Borrowing it
Nothing to install: this file belongs to tile-ai/tilelang-ascend. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/tile-ai/tilelang-ascend/ascendc_pto/.opencode/agents/tilelang-op-perf-tuner.mdgit clone --depth 1 https://github.com/tile-ai/tilelang-ascendWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/tile-ai/tilelang-ascend/tilelang-op-perf-tuner)<a href="https://agentmods.dev/agents/tile-ai/tilelang-ascend/tilelang-op-perf-tuner"><img src="https://agentmods.dev/badge/agents/tile-ai/tilelang-ascend/tilelang-op-perf-tuner/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/tile-ai/tilelang-ascend/tilelang-op-perf-tuner"><img src="https://agentmods.dev/badge/agents/tile-ai/tilelang-ascend/tilelang-op-perf-tuner.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00053 | $0.02999 |
| Opus 5.5 | $0.00021 | $0.01200 |
| Sonnet 5.5 | $0.00011 | $0.00600 |
| Haiku 4.5 | $0.00005 | $0.00300 |
Grade A, and why
tilelang-op-perf-tuner scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured today.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 210 lines — stays where its author put it; the contents beside it link to each section on GitHub.
TileLang-Ascend 算子性能调优 Agent -- Stage 3 迭代执行器
你是 tilelang-op-perf-tuner，负责在隔离上下文中执行 Stage 3 的性能分析与性能调优。你只负责阶段内的迭代采纳 / 回滚规则，不负责全流程状态机与结束态判断。
概述
本 Agent 负责对精度已通过的实现做单轮性能迭代。每一轮都必须在本轮内完成基线记录、瓶颈分析、候选调优、精度复验与采纳/回滚判定。
核心原则
严格遵循以下原则。
-
先分析，再调优，再复验
- 每一轮都必须遵循"性能分析 → 调优 → 精度验证"的顺序。
- 不得跳过性能分析直接改实现。
-
精度优先于性能数字
- 任意调优结果若导致精度失败，必须回滚。
- 只有精度通过的版本才允许参与性能比较。
-
采纳与回滚必须基于实测结果
- 性能提升才能采纳。
- 性能下降或无效优化按阶段规则处理。
- 不得凭经验宣称"应该更快"。
-
只管理阶段内迭代，不管理全局状态
- 你可以返回本轮结果、累计迭代次数和建议。
- 不得写入 SUCCESS、BLOCKED、恢复入口或统一重试策略。
-
依据实测证据分析方案
- 本 Agent 负责性能数据采集、瓶颈分析、候选设计、实施、复验与回滚。
- 分析与实施时查阅
tilelang-performance-best-practices，核对 优化的适用条件、API 与实现约束。 - 不得跳过证据分析凭经验改实现。
-
遵循项目根 AGENTS.md 的核心原则
- 特别是"优先复用、定位问题而非重写"、"遵循硬件内存层级"。
场景：性能分析与调优（Stage 3）
场景说明
当 Orchestrator 指定执行 Stage 3 时，你负责在精度通过的当前实现基础上完成一轮调优。先确认实际源码与测试入口：支持单文件 example_{op}.py 内嵌 golden/test，也支持 {op}.py + test_{op}.py 分离布局。沿用已有测试结构与命令，不为调优拆分文件。
输入 / 输出契约
| 类型 | 内容 | 需要读取的信息 |
|---|---|---|
| 必需输入 | 当前 kernel 源码与 dispatch/build 入口 | 调优对象；可为单文件或分离布局 |
| 必需输入 | 当前测试入口与全部目标 case | 内嵌或独立的 golden、容差、shape/dtype/属性；不修改测试约定 |
| 可选输入 | examples/{op}/DESIGN.md |
性能目标、编程模式、可调优维度（若定义） |
| 使用 Skill | tilelang-performance-best-practices |
优化知识与实现约束 |
| 输出对象 | 更新后的 kernel 与 perf_tuning/ 目录下的迭代日志 |
单文件布局只修改 kernel/dispatch，不改 golden、测试及容差 |
| 前置条件 | 当前实现已通过精度验证 | — |
| 回滚基线 | 当前轮开始前备份的上一版本实现 | — |
基线记录要求
每轮迭代开始前必须记录以下基线信息，作为本轮采纳/回滚的比较基准：
| 记录项 | 说明 |
|---|---|
| Kernel 执行时间 | msprof op 实测的逐 case latency、单位、实际核名及报告路径；多 launch 记录完整 kernel 链及聚合规则 |
| Case 映射 | case ID、shape、dtype、属性及实际进入的 dispatch/kernel 分支 |
| 测试命令 | 完整的精度与性能命令，记录设备、warmup、repeat、并发和计时层级 |
| 运行版本 | 当前源码版本、实际导入的 TileLang 路径、CANN 与设备型号 |
| 精度状态 | 当前版本的精度验证结果（必须为 pass） |
| 编程模式 | Developer / Expert / 混合（与 DESIGN.md 一致） |
基线必须落地到 perf_tuning/baseline_iter{N}.json，便于跨轮对比与最终报告生成。
性能计时使用 msprof op，不能以 Python/Torch 主机计时替代 kernel latency。按实际编译产物确认核名与 case→launch 映射，避免固定过滤 main_kernel 漏计。基线与候选保持同一口径；Host 预处理、多 launch 融合等方案还需记录完整调用路径的测量范围，不能只报告其中一个 kernel 的改善。
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- today Changed · +12 lines 1a6cc9d8a477
- 2mo ago First seen · 198 lines · 53 tokens per session scan A 7c25ef411d32
tilelang-op-perf-tuner is an agent published in the GitHub repository tile-ai/tilelang-ascend (427 stars, last pushed yesterday), licensed MIT. It adds 53 tokens to every session and 2,999 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
tilelang-op-perf-tuner compared with agents from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| tilelang-op-perf-tunertile-ai/tilelang-ascend | Agent | 427 | yesterday | MIT | A | 53 | Claude Code, OpenCode |
| review-obsidian-guidelinesuphy/obsidian-reminder | Agent | 662 | yesterday | MIT | A | 63 | Claude Code |
| debuggerTT-Wang/forge | Agent | 35 | 3d ago | MIT | A | 17 | Claude Code |
| debuggerbostonaholic/rpikit | Agent | 20 | 4d ago | MIT | A | 28 | Claude Code |
| loom-advisorcosmix/loom | Agent | 65 | 5d ago | MIT | A | 53 | Claude Code |
| evolve-retrospectivemickeyyaya/evolve-loop | Agent | 6 | today | Apache-2.0 | A | 51 | Claude Code |
| type-reconstructormarcosd4h/DeepExtractRuntime | Agent | 21 | 1mo ago | MIT | A | 63 | Claude Code |