RFR: 8336856: Efficient hidden class-based string concatenation strategy [v55]

[Claes Redestad]

On Wed, 14 Aug 2024 09:45:24 GMT, Shaojin Wen <duke at openjdk.org> wrote:

> In the `StringConcat.concat123String` scenario, setting the inlineThreshold to 150 still shows a performance improvement, although this is crazy. But this can be used as a performance optimization parameter.

I'd suggest looking at JIT compilation overheads when doing any such tuning, though: `-XX:CompileCommand=MemStat,*::concat123*,print` for example will show that cost of compiling `concat123String` goes up by orders of magnitude, both in time and memory use. It's less extreme than for the MH-based implementation, but still I think it's a feature of the new implementation that we do less force inlining at the high end.

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

PR Comment: https://git.openjdk.org/jdk/pull/20273#issuecomment-2288342800