Speedup sorting for inline views: 1.4x - 1.7x improvement #7856

Dandandan · 2025-07-03T06:34:56Z

Which issue does this PR close?

Closes Speedup sorting for inline views #7857

Rationale for this change

sort string_view[0-400] nulls to indices 2^12                                      1.00     45.2±1.37µs        ? ?/sec    1.01     45.8±1.74µs        ? ?/sec
sort string_view[0-400] to indices 2^12                                            1.00     69.1±1.98µs        ? ?/sec    1.00     69.1±4.24µs        ? ?/sec
sort string_view[10] nulls to indices 2^12                                         1.00     40.8±1.81µs        ? ?/sec    1.37     55.7±3.90µs        ? ?/sec
sort string_view[10] to indices 2^12                                               1.00     52.8±0.35µs        ? ?/sec    1.63     85.9±1.46µs        ? ?/sec
sort string_view_inlined[0-12] nulls to indices 2^12                               1.00     40.9±1.99µs        ? ?/sec    1.29     52.6±1.76µs        ? ?/sec
sort string_view_inlined[0-12] to indices 2^12                                     1.00     50.6±0.27µs        ? ?/sec    1.68    85.0±12.24µs        ? ?/sec

What changes are included in this PR?

Speedup by specializing on batches with only inline views.

Are these changes tested?, are they covered by existing tests)?

existing tests

Are there any user-facing changes?

no

Dandandan · 2025-07-03T10:29:54Z

FYI @zhuqi-lucas

zhuqi-lucas · 2025-07-03T10:55:02Z

Amazing work @Dandandan , reviewing now.

zhuqi-lucas

LGTM thank you @Dandandan !

zhuqi-lucas · 2025-07-03T10:58:59Z

arrow-ord/src/sort.rs

+        _ => value_indices.len(),
    };
+    // 3.a Check if all views are inline (no data buffers)
+    if values.data_buffers().is_empty() {


Thank you @Dandandan , it's very clear we optimize the no data buffers path!

alamb · 2025-07-03T15:30:38Z

🤖 ./gh_compare_arrow.sh Benchmark Script Running
Linux aal-dev 6.11.0-1016-gcp #16~24.04.1-Ubuntu SMP Wed May 28 02:40:52 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing speedup_byte_view (b2a1dd0) to 52ad7d7 diff
BENCH_NAME=sort_kernel
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental --bench sort_kernel
BENCH_FILTER=
BENCH_BRANCH_NAME=speedup_byte_view
Results will be posted here when complete

alamb · 2025-07-03T15:45:17Z

🤖: Benchmark completed

Details

group                                                   main                                   speedup_byte_view
-----                                                   ----                                   -----------------
lexsort (bool, bool) 2^12                               1.00    117.1±0.39µs        ? ?/sec    1.01    118.8±0.31µs        ? ?/sec
lexsort (bool, bool) nulls 2^12                         1.00    155.9±0.66µs        ? ?/sec    1.05    163.9±0.32µs        ? ?/sec
lexsort (f32, f32) 2^10                                 1.01     45.3±0.08µs        ? ?/sec    1.00     45.0±0.12µs        ? ?/sec
lexsort (f32, f32) 2^12                                 1.00    212.2±0.27µs        ? ?/sec    1.00    211.9±0.39µs        ? ?/sec
lexsort (f32, f32) 2^12 limit 10                        1.00     38.1±0.24µs        ? ?/sec    1.01     38.6±0.04µs        ? ?/sec
lexsort (f32, f32) 2^12 limit 100                       1.00     40.5±0.04µs        ? ?/sec    1.00     40.5±0.06µs        ? ?/sec
lexsort (f32, f32) 2^12 limit 1000                      1.00     78.6±0.77µs        ? ?/sec    1.00     78.3±0.11µs        ? ?/sec
lexsort (f32, f32) 2^12 limit 2^12                      1.00    212.3±0.42µs        ? ?/sec    1.00    211.8±0.29µs        ? ?/sec
lexsort (f32, f32) nulls 2^10                           1.00     53.5±0.14µs        ? ?/sec    1.03     55.3±0.13µs        ? ?/sec
lexsort (f32, f32) nulls 2^12                           1.00    252.3±0.33µs        ? ?/sec    1.03    259.3±0.61µs        ? ?/sec
lexsort (f32, f32) nulls 2^12 limit 10                  1.00     85.0±0.24µs        ? ?/sec    1.05     89.3±0.21µs        ? ?/sec
lexsort (f32, f32) nulls 2^12 limit 100                 1.00     86.2±0.17µs        ? ?/sec    1.05     90.2±0.19µs        ? ?/sec
lexsort (f32, f32) nulls 2^12 limit 1000                1.00     95.9±0.27µs        ? ?/sec    1.06    101.3±0.21µs        ? ?/sec
lexsort (f32, f32) nulls 2^12 limit 2^12                1.00    252.5±0.50µs        ? ?/sec    1.03    259.2±0.45µs        ? ?/sec
rank f32 2^12                                           1.00     68.1±0.34µs        ? ?/sec    1.00     68.4±0.16µs        ? ?/sec
rank f32 nulls 2^12                                     1.00     37.7±0.09µs        ? ?/sec    1.00     37.9±0.11µs        ? ?/sec
rank string[10] 2^12                                    1.00    235.9±0.65µs        ? ?/sec    1.02    239.5±0.64µs        ? ?/sec
rank string[10] nulls 2^12                              1.00    114.6±0.15µs        ? ?/sec    1.04    118.7±0.26µs        ? ?/sec
sort f32 2^12                                           1.08     65.0±0.37µs        ? ?/sec    1.00     60.4±0.45µs        ? ?/sec
sort f32 nulls 2^12                                     1.00     32.1±0.10µs        ? ?/sec    1.00     32.1±0.07µs        ? ?/sec
sort f32 nulls to indices 2^12                          1.00     70.3±0.61µs        ? ?/sec    1.04     73.4±0.27µs        ? ?/sec
sort f32 to indices 2^12                                1.07     77.3±0.25µs        ? ?/sec    1.00     72.2±0.17µs        ? ?/sec
sort i32 2^10                                           1.17      8.5±0.02µs        ? ?/sec    1.00      7.3±0.01µs        ? ?/sec
sort i32 2^12                                           1.17     42.1±0.11µs        ? ?/sec    1.00     36.1±0.12µs        ? ?/sec
sort i32 nulls 2^10                                     1.00      5.4±0.01µs        ? ?/sec    1.00      5.4±0.01µs        ? ?/sec
sort i32 nulls 2^12                                     1.01     22.9±0.07µs        ? ?/sec    1.00     22.6±0.07µs        ? ?/sec
sort i32 nulls to indices 2^10                          1.00     12.5±0.54µs        ? ?/sec    1.08     13.5±0.02µs        ? ?/sec
sort i32 nulls to indices 2^12                          1.00     52.8±0.12µs        ? ?/sec    1.10     58.1±0.13µs        ? ?/sec
sort i32 to indices 2^10                                1.01     11.5±0.03µs        ? ?/sec    1.00     11.4±0.02µs        ? ?/sec
sort i32 to indices 2^12                                1.02     54.9±0.18µs        ? ?/sec    1.00     53.7±0.27µs        ? ?/sec
sort primitive run 2^12                                 1.00      6.2±0.01µs        ? ?/sec    1.00      6.2±0.01µs        ? ?/sec
sort primitive run to indices 2^12                      1.00      9.0±0.25µs        ? ?/sec    1.04      9.4±0.01µs        ? ?/sec
sort string[10] dict nulls to indices 2^12              1.00    169.2±0.35µs        ? ?/sec    1.05    176.9±0.35µs        ? ?/sec
sort string[10] dict to indices 2^12                    1.00    300.2±0.46µs        ? ?/sec    1.02    305.3±0.59µs        ? ?/sec
sort string[10] nulls to indices 2^12                   1.00    137.1±0.40µs        ? ?/sec    1.05    143.4±7.07µs        ? ?/sec
sort string[10] to indices 2^12                         1.00    229.9±0.84µs        ? ?/sec    1.01    231.6±0.57µs        ? ?/sec
sort string_view[0-400] nulls to indices 2^12           1.00     82.7±0.20µs        ? ?/sec    1.06     87.4±0.28µs        ? ?/sec
sort string_view[0-400] to indices 2^12                 1.01    121.9±0.21µs        ? ?/sec    1.00    120.5±0.29µs        ? ?/sec
sort string_view[10] nulls to indices 2^12              1.47    108.6±0.22µs        ? ?/sec    1.00     73.7±0.32µs        ? ?/sec
sort string_view[10] to indices 2^12                    1.67    173.1±0.25µs        ? ?/sec    1.00    103.6±0.27µs        ? ?/sec
sort string_view_inlined[0-12] nulls to indices 2^12    1.45    103.3±0.14µs        ? ?/sec    1.00     71.1±0.63µs        ? ?/sec
sort string_view_inlined[0-12] to indices 2^12          1.74    163.3±0.21µs        ? ?/sec    1.00     93.7±0.36µs        ? ?/sec

Dandandan · 2025-07-03T16:07:37Z

sorting is about to become even more crazy fast!

alamb · 2025-07-03T16:18:02Z

sorting is about to become even more crazy fast!

Just when you think you can't make sorting faster.... Along comes @zhuqi-lucas and @Dandandan 🥳 🐶

alamb

Thank you @Dandandan and @zhuqi-lucas -- this looks great as well. I agree with @zhuqi-lucas that optimizing the short strings / no buffers case makes a lot of sense

arrow-ord/src/sort.rs

Co-authored-by: Andrew Lamb <[email protected]>

Dandandan added 2 commits July 3, 2025 08:00

speedup byteview comparisons

8b74554

revert sort, don't store len

04f2b1a

github-actions bot added the arrow Changes to the arrow crate label Jul 3, 2025

Dandandan added 2 commits July 3, 2025 09:12

Add all-inline case

9ed518f

Remove cmp changes

92cacf0

Dandandan changed the title ~~Speedup byte view ops some more~~ Speedup sorting for inline views Jul 3, 2025

Dandandan added 2 commits July 3, 2025 11:48

Cleanup

97e9265

Cleanup

eea04dc

Dandandan changed the title ~~Speedup sorting for inline views~~ Speedup sorting for inline views for 1.4x - 1.7x improvement Jul 3, 2025

Dandandan marked this pull request as ready for review July 3, 2025 10:28

Dandandan mentioned this pull request Jul 3, 2025

[EPIC] A collection of improvement for the performance for sort and compare and gc, etc #7802

Open

12 tasks

Remove comment

b2a1dd0

zhuqi-lucas approved these changes Jul 3, 2025

View reviewed changes

Dandandan changed the title ~~Speedup sorting for inline views for 1.4x - 1.7x improvement~~ Speedup sorting for inline views: 1.4x - 1.7x improvement Jul 3, 2025

alamb approved these changes Jul 3, 2025

View reviewed changes

arrow-ord/src/sort.rs Show resolved Hide resolved

Update arrow-ord/src/sort.rs

fb7e260

Co-authored-by: Andrew Lamb <[email protected]>

Dandandan merged commit e6cb61f into apache:main Jul 3, 2025
20 of 21 checks passed

zhuqi-lucas mentioned this pull request Jul 6, 2025

fix: Incorrect inlined string view comparison after Add prefix compar… #7875

Merged

alamb mentioned this pull request Jul 28, 2025

Speedup sorting for inline views #7857

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Speedup sorting for inline views: 1.4x - 1.7x improvement #7856

Speedup sorting for inline views: 1.4x - 1.7x improvement #7856

Uh oh!

Dandandan commented Jul 3, 2025 •

edited

Loading

Uh oh!

Dandandan commented Jul 3, 2025

Uh oh!

zhuqi-lucas commented Jul 3, 2025

Uh oh!

zhuqi-lucas left a comment

Uh oh!

zhuqi-lucas Jul 3, 2025

Uh oh!

alamb commented Jul 3, 2025

Uh oh!

alamb commented Jul 3, 2025

Uh oh!

Dandandan commented Jul 3, 2025

Uh oh!

alamb commented Jul 3, 2025

Uh oh!

alamb left a comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Speedup sorting for inline views: 1.4x - 1.7x improvement #7856

Speedup sorting for inline views: 1.4x - 1.7x improvement #7856

Uh oh!

Conversation

Dandandan commented Jul 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Are these changes tested?, are they covered by existing tests)?

Are there any user-facing changes?

Uh oh!

Dandandan commented Jul 3, 2025

Uh oh!

zhuqi-lucas commented Jul 3, 2025

Uh oh!

zhuqi-lucas left a comment

Choose a reason for hiding this comment

Uh oh!

zhuqi-lucas Jul 3, 2025

Choose a reason for hiding this comment

Uh oh!

alamb commented Jul 3, 2025

Uh oh!

alamb commented Jul 3, 2025

Uh oh!

Dandandan commented Jul 3, 2025

Uh oh!

alamb commented Jul 3, 2025

Uh oh!

alamb left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Dandandan commented Jul 3, 2025 •

edited

Loading