Add: port a2a3 eager prewarm arena to a5 tmr

[yanghaoran29]

• Align a5 runtime_maker with a2a3 Add: eager prebuilt runtime-arena prewarm at worker init #1322: shared build_and_cache_prebuilt_arena, strong prewarm_config_impl, decouple ensure_static_arenas from Runtime
• Add a5 prewarm regression test and onboard cold/hot benchmark

Onboard effect (a5, ring_task_window=64, dummy_task): cold first-run bind.prebuilt ~9.58 ms -> hot ~0.002 ms; ~9.97 ms build cost shifts to init simpler_prewarm.build (cache hit on first run).

[coderabbitai]

Important

Review skipped

Auto incremental reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 2dad4280-4655-40c2-9768-c349e1507f92

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

📝 Walkthrough

Walkthrough

The runtime arena miss path now uses a shared build/upload/cache helper, while cache hits and newly built arenas are wired directly. A new prewarm_config_impl entrypoint supports eager cache population, with benchmark and L2 initialization coverage.

Changes

Runtime arena prewarming

Layer / File(s)	Summary
Arena build and cache pipeline src/a5/runtime/tensormap_and_ringbuffer/host/runtime_maker.cpp	Static arena setup is separated from runtime wiring, and build_and_cache_prebuilt_arena centralizes sizing, reservation, image construction, upload, and cache storage.
Binding and prewarm integration src/a5/runtime/tensormap_and_ringbuffer/host/runtime_maker.cpp	Cache-hit wiring remains metadata-based, cache misses wire newly built arenas directly, and prewarm_config_impl eagerly populates the cache from ring sizing values.
Prewarm measurement and regression coverage tests/st/a5/tensormap_and_ringbuffer/bench_prewarm_timing.py, tests/st/a5/tensormap_and_ringbuffer/test_prewarm_config.py	The benchmark compares cold and hot timings, while the regression test validates L2 initialization with prewarm_config.

Estimated code review effort: 4 (Complex) | ~45 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Worker
  participant prewarm_config_impl
  participant build_and_cache_prebuilt_arena
  participant RuntimeArenaCache
  Worker->>prewarm_config_impl: provide ring sizing configuration
  prewarm_config_impl->>build_and_cache_prebuilt_arena: build and upload arena image
  build_and_cache_prebuilt_arena->>RuntimeArenaCache: store prebuilt runtime image
  Worker->>RuntimeArenaCache: bind configured runtime
  RuntimeArenaCache-->>Worker: return cached arena metadata

Loading

Possibly related PRs

• hw-native-sys/simpler#1322: Adds the worker initialization prewarm plumbing consumed by this runtime implementation.
• hw-native-sys/simpler#1397: Extends the worker initialization lifecycle with the prewarm_config argument validated here.

Poem

A rabbit watched the arenas bloom,
Built and cached to clear the room.
Cold runs hop, then hot runs race,
Prewarm puts speed in its place.
“Bind,” says the bun, “with less delay!” 🐇

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly reflects the main change: porting eager prewarm arena support from a2a3 to a5 tmr.
Description check	✅ Passed	The description matches the implemented runtime changes, regression test, and benchmark additions.
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.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

tests/st/a5/tensormap_and_ringbuffer/test_prewarm_config.py (1)

19-36: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Solid smoke test, but doesn't exercise the cache-hit path it's meant to protect.

This confirms worker.init(prewarm_config=...) doesn't crash, but never follows up with register/run using matching ring sizing to confirm the prewarmed image is actually consumed on bind (i.e., that the cache-hit branch in bind_callable_to_runtime_impl is reached rather than a silent fallback to the miss path).

     worker = Worker(
         level=2,
         device_id=int(st_device_ids[0]),
         platform=st_platform,
         runtime=_RUNTIME,
     )
     try:
         worker.init(prewarm_config=config)
+        # Exercise a bind with matching ring sizing and assert it takes the
+        # cache-hit path (e.g. via a callable register/run + STRACE assertion).
     finally:
         worker.close()

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/st/a5/tensormap_and_ringbuffer/test_prewarm_config.py` around lines 19
- 36, Extend test_l2_init_with_prewarm_config to register and run a workload
after worker.init using the same ring_task_window=64 configuration, ensuring the
prewarmed image is consumed during binding. Exercise the
bind_callable_to_runtime_impl cache-hit path and retain the existing worker
cleanup.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/st/a5/tensormap_and_ringbuffer/bench_prewarm_timing.py`:
- Around line 165-184: Make all timing output in the displayed per-run and
“both” summary paths None-safe. Reuse the existing formatting approach from
prewarm_text for parsed bind_prebuilt_strace_ms, cold, hot, and prewarm_ms
values, ensuring missing timings print a fallback such as “None” instead of
applying :.2f; preserve the existing guarded savings calculations.

---

Nitpick comments:
In `@tests/st/a5/tensormap_and_ringbuffer/test_prewarm_config.py`:
- Around line 19-36: Extend test_l2_init_with_prewarm_config to register and run
a workload after worker.init using the same ring_task_window=64 configuration,
ensuring the prewarmed image is consumed during binding. Exercise the
bind_callable_to_runtime_impl cache-hit path and retain the existing worker
cleanup.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 07025a8b-2d83-4aa5-8494-16509183e5af

📥 Commits

Reviewing files that changed from the base of the PR and between c032e07 and 01a3c43.

📒 Files selected for processing (3)

• src/a5/runtime/tensormap_and_ringbuffer/host/runtime_maker.cpp
• tests/st/a5/tensormap_and_ringbuffer/bench_prewarm_timing.py
• tests/st/a5/tensormap_and_ringbuffer/test_prewarm_config.py