Details
-
Enhancement
-
Resolution: Unresolved
-
P4
-
15
Description
Anecdotal evidence from an old discussion with [~tonyp] and a recent paper [1] suggest that it is advantageous to filter pre-values to be put in the SATB queues based on whether they have already been marked.
Existing overhead for the SATB barrier is already very low (~2% [2]), but maybe it is worth reducing it further.
Investigate this.
[1] Zhao, Blackburn, Deconstructing the Garbage-First Collector, in VEE ’20, March 17, 2020, Lausanne, Switzerland., 2020. http://users.cecs.anu.edu.au/~steveb/pubs/papers/g1-vee-2020.pdf
[2] Shipilev, Shenandoah GC - Version 2.0 (2019): The Great Revolution, JUG Berlin-Brandenburg, Sep 2019. https://shipilev.net/talks/jugbb-Sep2019-shenandoah.pdf
Existing overhead for the SATB barrier is already very low (~2% [2]), but maybe it is worth reducing it further.
Investigate this.
[1] Zhao, Blackburn, Deconstructing the Garbage-First Collector, in VEE ’20, March 17, 2020, Lausanne, Switzerland., 2020. http://users.cecs.anu.edu.au/~steveb/pubs/papers/g1-vee-2020.pdf
[2] Shipilev, Shenandoah GC - Version 2.0 (2019): The Great Revolution, JUG Berlin-Brandenburg, Sep 2019. https://shipilev.net/talks/jugbb-Sep2019-shenandoah.pdf