croq-dsl-tilelang

Automate performance tuning and validation of TileLang GPU kernels.

Updated Apr 10, 2026

Search Tags:#performance optimization #profiling #GPU #kernel tuning #TileLang #TVM

Ready to run

npx skills add https://github.com/LancerLab/croqtile-tuner --skill croq-dsl-tilelang

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: croq-dsl-tilelang
Source: https://github.com/LancerLab/croqtile-tuner/tree/main/.claude/skills/croq-dsl-tilelang
Command: npx skills add https://github.com/LancerLab/croqtile-tuner --skill croq-dsl-tilelang

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill streamlines the process of tuning TileLang kernels for GPU performance, reducing manual effort and trial-and-error in kernel optimization.

Core Features & Use Cases

Automated Kernel Tuning: Generate and execute multiple kernel configurations to optimize performance metrics such as TFLOPS and runtime.
Profiling and Validation: Incorporate performance profiling tools like NVIDIA NSight Compute and verify kernel correctness.
Use Case: A researcher wants to optimize a TileLang-based matrix multiplication kernel by tuning block sizes and pipeline stages to maximize throughput on a specific GPU.

Quick Start

Use the croq-dsl-tilelang skill to tune a TileLang kernel with default settings and profile the best-performing configuration.

Frequently Asked Questions about croq-dsl-tilelang

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

FAQPage Schema

How do I automate GPU kernel tuning for TileLang to maximize TFLOPS?▼

Automated TileLang GPU kernel tuning generates and executes multiple configurations to optimize performance metrics like TFLOPS and runtime. It iteratively tests block sizes and pipeline stages to identify the highest-throughput setup for your specific hardware.

What is the best way to profile and validate TileLang kernel performance on a GPU?▼

Profiling and validating TileLang kernel performance utilizes tools like NVIDIA NSight Compute to measure runtime metrics and verify correctness. This identifies optimal configurations and ensures high-performance computing tasks execute accurately.

Can I use NVIDIA NSight Compute with TVM and TileLang for iterative kernel experiments?▼

Yes, NVIDIA NSight Compute integrates with TVM and TileLang to support iterative experiments. This combination profiles generated kernel configurations and analyzes results to validate correctness during the optimization workflow.

Does automated kernel tuning reduce manual trial-and-error for TileLang matrix multiplication?▼

Automated kernel tuning significantly reduces manual trial-and-error for TileLang matrix multiplication. By systematically testing block sizes and pipeline stages, it streamlines optimization to maximize throughput on targeted GPU hardware.

What are the limitations of using automated workflows for GPU kernel tuning?▼

Automated GPU kernel tuning workflows require integrated scripting and profiling tools to function. Users must configure iterative experiments and analyze profiling results, as the workflow depends on validating correctness across multiple generated configurations.

When do I need to tune block sizes and pipeline stages in TileLang kernels?▼

Tuning block sizes and pipeline stages in TileLang kernels is needed when optimizing performance for specific GPU hardware. This process maximizes throughput and TFLOPS for high-performance computing tasks like matrix multiplication.