Near-term action items:

• DONE: OMWorld (external object to ObjectMonitor mapping): Combine LM_PLACEHOLDER into LM_LIGHTWEIGHT
  • The intention is that the OMWorld locking implementation will replace the Lightweigt locking. It will either go in with Lilliput or extracted into a separate, stand-alone RFE
  • Note: this means that the OMWorld implementation will still be used when the user runs with `-XX:-UseCompactObjectHeaders`
  • Owner: Axel
• DONE: Create a "patch queue" of all changes and maintain those patches as the top-most changes in https://github.com/openjdk/lilliput/tree/master
  • This will help
  • The "patch queue" will be rebased whenever the upstream (openjdk/jdk) is merged.
    • This will use non-destructive rebasing with "git merge -s ours", which "forgets" the old "patch queue"
  • Example patch queue:

    • * 4363dd8a1ef (HEAD -> lilliput-all, lilliput-stefank/lilliput-all) OMWorld Lilliput changes
      * 916d0958a7f JDK-8325104: Lilliput: Shrink Classpointers
      * 1346db1b546 Lilliput: Compact object headers
      * 8fb10dc6f1d 8305898: Alternative self-forwarding mechanism
      * 3726027944e 8305896: Alternative full GC forwarding
      * c1f2dfb8735 OMWorld
      * 998d0baab0f (tag: jdk-23+12, lilliput/jdk) 8324799: Use correct extension for C++ test headers

  • Owner: Roman

Workable items:

 + Lilliput Repository
 | <= + OMWorld Lilliput changes (Make UseCompactObjectHeaders imply LM_PLACEHOLDER)
 |    | <= Depends on OMWorld
 |    | <= Roman has initial patch
 |    | <= Stefan has a set on top of this
 |    | <= Maybe fold this into 'Lilliput: Compact object headers'
 |
 | <= + JDK-8325104: Lilliput: Shrink Classpointers (TinyCP)
 |    | <= Determine if 22-bit klass IDs are good enough
 |    |    | <= For Experimental UseCompactObjectHeaders
 |    |    | <= For productized UseCompactObjectHeaders
 |    | <= Maybe: Figure out if UseCompressedClassPointer should == UseCompactObjectHeaders
 |    | <= Figure out of hyperalignment is causing cache performance issues in anything other than microbenchmarks
 |    | <= Review CDS parts: Runtime
 |    | <= Review Metaspace parts: GC / Runtime
 |    | <= Review C2 parts. Esp. klass_offset_in_bytes and related discussion
 |
 | <= + Cleanups after 'Lilliput: Compact object headers'
 |    | <= Go over the GC code and clean out "safe" klass fetches.
 |
 | <= + Lilliput: Compact object headers
 |    | Review
 |
 | <= + 8305898: Alternative self-forwarding mechanism
 |    | <= Clean out preserved_marks for 32-bit JVMs
 |    | <= Review
 |
 | <= + 8305896: Alternative full GC forwarding
 |    | Review
 |
 | <= + OMWorld
 |    | <= + Monitor deflation heuristic
 |    |    | <= Rewrite
 |    |    | <= Interaction with ConcurrentHashTable sizing 
 |    | <= + Decide C2 monitor cache layout
 |    |    | <= should be loop unrolled or not
 |    |    | <= size?
 |    | <= Cleanup code used for various experiments (after evaluation)
 |    | <= + Full platform support?
 |    |    | <= Make non-primary platform call into Runtime until porters have added full support
 |    |    | <= PPC (optional)
 |    |    | <= RiscV (optional)
 |    |    | <= s390 (optional)
 |    |    | <= 32-bit arm (optional)
 |    |    | <= x86 (optional)

 + Testing (after "patch queue" has been created)
 | <= + Stability testing
 |
 | <= + Performance testing