tilelang-pass-workflow-analyzer

Analyzes TileLang-Ascend Pass pipeline relationships and execution order for placement recommendations.

346|151|Updated Sep 25, 2025
One-click install
npx skills add https://github.com/tile-ai/tilelang-ascend --skill tilelang-pass-workflow-analyzer
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: tilelang-pass-workflow-analyzer
Source: https://github.com/tile-ai/tilelang-ascend/tree/main/.agents/skills/tilelang-pass-workflow-analyzer
Command: npx skills add https://github.com/tile-ai/tilelang-ascend --skill tilelang-pass-workflow-analyzer

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Provides a focused analysis of TileLang-Ascend compiler Pass workflows to help engineers understand execution order, data dependencies, and where to safely insert new Passes without breaking the pipeline. It reduces guesswork when modifying pipeline order and ensures cross-phase attrs (like buffer_shapess, address_map, size_map, and buffer scope) are respected.

Core Features & Use Cases

  • Reference-first analysis: Prioritizes reading pass-pipeline-overview.md, pass-dependency-graph.md, and new-pass-placement-guide.md to produce accurate high-level reports.
  • Dependency mapping: Explains Pass-to-Pass data flow (e.g., buffer scope → CrossCorePipeline, CollectBufferShapes → AscendMemoryPlanning → AscendSyncInsert).
  • Placement recommendations: Recommends Phase 1 vs Phase 2 positions and exact insertion points (e.g., after AscendStorageRewrite and before AscendMemoryPlanning) with reasoning and implementation guidance.
  • Drill-down on demand: Only reads source files like tilelang/engine/phase.py when the user requests concrete implementation details.

Quick Start

Ask the skill to describe the TileLang-Ascend Pass pipeline, list key pass dependencies, and recommend where to insert a new memory optimization pass.

Frequently Asked Questions about tilelang-pass-workflow-analyzer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I analyze TileLang-Ascend Pass pipeline dependencies?

To analyze TileLang-Ascend Pass pipeline dependencies, prioritize reading reference documents like pass-dependency-graph.md to map Pass-to-Pass data flow such as buffer scope to CrossCorePipeline before checking source code.

Where should I insert a new pass in the TileLang-Ascend compiler workflow?

Inserting a new pass in the TileLang-Ascend compiler workflow requires diagnosing phase dependencies to recommend exact Phase 1 or Phase 2 positions, such as placing memory optimization passes after AscendStorageRewrite and before AscendMemoryPlanning.

How does cross-pass data flow work for buffer_shapes and address_map in TileLang?

Cross-pass data flow for buffer_shapes and address_map in TileLang follows dependency sequences like CollectBufferShapes to AscendMemoryPlanning to AscendSyncInsert, ensuring cross-phase attributes are respected during pipeline execution.

What is the two-phase workflow in the TileLang-Ascend compiler?

The two-phase workflow in the TileLang-Ascend compiler organizes Pass execution order into distinct phases, allowing engineers to diagnose phase dependencies and safely plan new pass placement without breaking the pipeline.

Can I read source code directly to understand TileLang pass placement?

Reading source code directly to understand TileLang pass placement is not recommended initially; you should prioritize reference documents like new-pass-placement-guide.md and only drill down into files like tilelang/engine/phase.py when concrete implementation details are needed.

What happens if I ignore pass dependencies when modifying the TileLang pipeline?

Ignoring pass dependencies when modifying the TileLang pipeline breaks cross-phase attributes like buffer scope and size_map, which is why dependency-aware placement recommendations are necessary to maintain valid execution order.