Skip to content

Panama implementation of float-byte vector ops #123270

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 10 commits into from
Mar 24, 2025
Merged

Panama implementation of float-byte vector ops #123270

merged 10 commits into from
Mar 24, 2025

Conversation

@thecoop
Copy link
Member

@thecoop thecoop commented Feb 24, 2025 •
edited
Loading

Followon from #122381, introduce vectorized operations for float-byte comparisons

JMH shows a ~16x speedup:

Benchmark                                  (dims)  (docType)  (function)  (queryType)  (type)  Mode  Cnt    Score   Error  Units
DistanceFunctionBenchmark.benchmark          1024       BYTE         DOT        FLOAT  BINARY  avgt    2  880.346          ns/op
DistanceFunctionBenchmark.vectorBenchmark    1024       BYTE         DOT        FLOAT  BINARY  avgt    2   55.105          ns/op

Unrolling the loop doesn't give any additional performance benefit here, so I've left it as-is (it looks like hotspot may already be unrolling it itself)

@thecoop thecoop force-pushed the float-byte-vector-op branch from e6a2e19 to ff2e0e2 Compare March 3, 2025 10:38
@thecoop thecoop added the :Search Relevance/Vectors Vector search label Mar 3, 2025
@thecoop thecoop marked this pull request as ready for review March 3, 2025 10:45
@thecoop thecoop mentioned this pull request Mar 3, 2025
Merged
@elasticsearchmachine elasticsearchmachine added the Team:Search Relevance Meta label for the Search Relevance team in Elasticsearch label Mar 3, 2025
@elasticsearchmachine
Copy link
Collaborator

elasticsearchmachine commented Mar 3, 2025

Pinging @elastic/es-search-relevance (Team:Search Relevance)

@benwtrent
Copy link
Member

benwtrent commented Mar 3, 2025

Not sure how to test as part of Elasticsearch, as this is new functionality we don't have performance tests for these operations yet...

We should at a minimum add some clauses here: ESVectorUtilTests (missing for some other things I think). This way we capture that we are actually doing the math correct with various random vector sizes.

For performance, we should add a new JMH benchmark if possible for folks to run in the future.

@benwtrent benwtrent requested a review from ChrisHegarty March 3, 2025 19:11
@thecoop thecoop requested a review from benwtrent March 7, 2025 11:52
@thecoop
Copy link
Member Author

thecoop commented Mar 7, 2025

The JMH tests need some rework to support heterogenous types - let me take a closer look at the infrastructure...

@thecoop
Copy link
Member Author

thecoop commented Mar 10, 2025

JMH tests refactored in #124351 ready for this implementation

@thecoop thecoop merged commit 2ba9e9f into elastic:main Mar 24, 2025
17 checks passed
@thecoop thecoop deleted the float-byte-vector-op branch March 24, 2025 10:30
omricohenn pushed a commit to omricohenn/elasticsearch that referenced this pull request Mar 28, 2025
@thecoop @omricohenn
Panama implementation of painless float-byte vector ops (elastic#123270)
683b56d
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@benwtrent benwtrent benwtrent approved these changes

@ChrisHegarty ChrisHegarty Awaiting requested review from ChrisHegarty

Assignees

No one assigned

Labels

>non-issue :Search Relevance/Vectors Vector search Team:Search Relevance Meta label for the Search Relevance team in Elasticsearch v9.1.0

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@thecoop @elasticsearchmachine @benwtrent