zhangxffff commented on PR #20694:
URL: https://github.com/apache/datafusion/pull/20694#issuecomment-4001877753
> I'm curious if the `null_count` short-circuit helps in practice -- can you
re-run the benchmarks when you get a chance?
Benchmark result (before vs after vs after_null_count):
before: datafusion/main
after: original patch
after_null_count: patch with `null_count` guard
For nulls=20% cases: after version showed ~3-5% regressions due to calling
true_count() on every iteration. after_null_count eliminates this, matching
before (e.g. list=28/match=100%/nulls=20%: 100.9µs vs 104.5µs).
For the in_list_cols/Utf8 cases: the benchmark implicitly contains ~20%
nulls, so the `null_count() == 0` similarly eliminates the regressions (e.g.
Utf8/list=3/match=50%: 92.3µs vs 105.2µs in after).
```
(zhangxffff) zhangxffff@95d3d60664da ~/W/datafusion ((bcc52cd4))> critcmp
before after after_null_count
group after
after_null_count before
----- -----
---------------- ------
in_list_cols/Int32/list=28/match=0%/nulls=0% 1.01 92.8±0.72µs
? ?/sec 1.02 93.3±1.48µs ? ?/sec 1.00 91.7±2.11µs
? ?/sec
in_list_cols/Int32/list=28/match=0%/nulls=20% 1.04 104.4±0.97µs
? ?/sec 1.00 100.2±2.29µs ? ?/sec 1.00 100.6±3.25µs
? ?/sec
in_list_cols/Int32/list=28/match=100%/nulls=0% 1.00 3.3±0.03µs
? ?/sec 1.00 3.3±0.03µs ? ?/sec 27.42 91.6±1.31µs
? ?/sec
in_list_cols/Int32/list=28/match=100%/nulls=20% 1.04 104.5±1.20µs
? ?/sec 1.01 100.9±2.39µs ? ?/sec 1.00 100.3±1.70µs
? ?/sec
in_list_cols/Int32/list=28/match=50%/nulls=0% 1.02 50.4±0.94µs
? ?/sec 1.00 49.7±0.51µs ? ?/sec 1.84 91.4±1.78µs
? ?/sec
in_list_cols/Int32/list=28/match=50%/nulls=20% 1.05 104.7±1.90µs
? ?/sec 1.00 99.7±0.86µs ? ?/sec 1.01 101.0±3.55µs
? ?/sec
in_list_cols/Int32/list=3/match=0%/nulls=0% 1.00 9.9±0.10µs
? ?/sec 1.00 9.9±0.13µs ? ?/sec 1.00 9.9±0.08µs
? ?/sec
in_list_cols/Int32/list=3/match=0%/nulls=20% 1.03 10.9±0.10µs
? ?/sec 1.00 10.6±0.17µs ? ?/sec 1.01 10.8±0.12µs
? ?/sec
in_list_cols/Int32/list=3/match=100%/nulls=0% 1.00 3.3±0.03µs
? ?/sec 1.00 3.3±0.08µs ? ?/sec 2.97 9.9±0.25µs
? ?/sec
in_list_cols/Int32/list=3/match=100%/nulls=20% 1.03 10.8±0.10µs
? ?/sec 1.00 10.5±0.10µs ? ?/sec 1.03 10.8±0.16µs
? ?/sec
in_list_cols/Int32/list=3/match=50%/nulls=0% 1.00 9.9±0.09µs
? ?/sec 1.02 10.1±0.24µs ? ?/sec 1.00 9.9±0.16µs
? ?/sec
in_list_cols/Int32/list=3/match=50%/nulls=20% 1.02 10.8±0.09µs
? ?/sec 1.00 10.5±0.15µs ? ?/sec 1.03 10.8±0.17µs
? ?/sec
in_list_cols/Int32/list=8/match=0%/nulls=0% 1.01 26.5±0.19µs
? ?/sec 1.02 26.8±0.48µs ? ?/sec 1.00 26.1±0.26µs
? ?/sec
in_list_cols/Int32/list=8/match=0%/nulls=20% 1.03 29.4±0.28µs
? ?/sec 1.00 28.6±0.58µs ? ?/sec 1.00 28.7±0.51µs
? ?/sec
in_list_cols/Int32/list=8/match=100%/nulls=0% 1.01 3.3±0.05µs
? ?/sec 1.00 3.3±0.03µs ? ?/sec 7.92 26.3±0.74µs
? ?/sec
in_list_cols/Int32/list=8/match=100%/nulls=20% 1.05 29.6±0.47µs
? ?/sec 1.00 28.2±0.40µs ? ?/sec 1.02 28.7±0.70µs
? ?/sec
in_list_cols/Int32/list=8/match=50%/nulls=0% 1.01 26.6±0.28µs
? ?/sec 1.01 26.6±0.53µs ? ?/sec 1.00 26.3±0.39µs
? ?/sec
in_list_cols/Int32/list=8/match=50%/nulls=20% 1.03 29.5±0.36µs
? ?/sec 1.00 28.6±0.55µs ? ?/sec 1.00 28.5±0.28µs
? ?/sec
in_list_cols/Utf8/list=28/match=0% 1.18 157.0±3.90µs
? ?/sec 1.10 146.1±2.62µs ? ?/sec 1.00 132.7±2.97µs
? ?/sec
in_list_cols/Utf8/list=28/match=100% 1.09 722.5±9.38µs
? ?/sec 1.00 665.8±9.49µs ? ?/sec 1.08 722.0±6.94µs
? ?/sec
in_list_cols/Utf8/list=28/match=50% 1.01 1068.6±16.04µs
? ?/sec 1.01 1064.5±19.03µs ? ?/sec 1.00 1053.9±14.52µs
? ?/sec
in_list_cols/Utf8/list=3/match=0% 1.14 16.2±0.38µs
? ?/sec 1.07 15.3±0.28µs ? ?/sec 1.00 14.2±0.24µs
? ?/sec
in_list_cols/Utf8/list=3/match=100% 1.03 67.7±1.22µs
? ?/sec 1.00 65.6±0.80µs ? ?/sec 1.03 67.8±2.21µs
? ?/sec
in_list_cols/Utf8/list=3/match=50% 1.14 105.2±1.65µs
? ?/sec 1.00 92.3±1.64µs ? ?/sec 1.04 96.3±5.61µs
? ?/sec
in_list_cols/Utf8/list=8/match=0% 1.19 44.9±1.11µs
? ?/sec 1.09 41.0±0.64µs ? ?/sec 1.00 37.7±0.87µs
? ?/sec
in_list_cols/Utf8/list=8/match=100% 1.01 194.3±2.14µs
? ?/sec 1.00 191.7±2.73µs ? ?/sec 1.02 195.9±2.36µs
? ?/sec
in_list_cols/Utf8/list=8/match=50% 1.02 294.0±2.76µs
? ?/sec 1.00 287.3±2.73µs ? ?/sec 1.02 294.0±3.57µs
? ?/sec
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
