MAINT Division-less alternative for fused sparse-dense support #15
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
The modulo trick is costly because it uses integer division under the hood which are up to 80 times slower. An alternative is to pass the address of the `indices` array and shift this address at the callers level with the information that we have there so that the dereferencing works properly. This commit replaces the modulo trick by this alternative and adapt documentation and comments in this regard. See: scikit-learn#23585 (comment)
Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
6633d2f to
a3cf4d8
Compare
|
This alternative performs better than the one used in the base branch. 🎉 Here are the results using a10dbcb. |
Merged
4 tasks
ogrisel
approved these changes
Aug 10, 2022
Co-authored-by: Olivier Grisel <olivier.grisel@ensta.org>
|
I think that I would wait for @thomasjpfan's opinion before merge this PR as it's easier to review this diff independently. |
thomasjpfan
reviewed
Aug 11, 2022
sklearn/metrics/_pairwise_distances_reduction/_datasets_pair.pyx
Outdated
Show resolved
Hide resolved
|
BTW I think we can merge this directly into |
MNT Pushing data up instead of indices
|
I merged a few commit in scikit-learn#23585, IIRC and we still need to perform benchmark against So let's merge this first in this scikit-learn#23585 first. |
jjerphan
added a commit
that referenced
this pull request
Aug 22, 2022
To use the a more efficient implementation for future callers implementations. See: #15
jjerphan
added a commit
that referenced
this pull request
Aug 22, 2022
To use the a more efficient implementation for future callers implementations. See: #15
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.
Reference Issues/PRs
Test trick for scikit-learn#23585.
What does this implement/fix? Explain your changes.
The modulo trick is costly because it uses integer division
under the hood which uses
idivinstruction that takes up to80 CPU cycles each.
An alternative is to pass the address of the
indicesarray and shiftthis address at the callers level with the information that we have
there so that the dereferencing works properly.
This commit replaces the modulo trick by this alternative and adapt
documentation and comments in this regard.
See: scikit-learn#23585 (comment)