New VectorAPI benchmark to accelerate JsonParsing through fast table lookup using Vector.selectFrom API

[jatin-bhateja]

Benchmark performs fast table lookup to match following JSON structural code points to accelerate JSON parsing scheme described in [1] using two vector selectFrom API[2] added in JDK-24[3]
, -> 0x2c
: -> 0x3a
[ -> 0x5b
] -> 0x5d
{ -> 0x7b
} -> 0x7d
tab, line feed, carriage return -> 0x9, 0xa, 0xd
space -> 0x20

Following are the perfromance numbers on Ryzen 7 7840HS - AVX512 system.

Benchmark                                     (size)   Mode  Cnt      Score   Error   Units
FastJSONCodePointMatchBenchmark.match_avx2      1024  thrpt    2   3383.359          ops/ms
FastJSONCodePointMatchBenchmark.match_avx512    1024  thrpt    2  60911.407          ops/ms
FastJSONCodePointMatchBenchmark.match_scalar    1024  thrpt    2   1445.365          ops/ms

[1] Parsing Gigabytes of JSON per Second : https://arxiv.org/abs/1902.08318
[2] Vector.selectFrom
[3] JEP 489: Vector API (Ninth Incubator): https://openjdk.org/jeps/489

[ionutbalosin]

Could you please use @BenchmarkMode(Mode.AverageTime)?

The main reason is that after all the benchmarks run, I have some additional logic that calculates the geometric mean, which relies on AverageTime and keeps it consistent across all benchmarks.

E.g., https://github.com/ionutbalosin/jvm-performance-benchmarks/blob/main/scripts/ggplot2/geomean-utils.r#L69

[jatin-bhateja]

Done

Benchmark                                     (size)  Mode  Cnt    Score   Error  Units
FastJSONCodePointMatchBenchmark.match_avx2      1024  avgt    2  303.377          ns/op
FastJSONCodePointMatchBenchmark.match_avx512    1024  avgt    2   15.549          ns/op
FastJSONCodePointMatchBenchmark.match_scalar    1024  avgt    2  694.167          ns/op

[jatin-bhateja]

Thanks @chriswhocodes and @ionutbalosin , kindly merge this PR.