RFR: 8317007: Add bulk removal of dead nmethods during class unloading [v3]

Thomas Schatzl tschatzl at openjdk.org
Thu Dec 7 07:39:34 UTC 2023


On Wed, 6 Dec 2023 17:59:02 GMT, Albert Mingkun Yang <ayang at openjdk.org> wrote:

> How much perf impact does "bulk removal" introduce? (If I understand this PR correctly, this IS an optimization after all.)

On G1, bulk removal reduces time spent in unregistering nmethods from 600ms to 30ms (and only that bad because in the test the code roots are very unbalanced and distribution is per-region) for a stress test, i.e. a 20x improvement. Sorry for not mentioning (again).

I think this is sufficient enough to warrant improvement.

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

PR Comment: https://git.openjdk.org/jdk/pull/16767#issuecomment-1844819222


More information about the hotspot-gc-dev mailing list