RFR: 8358696: Assert with extreme values for -XX:BciProfileWidth [v11]

Martin Doerr mdoerr at openjdk.org
Thu Aug 21 10:47:59 UTC 2025


On Wed, 20 Aug 2025 23:39:50 GMT, Dean Long <dlong at openjdk.org> wrote:

> If we wanted to decrease the code size, we could change the unrolled loop to a real loop. But I think first we should answer the question, why are we profiling "ret" instructions at all? As far as I can tell, the compilers are not using the profiling data for anything, so maybe we could just remove it.

Interesting point. I wonder if it makes sense to collect as much profiling data in the interpreter as we currently do. C1 tier 3 can still collect it.

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

PR Comment: https://git.openjdk.org/jdk/pull/26139#issuecomment-3210013114


More information about the hotspot-compiler-dev mailing list