# tilegym-adding-cutile-kernel

Guides AI coding agents through adding cuTile GPU kernel operators to TileGym. Useful when implementing custom kernels, tests, and benchmarks.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode · 1800 installs · ★ 3735
- Author: TileGym Team
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/NVIDIA/skills/tree/HEAD/skills/tilegym-adding-cutile-kernel/SKILL.md
- Updated: 2026-10-23
- HTML: https://affpapa.org/en/nevibecode/skills/tilegym-adding-cutile-kernel

## What it does

This skill defines an ordered workflow for implementing cuTile backend operators. The agent first tracks the six-step checklist in TodoWrite before editing code. Work starts in `src/tilegym/ops/ops.py`, defining the `@dispatch` interface that throws `NotImplementedError` for unhandled backends.

Next, the kernel is written with `cuda.tile` and registered via `@register_impl`. Without explicit exports in `src/tilegym/ops/cutile/__init__.py`, the cuTile implementation fails to load at runtime. The workflow finishes with PyTorch reference tests and Triton-based performance benchmarks measuring TFLOPS or bandwidth.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-adding-cutile-kernel -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/tilegym-adding-cutile-kernel/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/tilegym-adding-cutile-kernel/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-adding-cutile-kernel -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/tilegym-adding-cutile-kernel/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-adding-cutile-kernel -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/tilegym-adding-cutile-kernel/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install NVIDIA/skills tilegym-adding-cutile-kernel
```

2. **Manually: put the files in place**
- File: `.github/skills/tilegym-adding-cutile-kernel/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-adding-cutile-kernel -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/NVIDIA/skills --path skills/tilegym-adding-cutile-kernel
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/tilegym-adding-cutile-kernel/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-adding-cutile-kernel -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/NVIDIA/skills/tree/HEAD/skills/tilegym-adding-cutile-kernel/SKILL.md

## When to use it
- register dispatch interfaces in `ops.py`
- implement GPU kernels using `cuda.tile`
- export cuTile modules in `__init__.py`
- write correctness tests against PyTorch references
- build operator benchmarks using Triton testing tools

## Example prompts
- Add a new cuTile kernel operator named my_op to TileGym
- Register a cuTile backend implementation and create benchmarks
- Set up tests and dispatch interface for a new cuTile op

## Similar
- [model-serving-kubernetes](https://affpapa.org/en/nevibecode/skills/model-serving-kubernetes.md) — skill · 327 installs
- [runpod-templates](https://affpapa.org/en/nevibecode/skills/runpod-templates.md) — skill · 10100 installs
- [deep-learning](https://affpapa.org/en/nevibecode/skills/deep-learning.md) — skill · 946 installs
- [tracelens-analysis-orchestrator](https://affpapa.org/en/nevibecode/skills/tracelens-analysis-orchestrator.md) — skill · 877 installs
- [quark-install](https://affpapa.org/en/nevibecode/skills/quark-install.md) — skill · 526 installs
- [vast-gpu](https://affpapa.org/en/nevibecode/skills/vast-gpu.md) — skill · 453 installs

From the same set: [nvidia/skills](https://affpapa.org/en/nevibecode/by/nvidia/skills.md) — 194 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md