On Tue, 23 Jul 2024 13:42:20 GMT, Shaojin Wen <d...@openjdk.org> wrote:
>> src/java.base/share/classes/java/lang/StringLatin1.java line 832:
>>
>>> 830: static void putCharsAt(byte[] val, int index, int c1, int c2, int
>>> c3, int c4) {
>>> 831: assert index >= 0 && index + 3 < length(val) : "Trusted caller
>>> missed bounds check";
>>> 832: // Don't use the putChar method, Its instrinsic will cause C2
>>> unable to combining values into larger stores.
>>
>> `StringLatin1` has no `putChar` method. I think you can just convert this to
>> explicit array stores for the merge store optimization.
>
> explicit array stores has the overhead of boundary checking. If putCharsAt
> of StringLatin1 is not implemented based on Unsafe, the performance will be
> worse than StringUTF16.
>
> Of course, this is a common problem. StringUTF16.putChar is equivalent to
> Unsafe.putChar, without boundary checking. I found in many test scenarios
> that the UTF16 version performs better than the StringLatin1 version.
>
> We may need to change some StringLatin1 related implementations to use
> Unsafe, otherwise users will turn off COMPACT_STRINGS to improve performance.
Thank you for telling what the blocking issue is here. Does C2 not merge the
bound checks when it does the merge stores? Interesting, and I think a fix from
their side should be the way to go.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/19626#discussion_r1688102742
