RFR: 8372701: Randomized profile counters [v5]

[Tobias Hartmann]

On Mon, 15 Dec 2025 15:49:57 GMT, Andrew Haley <aph at openjdk.org> wrote:

> I submitted benchmarks and will report back once it finished.

Unfortunately, I see various regressions. Here are the most severe ones:

Benchmark, Linux aarch64, Linux x64, MacOSX aarch64, Windows x64
DaCapo-jython-large, -23.96%, -12.88%, -12.12%, -11.40%
SPECjbb2005-G1, -2.13%, -9.80%, -1.60%, -0.78%
Crypto-AESBench_decrypt, -4.86%, -6.48%, -3.61%, -8.33
Crypto-AESGCMBench.decrypt, -2.19%, -2.94%, -0.55%, -3.68%
Crypto-AESGCMBench.encrypt, -2.79%, -2.67%, -1.60%, -1.82%


I currently don't have time to look into this but hopefully you can reproduce on your end.

> Can you please run benchmarks with -XX:+UnlockExperimentalVMOptions -XX:ProfileCaptureRatio=64? Thanks.

Missed that, let me add a run with these settings.

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

PR Comment: https://git.openjdk.org/jdk/pull/28541#issuecomment-3660054383
PR Comment: https://git.openjdk.org/jdk/pull/28541#issuecomment-3660056989