Repository navigation
Conversation
Count each cached fused-PQ source object and its code-buffer storage. Sum InlineSource byte estimates directly instead of multiplying them by the integer element width. Keep compressor ownership separate. Add PQ-only regression tests for source storage overhead and byte units through an actual serialized/loaded fused graph. Both tests fail on the unmodified main branch and pass with this fix. Validation: JDK 22, 65 tests each under scalar and Panama profiles; new tests also pass with compressed and uncompressed object references.
tlwillke
requested review from
MarkWolters,
ashkrisk and
jshook
as code owners
November 2, 2026 21:08
Contributor
|
Before you submit for review:
If you did not complete any of these, then please explain below. |
Handle independent reference/class-pointer settings and compact headers. Use VM field and array offsets; include full byte-sequence, segment, scope, and backing-array storage.
Count graph metadata, allocated cache capacity, adjacency arrays, and fused source storage once. Exclude separately owned models, worker scratch, and shared JVM metadata. Add independent instrumentation checks across JDK 22 and 25 layouts and document the reporting cost.
tlwillke
changed the title
Skip the compressed-class-pointer/wide-reference combination unavailable before JDK 15. Preserve all eight JDK 22 and twelve JDK 25 layouts; verify the six supported JDK 11 layouts.
Collaborator
Author
|
Replaced repeated object walks with cached totals. Graph payloads are counted while loading; PQ totals update only when chunk storage grows. Reporting performs a field read, and native buffer reporting creates no temporary Validation: 74 tests pass per provider on JDK 22/25, with 60 independent JVM-layout checks; final JDK 11/22/25 checks also pass. Repeated native exercises measured zero allocation across 180 million steady-state reads and low single-digit-nanosecond medians. The warmed metadata-sizing exercise used about 1 KB of temporary storage; cold loading was measured separately. Extra retained state is per graph/collection, not per vector. The documentation and standard release note now describe cached reporting and read-only loaded feature metadata. Fresh CI is running on the pushed revision. |
Count on-disk cache payloads while loading and size only bounded header metadata. Cache built-in PQ totals at chunk/reference-array growth, preserve shared immutable backing storage, and reuse native buffer fields for allocation-free size reporting.
Check independent JVM footprints, concurrent chunk growth, capacity reuse, and zero-allocation reporting. Add a reproducible timing/allocation exercise and document immutable cache ownership and initialization cost. Update the 4.0.3 bug-fix release note.
tlwillke
added a commit
that referenced
this pull request
Nov 3, 2026
Count retained ASH objects, backing arrays, headers, model matrices, and initialized packed/query/symmetric representations. Cache immutable totals and account lazy block storage at allocation; polling does not scan vectors or allocate. Align individual BlockRealMatrix tail arrays. Keep common graph, PQ, and buffer accounting changes in PR #781. Validated 90 ASH/fused tests per provider on JDK 22 and JDK 25 (native, Panama, scalar), plus 34 native module tests per JDK. Four native layout checks pass at 8/16-byte alignment; allocation regression checks verify zero-allocation polling. Tests and documentation describe ownership boundaries and conservative caller-alias/map estimates.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Correct fused-PQ retained-memory estimates, including source/buffer wrappers, native heap scopes, cache capacity, adjacency arrays, and JVM layout/alignment. Shared storage is counted once.
Graph totals are accumulated during loading, with only bounded header metadata sizing. Built-in PQ totals update under the existing chunk-allocation lock. Reporting is a cached field read; native buffer reporting also avoids temporary
Optionalobjects. Loaded feature metadata is read-only, and internal source caches are read-only by contract. Model/workspace ownership boundaries are documented.Validation: 74 regression tests pass per scalar/Panama/native configuration on JDK 22 and 25; independent measurements cover 60 provider/layout combinations. Final lifecycle/scoring/allocation checks also pass on JDK 11, 22, and 25. Repeated native exercises on fixed CPU/NUMA placement measured zero steady-state allocation and low single-digit-nanosecond reporting. Cold loading and one-time metadata sizing are measured separately. Includes the standard 4.0.3 bug-fix release note.