Perf: use GPU-only searchsorted instead of numpy.repeat#253
Open
eriknw wants to merge 2 commits intorapidsai:mainfrom
Open
Perf: use GPU-only searchsorted instead of numpy.repeat#253eriknw wants to merge 2 commits intorapidsai:mainfrom
eriknw wants to merge 2 commits intorapidsai:mainfrom
Conversation
The GPU-only searchsorted for doing indptr-to-COO expansion (which eliminates D2H transfers) is faster than numpy.repeat. `cupy.repeat` does not yet support ndarray as `repeats` argument, so searchsorted is the appropriate recipe to do instead for most data. For very large data, a "cumsum+scatter" approach may be faster. `cp.repeat` is being updated in cupy/cupy#9828 Example speed improvement: from_csr (1M nodes, 20M edges): 2.4ms vs 72ms (30x faster) This PR was motivated by work I am doing in cupy/cupy#9825
eriknw
added
improvement
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The GPU-only searchsorted for doing indptr-to-COO expansion (which eliminates D2H transfers) is faster than numpy.repeat.
cupy.repeatdoes not yet support ndarray asrepeatsargument, so searchsorted is the appropriate recipe to do instead for most data. For very large data, a "cumsum+scatter" approach may be faster.cp.repeatis being updated in cupy/cupy#9828Example speed improvement:
This PR was motivated by work I am doing in cupy/cupy#9825