tracelens-analysis-orchestrator

Files (55)

• .federated.json
• SKILL.md
• convolution-analyzer.md · agents/
• cpu-idle-analyzer.md · agents/
• elementwise-analyzer.md · agents/
• gemm-analyzer.md · agents/
• generic-op-analyzer.md · agents/
• kernel-coherence-agent.md · agents/
• kernel-fusion-analyzer.md · agents/
• kernel-stem-preprocessing-agent.md · agents/
• kernel-unification-agent.md · agents/
• model-identification-agent.md · agents/
• moe-analyzer.md · agents/
• multi-kernel-analyzer.md · agents/
• norm-analyzer.md · agents/
• reduce-analyzer.md · agents/
• sdpa-analyzer.md · agents/
• semantic-comparison-agent.md · agents/
• triton-analyzer.md · agents/
• .gitignore · evals/
• EVAL_RUBRICS.md · evals/
• README.md · evals/
• combined_traces_comparative.csv · evals/analysis_tests/
• combined_traces_standalone.csv · evals/analysis_tests/
• generate_ref.sh · evals/eval_scripts/
• kill_eval_jobs.sh · evals/eval_scripts/
• pack_test_archives.sh · evals/eval_scripts/
• run_pi_analysis_in_docker.sh · evals/eval_scripts/
• run_repeatability_parallel.sh · evals/eval_scripts/
• aggregate_repeatability.py · evals/eval_utils/
• compare_lca_partitions.py · evals/eval_utils/
• manual_eval_checklist.md · evals/eval_utils/
• merge_results.py · evals/eval_utils/
• quality_scripted_evals.py · evals/eval_utils/
• report_section_rules.yaml · evals/eval_utils/
• workflow_scripted_evals.py · evals/eval_utils/
• evals.json · evals/
• machine.yml · evals/
• README.md · evals/partial_tests/
• semantic_partition_scripted_evals.py · evals/partial_tests/eval_utils/
• semantic_purity_aggregate.py · evals/partial_tests/eval_utils/
• generate_gold_ref.sh · evals/partial_tests/
• partial_test_cases.csv · evals/partial_tests/
• run_partial_tests.sh · evals/partial_tests/
• SKILL.md · evals/skills/eval-post-processing/
• reference.md · evals/skills/eval-post-processing/
• run_post_processing.py · evals/skills/eval-post-processing/
• SKILL.md · evals/skills/quality-llm-eval/
• reference.md · evals/skills/quality-llm-eval/
• SKILL.md · evals/skills/workflow-llm-eval/
• reference.md · evals/skills/workflow-llm-eval/
• reference.md
• skill-card.md
• analysis_template.md · templates/
• sub_agent_spec.md · templates/

SKILL.md (first lines)

<!--
Copyright (c) 2026 Advanced Micro Devices, Inc. All rights reserved.

See LICENSE for license information.
-->

# Analysis orchestrator

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`.

## Full procedure

Follow **[reference.md](reference.md)** for every step (user prompts, `<prefix>` / `{CMD}` usage, CLI commands, subagent launch text, validation, report `tee` order, and plot embedding).

## Workflow index

```
0. Query User Inputs (interactive answers or a run YAML, Environment Setup) → cache/run_config.yaml
1. Generate Performance Report (one wrapper call on cache/run_config.yaml; semantic pre-processing first)
2. Trace-Quality Gate → Deterministic fallback analysis
3-6. Prepare Category Data (GPU Util, Top Ops, Tree Data, Multi-Kernel Data, Category Filtering)
7. System-Level Analysis (PARALLEL) → system_findings/
8. Compute Kernel Subagents (PARALLEL) → category_findings/
  8.5. Aggregate → priority_data.json::findings[]
9. Validate Subagent Outputs
10. load_findings + Model Identification (subagent) → metadata/model_info.json
11. Render performance PNG if agent_extension.py is absent
12. Generate analysis.md (orchestrator writes via <prefix> tee), optional extension, embed PNG
```

## Rules

- **Subagents:** Use your runner's subagent primitive (the Task tool in Claude Code or Cursor, `spawn_agent` under Codex) **only** where reference.md says “subagent” (Steps **7**, **8**, **10**). The orchestrator runs everything else, including Step 8.5, 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).

## Primary outputs

- **Deliverable:** `<output_dir>/analysis.md`
- **Internals:** `system_findings/`, `category_findings/`, `category_data/`, `metadata/`, `perf_report*.xlsx`, CSV folders — see package README for layout.

## Agent layout

Project subagents ship with this skill: `TraceLens/Agent/Analysis/skills/analysis-orchestrator/agents/*.md`.

From amd/skills, MIT.

More from amd/skills