Improved memory efficiency in UMAP given precomputed knn graphs #7481
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
jinsolp
added
non-breaking
tarang-jain
reviewed
Nov 13, 2025
cpp/include/cuml/manifold/common.hpp
Outdated
| cudaGetLastError(); | ||
| return false; // Assume host pointer if query fails | ||
| } | ||
| return attr.type == cudaMemoryTypeDevice || attr.type == cudaMemoryTypeManaged; |
Contributor
There was a problem hiding this comment.
this pattern of checking if its a device pointer is being repeated way too much. In cuvs and raft we have abstracted it away into functions such as check_pointer_residency. Can you check if you can use those directly? If you cannot use those directly, we should create an issue for this -- ideally we'd have a function in raft that does this.
Contributor
Author
There was a problem hiding this comment.
Changes to using check_pointer_residency!
csadorf
approved these changes
Nov 19, 2025
Contributor
csadorf
left a comment
csadorf
left a comment
There was a problem hiding this comment.
LGTM. Just a few minor suggestions for improved wording in the docs.
csadorf
approved these changes
Nov 19, 2025
Contributor
Author
|
Had to change the default based on the changes of this PR: #7501 |
tarang-jain
reviewed
Nov 21, 2025
tarang-jain
approved these changes
Nov 21, 2025
Contributor
Author
|
/merge |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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Outdated suggestions cannot be applied.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.
Closes #7143
This PR improves memory usage in UMAP when given a precomputed knn graph.
Previously, a user-given knn graph will occupy GPU memory throughout the full UMAP pipeline even though it is not needed in later steps of UMAP.
In this PR, if the user-given knn graph is on host memory, we keep it on host memory and copy to device at the cpp level to allow better memory management.
This PR with precomputed knn graph on CPU
Before with precomputed knn graph on CPU