RFR: 8252752: Clear card table for old regions during scan in G1 [v2]

[Thomas Schatzl]

> Hi all,
> 
>   can I get reviews for this change that removes the need for explicit clearing of the card table for typically a large
>   amount of regions in most cases?
> 
> Currently g1 unconditionally marks scanned dirty cards as "scanned" to remember that they had already been scanned in
> earlier evacuation passes. Then in the "clear card table phase" g1 clears all these marks including the card table of
> evacuated regions.  This change modifies g1 so that if possible, it clears dirty cards after scanning them instead of
> setting them to "scanned" if there is only one evacuation pass. This saves a lot of work later (e.g. the amount of
> regions to clear in the clear card table phase may be reduced by a magnitude or more, depending on the ration between
> evacuated and scanned regions).  The main change here is which type of regions (ones in the collection set, ones that
> are only scanned) are put into which list of regions g1 already manages: there is one "all" list containing all regions
> that need clearing, and a "next" list containing the regions to be scanned in the current evacuation pass. Previously
> g1 put regions into at least one list; now regions in the current collection set are always directly put into the "all"
> list (as they independently of other reasons require card table cleaning), while to-be-scanned regions are always only
> put into the "next" list.  Then, to achieve the desired effect that only regions that absolutely require clearing are
> in the "all" list, g1 does not merge the "next" list into the "all" list when it detects that there will only be one
> evacuation pass.  Performance impact: Decreases clear card table time significantly if the number of regions in the
> collection set is much smaller than the number of scanned regions (as it decreases the number of regions to clear).
> Some cursory perf testing showed slight overall pause time improvements.  Testing: hs-tier1-5 ("Initial import" change)
> hs-tier1-2 (latest)

Thomas Schatzl has updated the pull request incrementally with one additional commit since the last revision:

  Comments by ayang, kbarrett, iwalulya

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

Changes:
  - all: https://git.openjdk.java.net/jdk/pull/343/files
  - new: https://git.openjdk.java.net/jdk/pull/343/files/260e5890..fd0d2835

Webrevs:
 - full: https://webrevs.openjdk.java.net/?repo=jdk&pr=343&range=01
 - incr: https://webrevs.openjdk.java.net/?repo=jdk&pr=343&range=00-01

  Stats: 27 lines in 5 files changed: 13 ins; 0 del; 14 mod
  Patch: https://git.openjdk.java.net/jdk/pull/343.diff
  Fetch: git fetch https://git.openjdk.java.net/jdk pull/343/head:pull/343

PR: https://git.openjdk.java.net/jdk/pull/343