Add benchmarks for FromIter (PrimitiveArray and BooleanArray)

[tobixdev]

Which issue does this PR close?

• Relates to Incorrect Behavior of Collecting a filtered iterator to a BooleanArray #8505 .

I want to see any performance regressions to BooleanArray::from_iter.

Rationale for this change

Add microbenchmarks for observing the performance of XYZArray::from_iter.

On my machine, executing the benchmarks back to back results in deviations within 1% .

Int64Array::from_iter   time:   [14.292 µs 14.297 µs 14.303 µs]
                        change: [-0.0049% +0.1290% +0.2631%] (p = 0.06 > 0.05)
                        No change in performance detected.
Found 26 outliers among 100 measurements (26.00%)
  1 (1.00%) low severe
  3 (3.00%) low mild
  9 (9.00%) high mild
  13 (13.00%) high severe

Int64Array::from_trusted_len_iter
                        time:   [6.7355 µs 6.7472 µs 6.7628 µs]
                        change: [+0.0215% +0.1868% +0.3739%] (p = 0.03 < 0.05)
                        Change within noise threshold.
Found 11 outliers among 100 measurements (11.00%)
  4 (4.00%) high mild
  7 (7.00%) high severe

BooleanArray::from_iter time:   [7.3389 µs 7.3596 µs 7.3861 µs]
                        change: [-1.3820% -0.8065% -0.2803%] (p = 0.00 < 0.05)
                        Change within noise threshold.
Found 16 outliers among 100 measurements (16.00%)
  9 (9.00%) high mild
  7 (7.00%) high severe

What changes are included in this PR?

Only benchmarks

Are these changes tested?

Functionality is tested in the implementation file.

Are there any user-facing changes?

None

[alamb]

Thanks for this @tobixdev -- looks great to me

[alamb]

this looks similar to the existing benchmark here: https://github.com/apache/arrow-rs/blob/31ea84453b2f7ed7aa4e85825bd6cbf7ecd45f3a/arrow/benches/buffer_create.rs#L179-L178

Perhaps you could extend that benchmark rather than adding a new one?

[tobixdev]

It certainly makes sense to add them to another benchmark. I wasn't aware of the suite at the top-level crate.

What do think about renaming array_from_vec to array_from and adding them there? I think this would also be related code but we wouldn't mix Array::from_xyz with Buffer::from_xyz. But I am also fine with adding it to buffer_create.

https://github.com/apache/arrow-rs/blob/31ea84453b2f7ed7aa4e85825bd6cbf7ecd45f3a/arrow/benches/array_from_vec.rs

[tobixdev]

I've update it now to use array_from. I can change it if you prefer buffer_create.

[alamb]

Thank you @tobixdev

[alamb]

👍