fix(a2a3/tensormap): propagate early-dispatch from the sync_start drain path

[ChaoWao]

Summary

propagate_dispatch_fanin — the event that turns a flagged producer's
dispatch into early-dispatch candidacy for its consumers — was only called on
the normal ready-pop dispatch path (scheduler_dispatch.cpp:457). A
sync_start task dispatches through the separate drain path
(handle_drain_mode → drain_worker_dispatch in scheduler_completion.cpp),
which never propagated.

Consequence: any consumer whose only flagged producer is a sync_start
(drain-dispatched) task never became an early-dispatch candidate — it wasn't
gated and it had spare (idle/pending) cores to stage onto; the candidate-enqueue
event itself was simply missing.

Fix

After drain_worker_dispatch publishes all of the drained producer's blocks
(the producer is now dispatched), call propagate_dispatch_fanin once —
mirroring the normal path's post-publish propagate.

• Idempotent via propagate's dispatch_propagated once-guard.
• The internal gate no-ops for an unflagged producer, so calling it
  unconditionally is cheap (same as the normal path).
• Scoped exactly to the gap: only sync_start tasks reach the drain path
  (scheduler_dispatch.cpp re-queues a non-sync_start task that doesn't fit, so
  it propagates via the normal path on eventual dispatch).

Testing

• a2a3 hardware — qwen3-14B decode_fwd (2-layer) early-dispatched-task
  count 35 → 83 with the fix. The +48 are exactly the 24 out_proj
  tiles/layer whose sole producer is the sync_start fa_fused — previously
  stuck out of early-dispatch, now pre-staged onto pending slots while fa_fused
  runs.
• spmd_sync_start / _aiv / _edge / _stress scene tests pass (no drain
  regression).
• cpput WiringTest 19/19 pass.

Testing checklist

• Simulation / cpput tests pass
• Hardware (a2a3) tests pass

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 227c916c-e017-49ba-b682-50703666eaf1

📥 Commits

Reviewing files that changed from the base of the PR and between a5895d8 and b5d3f3f.

📒 Files selected for processing (1)

• src/a2a3/runtime/tensormap_and_ringbuffer/runtime/scheduler/scheduler_completion.cpp

---

📝 Walkthrough

Walkthrough

This change adds a call to propagate_dispatch_fanin within SchedulerContext::drain_worker_dispatch during sync_start dispatch, so consumer dispatch_fanin state updates consistently, matching behavior already present in the normal scheduler dispatch path.

Changes

Drain Dispatch Fanin Propagation

Layer / File(s)	Summary
Propagate dispatch_fanin on sync_start drain src/a2a3/runtime/tensormap_and_ringbuffer/runtime/scheduler/scheduler_completion.cpp	Adds explicit sched_->propagate_dispatch_fanin(*slot_state) call during sync_start drain dispatch so consumers' dispatch_fanin is updated as in the normal dispatch path.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Possibly related PRs

• hw-native-sys/simpler#1285: Both PRs add/ensure propagate_dispatch_fanin calls for flagged producers to update early-dispatch eligibility.
• hw-native-sys/simpler#1292: Both PRs address correctness of consumers' dispatch-fanin state in the early-dispatch code path.

Poem

A drain, a hop, a fanin cheer,
Consumers now know producers are near!
Sync_start whispers down the line,
"Your turn is coming, right on time!" 🐇
Thump-thump, the scheduler's clear.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the fix to propagate early-dispatch from the sync_start drain path.
Description check	✅ Passed	The description accurately explains the same fix and related validation.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.

---

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

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

[gemini-code-assist]

Code Review

This pull request updates the drain_worker_dispatch function in scheduler_completion.cpp to call sched_->propagate_dispatch_fanin(*slot_state). This ensures that the drain path correctly increments its consumers' dispatch fan-in, allowing consumers whose only flagged producer is a sync_start task to become early-dispatch candidates. There are no review comments, so I have no feedback to provide.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.