On Thu, 12 Oct 2023 09:55:15 GMT, Shaojin Wen <d...@openjdk.org> wrote:
>> CharacterDataLatin1.digit is for multi-radix, and the performance when radix
>> = 10 is not good enough.
>
> If we reuse CharacterDataLatin1#DIGITS, the performance will be slower, the
> performance numbers are as follows
>
>
> class CharacterDataLatin1 {
> static int digit(int ch) {
> int value = DIGITS[ch & 0xFF];
> return (value >= 0 && value < 10) ? value : -1;
> }
> }
>
>
>
> Benchmark (size) Mode Cnt Score Error Units (baseline)
> Integers.parseInt 500 avgt 15 2.614 ? 0.003 us/op
>
> Benchmark (size) Mode Cnt Score Error Units (use
> CharacterDataLatin1#digit)
> Integers.parseInt 500 avgt 15 2.399 ? 0.008 us/op
>
> Benchmark (size) Mode Cnt Score Error Units (c3e878a)
> Integers.parseInt 500 avgt 15 2.251 ? 0.010 us/op
Maybe there's something wrong with the `value < radix` check that JIT cannot
efficiently optimize it (so `digit <= 9` has a similar performance to `value <
10` in loops). We might have to check the hotspot assembly to find out the
reason.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/16112#discussion_r1356591852
