RFR: 8339290: Optimize ClassFile Utf8EntryImpl#writeTo [v10]

[Claes Redestad]

On Tue, 3 Sep 2024 00:38:40 GMT, Shaojin Wen <swen at openjdk.org> wrote:

>> Use fast path for ascii characters 1 to 127 to improve the performance of writing Utf8Entry to BufferWriter.
>
> Shaojin Wen has updated the pull request incrementally with two additional commits since the last revision:
> 
>  - code style
>  - remove unsafe

I think this is looking much better now, thanks! There's some polish I'd like to see made, though.

src/java.base/share/classes/java/lang/System.java line 2597:

> 2595:             }
> 2596: 
> 2597:             public byte stringCoder(String s) {

I think it'd be better to do `boolean isLatin1(String s)` here. It's still somewhat leaky, but keeps things a bit more high-level, containing more of the low-level details. Seems all uses basically just checks `coder == 0` anyhow.

src/java.base/share/classes/jdk/internal/classfile/impl/BufWriterImpl.java line 182:

> 180:         byte[] elems = this.elems;
> 181: 
> 182:         str.getBytes(0, countGreaterThanZero, elems, offset);

Might be beneficial to non-latin1 case to move this up into the latin1 block.

Could you verify the 256 limit is useful? Add a couple of cases to the micro with much longer latin1 strings; both ASCII and with non-ASCII at the end. Would be nice if we can keep this simple and the "too long" check feels a bit premature.

src/java.base/share/classes/jdk/internal/classfile/impl/BufWriterImpl.java line 201:

> 199:         }
> 200:         int utflen = offset - start - 2;
> 201:         if (utflen > 65535) {

Previously we'd count bytes and throw before writing to the byte array. Perhaps this is still preferable, even if it comes at a cost.

test/micro/org/openjdk/bench/java/lang/classfile/Utf8EntryWriteTo.java line 76:

> 74:         byte[] bytes = HexFormat.of().parseHex(
> 75:                 switch (charType) {
> 76:                     case "ascii"        -> "78";

Might be good to randomize these inputs a bit.

-------------

Changes requested by redestad (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/20772#pullrequestreview-2276820291
PR Review Comment: https://git.openjdk.org/jdk/pull/20772#discussion_r1741719346
PR Review Comment: https://git.openjdk.org/jdk/pull/20772#discussion_r1741713216
PR Review Comment: https://git.openjdk.org/jdk/pull/20772#discussion_r1741717901
PR Review Comment: https://git.openjdk.org/jdk/pull/20772#discussion_r1741688429