RFR: 8354996: Reduce dynamic code generation for a single downcall [v3]

[Jorn Vernee]

On Mon, 28 Apr 2025 14:36:33 GMT, Chen Liang <liach at openjdk.org> wrote:

>> Perf numbers for simple main:
>> Linking of `Class::forName0` down from ~152 to ~83
>> 
>> For calling little color management system https://bugs.openjdk.org/browse/JDK-8313344:
>> JNI: ~45
>> baseline panama: ~164
>> patch: ~103
>> 
>> Also see #24705.
>
> Chen Liang has updated the pull request with a new target base due to a merge or a rebase. The incremental webrev excludes the unrelated changes brought in by the merge/rebase. The pull request contains five additional commits since the last revision:
> 
>  - Merge branch 'master' of https://github.com/openjdk/jdk into exp/ffm-lcms
>  - Use more simple function descriptor
>  - Simplify FFM linker warmup
>  - Merge branch 'master' of https://github.com/openjdk/jdk into exp/ffm-lcms
>  - Wip work trying to speed up FFM

That stack trace is showing that we're going through the `LinuxRISCV64CallArranger`, but I wonder if we are ending up in the `ForeignGlobals::parse_abi_descriptor` of another platform? That would result in a mismatch between the kind of ABIDescriptor which the Java code passes, and the one which the VM code expects.

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

PR Comment: https://git.openjdk.org/jdk/pull/24742#issuecomment-2841582780