-
Enhancement
-
Resolution: Won't Fix
-
P4
-
None
-
8-shenandoah, 11-shenandoah, 14, 15
Looking at init-mark root statistics and profiles, it seems we bottleneck on MarkBitMap CAS updates. Maybe we can just push the roots on the scan queue without marks, and then update MarkBitMap at very beginning of concurrent cycle.
For example: http://cr.openjdk.java.net/~shade/8245849/8245849-poc.patch
For example: http://cr.openjdk.java.net/~shade/8245849/8245849-poc.patch