ARROW-15005: [C++] Improve csv parser with Neon

[cyb70289]

No description provided.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-15005

[cyb70289]

Neoverse N1, clang-12

-----------------------------------------------------------------
Non-regressions: (8)
-----------------------------------------------------------------
              benchmark        baseline       contender  change %
ParseCSVVehiclesExample 943.178 MiB/sec   1.411 GiB/sec    53.152
  ParseCSVStocksExample 817.026 MiB/sec 888.977 MiB/sec     8.806
    ParseCSVQuotedBlock 497.098 MiB/sec 514.735 MiB/sec     3.548
 ParseCSVFlightsExample 365.120 MiB/sec 370.772 MiB/sec     1.548
   ParseCSVEscapedBlock 451.707 MiB/sec 452.478 MiB/sec     0.171
ChunkCSVNoNewlinesBlock      193.220245      193.543142     0.167
    ChunkCSVQuotedBlock 720.860 MiB/sec 720.979 MiB/sec     0.017
   ChunkCSVEscapedBlock 774.219 MiB/sec 774.157 MiB/sec    -0.008

M1, AppleClang-12

------------------------------------------------------------------
Non-regressions: (8)
------------------------------------------------------------------
              benchmark         baseline       contender  change %
ParseCSVVehiclesExample 1006.916 MiB/sec   1.765 GiB/sec    79.447
  ParseCSVStocksExample    1.005 GiB/sec   1.242 GiB/sec    23.544
 ParseCSVFlightsExample  483.309 MiB/sec 484.854 MiB/sec     0.320
    ChunkCSVQuotedBlock    1.505 GiB/sec   1.508 GiB/sec     0.238
   ParseCSVEscapedBlock  527.498 MiB/sec 528.321 MiB/sec     0.156
    ParseCSVQuotedBlock  521.683 MiB/sec 522.305 MiB/sec     0.119
   ChunkCSVEscapedBlock    1.615 GiB/sec   1.614 GiB/sec    -0.021
ChunkCSVNoNewlinesBlock        83.988479       81.389943    -3.094

[pitrou]

Interesting!

[cyb70289]

New benchmark result after templating special options.

Neoverse N1, clang-12

------------------------------------------------------------------
Non-regressions: (8)
------------------------------------------------------------------
              benchmark        baseline        contender  change %
ParseCSVVehiclesExample 936.272 MiB/sec    1.695 GiB/sec    85.394
  ParseCSVStocksExample 832.577 MiB/sec 1010.333 MiB/sec    21.350
    ParseCSVQuotedBlock 433.361 MiB/sec  499.383 MiB/sec    15.235
 ParseCSVFlightsExample 344.455 MiB/sec  358.278 MiB/sec     4.013
   ParseCSVEscapedBlock 438.008 MiB/sec  447.634 MiB/sec     2.198
   ChunkCSVEscapedBlock 774.220 MiB/sec  774.256 MiB/sec     0.005
    ChunkCSVQuotedBlock 721.149 MiB/sec  720.960 MiB/sec    -0.026
ChunkCSVNoNewlinesBlock      192.677073       192.158918    -0.269

M1, AppleClang-12

------------------------------------------------------------------
Non-regressions: (8)
------------------------------------------------------------------
              benchmark         baseline       contender  change %
ParseCSVVehiclesExample 1011.496 MiB/sec   1.765 GiB/sec    78.723
  ParseCSVStocksExample    1.009 GiB/sec   1.243 GiB/sec    23.128
   ChunkCSVEscapedBlock    1.614 GiB/sec   1.615 GiB/sec     0.037
    ChunkCSVQuotedBlock    1.510 GiB/sec   1.510 GiB/sec     0.029
   ParseCSVEscapedBlock  528.560 MiB/sec 528.450 MiB/sec    -0.021
    ParseCSVQuotedBlock  523.418 MiB/sec 522.261 MiB/sec    -0.221
 ParseCSVFlightsExample  486.159 MiB/sec 484.643 MiB/sec    -0.312
ChunkCSVNoNewlinesBlock        83.910285       81.432684    -2.953

[pitrou]

That's a very nice improvement, thank you :-)

[ursabot]

Benchmark runs are scheduled for baseline = f0110cf and contender = 464ccde. 464ccde 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 ⬇️0.0% ⬆️0.9%] ursa-i9-9960x
[Finished ⬇️0.71% ⬆️0.0%] ursa-thinkcentre-m75q
Supported benchmarks:
ec2-t3-xlarge-us-east-2: Supported benchmark langs: Python. Runs only benchmarks with cloud = True
ursa-i9-9960x: Supported benchmark langs: Python, R, JavaScript
ursa-thinkcentre-m75q: Supported benchmark langs: C++, Java