Feature/speed up binary vector decoding #96716
Merged
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
benwtrent
added
>enhancement
:Search/Search
|
Hi @benwtrent, I've created a changelog YAML for you. |
|
Pinging @elastic/es-search (Team:Search) |
|
Baseline is current main (with Panama Vector Module) |
…benwtrent/elasticsearch into feature/speed-up-binary-vector-decoding
|
@elasticmachine update branch |
jdconrad
approved these changes
Jun 12, 2023
| @@ -64,6 +65,8 @@ | |||
| * A {@link FieldMapper} for indexing a dense vector of floats. | |||
| */ | |||
| public class DenseVectorFieldMapper extends FieldMapper { | |||
| public static final Version MAGNITUDE_STORED_INDEX_VERSION = Version.V_7_5_0; | |||
| public static final Version LITTLE_ENDIAN_FLOAT_STORED_INDEX_VERSION = Version.V_8_9_0; | |||
There was a problem hiding this comment.
Should this be the last version prior to your change's version using the new TransportVersion constants?
There was a problem hiding this comment.
@jdconrad this is a transport/wire serialization thing. its an index version thing. From my understanding index versioning is different. I will see what I can find.
There was a problem hiding this comment.
I asked @thecoop and me just using Version here is OK. We may need to update it to IndexVersion depending on which commits make it in first :)
| @@ -890,18 +907,18 @@ private Field parseKnnVector(DocumentParserContext context) throws IOException { | |||
| private Field parseBinaryDocValuesVector(DocumentParserContext context) throws IOException { | |||
| // encode array of floats as array of integers and store into buf | |||
| // this code is here and not int the VectorEncoderDecoder so not to create extra arrays | |||
There was a problem hiding this comment.
Not related to your change, but would you mind fixing the not int -> not in?
Comment on lines
+78
to
+86
| FloatBuffer fb = ByteBuffer.wrap(vectorBR.bytes, vectorBR.offset, vectorBR.length) | ||
| .order(ByteOrder.LITTLE_ENDIAN) | ||
| .asFloatBuffer(); | ||
| fb.get(vector); | ||
| } else { | ||
| ByteBuffer byteBuffer = ByteBuffer.wrap(vectorBR.bytes, vectorBR.offset, vectorBR.length); | ||
| for (int dim = 0; dim < vector.length; dim++) { | ||
| vector[dim] = byteBuffer.getFloat((dim * Float.BYTES) + vectorBR.offset); | ||
| } |
There was a problem hiding this comment.
Could .asFloatBuffer() not be used for both little and big endian?
There was a problem hiding this comment.
.asFloatBuffer() is marginally slower for BE. These implementations are the fastest I could get them.
…nary-vector-decoding
…nary-vector-decoding
benwtrent
added
the
auto-merge
ChrisHegarty
approved these changes
Jun 13, 2023
| return indexVersion.onOrAfter(LITTLE_ENDIAN_FLOAT_STORED_INDEX_VERSION) | ||
| ? ByteBuffer.wrap(new byte[numBytes]).order(ByteOrder.LITTLE_ENDIAN) | ||
| : ByteBuffer.wrap(new byte[numBytes]); | ||
| } |
HiDAl
pushed a commit
to HiDAl/elasticsearch
that referenced
this pull request
Jun 14, 2023
Encoding floats in little endian format provides much faster decoding. This commit takes all indices created in 8.9.0+ and stores binary vectors as little endian. closes: elastic#96710
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.
Encoding floats in little endian format provides much faster decoding.
This commit takes all indices created in 8.9.0+ and stores binary vectors as little endian.
closes: #96710