cuda-kernel-autodev

Name: cuda-kernel-autodev
Availability: InStock
Author: romaosir

Optimize CUDA kernels through correctness-checked keep-or-revert autotuning loops with NCU profiling.

3|Updated May 14, 2026

Search Tags:#performance profiling #cuda #kernel optimization #autotuning #nsight compute #correctness gating #keep-revert workflow

Ready to run

npx skills add https://github.com/Romaosir/IF_Romao_kernel_optimize --skill cuda-kernel-autodev

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: cuda-kernel-autodev
Source: https://github.com/Romaosir/IF_Romao_kernel_optimize/tree/main/skills/b200_skill/cuda-kernel-autodev
Command: npx skills add https://github.com/Romaosir/IF_Romao_kernel_optimize --skill cuda-kernel-autodev

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires nvidia-smi, nvcc, torch, ncu-cuda-profiling, git, web-search, and includes references (resource) components.

What problem does it solve?

It helps you turn a correctness-verified CUDA kernel into a faster one by running an iterative hypothesis → implement → measure → keep-or-revert optimization loop with disciplined logging.

Core Features & Use Cases

End-to-end CUDA kernel dev loop: Establishes a correct baseline, then iterates with keep/revert discipline while tracking speedups and correctness across workloads.
Roofline- and technique-driven optimization planning: Uses sibling skills to choose what to try next and to apply proven technique patterns for the identified bottleneck.
Profiling and experiment governance: Integrates Nsight Compute collection (NCU) and enforces strict experiment hygiene (one focused change per iteration, commit-before-run, monotonic progress expectations).
Submission-ready workflow: Wraps up with a final correctness check and produces a reviewable performance report.

Quick Start

Ask an AI to optimize your CUDA kernel to beat a reference baseline on your target GPU using a correctness-gated, keep-or-revert autotuning loop with NCU profiling.

Frequently Asked Questions about cuda-kernel-autodev

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

FAQPage Schema

How do I optimize a CUDA kernel to beat a reference baseline?▼

To optimize a CUDA kernel safely, establish a correctness-verified baseline, then iterate using a keep-or-revert discipline to track speedups across workloads on your target GPU.

How does Nsight Compute profiling integrate with CUDA autotuning workflows?▼

Nsight Compute profiling integrates with CUDA autotuning by collecting performance metrics within an experiment governance loop. This enforces strict experiment hygiene, ensuring one focused change per iteration with commit-before-run discipline for measured performance planning.

Do I need a correctness-verified baseline before starting kernel autotuning?▼

Yes, a correctness-verified baseline is required before kernel autotuning. The workflow orchestrates an experiment-by-experiment optimization loop to ensure any speedup achieved does not break kernel correctness.

What is the best way to plan CUDA kernel optimizations using roofline strategy?▼

The best way to plan CUDA kernel optimizations is by using roofline strategy and a technique catalog to identify bottlenecks. This approach helps choose what to try next and applies proven technique patterns for the identified performance bottlenecks.

What are the limitations of a keep-revert workflow for CUDA kernel development?▼

The keep-revert workflow for CUDA kernel development requires strict experiment hygiene, enforcing one focused change per iteration and commit-before-run practices. It demands monotonic progress expectations, which may limit exploring multiple simultaneous optimizations.