GH-15200: [C++] Created benchmarks for round kernels.

[EpsilonPrime]

The four existing kernel functions Ceil, Floor, Round, and Trunc gain benchmarks with this change.

• Closes: [C++][Arrow] Add benchmarks for the round functions. #15200

[github-actions]

• Closes: [C++][Arrow] Add benchmarks for the round functions. #15200

[github-actions]

⚠️ GitHub issue #15200 has been automatically assigned in GitHub to PR creator.

[github-actions]

⚠️ GitHub issue #15200 has no components, please add labels for components.

[EpsilonPrime]

One thought about the number of benchmarks -- there is a benchmark for every type and round mode (10 each). Some alternatives could be to try to fold the round mode tests into the type (for less visibility) or come up with a "less frequently used" benchmark concept if having less benchmarks is ideal.

[cyb70289]

One thought about the number of benchmarks -- there is a benchmark for every type and round mode (10 each). Some alternatives could be to try to fold the round mode tests into the type (for less visibility) or come up with a "less frequently used" benchmark concept if having less benchmarks is ideal.

It does looks too many benchmarks are generated. And I guess most of them are similar.
How long will all these cases run? If the round modes don't impact performance, I agree that we can only test one of them. And probably few data types, e.g., int64/uint64/double.

[cyb70289]

Do you plan to add benchmark for decimal types? It can be in a followup PR.

[EpsilonPrime]

Do you plan to add benchmark for decimal types? It can be in a followup PR.

Done.

[EpsilonPrime]

All of the tests together are about 3 minutes. I've selected 3 of the round modes to run normally and disabled the rest using a command line flag.

[EpsilonPrime]

I've updated the arithmetic benchmarks to also have labels. Thanks for the suggestion!

[cyb70289]

Travis CI error is not related.

[cyb70289]

Not related to this PR, but the benchmark result is not obvious to me.
Why is double faster than int64? I suppose integer rounding does nothing.

Ceil<Int64Type, RoundMode::DOWN>/size:1048576/inverse_null_proportion:0                      323253 ns       323251 ns        
2161 bytes_per_second=3.02107G/s items_per_second=405.481M/s null_percent=0 size=1048.58k

Ceil<DoubleType, RoundMode::DOWN>/size:1048576/inverse_null_proportion:0                      89834 ns        89832 ns         
7795 bytes_per_second=10.871G/s items_per_second=1.45908G/s null_percent=0 size=1048.58k

[ursabot]

Benchmark runs are scheduled for baseline = 33d677c and contender = 85b167c. 85b167c is a master commit associated with this PR. Results will be available as each benchmark for each run completes.
Conbench compare runs links:
[Finished ⬇️0.0% ⬆️0.0%] ec2-t3-xlarge-us-east-2
[Failed ⬇️0.33% ⬆️0.0%] test-mac-arm
[Finished ⬇️0.0% ⬆️0.0%] ursa-i9-9960x
[Finished ⬇️0.14% ⬆️0.06%] ursa-thinkcentre-m75q
Buildkite builds:
[Finished] 85b167c0 ec2-t3-xlarge-us-east-2
[Failed] 85b167c0 test-mac-arm
[Finished] 85b167c0 ursa-i9-9960x
[Finished] 85b167c0 ursa-thinkcentre-m75q
[Finished] 33d677c4 ec2-t3-xlarge-us-east-2
[Finished] 33d677c4 test-mac-arm
[Finished] 33d677c4 ursa-i9-9960x
[Finished] 33d677c4 ursa-thinkcentre-m75q
Supported benchmarks:
ec2-t3-xlarge-us-east-2: Supported benchmark langs: Python, R. Runs only benchmarks with cloud = True
test-mac-arm: Supported benchmark langs: C++, Python, R
ursa-i9-9960x: Supported benchmark langs: Python, R, JavaScript
ursa-thinkcentre-m75q: Supported benchmark langs: C++, Java