RFR: 8323791: Parallel: Using non-atomic for live bytes update during Full GC

Albert Mingkun Yang ayang at openjdk.org
Tue Jan 16 10:45:41 UTC 2024


Simple optimization to avoid atomic-operation during Full GC marking.

Test: Full GC pause length shows ~1% regresion when using single gc-thread and ~20% improvement when using multiple gc-threads using the attached program.

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

Commit messages:
 - pgc-fullgc-livebyte

Changes: https://git.openjdk.org/jdk/pull/17442/files
 Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=17442&range=00
  Issue: https://bugs.openjdk.org/browse/JDK-8323791
  Stats: 71 lines in 3 files changed: 67 ins; 3 del; 1 mod
  Patch: https://git.openjdk.org/jdk/pull/17442.diff
  Fetch: git fetch https://git.openjdk.org/jdk.git pull/17442/head:pull/17442

PR: https://git.openjdk.org/jdk/pull/17442


More information about the hotspot-gc-dev mailing list