GH-36059: [C++][Compute] Reserve space for hashtable for scalar lookup functions

[js8544]

Rationale for this change

Performance improvement for scalar lookup functions.

What changes are included in this PR?

1. Reserve space for hashtable for scalar lookup functions before the insertion process.
2. A benchmark that demonstrates the improvement. Data is pasted in the comments.

Are these changes tested?

By existing unit tests.

Are there any user-facing changes?

No.

• Closes: [C++] Performance of building up HashTable (MemoTable) in is_in kernel #36059

[github-actions]

⚠️ GitHub issue #36059has been automatically assigned in GitHub to PR creator.

[js8544]

Benchmark result:
Before:

----------------------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
----------------------------------------------------------------------------------------
IndexInInt32LargeSet/100 2883 ns 2883 ns 246728 bytes_per_second=13.2302M/s items_per_second=3.46823M/s
IndexInInt32LargeSet/1000 10322 ns 10322 ns 63804 bytes_per_second=3.6958M/s items_per_second=968.833k/s
IndexInInt32LargeSet/10000 139318 ns 139316 ns 4987 bytes_per_second=280.389k/s items_per_second=71.7795k/s
IndexInInt32LargeSet/100000 4588807 ns 4588452 ns 165 bytes_per_second=8.51322k/s items_per_second=2.17938k/s
IndexInInt32LargeSet/1000000 90538493 ns 90536500 ns 10 bytes_per_second=441.811/s items_per_second=110.453/s
IndexInInt32LargeSet/10000000 1428412088 ns 1428369286 ns 1 bytes_per_second=28.004/s items_per_second=7.00099/s
IsInInt32LargeSet/100 2833 ns 2833 ns 234016 bytes_per_second=13.4639M/s items_per_second=3.52948M/s
IsInInt32LargeSet/1000 10376 ns 10376 ns 67001 bytes_per_second=3.6765M/s items_per_second=963.773k/s
IsInInt32LargeSet/10000 148294 ns 148284 ns 4769 bytes_per_second=263.43k/s items_per_second=67.4381k/s
IsInInt32LargeSet/100000 4463505 ns 4463293 ns 147 bytes_per_second=8.75195k/s items_per_second=2.2405k/s
IsInInt32LargeSet/1000000 75740587 ns 75738687 ns 10 bytes_per_second=528.132/s items_per_second=132.033/s
IsInInt32LargeSet/10000000 1317214434 ns 1317158744 ns 1 bytes_per_second=30.3684/s items_per_second=7.5921/s

After

----------------------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
----------------------------------------------------------------------------------------
IndexInInt32LargeSet/100 2659 ns 2658 ns 266800 bytes_per_second=14.3493M/s items_per_second=3.76159M/s
IndexInInt32LargeSet/1000 9759 ns 9759 ns 70169 bytes_per_second=3.9091M/s items_per_second=1024.75k/s
IndexInInt32LargeSet/10000 93311 ns 93309 ns 7477 bytes_per_second=418.634k/s items_per_second=107.17k/s
IndexInInt32LargeSet/100000 1729160 ns 1729162 ns 453 bytes_per_second=22.5904k/s items_per_second=5.78315k/s
IndexInInt32LargeSet/1000000 79806851 ns 79805041 ns 9 bytes_per_second=501.221/s items_per_second=125.305/s
IndexInInt32LargeSet/10000000 1112193163 ns 1112056579 ns 1 bytes_per_second=35.9694/s items_per_second=8.99235/s
IsInInt32LargeSet/100 2518 ns 2518 ns 280148 bytes_per_second=15.1526M/s items_per_second=3.97215M/s
IsInInt32LargeSet/1000 9625 ns 9624 ns 72600 bytes_per_second=3.96356M/s items_per_second=1039.02k/s
IsInInt32LargeSet/10000 93759 ns 93754 ns 7256 bytes_per_second=416.647k/s items_per_second=106.662k/s
IsInInt32LargeSet/100000 1608221 ns 1608152 ns 495 bytes_per_second=24.2903k/s items_per_second=6.21832k/s
IsInInt32LargeSet/1000000 73113133 ns 73112833 ns 9 bytes_per_second=547.1/s items_per_second=136.775/s
IsInInt32LargeSet/10000000 1207785785 ns 1207731078 ns 1 bytes_per_second=33.12/s items_per_second=8.27999/s

[westonpace]

Just one question, otherwise this makes sense to me.

[conbench-apache-arrow]

Conbench analyzed the 6 benchmark runs on commit 2ce4a38f.

There were 9 benchmark results indicating a performance regression:

• Commit Run on ursa-thinkcentre-m75q at 2023-06-25 05:44:29Z
  • params=<Add, DoubleType>/size:524288/inverse_null_proportion:100, source=cpp-micro, suite=arrow-compute-scalar-arithmetic-benchmark
  • params=<Multiply, Int8Type>/size:524288/inverse_null_proportion:100, source=cpp-micro, suite=arrow-compute-scalar-arithmetic-benchmark
• and 7 more (see the report linked below)

The full Conbench report has more details.