ARROW-3321: [C++] Improve integer parsing performance

[pitrou]

Before:

---------------------------------------------------------------------
Benchmark                              Time           CPU Iterations
---------------------------------------------------------------------
BM_IntegerParsing<Int8Type>         1388 ns       1387 ns     502861   5.49911M items/s
BM_IntegerParsing<Int16Type>        1475 ns       1475 ns     468724   5.17179M items/s
BM_IntegerParsing<Int32Type>        1730 ns       1729 ns     405693   4.41194M items/s
BM_IntegerParsing<Int64Type>        2131 ns       2131 ns     328192   3.58034M items/s
BM_IntegerParsing<UInt8Type>        1238 ns       1238 ns     572573   6.16483M items/s
BM_IntegerParsing<UInt16Type>       1302 ns       1301 ns     537960   5.86206M items/s
BM_IntegerParsing<UInt32Type>       1391 ns       1391 ns     502859    5.4857M items/s
BM_IntegerParsing<UInt64Type>       1637 ns       1637 ns     427832     4.661M items/s
BM_FloatParsing<FloatType>          4437 ns       4436 ns     156887   1.71973M items/s
BM_FloatParsing<DoubleType>         4593 ns       4592 ns     152459   1.66129M items/s

After:

---------------------------------------------------------------------
Benchmark                              Time           CPU Iterations
---------------------------------------------------------------------
BM_IntegerParsing<Int8Type>           23 ns         23 ns   29800687   324.788M items/s
BM_IntegerParsing<Int16Type>          27 ns         27 ns   26593165   287.438M items/s
BM_IntegerParsing<Int32Type>          34 ns         34 ns   20689813   226.211M items/s
BM_IntegerParsing<Int64Type>          49 ns         49 ns   14256379   155.424M items/s
BM_IntegerParsing<UInt8Type>          17 ns         17 ns   42295211   454.911M items/s
BM_IntegerParsing<UInt16Type>         16 ns         16 ns   42663172   464.397M items/s
BM_IntegerParsing<UInt32Type>         21 ns         21 ns   33372432   363.209M items/s
BM_IntegerParsing<UInt64Type>         33 ns         33 ns   21502295   234.255M items/s
BM_FloatParsing<FloatType>          4554 ns       4553 ns     153207   1.67565M items/s
BM_FloatParsing<DoubleType>         4579 ns       4578 ns     152304   1.66651M items/s

[codecov-io]

Codecov Report

Merging #2619 into master will increase coverage by 1.06%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #2619      +/-   ##
==========================================
+ Coverage   87.15%   88.22%   +1.06%     
==========================================
  Files         380      318      -62     
  Lines       59037    55408    -3629     
==========================================
- Hits        51456    48883    -2573     
+ Misses       7507     6525     -982     
+ Partials       74        0      -74

Impacted Files	Coverage Δ
cpp/src/arrow/util/parsing-util-test.cc	99.33% <100%> (+0.11%)	⬆️
cpp/src/arrow/util/parsing.h	100% <100%> (ø)	⬆️
rust/src/record_batch.rs
go/arrow/datatype_nested.go
rust/src/util/bit_util.rs
go/arrow/math/uint64_amd64.go
go/arrow/internal/testing/tools/bool.go
go/arrow/internal/bitutil/bitutil.go
go/arrow/memory/memory_avx2_amd64.go
... and 55 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 433a898...9834b28. Read the comment docs.

[kszucs]

That's an impressive speedup!

[wesm]

+1, very nice! If you ever want to write a blog post illustrating your work on this (how does it compare with parsers in other performance-sensitive projects, like RapidJSON?), please go ahead

[wesm]

At some point do you think we might want to return a "reason" as an out argument when the parser returns false? Could be follow up work

[pitrou]

Not sure, because if we're calling out to an external parser (as we'll keep doing for floating-point parsing), it may not give us a reason.