[yifjiang]

Summary

GpuDirect stays distinct from native Device/G1. Preserve salted roots and GPU MM exclusions across native clears; never credit retained rows as execution hits.

Review slice D04 of #16748; part of #16747 and #16746. This draft targets codex/kv-reuse-d03-planner-parser for a complete incremental review diff of 459 additions plus deletions, including tests, documentation and extraction glue.

Extraction details:

• GPU capability metadata derive/docs move to D09 with their type. Enum readers and mocker/consolidator cases stay atomic.

Reuse remains experimental and default off. Retained inventory is advisory, complete compatible rank parts must precede readiness, and native operations retain ownership until quiescence. Supported GPU activation also requires the companion KVCR/native slices and the declared deployment profile.

Validation

All configured pre-commit hooks passed for this exact slice on allocated SLURM. Stack-level CPU regression coverage is summarized on D31; each draft still requires its own applicable CI in the Checks tab. Historical combined-head and hardware results retain their original source scope.

[devin-ai-integration]

✅ Dynamo PR CI passed — run 41941613819 (attempt 2) on 969eaf638a

Gate checks: ✅ backend-status-check · ✅ deploy-status-check · ✅ dynamo-status-check

_{Posted automatically by Devin for run 41941613819. Updated on every full-CI run of this PR.}

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[yifjiang]

/ok to test eb4033e

[yifjiang]

/ok to test 969eaf6