perf: fast FixedSizeListArray canonicalization for chunked arrays#8161
Merged
Conversation
Canonicalizing a `ChunkedArray` of `FixedSizeList` chunks previously fell through to the generic builder path, copying every element into a fresh contiguous buffer. Because each chunk's `elements` child is already exactly `list_size * chunk.len()` long and starts at the first list, the children can be reused directly as the chunks of a combined `ChunkedArray` of `elements`, mirroring the existing `swizzle_list_chunks` approach for `List`. Route `DType::FixedSizeList` through `_canonicalize` and add `swizzle_fixed_size_list_chunks`, which builds the combined elements and wraps them in a single `FixedSizeListArray` in O(nchunks) without copying element data. Adds a `chunked_fsl_canonicalize` divan benchmark and a regression test. The benchmark shows the swizzle is effectively O(1) in list size (e.g. 1024 elements/list across 32 chunks drops from ~175 ms to ~3 µs). Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>
joseph-isaacs
changed the title
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
Contributor
Polar Signals Profiling ResultsLatest Run
Powered by Polar Signals Cloud |
Merging this PR will not alter performance
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ⚡ | Simulation | chunked_varbinview_opt_canonical_into[(1000, 10)] |
225.4 µs | 188.1 µs | +19.85% |
| ❌ | Simulation | chunked_varbinview_canonical_into[(100, 100)] |
273.1 µs | 307.8 µs | -11.28% |
| 🆕 | Simulation | canonicalize[1024, 2] |
N/A | 18.7 µs | N/A |
| 🆕 | Simulation | canonicalize[1024, 8] |
N/A | 20.4 µs | N/A |
| 🆕 | Simulation | canonicalize[16, 2] |
N/A | 23.9 µs | N/A |
| 🆕 | Simulation | canonicalize[256, 2] |
N/A | 18.9 µs | N/A |
| 🆕 | Simulation | canonicalize[256, 32] |
N/A | 24.7 µs | N/A |
| 🆕 | Simulation | canonicalize[1024, 32] |
N/A | 24.6 µs | N/A |
| 🆕 | Simulation | canonicalize[16, 32] |
N/A | 25.6 µs | N/A |
| 🆕 | Simulation | canonicalize[16, 8] |
N/A | 20.6 µs | N/A |
| 🆕 | Simulation | canonicalize[256, 8] |
N/A | 20.4 µs | N/A |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing claude/admiring-keller-4Sz1N (e0e8c96) with develop (30103b8)
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 0.994x ➖ How to read Verdict and Engines
datafusion / vortex-file-compressed (0.994x ➖, 0↑ 1↓)
No file size changes detected. |
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.025x ➖, 0↑ 1↓)
datafusion / vortex-compact (0.995x ➖, 0↑ 0↓)
datafusion / parquet (0.989x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (0.999x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.019x ➖, 0↑ 0↓)
duckdb / parquet (1.005x ➖, 0↑ 0↓)
No file size changes detected. Full attributed analysis
|
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.194x ❌, 0↑ 15↓)
datafusion / vortex-compact (1.240x ❌, 0↑ 22↓)
datafusion / parquet (1.181x ❌, 0↑ 16↓)
datafusion / arrow (1.461x ❌, 0↑ 22↓)
duckdb / vortex-file-compressed (1.168x ❌, 0↑ 19↓)
duckdb / vortex-compact (1.134x ❌, 0↑ 19↓)
duckdb / parquet (1.111x ❌, 0↑ 15↓)
duckdb / duckdb (1.113x ❌, 0↑ 13↓)
No file size changes detected. Full attributed analysis
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.013x ➖, 0↑ 3↓)
datafusion / vortex-compact (1.009x ➖, 0↑ 5↓)
datafusion / parquet (1.004x ➖, 0↑ 2↓)
duckdb / vortex-file-compressed (1.011x ➖, 0↑ 4↓)
duckdb / vortex-compact (1.000x ➖, 0↑ 1↓)
duckdb / parquet (1.013x ➖, 0↑ 5↓)
duckdb / duckdb (1.005x ➖, 2↑ 1↓)
No file size changes detected. Full attributed analysis
|
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.999x ➖, 1↑ 0↓)
datafusion / vortex-compact (1.006x ➖, 0↑ 0↓)
datafusion / parquet (0.969x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.963x ➖, 1↑ 0↓)
duckdb / vortex-compact (0.971x ➖, 0↑ 0↓)
duckdb / parquet (1.017x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) How to read Verdict and Engines
duckdb / vortex-file-compressed (1.030x ➖, 0↑ 1↓)
duckdb / vortex-compact (1.025x ➖, 0↑ 0↓)
duckdb / parquet (1.045x ➖, 0↑ 1↓)
No file size changes detected. Full attributed analysis
|
Contributor
Benchmarks: Random AccessVortex (geomean): 1.002x ➖ How to read Verdict and Engines
unknown / unknown (1.037x ➖, 2↑ 2↓)
|
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.009x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.005x ➖, 0↑ 0↓)
datafusion / parquet (1.001x ➖, 0↑ 0↓)
datafusion / arrow (0.985x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.005x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.999x ➖, 0↑ 0↓)
duckdb / parquet (0.992x ➖, 0↑ 0↓)
duckdb / duckdb (1.004x ➖, 0↑ 0↓)
No file size changes detected. Full attributed analysis
|
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.945x ➖, 8↑ 2↓)
datafusion / parquet (0.960x ➖, 2↑ 0↓)
duckdb / vortex-file-compressed (0.967x ➖, 5↑ 1↓)
duckdb / parquet (0.961x ➖, 3↑ 0↓)
duckdb / duckdb (0.979x ➖, 0↑ 0↓)
File Size Changes (1 files changed, -0.0% overall, 0↑ 1↓)
Totals:
Full attributed analysis
|
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
Contributor
Benchmarks: Appian on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.002x ➖, 0↑ 0↓)
datafusion / parquet (0.997x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.028x ➖, 0↑ 0↓)
duckdb / parquet (1.007x ➖, 0↑ 0↓)
duckdb / duckdb (1.030x ➖, 0↑ 0↓)
File Size Changes (1 files changed, -0.0% overall, 0↑ 1↓)
Totals:
Full attributed analysis
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.980x ➖, 4↑ 4↓)
datafusion / vortex-compact (1.030x ➖, 0↑ 3↓)
datafusion / parquet (0.830x ➖, 7↑ 1↓)
duckdb / vortex-file-compressed (0.994x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.019x ➖, 0↑ 1↓)
duckdb / parquet (0.927x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: CompressionVortex (geomean): 0.990x ➖ How to read Verdict and Engines
unknown / unknown (1.001x ➖, 5↑ 6↓)
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.066x ➖, 0↑ 3↓)
datafusion / vortex-compact (1.137x ➖, 0↑ 4↓)
datafusion / parquet (1.195x ➖, 0↑ 7↓)
duckdb / vortex-file-compressed (1.015x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.955x ➖, 0↑ 0↓)
duckdb / parquet (1.149x ➖, 0↑ 3↓)
Full attributed analysis
|
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
3 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.
Summary
This PR adds support for canonicalizing
ChunkedArrayofFixedSizeListArraychunks, similar to existing support forListandStructarrays. The canonicalization process combines multipleFixedSizeListArraychunks into a single array with aChunkedArrayof elements, enabling efficient processing of chunked fixed-size list data.Changes
Core canonicalization logic (
canonical.rs):swizzle_fixed_size_list_chunks()function that combines multipleFixedSizeListArraychunks by reusing their element children directly as chunks of a combined elements array_canonicalize()to handleDType::FixedSizeListcaseFixedSizeListArrayandFixedSizeListArrayExtVTable routing (
mod.rs):execute()method to routeFixedSizeListArraychunks through the canonicalization path (alongsideStruct,List, andVariant)Testing:
pack_fixed_size_lists()unit test verifying that multipleFixedSizeListArraychunks are correctly combined into a canonical formBenchmarking:
chunked_fsl_canonicalizebenchmark to measure canonicalization performance across different chunk counts (2, 8, 32) and list sizes (16, 256, 1024 elements)Testing
pack_fixed_size_lists()that verifies canonicalization of chunked fixed-size lists produces correct resultshttps://claude.ai/code/session_01TY5GScKyQ135hphErzLhpV