ENH Reuse TopK buffers #110
Open
Conversation
This file contains 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
ogrisel
approved these changes
Apr 26, 2023
There was a problem hiding this comment.
The docstring for the _make_get_topk_kernel kernel looks good but as far as I can see, you did not add a unit test that checks that the closure tricks works as expected when doing repeated calls to the same closure, in particular for both possible values of reuse_result_buffer.
Other than that, LGTM (I just did a very shallow review).
|
|
||
|
|
||
| @lru_cache | ||
| def make_range_kernel(n_items, work_group_size): |
There was a problem hiding this comment.
Maybe range => arange to match the numpy function name?
Otherwise one can get confused with the "range" name from SYCL terminology.
|
The unit test is a bit technical with the closure-style definition of things everywhere but it should be possible, I'll look into it. |
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Outdated suggestions cannot be applied.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.
Before going forward on the KNN I'd like to add this enhancement for the TopK, this PR exposes a function
_make_get_topk_kernelthat returns a closure that will held references definitions of kernels AND to buffers that will be re-used if successive calls to_make_get_topk_kernelare performed.The current top-k top level functions in main include an allocation and de-allocation at each calls. (The kernel definitions are still cached by the lru_cache).
This fit the usecase for the KNN that will consist in slicing the input data and running the same pairwise distance + TopK kernels for each slice, and aligns with the behavior of the sum and argmin reduction kernels in the reduction .