Update: progress-aware ring buffer spin detection with env-configurable sizes

[hw-native-sys-bot]

Summary

• Add progress-aware spin detection to all three ring buffers (TaskRing, HeapRing, DepPool): reset spin counter when last_task_alive or heap_tail advances, only report deadlock when truly stuck
• Revert PTO2_TASK_WINDOW_SIZE from 131072 to 65536 (progress-aware detection eliminates the doubled window workaround from PR Fix: support batch=256 by doubling task window and fixing deferred release #273)
• Add PTO2_RING_DEP_POOL env var for runtime dep pool size control, completing the set with PTO2_RING_TASK_WINDOW and PTO2_RING_HEAP
• Thread dep_pool_capacity through runtime creation APIs
• Extract dep pool reclamation into reusable pto2_dep_pool_reclaim()
• Guard against division by zero in dep pool diagnostic logging
• Update all deadlock/overflow Solution messages to show both compile-time macro and runtime env var (PTO2_RING_HEAP/PTO2_RING_TASK_WINDOW/PTO2_RING_DEP_POOL) overrides for easier on-site troubleshooting
• Add paged_attention_ringbuffer test with small ring sizes (TW=1024, HP=1MB, DP=1024) to guard rotation/reclamation logic
• Add 3 new paged attention test cases (batch=512, context=16384, batch=32)

Testing

• 22 ring buffer size combinations tested on hardware (device 6), all pass
• paged_attention_ringbuffer Case1 (batch=32) passes with 1MB heap (empirically verified minimum)
• Extreme values (TW=256, HP=4MB, DP=256) work correctly with batch=256/512
• Deadlock detection correctly triggers on insufficient heap (verified via device logs)

[gemini-code-assist]

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request significantly enhances the robustness and diagnostic capabilities of the dependency pool management system. By introducing explicit reclamation and deadlock detection mechanisms, it addresses potential silent hangs and provides clearer guidance for capacity issues. Additionally, the expanded test suite for paged attention ensures that the system is thoroughly validated under diverse and demanding conditions, improving overall stability and developer experience.

Highlights

• Dependency Pool Reclamation: Extracted the dependency pool reclamation logic into a reusable helper function, pto2_dep_pool_reclaim(), to improve code modularity and reuse.
• Deadlock Detection and Prevention: Implemented pto2_dep_pool_ensure_space() with a spin-wait mechanism and deadlock detection. This prevents silent hangs when the dependency pool capacity is insufficient and provides detailed diagnostic information upon detection, including recommended capacity adjustments.
• Improved Error Messaging: Enhanced the alloc() overflow error message in the dependency pool to include actionable suggestions, guiding users to increase PTO2_DEP_LIST_POOL_SIZE for resolution.
• Expanded Paged Attention Test Coverage: Added three new paged attention test cases (Case4, Case5, Case6) to stress-test the ring buffer under various workload profiles, including high-pressure wrapping and high dependency pool watermark scenarios.
• Runtime Variant Application: Applied all changes to both a2a3 and a5 runtime variants, ensuring consistent behavior and improvements across different architectures.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

• src/a2a3/runtime/tensormap_and_ringbuffer/runtime/pto_orchestrator.cpp
  • Added pto2_dep_pool_reclaim function to centralize dependency pool reclamation logic.
  • Added pto2_dep_pool_ensure_space function to manage dependency pool capacity, including spin-waiting and deadlock detection.
  • Replaced inline dependency pool reclamation logic with a call to pto2_dep_pool_reclaim.
  • Integrated pto2_dep_pool_ensure_space before allocating dependencies to prevent overflow.
• src/a2a3/runtime/tensormap_and_ringbuffer/runtime/pto_ring_buffer.h
  • Defined PTO2_DEP_POOL_SPIN_LIMIT to control the maximum spin count before declaring a dependency pool deadlock.
  • Updated the PTO2DepListPool::alloc() error message to provide solution suggestions for increasing PTO2_DEP_LIST_POOL_SIZE.
• src/a5/runtime/tensormap_and_ringbuffer/runtime/pto_orchestrator.cpp
  • Added pto2_dep_pool_reclaim function to centralize dependency pool reclamation logic.
  • Added pto2_dep_pool_ensure_space function to manage dependency pool capacity, including spin-waiting and deadlock detection.
  • Replaced inline dependency pool reclamation logic with a call to pto2_dep_pool_reclaim.
  • Integrated pto2_dep_pool_ensure_space before allocating dependencies to prevent overflow.
• src/a5/runtime/tensormap_and_ringbuffer/runtime/pto_ring_buffer.h
  • Defined PTO2_DEP_POOL_SPIN_LIMIT to control the maximum spin count before declaring a dependency pool deadlock.
  • Updated the PTO2DepListPool::alloc() error message to provide solution suggestions for increasing PTO2_DEP_LIST_POOL_SIZE.
• tests/device_tests/a2a3/tensormap_and_ringbuffer/paged_attention/golden.py
  • Added three new paged attention test cases (Case4, Case5, Case6) with varying batch sizes, context lengths, and other parameters to enhance test coverage.

Activity

• No human activity has been recorded on this pull request yet.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces important improvements for dependency pool management, including deadlock detection and more robust error reporting. The core change is the pto2_dep_pool_ensure_space function, which proactively checks for available space in the dependency pool and spin-waits if necessary, with a timeout to detect and report deadlocks. The reclamation logic has also been nicely refactored into a reusable helper. The changes are applied to both a2a3 and a5 runtime variants, and I've noted the significant code duplication between them as a point for future improvement. Additionally, the new paged attention test cases are a valuable addition for stress-testing the system under various loads. My review includes suggestions to prevent potential division-by-zero issues in the new error logging paths.

[jvjhfhg]

新的死锁/overflow “Solution” 文案没有体现你新增的运行时 knob

现在 dep pool / task window / heap 的 fatal 日志仍主要提示改宏（PTO2_*_SIZE），但你已经引入了运行时覆盖（至少 a2a3 的 PTO2_RING_DEP_POOL / PTO2_RING_TASK_WINDOW / PTO2_RING_HEAP），对于现场排障的人不友好。

• 位置：Dep pool deadlock/overflow 日志在 pto_orchestrator.cpp:218-242 和 pto_ring_buffer.h:467-479；TaskRing/HeapRing deadlock 日志在 pto_ring_buffer.h:144-160、pto_ring_buffer.h:335-360。
• 建议：Solution 里同时提示环境变量覆盖（例如 PTO2_RING_DEP_POOL=...），这样压测和线上问题会更快收敛。

[jvjhfhg]

active_count * 2 -> total_count * 2, to be fixed later