Skip to content

WIP: Special case apply_bitwise_*_op operation when buffers are u64 aligned #8835

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
wants to merge 1 commit into from
Closed

WIP: Special case apply_bitwise_*_op operation when buffers are u64 aligned #8835

wants to merge 1 commit into from

Conversation

@alamb
Copy link
Contributor

@alamb alamb commented Nov 13, 2025 •
edited
Loading

Which issue does this PR close?

Rationale for this change

Add a special case for u64 (word aligned) buffers in apply_bitwise_*_op operations

I have been thinking (dreaming?) about this since the initial PR from @ruluvaton (see #8619 (comment))

If this shows reasonable performance improvements I will polish this up

What changes are included in this PR?

Add special case for apply_bitwise_*_op

Are these changes tested?

Functionally covered by existing tests.

I will also run benchmarks on this

Are there any user-facing changes?

TBD

@github-actions github-actions bot added the arrow Changes to the arrow crate label Nov 13, 2025
@alamb alamb force-pushed the alamb/test_word_aligned branch from a581f4d to 5020953 Compare November 13, 2025 14:20
@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖 ./gh_compare_arrow.sh Benchmark Script Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing alamb/test_word_aligned (5020953) to f8d9572 diff
BENCH_NAME=boolean_append_packed
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental --bench boolean_append_packed
BENCH_FILTER=
BENCH_BRANCH_NAME=alamb_test_word_aligned
Results will be posted here when complete

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖: Benchmark completed

Details 

group                    alamb_test_word_aligned                main
-----                    -----------------------                ----
boolean_append_packed    1.00      5.1±0.02µs        ? ?/sec    1.11      5.7±0.01µs        ? ?/sec

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖 ./gh_compare_arrow.sh Benchmark Script Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing alamb/test_word_aligned (5020953) to f8d9572 diff
BENCH_NAME=concatenate_kernel
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental --bench concatenate_kernel
BENCH_FILTER=
BENCH_BRANCH_NAME=alamb_test_word_aligned
Results will be posted here when complete

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖: Benchmark completed

Details 

group                                                          alamb_test_word_aligned                main
-----                                                          -----------------------                ----
concat 1024 arrays boolean 4                                   1.01     22.7±0.07µs        ? ?/sec    1.00     22.5±0.09µs        ? ?/sec
concat 1024 arrays i32 4                                       1.02     14.8±0.03µs        ? ?/sec    1.00     14.5±0.02µs        ? ?/sec
concat 1024 arrays str 4                                       1.06     37.4±0.35µs        ? ?/sec    1.00     35.1±0.31µs        ? ?/sec
concat boolean 1024                                            1.01    340.0±0.77ns        ? ?/sec    1.00    337.5±5.23ns        ? ?/sec
concat boolean 8192 over 100 arrays                            1.00      5.1±0.02µs        ? ?/sec    1.00      5.1±0.02µs        ? ?/sec
concat boolean nulls 1024                                      1.02    593.4±3.52ns        ? ?/sec    1.00    580.2±0.90ns        ? ?/sec
concat boolean nulls 8192 over 100 arrays                      1.00     18.2±0.07µs        ? ?/sec    1.00     18.2±0.07µs        ? ?/sec
concat fixed size lists                                        1.09   769.4±27.33µs        ? ?/sec    1.00   705.2±22.59µs        ? ?/sec
concat i32 1024                                                1.00    387.8±1.47ns        ? ?/sec    1.00    387.1±1.10ns        ? ?/sec
concat i32 8192 over 100 arrays                                1.05   222.3±10.02µs        ? ?/sec    1.00    211.8±5.15µs        ? ?/sec
concat i32 nulls 1024                                          1.00    609.7±1.31ns        ? ?/sec    1.01    613.3±1.32ns        ? ?/sec
concat i32 nulls 8192 over 100 arrays                          1.09    258.4±9.18µs        ? ?/sec    1.00    237.9±4.31µs        ? ?/sec
concat str 1024                                                1.00     12.3±1.07µs        ? ?/sec    1.12     13.8±1.22µs        ? ?/sec
concat str 8192 over 100 arrays                                1.03    106.1±0.89ms        ? ?/sec    1.00    103.2±1.04ms        ? ?/sec
concat str nulls 1024                                          1.00      5.9±0.66µs        ? ?/sec    1.04      6.1±0.63µs        ? ?/sec
concat str nulls 8192 over 100 arrays                          1.03     53.3±0.49ms        ? ?/sec    1.00     52.0±0.56ms        ? ?/sec
concat str_dict 1024                                           1.00      2.8±0.01µs        ? ?/sec    1.00      2.9±0.01µs        ? ?/sec
concat str_dict_sparse 1024                                    1.00      6.9±0.02µs        ? ?/sec    1.01      7.0±0.03µs        ? ?/sec
concat struct with int32 and dicts size=1024 count=2           1.04      6.9±0.17µs        ? ?/sec    1.00      6.6±0.03µs        ? ?/sec
concat utf8_view  max_str_len=128 null_density=0               1.09     84.7±0.52µs        ? ?/sec    1.00     78.0±0.44µs        ? ?/sec
concat utf8_view  max_str_len=128 null_density=0.2             1.00     79.7±0.30µs        ? ?/sec    1.00     79.5±0.49µs        ? ?/sec
concat utf8_view  max_str_len=20 null_density=0                1.00     90.2±0.50µs        ? ?/sec    1.00     89.9±0.39µs        ? ?/sec
concat utf8_view  max_str_len=20 null_density=0.2              1.01     92.4±0.36µs        ? ?/sec    1.00     91.8±0.40µs        ? ?/sec
concat utf8_view all_inline max_str_len=12 null_density=0      1.05     50.2±3.55µs        ? ?/sec    1.00     47.6±3.06µs        ? ?/sec
concat utf8_view all_inline max_str_len=12 null_density=0.2    1.03     50.6±3.38µs        ? ?/sec    1.00     48.9±3.10µs        ? ?/sec

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖 ./gh_compare_arrow.sh Benchmark Script Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing alamb/test_word_aligned (5020953) to f8d9572 diff
BENCH_NAME=filter_kernels
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental --bench filter_kernels
BENCH_FILTER=
BENCH_BRANCH_NAME=alamb_test_word_aligned
Results will be posted here when complete

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖: Benchmark completed

Details 

group                                                                         alamb_test_word_aligned                main
-----                                                                         -----------------------                ----
filter context decimal128 (kept 1/2)                                          1.00     42.4±2.46µs        ? ?/sec    1.03     43.6±3.56µs        ? ?/sec
filter context decimal128 high selectivity (kept 1023/1024)                   1.01     49.7±1.57µs        ? ?/sec    1.00     49.3±0.68µs        ? ?/sec
filter context decimal128 low selectivity (kept 1/1024)                       1.00    238.9±0.30ns        ? ?/sec    1.05    250.0±0.39ns        ? ?/sec
filter context f32 (kept 1/2)                                                 1.00     98.1±0.66µs        ? ?/sec    1.00     98.0±0.31µs        ? ?/sec
filter context f32 high selectivity (kept 1023/1024)                          1.00      9.9±0.32µs        ? ?/sec    1.02     10.1±0.39µs        ? ?/sec
filter context f32 low selectivity (kept 1/1024)                              1.19    543.9±1.18ns        ? ?/sec    1.00    455.9±0.51ns        ? ?/sec
filter context fsb with value length 20 (kept 1/2)                            1.00     79.7±0.15µs        ? ?/sec    1.00     79.6±0.17µs        ? ?/sec
filter context fsb with value length 20 high selectivity (kept 1023/1024)     1.00     79.5±0.16µs        ? ?/sec    1.00     79.6±0.14µs        ? ?/sec
filter context fsb with value length 20 low selectivity (kept 1/1024)         1.00     79.5±0.13µs        ? ?/sec    1.00     79.6±0.48µs        ? ?/sec
filter context fsb with value length 5 (kept 1/2)                             1.00     79.5±0.11µs        ? ?/sec    1.00     79.5±0.14µs        ? ?/sec
filter context fsb with value length 5 high selectivity (kept 1023/1024)      1.00     79.5±0.19µs        ? ?/sec    1.00     79.6±0.14µs        ? ?/sec
filter context fsb with value length 5 low selectivity (kept 1/1024)          1.00     79.6±0.86µs        ? ?/sec    1.00     79.6±0.62µs        ? ?/sec
filter context fsb with value length 50 (kept 1/2)                            1.00     79.6±0.11µs        ? ?/sec    1.00     79.6±0.12µs        ? ?/sec
filter context fsb with value length 50 high selectivity (kept 1023/1024)     1.00     79.5±0.13µs        ? ?/sec    1.00     79.5±0.13µs        ? ?/sec
filter context fsb with value length 50 low selectivity (kept 1/1024)         1.00     79.5±0.12µs        ? ?/sec    1.00     79.6±0.26µs        ? ?/sec
filter context i32 (kept 1/2)                                                 1.00     18.1±0.05µs        ? ?/sec    1.00     18.1±0.06µs        ? ?/sec
filter context i32 high selectivity (kept 1023/1024)                          1.01      6.5±0.32µs        ? ?/sec    1.00      6.4±0.47µs        ? ?/sec
filter context i32 low selectivity (kept 1/1024)                              1.00    243.4±0.45ns        ? ?/sec    1.03    251.8±0.35ns        ? ?/sec
filter context i32 w NULLs (kept 1/2)                                         1.00     98.2±0.18µs        ? ?/sec    1.00     98.6±1.34µs        ? ?/sec
filter context i32 w NULLs high selectivity (kept 1023/1024)                  1.00     10.3±0.25µs        ? ?/sec    1.02     10.4±0.37µs        ? ?/sec
filter context i32 w NULLs low selectivity (kept 1/1024)                      1.00    445.0±1.02ns        ? ?/sec    1.02    453.6±3.31ns        ? ?/sec
filter context mixed string view (kept 1/2)                                   1.00    122.2±5.52µs        ? ?/sec    1.04    126.9±7.06µs        ? ?/sec
filter context mixed string view high selectivity (kept 1023/1024)            1.00     53.2±1.35µs        ? ?/sec    1.01     53.9±1.55µs        ? ?/sec
filter context mixed string view low selectivity (kept 1/1024)                1.00    628.6±1.58ns        ? ?/sec    1.03    649.9±0.80ns        ? ?/sec
filter context short string view (kept 1/2)                                   1.03    124.1±5.32µs        ? ?/sec    1.00    121.0±0.77µs        ? ?/sec
filter context short string view high selectivity (kept 1023/1024)            1.00     52.8±1.38µs        ? ?/sec    1.04     55.1±2.01µs        ? ?/sec
filter context short string view low selectivity (kept 1/1024)                1.00    453.2±0.46ns        ? ?/sec    1.02    463.5±0.67ns        ? ?/sec
filter context string (kept 1/2)                                              1.01   602.2±14.24µs        ? ?/sec    1.00   595.7±12.99µs        ? ?/sec
filter context string dictionary (kept 1/2)                                   1.01     18.7±0.07µs        ? ?/sec    1.00     18.6±0.06µs        ? ?/sec
filter context string dictionary high selectivity (kept 1023/1024)            1.01      7.6±0.32µs        ? ?/sec    1.00      7.6±0.41µs        ? ?/sec
filter context string dictionary low selectivity (kept 1/1024)                1.00    826.1±2.29ns        ? ?/sec    1.01    833.6±1.87ns        ? ?/sec
filter context string dictionary w NULLs (kept 1/2)                           1.00     98.9±0.24µs        ? ?/sec    1.00     98.8±0.18µs        ? ?/sec
filter context string dictionary w NULLs high selectivity (kept 1023/1024)    1.02     10.7±0.35µs        ? ?/sec    1.00     10.5±0.47µs        ? ?/sec
filter context string dictionary w NULLs low selectivity (kept 1/1024)        1.00   1050.3±1.75ns        ? ?/sec    1.01   1064.8±2.03ns        ? ?/sec
filter context string high selectivity (kept 1023/1024)                       1.10   691.0±18.09µs        ? ?/sec    1.00   627.6±13.75µs        ? ?/sec
filter context string low selectivity (kept 1/1024)                           1.00    908.4±1.36ns        ? ?/sec    1.23   1121.7±1.92ns        ? ?/sec
filter context u8 (kept 1/2)                                                  1.00     22.5±0.04µs        ? ?/sec    1.00     22.5±0.06µs        ? ?/sec
filter context u8 high selectivity (kept 1023/1024)                           1.00   1832.4±8.72ns        ? ?/sec    1.10      2.0±0.01µs        ? ?/sec
filter context u8 low selectivity (kept 1/1024)                               1.00    241.1±0.23ns        ? ?/sec    1.04    250.4±0.55ns        ? ?/sec
filter context u8 w NULLs (kept 1/2)                                          1.00    102.4±0.21µs        ? ?/sec    1.00    102.5±0.68µs        ? ?/sec
filter context u8 w NULLs high selectivity (kept 1023/1024)                   1.05      5.3±0.01µs        ? ?/sec    1.00      5.1±0.02µs        ? ?/sec
filter context u8 w NULLs low selectivity (kept 1/1024)                       1.21    546.5±1.07ns        ? ?/sec    1.00    452.9±0.92ns        ? ?/sec
filter decimal128 (kept 1/2)                                                  1.07     52.7±4.33µs        ? ?/sec    1.00     49.2±0.54µs        ? ?/sec
filter decimal128 high selectivity (kept 1023/1024)                           1.01     53.6±1.44µs        ? ?/sec    1.00     52.9±1.84µs        ? ?/sec
filter decimal128 low selectivity (kept 1/1024)                               1.00      3.0±0.02µs        ? ?/sec    1.00      3.0±0.01µs        ? ?/sec
filter f32 (kept 1/2)                                                         1.00    155.4±0.62µs        ? ?/sec    1.00    155.2±0.54µs        ? ?/sec
filter fsb with value length 20 (kept 1/2)                                    1.00    125.6±1.50µs        ? ?/sec    1.00    125.2±0.43µs        ? ?/sec
filter fsb with value length 20 high selectivity (kept 1023/1024)             1.02     70.9±2.95µs        ? ?/sec    1.00     69.5±1.78µs        ? ?/sec
filter fsb with value length 20 low selectivity (kept 1/1024)                 1.00      3.4±0.02µs        ? ?/sec    1.02      3.4±0.01µs        ? ?/sec
filter fsb with value length 5 (kept 1/2)                                     1.00    120.5±0.22µs        ? ?/sec    1.00    120.9±0.32µs        ? ?/sec
filter fsb with value length 5 high selectivity (kept 1023/1024)              1.00     11.1±0.54µs        ? ?/sec    1.02     11.3±0.65µs        ? ?/sec
filter fsb with value length 5 low selectivity (kept 1/1024)                  1.01      3.3±0.01µs        ? ?/sec    1.00      3.3±0.01µs        ? ?/sec
filter fsb with value length 50 (kept 1/2)                                    1.00    161.2±8.21µs        ? ?/sec    1.03    166.2±8.98µs        ? ?/sec
filter fsb with value length 50 high selectivity (kept 1023/1024)             1.00    208.2±9.23µs        ? ?/sec    1.01    210.9±3.54µs        ? ?/sec
filter fsb with value length 50 low selectivity (kept 1/1024)                 1.00      3.3±0.01µs        ? ?/sec    1.00      3.4±0.01µs        ? ?/sec
filter i32 (kept 1/2)                                                         1.00     45.8±0.55µs        ? ?/sec    1.00     45.7±0.09µs        ? ?/sec
filter i32 high selectivity (kept 1023/1024)                                  1.01      8.9±0.32µs        ? ?/sec    1.00      8.9±0.36µs        ? ?/sec
filter i32 low selectivity (kept 1/1024)                                      1.00      3.0±0.01µs        ? ?/sec    1.00      3.0±0.01µs        ? ?/sec
filter optimize (kept 1/2)                                                    1.00     53.6±0.06µs        ? ?/sec    1.00     53.6±0.08µs        ? ?/sec
filter optimize high selectivity (kept 1023/1024)                             1.01      3.1±0.01µs        ? ?/sec    1.00      3.1±0.01µs        ? ?/sec
filter optimize low selectivity (kept 1/1024)                                 1.00      3.1±0.00µs        ? ?/sec    1.00      3.1±0.01µs        ? ?/sec
filter run array (kept 1/2)                                                   1.00    424.7±0.84µs        ? ?/sec    1.00    425.2±0.97µs        ? ?/sec
filter run array high selectivity (kept 1023/1024)                            1.00    449.2±1.64µs        ? ?/sec    1.00    447.9±1.17µs        ? ?/sec
filter run array low selectivity (kept 1/1024)                                1.00    335.0±0.89µs        ? ?/sec    1.00    335.2±0.99µs        ? ?/sec
filter single record batch                                                    1.00     46.1±0.09µs        ? ?/sec    1.00     46.1±0.11µs        ? ?/sec
filter u8 (kept 1/2)                                                          1.00     45.7±0.19µs        ? ?/sec    1.00     45.6±0.04µs        ? ?/sec
filter u8 high selectivity (kept 1023/1024)                                   1.02      4.0±0.02µs        ? ?/sec    1.00      4.0±0.01µs        ? ?/sec
filter u8 low selectivity (kept 1/1024)                                       1.00      3.1±0.01µs        ? ?/sec    1.00      3.1±0.01µs        ? ?/sec

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖 ./gh_compare_arrow.sh Benchmark Script Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing alamb/test_word_aligned (5020953) to f8d9572 diff
BENCH_NAME=boolean_append_packed
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental --bench boolean_append_packed
BENCH_FILTER=
BENCH_BRANCH_NAME=alamb_test_word_aligned
Results will be posted here when complete

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖: Benchmark completed

Details 

group                    alamb_test_word_aligned                main
-----                    -----------------------                ----
boolean_append_packed    1.00      5.5±0.02µs        ? ?/sec    1.10      6.1±0.02µs        ? ?/sec

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖 ./gh_compare_arrow.sh Benchmark Script Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing alamb/test_word_aligned (5020953) to f8d9572 diff
BENCH_NAME=concatenate_kernel
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental --bench concatenate_kernel
BENCH_FILTER=
BENCH_BRANCH_NAME=alamb_test_word_aligned
Results will be posted here when complete

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖: Benchmark completed

Details 

group                                                          alamb_test_word_aligned                main
-----                                                          -----------------------                ----
concat 1024 arrays boolean 4                                   1.01     22.7±0.08µs        ? ?/sec    1.00     22.5±0.10µs        ? ?/sec
concat 1024 arrays i32 4                                       1.01     14.6±0.11µs        ? ?/sec    1.00     14.6±0.12µs        ? ?/sec
concat 1024 arrays str 4                                       1.00     36.5±0.28µs        ? ?/sec    1.00     36.4±0.30µs        ? ?/sec
concat boolean 1024                                            1.00    329.4±2.12ns        ? ?/sec    1.05    345.8±0.49ns        ? ?/sec
concat boolean 8192 over 100 arrays                            1.00      5.0±0.02µs        ? ?/sec    1.03      5.1±0.02µs        ? ?/sec
concat boolean nulls 1024                                      1.02    572.3±0.80ns        ? ?/sec    1.00    560.8±4.79ns        ? ?/sec
concat boolean nulls 8192 over 100 arrays                      1.00     18.1±0.06µs        ? ?/sec    1.00     18.2±0.04µs        ? ?/sec
concat fixed size lists                                        1.00   697.2±22.65µs        ? ?/sec    1.09   760.6±29.42µs        ? ?/sec
concat i32 1024                                                1.02    392.3±0.59ns        ? ?/sec    1.00    385.7±2.95ns        ? ?/sec
concat i32 8192 over 100 arrays                                1.00    201.4±5.08µs        ? ?/sec    1.03    206.9±4.52µs        ? ?/sec
concat i32 nulls 1024                                          1.00    609.3±0.96ns        ? ?/sec    1.01    616.0±3.91ns        ? ?/sec
concat i32 nulls 8192 over 100 arrays                          1.04   248.8±13.46µs        ? ?/sec    1.00    238.9±6.25µs        ? ?/sec
concat str 1024                                                1.00     13.1±0.97µs        ? ?/sec    1.06     13.8±1.03µs        ? ?/sec
concat str 8192 over 100 arrays                                1.00    105.0±0.85ms        ? ?/sec    1.00    105.0±0.82ms        ? ?/sec
concat str nulls 1024                                          1.17      6.2±0.86µs        ? ?/sec    1.00      5.3±0.60µs        ? ?/sec
concat str nulls 8192 over 100 arrays                          1.00     52.7±0.40ms        ? ?/sec    1.00     52.8±0.50ms        ? ?/sec
concat str_dict 1024                                           1.04      3.1±0.01µs        ? ?/sec    1.00      3.0±0.01µs        ? ?/sec
concat str_dict_sparse 1024                                    1.00      6.9±0.03µs        ? ?/sec    1.00      6.9±0.07µs        ? ?/sec
concat struct with int32 and dicts size=1024 count=2           1.00      6.8±0.02µs        ? ?/sec    1.02      6.9±0.05µs        ? ?/sec
concat utf8_view  max_str_len=128 null_density=0               1.14     88.4±0.57µs        ? ?/sec    1.00     77.7±0.46µs        ? ?/sec
concat utf8_view  max_str_len=128 null_density=0.2             1.00     79.2±0.47µs        ? ?/sec    1.00     79.5±0.36µs        ? ?/sec
concat utf8_view  max_str_len=20 null_density=0                1.00     77.4±0.59µs        ? ?/sec    1.14     88.5±0.55µs        ? ?/sec
concat utf8_view  max_str_len=20 null_density=0.2              1.00     79.3±0.37µs        ? ?/sec    1.14     90.8±0.79µs        ? ?/sec
concat utf8_view all_inline max_str_len=12 null_density=0      1.00     46.6±3.38µs        ? ?/sec    1.00     46.7±3.60µs        ? ?/sec
concat utf8_view all_inline max_str_len=12 null_density=0.2    1.00     48.9±3.18µs        ? ?/sec    1.02     49.9±2.92µs        ? ?/sec

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖 ./gh_compare_arrow.sh Benchmark Script Running
Linux aal-dev 6.14.0-1018-gcp #19~24.04.1-Ubuntu SMP Wed Sep 24 23:23:09 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing alamb/test_word_aligned (5020953) to f8d9572 diff
BENCH_NAME=filter_kernels
BENCH_COMMAND=cargo bench --features=arrow,async,test_common,experimental --bench filter_kernels
BENCH_FILTER=
BENCH_BRANCH_NAME=alamb_test_word_aligned
Results will be posted here when complete

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

🤖: Benchmark completed

Details 

group                                                                         alamb_test_word_aligned                main
-----                                                                         -----------------------                ----
filter context decimal128 (kept 1/2)                                          1.00     43.4±4.61µs        ? ?/sec    1.14     49.3±8.64µs        ? ?/sec
filter context decimal128 high selectivity (kept 1023/1024)                   1.00     49.2±1.37µs        ? ?/sec    1.00     49.5±1.37µs        ? ?/sec
filter context decimal128 low selectivity (kept 1/1024)                       1.06    253.0±0.41ns        ? ?/sec    1.00    239.7±0.59ns        ? ?/sec
filter context f32 (kept 1/2)                                                 1.00     98.1±0.27µs        ? ?/sec    1.00     97.9±0.26µs        ? ?/sec
filter context f32 high selectivity (kept 1023/1024)                          1.00      9.8±0.30µs        ? ?/sec    1.01      9.9±0.36µs        ? ?/sec
filter context f32 low selectivity (kept 1/1024)                              1.00    461.7±0.58ns        ? ?/sec    1.18    546.8±0.98ns        ? ?/sec
filter context fsb with value length 20 (kept 1/2)                            1.00     79.5±0.17µs        ? ?/sec    1.00     79.5±0.16µs        ? ?/sec
filter context fsb with value length 20 high selectivity (kept 1023/1024)     1.00     79.5±0.13µs        ? ?/sec    1.00     79.5±0.22µs        ? ?/sec
filter context fsb with value length 20 low selectivity (kept 1/1024)         1.00     79.6±0.18µs        ? ?/sec    1.00     79.5±0.10µs        ? ?/sec
filter context fsb with value length 5 (kept 1/2)                             1.00     79.5±0.17µs        ? ?/sec    1.00     79.6±0.13µs        ? ?/sec
filter context fsb with value length 5 high selectivity (kept 1023/1024)      1.00     79.5±0.12µs        ? ?/sec    1.00     79.5±0.13µs        ? ?/sec
filter context fsb with value length 5 low selectivity (kept 1/1024)          1.00     79.5±0.14µs        ? ?/sec    1.00     79.5±0.13µs        ? ?/sec
filter context fsb with value length 50 (kept 1/2)                            1.00     79.7±0.19µs        ? ?/sec    1.00     79.7±0.19µs        ? ?/sec
filter context fsb with value length 50 high selectivity (kept 1023/1024)     1.00     79.5±0.14µs        ? ?/sec    1.00     79.5±0.20µs        ? ?/sec
filter context fsb with value length 50 low selectivity (kept 1/1024)         1.00     79.5±0.17µs        ? ?/sec    1.00     79.5±0.08µs        ? ?/sec
filter context i32 (kept 1/2)                                                 1.00     17.9±0.04µs        ? ?/sec    1.01     18.1±0.11µs        ? ?/sec
filter context i32 high selectivity (kept 1023/1024)                          1.00      6.1±0.26µs        ? ?/sec    1.05      6.5±0.43µs        ? ?/sec
filter context i32 low selectivity (kept 1/1024)                              1.07    261.9±2.24ns        ? ?/sec    1.00    244.0±1.82ns        ? ?/sec
filter context i32 w NULLs (kept 1/2)                                         1.00     98.2±0.64µs        ? ?/sec    1.00     98.0±0.16µs        ? ?/sec
filter context i32 w NULLs high selectivity (kept 1023/1024)                  1.00     10.2±0.43µs        ? ?/sec    1.03     10.4±0.42µs        ? ?/sec
filter context i32 w NULLs low selectivity (kept 1/1024)                      1.00    463.0±0.81ns        ? ?/sec    1.18    548.5±2.32ns        ? ?/sec
filter context mixed string view (kept 1/2)                                   1.01    122.5±4.50µs        ? ?/sec    1.00    121.5±3.09µs        ? ?/sec
filter context mixed string view high selectivity (kept 1023/1024)            1.01     54.4±1.57µs        ? ?/sec    1.00     53.9±1.31µs        ? ?/sec
filter context mixed string view low selectivity (kept 1/1024)                1.00    646.7±1.06ns        ? ?/sec    1.00    645.6±0.89ns        ? ?/sec
filter context short string view (kept 1/2)                                   1.03    124.2±6.53µs        ? ?/sec    1.00    120.9±2.13µs        ? ?/sec
filter context short string view high selectivity (kept 1023/1024)            1.00     54.5±1.28µs        ? ?/sec    1.00     54.5±1.97µs        ? ?/sec
filter context short string view low selectivity (kept 1/1024)                1.00    452.1±0.49ns        ? ?/sec    1.02    459.0±1.25ns        ? ?/sec
filter context string (kept 1/2)                                              1.00    591.9±5.43µs        ? ?/sec    1.01   597.4±11.71µs        ? ?/sec
filter context string dictionary (kept 1/2)                                   1.00     18.6±0.07µs        ? ?/sec    1.00     18.6±0.08µs        ? ?/sec
filter context string dictionary high selectivity (kept 1023/1024)            1.00      7.1±0.38µs        ? ?/sec    1.04      7.3±0.44µs        ? ?/sec
filter context string dictionary low selectivity (kept 1/1024)                1.04    860.1±1.90ns        ? ?/sec    1.00    823.1±2.90ns        ? ?/sec
filter context string dictionary w NULLs (kept 1/2)                           1.00     98.9±0.23µs        ? ?/sec    1.00     99.1±0.28µs        ? ?/sec
filter context string dictionary w NULLs high selectivity (kept 1023/1024)    1.01     10.6±0.36µs        ? ?/sec    1.00     10.5±0.33µs        ? ?/sec
filter context string dictionary w NULLs low selectivity (kept 1/1024)        1.03   1091.8±2.24ns        ? ?/sec    1.00   1061.9±3.23ns        ? ?/sec
filter context string high selectivity (kept 1023/1024)                       1.00   644.5±13.17µs        ? ?/sec    1.00   642.1±12.30µs        ? ?/sec
filter context string low selectivity (kept 1/1024)                           1.00   1004.5±3.09ns        ? ?/sec    1.08   1087.7±4.80ns        ? ?/sec
filter context u8 (kept 1/2)                                                  1.00     22.4±0.05µs        ? ?/sec    1.00     22.5±0.06µs        ? ?/sec
filter context u8 high selectivity (kept 1023/1024)                           1.00   1841.0±7.94ns        ? ?/sec    1.03  1900.9±13.28ns        ? ?/sec
filter context u8 low selectivity (kept 1/1024)                               1.06    259.7±0.27ns        ? ?/sec    1.00    245.0±0.56ns        ? ?/sec
filter context u8 w NULLs (kept 1/2)                                          1.00    102.4±0.18µs        ? ?/sec    1.00    102.4±0.31µs        ? ?/sec
filter context u8 w NULLs high selectivity (kept 1023/1024)                   1.02      5.2±0.02µs        ? ?/sec    1.00      5.1±0.02µs        ? ?/sec
filter context u8 w NULLs low selectivity (kept 1/1024)                       1.03    564.4±1.25ns        ? ?/sec    1.00    547.6±3.38ns        ? ?/sec
filter decimal128 (kept 1/2)                                                  1.00     50.4±1.84µs        ? ?/sec    1.02     51.2±3.06µs        ? ?/sec
filter decimal128 high selectivity (kept 1023/1024)                           1.00     52.0±0.99µs        ? ?/sec    1.03     53.3±1.62µs        ? ?/sec
filter decimal128 low selectivity (kept 1/1024)                               1.00      3.0±0.01µs        ? ?/sec    1.00      3.0±0.02µs        ? ?/sec
filter f32 (kept 1/2)                                                         1.00    155.3±1.36µs        ? ?/sec    1.00    155.1±0.22µs        ? ?/sec
filter fsb with value length 20 (kept 1/2)                                    1.00    124.7±0.50µs        ? ?/sec    1.00    124.5±0.41µs        ? ?/sec
filter fsb with value length 20 high selectivity (kept 1023/1024)             1.00     71.7±2.07µs        ? ?/sec    1.00     71.8±2.87µs        ? ?/sec
filter fsb with value length 20 low selectivity (kept 1/1024)                 1.00      3.4±0.01µs        ? ?/sec    1.00      3.4±0.01µs        ? ?/sec
filter fsb with value length 5 (kept 1/2)                                     1.00    120.7±0.33µs        ? ?/sec    1.00    120.8±0.32µs        ? ?/sec
filter fsb with value length 5 high selectivity (kept 1023/1024)              1.00     11.0±0.39µs        ? ?/sec    1.05     11.5±0.55µs        ? ?/sec
filter fsb with value length 5 low selectivity (kept 1/1024)                  1.00      3.3±0.01µs        ? ?/sec    1.00      3.3±0.01µs        ? ?/sec
filter fsb with value length 50 (kept 1/2)                                    1.01    163.8±5.84µs        ? ?/sec    1.00   162.5±13.84µs        ? ?/sec
filter fsb with value length 50 high selectivity (kept 1023/1024)             1.00    207.8±3.34µs        ? ?/sec    1.05    217.5±6.14µs        ? ?/sec
filter fsb with value length 50 low selectivity (kept 1/1024)                 1.00      3.3±0.01µs        ? ?/sec    1.01      3.4±0.01µs        ? ?/sec
filter i32 (kept 1/2)                                                         1.00     45.6±0.07µs        ? ?/sec    1.00     45.7±0.11µs        ? ?/sec
filter i32 high selectivity (kept 1023/1024)                                  1.01      8.7±0.43µs        ? ?/sec    1.00      8.6±0.34µs        ? ?/sec
filter i32 low selectivity (kept 1/1024)                                      1.00      3.0±0.01µs        ? ?/sec    1.00      3.0±0.01µs        ? ?/sec
filter optimize (kept 1/2)                                                    1.00     53.6±0.09µs        ? ?/sec    1.00     53.6±0.07µs        ? ?/sec
filter optimize high selectivity (kept 1023/1024)                             1.00      3.0±0.01µs        ? ?/sec    1.00      3.0±0.02µs        ? ?/sec
filter optimize low selectivity (kept 1/1024)                                 1.00      3.1±0.01µs        ? ?/sec    1.00      3.1±0.01µs        ? ?/sec
filter run array (kept 1/2)                                                   1.00    424.5±1.01µs        ? ?/sec    1.00    424.1±0.76µs        ? ?/sec
filter run array high selectivity (kept 1023/1024)                            1.00    447.7±1.43µs        ? ?/sec    1.00    448.9±1.61µs        ? ?/sec
filter run array low selectivity (kept 1/1024)                                1.00    335.2±1.01µs        ? ?/sec    1.00    335.1±0.83µs        ? ?/sec
filter single record batch                                                    1.00     46.1±0.09µs        ? ?/sec    1.00     46.1±0.08µs        ? ?/sec
filter u8 (kept 1/2)                                                          1.00     45.7±0.06µs        ? ?/sec    1.00     45.7±0.10µs        ? ?/sec
filter u8 high selectivity (kept 1023/1024)                                   1.02      4.1±0.02µs        ? ?/sec    1.00      4.0±0.01µs        ? ?/sec
filter u8 low selectivity (kept 1/1024)                                       1.00      3.1±0.01µs        ? ?/sec    1.00      3.1±0.02µs        ? ?/sec

@alamb
Copy link
Contributor Author

alamb commented Nov 13, 2025

My analysis of the results is that there are no obvious changes in performance

@alamb alamb closed this Nov 14, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

arrow Changes to the arrow crate

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@alamb