fix(a2a3/tensormap): flag alloc-tensor creators so they don't block early-dispatch

[ChaoWao]

Follow-up to #1285 (direct-only early-dispatch).

Problem

Under the direct-only model, an unflagged producer disqualifies its consumer
from early-dispatch, and a pre-completed producer only seeds dispatch_fanin
when flagged.

alloc_tensors builds a hidden task that completes INLINE at submit
(pto_orchestrator.cpp), so it is a pre-completed producer — and it carried no
allow_early_resolve flag (defaulted false). Every consumer that reads an
alloc'd buffer therefore had an unflagged pre-completed producer, which
disqualified it from early-dispatch entirely. Because alloc'd buffers are
pervasive inputs, this silently defeats early-dispatch for most real graphs.

A buffer allocation is pure memory whose output is ready at creation — it must
be transparent, never a barrier.

Fix

Flag the creator unconditionally (allow_early_resolve = true) on the
inline-complete path. Unlike a codegen task there is no Arg-driven hint to
honor: an alloc creator should never suppress its consumers' early-dispatch.

This keeps the "unflagged = suppress" invariant clean and deterministic
— a consumer early-dispatches iff every producer is flagged, independent of
which producers happened to race to completion before wiring. Intentional
barriers still use a dummy task, which is scheduled (not inline-completed)
and can be left unflagged to suppress.

Testing

• WiringTest.FlaggedPrecompletedCreatorTransparentToEarlyDispatch: a flagged
  pre-completed creator + a flagged pending compute producer — the creator is
  seeded (transparent) and the consumer reaches the candidate once the compute
  producer dispatches. cpput WiringTest 19/19 pass.
• a2a3 tensormap_and_ringbuffer onboard smoke passes; runtime compiles clean.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[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: f8e08447-4d8a-4c23-b34f-0da2e8ef9818

📥 Commits

Reviewing files that changed from the base of the PR and between 414c778 and 63a8ecb.

📒 Files selected for processing (2)

• src/a2a3/runtime/tensormap_and_ringbuffer/runtime/pto_orchestrator.cpp
• tests/ut/cpp/a2a3/test_wiring.cpp

---

📝 Walkthrough

Walkthrough

The orchestrator now marks hidden/inline alloc task slot states with allow_early_resolve = true before completion, allowing early-dispatch gating to treat allocation outputs as transparent. A new unit test validates that a flagged pre-completed creator does not block early-dispatch fanin propagation.

Changes

Early Resolve Flag for Alloc Tasks

Layer / File(s)	Summary
Set allow_early_resolve for hidden alloc tasks and validate src/a2a3/runtime/tensormap_and_ringbuffer/runtime/pto_orchestrator.cpp, tests/ut/cpp/a2a3/test_wiring.cpp	Orchestrator sets slot_state->allow_early_resolve = true for hidden/inline alloc tasks before marking PTO2_TASK_COMPLETED; a new test confirms a pre-completed flagged creator seeds dispatch_fanin transparently and that fanin reaches fanin_actual_count after propagation from a compute producer.

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

Sequence Diagram(s)

sequenceDiagram
    participant Orchestrator as PTO2OrchestratorState
    participant SlotState
    participant Consumer

    Orchestrator->>SlotState: allow_early_resolve = true
    Orchestrator->>SlotState: mark PTO2_TASK_COMPLETED
    Consumer->>SlotState: seed dispatch_fanin (transparent creator)
    Consumer->>Consumer: propagate_dispatch_fanin(compute producer)
    Consumer->>Consumer: dispatch_fanin == fanin_actual_count

Loading

Poem

A flag flipped true, so small, so neat,
lets alloc tasks skip the waiting seat.
The creator's done, no need to stall,
dispatch flows on and clears them all.
🐇 hop hop, the fanin's complete!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly matches the change: alloc-tensor creators are flagged to avoid blocking early-dispatch.
Description check	✅ Passed	The description directly explains the bug, fix, and tests for alloc-tensor early-dispatch behavior.
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.}