[Parquet] Optimize appending max level comparison in DefinitionLevelDecoder #9217
Conversation
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
jhorstmann
commented
Jan 18, 2026
| None => { | ||
| for def_level in def_levels { | ||
| bitmap_builder.append(*def_level >= self.struct_def_level) | ||
| // Safety: slice iterator has a trusted length |
Contributor
Author
There was a problem hiding this comment.
My benchmark did not actually cover this code, since it reads the leaf column as a primitive array. That would mean the performance might be even worse than measured, but also the improvement would be a bit bigger.
Contributor
|
run benchmark arrow_reader arrow_reader_row_filter arrow_reader_clickbench |
jhorstmann
force-pushed
the
optimize-struct-reading
branch
from
7adb6ed to
f447bac
Compare
|
🤖 |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
🤖: Benchmark completed Details
|
Dandandan
approved these changes
Jan 19, 2026
alamb
approved these changes
Jan 19, 2026
Contributor
alamb
left a comment
alamb
left a comment
There was a problem hiding this comment.
Love it -- thank you @jhorstmann and @Dandandan
| for i in &levels[start..] { | ||
| nulls.append(i == max_level); | ||
| // Safety: slice iterator has a trusted length | ||
| unsafe { |
Contributor
There was a problem hiding this comment.
Another thing that we could potentially try is to evaluate to a bitmap first and then append that bitmap directly via set_bits or something -- though now that I type this I am not sure that would be be any better 🤔
Contributor
Author
There was a problem hiding this comment.
That would be interesting to try out. It probably will only be faster when using explicit simd intrinsics to compute that bitmap.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Which issue does this PR close?
Rationale for this change
Profiling showed this loop to be a clear performance hotspot and thanks to the new
BooleanBufferBuilder::extend_trusted_lenmethod introduced in #9137 there is a very simple improvement.Benchmark results for optional structs after the fix:
This is a big improvement, but still significantly slower than reading non-nested data. The main hotspot still is the
extend_trusted_lenand manual simd code for comparing chunks and appending 64-bits at a time could potentially speed it up even more. But that would require either architecture specific intrinsics or unstable features.What changes are included in this PR?
Are these changes tested?
Tested by existing tests.
Are there any user-facing changes?
no