RFR (S) 8244510: Shenandoah: invert SHC2Support::is_in_cset condition

Aleksey Shipilev shade at redhat.com
Wed May 6 09:51:37 UTC 2020


RFE:
  https://bugs.openjdk.java.net/browse/JDK-8244510

Current code in SHC2Support::is_in_cset treats "in cset" branch as frequent. We should really invert
it after SFX barriers make that condition much less likely. It improves some benchmarks without any
observable regressions on others. Note I opted to invert the condition instead of changing the
probabilities, because I think it flows better: test_in_cset is UNLIKELY to be true now.

Webrev:
  https://cr.openjdk.java.net/~shade/8244510/webrev.01/

Testing: hotspot_gc_shenandoah, benchmarks

-- 
Thanks,
-Aleksey



More information about the hotspot-gc-dev mailing list