saadtajwar commented on PR #24476:
URL: https://github.com/apache/datafusion/pull/24476#issuecomment-5363361008
> ```
> dict_take_n/take_n/size_65536_card_1000_null_0.10
1.03 2.6±0.00ms 97.7 MElem/sec 1.00 2.5±0.00ms 100.2
MElem/sec
> dict_take_n/take_n/size_65536_card_20_null_0.10
1.03 2.2±0.01ms 113.5 MElem/sec 1.00 2.1±0.00ms 117.0
MElem/sec
> dict_take_n/take_n/size_65536_card_300_null_0.10
1.02 2.3±0.00ms 107.1 MElem/sec 1.00 2.3±0.00ms 109.0
MElem/sec
> dict_take_n/take_n/size_65536_card_65536_null_0.10
1.00 25.5±0.07ms 9.8 MElem/sec 1.04 26.4±0.13ms 9.5
MElem/sec
> dict_take_n/take_n/size_65536_card_75_null_0.10
1.02 2.3±0.01ms 109.8 MElem/sec 1.00 2.2±0.00ms 112.4
MElem/sec
> ```
>
> @saadtajwar
Oof, if I'm reading this correctly this doesn't look like a huge win for us
here...any thoughts on how we want move forward here? I feel like we could
either:
- decide the current benchmarks capture the common use cases and don't
warrant these changes
- decide the current benchmarks capture the common use cases and still
warrant changes but maybe not the specific changes we have - we could look at
the actual change set and either modify or more likely remove some of these
optimizations
- decide to modify the benchmarks to try and better exercise the paths that
our current changes would optimize for (maybe increase the number of emit
cycles to be >>> 4/5, use more distinct leftovers, use more expensive values
with longer strings/Utf8View)
My preference would be the last option, but I'm obviously biased in wanting
to see these changes move the needle 😆 open to any feedback!
--
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]
