Tilegym Converting Cutile To Triton

This is a phase-gated conversion pipeline for translating NVIDIA cuTile GPU kernels to Triton. It walks you through analyze, convert, validate, test, and benchmark with explicit stop gates at each step. The workflow enforces TMA descriptors for 2D tile operations instead of raw pointer arithmetic, handles dual-layout kernels that need separate @triton.jit definitions for transpose flags, and includes a mandatory optimization checklist for attention and FMHA ops before you call the conversion done. The pre-flight grep commands catch layout flags and TMA-eligible loads before you write any Triton code. If you are porting cuTile code to Triton or debugging a translated kernel that regressed 10x on performance, this gives you the structured process and API mapping tables to do it correctly.

skills.sh installs: 1.5K
GitHub stars: 3.9K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add nvidia/skills --skill tilegym-converting-cutile-to-triton --agent claude-code

Installs into .claude/skills of the current project.

Files

cuTile → Triton Conversion

Convert @ct.kernel kernels to @triton.jit. API mapping: references/api-mapping.md (cuTile → Triton).

In this skill’s Markdown, Triton launch syntax kernel［grid］(…) uses Unicode brackets so link checkers do not parse [grid](…) as a hyperlink; use normal ASCII brackets in real Triton code.

Instructions

Follow the phase-gated workflow in translations/workflow.md. Every conversion should go through analyze → convert → validate → test → benchmark, with explicit gates before moving on. Use the documents in Workflow Selection when the task matches a special case (errors, layout flags, perf).

SKILL.md on GitHub

Details

Repository: github.com/nvidia/skills
Categories: Frontend Development, Debugging, Marketing & SEO
First seen: Jul 14, 2026