadams commented Feb 20, 2026

Their strides are meaningless anyway

Fixes #8956

Their strides are meaningless anyway

Fixes #8956
@alexreinking
Copy link
Copy Markdown
Member

Can you include a test based on the issue? Maybe that the device copy doesn't get slower or something?

@abadams
Copy link
Copy Markdown
Member Author

abadams commented Feb 20, 2026

Done.

Comment thread test/performance/device_copy.cpp Outdated
@alexreinking
Copy link
Copy Markdown
Member

The win64 failure is my fault. Troubleshooting.

@abadams abadams merged commit 1e831cc into main Feb 21, 2026
16 of 17 checks passed
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.

copy_to_device is extremely slow on CUDA with storage order {3, 2, 0, 1} and 3rd extent = 1

3 participants