Optimise text.lines for longer lines, larger chunks use-cases #3481
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
The main idea is to avoid copying single characters, and instead copy string slices, that can use avx2 instructions. The copy is still unavoidable unfortunately in JVM - AFAIK. Before: ``` LinesBenchmark.linesBenchmark 0 4 thrpt 3 122139.570 ± 236181.395 ops/s LinesBenchmark.linesBenchmark 0 16 thrpt 3 134204.974 ± 76569.841 ops/s LinesBenchmark.linesBenchmark 0 64 thrpt 3 133776.725 ± 137902.948 ops/s LinesBenchmark.linesBenchmark 1 4 thrpt 3 44078.282 ± 46048.728 ops/s LinesBenchmark.linesBenchmark 1 16 thrpt 3 95173.337 ± 31919.684 ops/s LinesBenchmark.linesBenchmark 1 64 thrpt 3 91207.064 ± 134077.287 ops/s LinesBenchmark.linesBenchmark 10 4 thrpt 3 11310.247 ± 6860.349 ops/s LinesBenchmark.linesBenchmark 10 16 thrpt 3 53543.590 ± 12547.417 ops/s LinesBenchmark.linesBenchmark 10 64 thrpt 3 66603.479 ± 66765.096 ops/s LinesBenchmark.linesBenchmark 100 4 thrpt 3 1759.697 ± 1187.625 ops/s LinesBenchmark.linesBenchmark 100 16 thrpt 3 10084.010 ± 6844.210 ops/s LinesBenchmark.linesBenchmark 100 64 thrpt 3 15806.198 ± 3290.178 ops/s ``` After: ``` LinesBenchmark.linesBenchmark 0 4 thrpt 3 129302.595 ± 42736.093 ops/s LinesBenchmark.linesBenchmark 0 16 thrpt 3 130886.744 ± 47487.794 ops/s LinesBenchmark.linesBenchmark 0 64 thrpt 3 123490.463 ± 156930.971 ops/s LinesBenchmark.linesBenchmark 1 4 thrpt 3 41200.649 ± 10262.513 ops/s LinesBenchmark.linesBenchmark 1 16 thrpt 3 87733.636 ± 111704.611 ops/s LinesBenchmark.linesBenchmark 1 64 thrpt 3 94767.345 ± 62120.286 ops/s LinesBenchmark.linesBenchmark 10 4 thrpt 3 12385.057 ± 7146.545 ops/s LinesBenchmark.linesBenchmark 10 16 thrpt 3 49429.892 ± 33859.569 ops/s LinesBenchmark.linesBenchmark 10 64 thrpt 3 75037.094 ± 107270.451 ops/s LinesBenchmark.linesBenchmark 100 4 thrpt 3 1699.683 ± 544.435 ops/s LinesBenchmark.linesBenchmark 100 16 thrpt 3 13667.955 ± 3294.870 ops/s LinesBenchmark.linesBenchmark 100 64 thrpt 3 31498.156 ± 13943.473 ops/s ```
mpilquist
reviewed
Oct 6, 2024
| @@ -542,9 +548,9 @@ object text { | |||
| else Pull.output1(result) | |||
| } | |||
| case Some((chunk, stream)) => | |||
| val linesBuffer = ArrayBuffer.empty[String] | |||
| linesBuffer.clear() | |||
There was a problem hiding this comment.
Can we clear this earlier? To avoid holding on to lines after they've been emitted?
There was a problem hiding this comment.
I'll check, thx
There was a problem hiding this comment.
actually I've reverted this part back to creating a new array buffer for each output chunk
it's a mutable array and it is not safe to clear it because of unexpected consequences if the chunk is used after we pull the next one
thanks for pointing my attention to this line! :)
underlying array is mutable and can cause unexpected consequences when clearing it - e.g. if a chunk is used after pulling the next chunk
svalaskevicius
force-pushed
the
text-lines-performance
branch
from
feeeb91
to
3d7e124
Compare
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.
The main idea is to avoid copying single characters, and instead copy string slices, that can use avx2 instructions.
The copy is still unavoidable unfortunately in JVM - AFAIK.
benchmarks running locally
-wi 3 -i 3 -f 1 -t 1Before:
After: