perf: improve range and generate_series for Int64
#19431
Conversation
github-actions
bot
added
sqllogictest
|
run benchmark range_and_generate_series |
|
🤖 |
|
@alamb I might increased the benchmark too much, is it stuck? |
|
show benchmark queue |
| } | ||
| } else { | ||
| // step < 0 | ||
| let cur = series_state.current as i128; |
There was a problem hiding this comment.
Can the same effect be achieved using a reverse iterator like:
(series_state.end..=series_state.current).rev()
Depending on whether end is exclusive or not, you may have to offset it by one.
There was a problem hiding this comment.
I gave it a go locally. This seems to do the trick:
if series_state.include_end {
Int64Array::from_iter_values(
(series_state.end..=series_state.current)
.rev()
.step_by(-series_state.step as usize)
.take(series_state.batch_size),
)
} else {
Int64Array::from_iter_values(
((series_state.end + 1)..=series_state.current)
.rev()
.step_by(-series_state.step as usize)
.take(series_state.batch_size),
)
}
|
🤖 |
I don't know what happened -- the runner had died. I restarted t |
|
🤖: Benchmark completed Details
|
|
These cases seems to have slowed down Is that expected? |
|
run benchmark range_and_generate_series |
|
|
||
| # generate_series equal batch size * 2 with starting value and step (including end) | ||
| query I | ||
| SELECT * FROM generate_series(1, 39, 2) |
There was a problem hiding this comment.
should we also add a test for a starting value other than 1? Perhaps 100 and -100?
|
🤖 |
|
🤖: Benchmark completed Details
|
|
Closing as it is fast enough and this have some regression |
4 participants
Which issue does this PR close?
N/A
Rationale for this change
make
rangeandgenerate_seriesfasterWhat changes are included in this PR?
manually implement specialized implementation for Int64
rangeandgenerate_seriesAre these changes tested?
Existing tests + added more tests for edge cases
Are there any user-facing changes?
Yes, added to public trait
SeriesValuea function calledgenerate_array_for_serieswith default implementation to avoid breaking changes