AgentSkillsCN

kebab-build-and-env

在 Linux 上使用 CUDA、CMake 和 Makefile 工作流构建并引导 Kebab 项目。适用于被要求设置依赖项、检测 CUDA/GPU 架构、配置构建，或修复编译环境问题时使用。

SKILL.md
--- frontmatter
name: kebab-build-and-env
description: Build and bootstrap the Kebab project on Linux with CUDA, CMake, and Makefile workflows. Use when asked to set up dependencies, detect CUDA/GPU architecture, configure builds, or fix compilation environment issues.

Kebab Build and Environment

When to Use This Skill

  • •User asks to build project from scratch
  • •User sees CUDA detection failures (CUDA_PATH / CUDA_ARCH / nvidia-smi)
  • •User asks for debug/release build instructions
  • •User asks to clean and rebuild reliably

Prerequisites

  • •Linux with NVIDIA driver and visible GPU (nvidia-smi works)
  • •CUDA toolkit and nvcc
  • •CMake (3.18+) and GNU Make
  • •yaml-cpp development headers

Step-by-Step Workflows

Workflow A: Standard Build (preferred)

  1. •Check environment:
    • •make gpu-info
  2. •Prepare dependencies:
    • •make setup
  3. •Build:
    • •make build

Workflow B: Explicit CMake Build

  1. •Configure:
    • •cmake -S kebab -B build -DCUDA_ARCH=sm_90a
  2. •Compile:
    • •cmake --build build -j$(nproc)

Workflow C: Clean Rebuild

  1. •make clean
  2. •make build

Environment Guidance

  • •If auto detection fails, set:
    • •export CUDA_PATH=/usr/local/cuda
    • •export CUDA_ARCH=sm_90a
  • •Hopper WGMMA workloads should use sm_90a when possible.

Troubleshooting

IssueMitigation
GPU detection failed from MakefileRun nvidia-smi; then set CUDA_ARCH manually
CUDA toolkit not foundSet CUDA_PATH and ensure ${CUDA_PATH}/bin/nvcc exists
CMake configure failsRemove build/ and rerun configure+build

References

  • •Makefile (setup, gpu-info, cmake-configure, cmake-build, clean)
  • •kebab/CMakeLists.txt