skills.rest
FutureMLS-Lab
FutureMLS-LabOFFICIAL
Explore all skills by this author →
GitHub

add-sgl-kernel

Guide adding AOT CUDA or C++ kernels with Torch registration, CMake, tests, and benchmarks.

556|81|Updated May 19, 2026
Search Tags:#benchmarking#c++#pytorch#cuda#sgl-kernel#kernel integration#aot kernels
One-click install
npx skills add https://github.com/FutureMLS-Lab/OSCAR --skill add-sgl-kernel-futuremls-lab
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: add-sgl-kernel
Source: https://github.com/FutureMLS-Lab/OSCAR/tree/main/sglang-research/.claude/skills/add-sgl-kernel
Command: npx skills add https://github.com/FutureMLS-Lab/OSCAR --skill add-sgl-kernel-futuremls-lab

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps developers add heavyweight ahead-of-time CUDA or C++ kernels to sgl-kernel without missing the required integration, testing, benchmarking, or build steps.

Core Features & Use Cases

  • Kernel Implementation: Guides creation of CUDA source files with tensor validation, dtype dispatch, stream handling, and launch error checks.
  • Framework Integration: Covers C++ declarations, Torch operator registration, CMake source inclusion, and Python API exposure.
  • Verification Workflow: Provides pytest correctness tests, Triton benchmarking, build commands, and troubleshooting guidance.
  • Use Case: Add a new FP16, BF16, and FP32 CUDA elementwise operation to SGLang's AOT kernel library with a complete test and performance benchmark.

Quick Start

Use the add-sgl-kernel skill to implement, register, build, test, and benchmark a new heavyweight CUDA kernel in sgl-kernel.

Frequently Asked Questions about add-sgl-kernel

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

FAQPage Schema
How do I add a new CUDA kernel to sgl-kernel?▼

Adding a CUDA kernel to sgl-kernel requires creating the source file with tensor validation, registering the Torch operator, updating CMake, exposing the Python API, and writing pytest tests.

What is the process for integrating C++ operations into SGLang's AOT kernel library?▼

Integrating C++ operations into SGLang involves Torch operator registration, CMake source inclusion, Python API exposure, and controlled build validation for the ahead-of-time kernel.

Does sgl-kernel support FP16, BF16, and FP32 dtype dispatch for new elementwise operations?▼

Yes, sgl-kernel supports FP16, BF16, and FP32 dtype dispatch for new elementwise operations, requiring CUDA-aware tensor validation and kernel launch error checking.

How do I benchmark PyTorch CUDA kernels using Triton?▼

You benchmark PyTorch CUDA kernels in sgl-kernel by executing Triton performance measurements alongside pytest correctness tests to validate speed and accuracy.

What are common build validation issues when registering Torch operators for CUDA kernels?▼

Common build validation issues when registering Torch operators include missing CMake source inclusion, incorrect dtype dispatch, or lacking kernel launch error checks, which troubleshooting addresses.

Related Skills

Explore More Software Engineering Skills→
scottgl9scottgl9
12

add-sgl-kernel

Implement and integrate CUDA/C++ kernels into sgl-kernel for tensor operations.

Community
Advanced
xchang1121xchang1121
6

triton-ascend-example-matmul

Implement a Triton Ascend matmul example with 2D tiling and K-loop accumulation.

Community
Advanced
Nabilhassan12345Nabilhassan12345

add-sgl-kernel

Implement CUDA/C++ AOT kernels in sgl-kernel with Python and Torch extensions.

Community
Advanced
NVIDIANVIDIA
2.8k

kernel-triton-writing

Design, implement, verify and benchmark Triton kernels with fixed contracts and built-in harnesses.

Official
Advanced