On Sat, 18 Jan 2025 00:58:36 GMT, Shaojin Wen <s...@openjdk.org> wrote:
>> Sorry, I was just reading the comment and not how DIGITS is initialized and
>> used.
>>
>> The _correct_ comment should be something like
>>
>> * 97 -> '9' | ('7' << 8) -> 0x3739
>>
>> so the `short` value was correct before, but not the expression to construct
>> it.
>> Again, sorry for the confusion.
>
> It was my mistake. I made the change without checking it carefully. It has
> been fixed.
Thanks @wenshao.
I ran the benchmarks on a Apple M1 CPU. I can observe a performance increase
for the Latin1 cases but _not_ for the Utf16 cases, which is the same within
the error bounds.
before
StringBuilders.appendWithIntLatin1 N/A avgt 15 181.071 ± 2.392
ns/op
StringBuilders.appendWithIntUtf16 N/A avgt 15 160.507 ± 7.902
ns/op
StringBuilders.appendWithLongLatin1 N/A avgt 15 232.727 ± 2.879
ns/op
StringBuilders.appendWithLongUtf16 N/A avgt 15 222.765 ± 3.208
ns/op
after
StringBuilders.appendWithIntLatin1 N/A avgt 15 129.019 ± 0.097
ns/op
StringBuilders.appendWithIntUtf16 N/A avgt 15 166.260 ± 2.144
ns/op
StringBuilders.appendWithLongLatin1 N/A avgt 15 192.300 ± 1.520
ns/op
StringBuilders.appendWithLongUtf16 N/A avgt 15 219.962 ± 0.850
ns/op
Can you please rerun the benchmarks on your platforms and report here?
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/22023#discussion_r1922611115
