tracelens-analysis-orchestrator

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Analysis orchestrator

分析编排器

Coordinate system-level analysis (CPU/idle, kernel fusion, multi-kernel / comm / memcpy) and compute-kernel analysis (GEMM, SDPA, elementwise, etc.): one trace load, shared prep, parallel subagents, then aggregation into

analysis.md

协调系统级分析（CPU/空闲状态、内核融合、多内核/通信/内存复制）与计算内核分析（GEMM、SDPA、逐元素运算等）：单次加载追踪数据、共享预处理流程、并行运行子Agent，最终将结果汇总至

analysis.md

。

Full procedure

完整流程

Follow reference.md for every step (user prompts,

<prefix>

{CMD}

usage, CLI commands, subagent launch text, validation, report

tee

order, plot embedding, and trace diagnostics).

每一步均需遵循**reference.md**（包括用户提示、

<prefix>

{CMD}

用法、CLI命令、子Agent启动文本、验证流程、报告

tee

命令顺序、图表嵌入及追踪诊断）。

Workflow index

工作流索引

0. Query User Inputs (Platform, Trace Path(s), Analysis Mode, Environment Setup)
1. Generate Performance Report (branches on analysis mode: training vs inference then, comparison scope)
2-5. Prepare Category Data (GPU Util, Top Ops, Tree Data, Multi-Kernel Data, Category Filtering)
6. System-Level Analysis (PARALLEL) → system_findings/
7. Compute Kernel Subagents (PARALLEL) → category_findings/
   7.5. Aggregate → priority_data.json::findings[]
8. Validate Subagent Outputs
9. load_findings + Model Identification (subagent) → metadata/model_info.json
10. Render performance PNG if agent_extension.py is absent
11. Generate analysis.md (orchestrator writes via <prefix> tee), optional extension, embed PNG

0. 查询用户输入（平台、追踪文件路径、分析模式、环境配置）
1. 生成性能报告（根据分析模式分支：训练 vs 推理，再区分对比范围）
2-5. 准备分类数据（GPU利用率、Top算子、树形数据、多内核数据、分类过滤）
6. 系统级分析（并行执行）→ system_findings/
7. 计算内核子Agent（并行执行）→ category_findings/
   7.5. 汇总结果 → priority_data.json::findings[]
8. 验证子Agent输出
9. 加载结果 + 模型识别（子Agent）→ metadata/model_info.json
10. 若agent_extension.py不存在则渲染性能PNG图
11. 生成analysis.md（编排器通过<prefix> tee命令写入），可选扩展功能，嵌入PNG图

Rules

规则

Subagents: Use the Task tool only where reference.md says “subagent” (Steps 6, 7, 9). The orchestrator runs everything else, including Step 7.5, using the command prefix from
```
<output_dir>/cache/cmd_prefix.txt
```
(
```
{CMD}
```
substitution).
Language: Prefer vendor-agnostic terms (GPU kernels, collective communication, vendor GEMM library, DNN primitives, GPU graph). When quoting trace data, real kernel names are fine.
Subagent prompts: Point each subagent at the checked-in agent file under
```
TraceLens/Agent/Analysis/skills/analysis-orchestrator/agents/<name>.md
```
(see reference.md for exact paths and prompt shells).

子Agent：仅在reference.md标注“subagent”的步骤（步骤6、7、9）中使用Task工具。编排器负责运行其他所有步骤，包括步骤7.5，使用
```
<output_dir>/cache/cmd_prefix.txt
```
中的命令前缀（替换
```
{CMD}
```
）。
**术语规范：**优先使用厂商无关术语（GPU内核、集合通信、厂商GEMM库、DNN原语、GPU图）。引用追踪数据时，可使用真实内核名称。
**子Agent提示词：**将每个子Agent指向已签入的代理文件，路径为
```
TraceLens/Agent/Analysis/skills/analysis-orchestrator/agents/<name>.md
```
（具体路径及提示词模板请参考reference.md）。

Primary outputs

主要输出

Deliverable:
```
<output_dir>/analysis.md
```

Internals:

system_findings/

category_findings/

category_data/

metadata/

perf_report*.xlsx

, CSV folders — see package README for layout.

交付物：
```
<output_dir>/analysis.md
```
内部文件：
```
system_findings/
```
、
```
category_findings/
```
、
```
category_data/
```
、
```
metadata/
```
、
```
perf_report*.xlsx
```
、CSV文件夹 — 目录结构请参考包内README。

Agent layout

Agent架构

Project subagents ship with this skill:

TraceLens/Agent/Analysis/skills/analysis-orchestrator/agents/*.md

本技能附带项目子Agent：

TraceLens/Agent/Analysis/skills/analysis-orchestrator/agents/*.md

。