Fix: memset TMR payload/slab buffers before orch starts

[yanghaoran29]

Summary

• Move the cold first-touch memset of live payload_per_core_ / deferred_slab_per_core_ rows into TMR pre_handshake_init (A5 + a2a3), still before peers observe hs_setup_done_, so Orch can start without paying round-1 page-fault cost inside the Orch→Sched window.
• assign_own_clusters still rewires fields on already-resident pages; deinit comments updated to match.

Test plan

• Onboard A5: rebuild with the change, run TMR vector_example for 3 rounds under task-submit with device-wall STRACE enabled; compare against a clean main build of the same case.
• Acceptance: round-1 SchedWindow.start − OrchWindow.start should collapse toward round-2+ levels (cold cost moves into setup before orch, not the Orch→Sched gap).
• Observed on A5 with --rounds 3 (clean main vs this change): round-1 gap ~256 µs → ~0.6 µs; round-2+ gap stays on the order of a few µs (warm path; small flips vs Sched post-handshake race are expected once pages are resident).
• a2a3: code mirrored only; no onboard validation on this change.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Advanced

Run ID: c4d68333-19d8-4ce7-89d9-fde32c8c7251

📥 Commits

Reviewing files that changed from the base of the PR and between 7bcdc95 and 5b3ef1d.

📒 Files selected for processing (2)

• src/a2a3/runtime/tensormap_and_ringbuffer/runtime/scheduler/scheduler_cold_path.cpp
• src/a5/runtime/tensormap_and_ringbuffer/runtime/scheduler/scheduler_cold_path.cpp

Included review availability: Your plan provides up to 1 included review per hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

The A5 and a2a3 scheduler paths now first-touch and zero live per-core dispatch buffers during pre_handshake_init before handshake setup is published. The deinit() comments document this lifecycle.

Changes

Scheduler buffer initialization

Layer / File(s)	Summary
Pre-handshake buffer zeroing src/a2a3/.../scheduler/scheduler_cold_path.cpp, src/a5/.../scheduler/scheduler_cold_path.cpp	Both scheduler paths include <cstring>, zero live payload and deferred-slab rows before publishing hs_setup_done_, and update the deinit() comments to describe the new initialization lifecycle.

Priority: ⬇️ Low — Defer the scheduler first-touch optimization because it is a narrow cold-path performance change with no supplied customer or external urgency evidence.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to 5b3ef

This change moves first-touch buffer initialization ahead of scheduler handoff to remove cold-start latency without changing the later dispatch lifecycle. No merge-blocking risk is currently identified.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 50.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 4 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description check	✅ Passed	The description clearly explains the relocation of the TMR buffer initialization, the affected paths, and the validation results. It is directly related to the changeset.
Title check	✅ Passed	The title clearly summarizes the main change: zeroing the TMR payload and slab buffers before the orchestrator starts.

• Fix all pre-merge checks with AI

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{I twitch my nose at buffers bright
Before the handshake wakes to light
Each core row is touched and clear
No tiny fault shall hop near
The scheduler starts with cheer}

_{Comment @coderabbitai help to get the list of available commands.}