Getting Started

This document covers installation and build instructions for CUDA Kernel Academy.

System Requirements

Component	Minimum	Recommended
CUDA Toolkit	12.0	12.x latest
CMake	3.20	3.24+
GCC / Clang	GCC 9 / Clang 10	GCC 11+
Python	3.8	3.10+
GPU	Volta (sm_70)	Ampere / Ada / Hopper

Clone Repository

git clone https://github.com/AICL-Lab/cuda-kernel-academy.git
cd cuda-kernel-academy

Build with CMake Presets

cmake --list-presets
cmake --preset default
cmake --build --preset default
ctest --preset default

The root presets intentionally export /usr/bin/gcc and /usr/bin/g++ so CUDA 12 builds do not accidentally pick up newer Conda toolchains that sit earlier in PATH.

Build Individual Modules

01-sgemm-tutorial

cd 01-sgemm-tutorial
make GPU_ARCH=sm_86
./build/sgemm_benchmark

02-tensorcraft-core

cd 02-tensorcraft-core
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build -j$(nproc)

03-hpc-advanced

cd 03-hpc-advanced
cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release
cmake --build build -j$(nproc)

04-inference-engine

cmake --preset default
cmake --build --preset default
ctest --preset default

04-inference-engine now relies on the parent build to provide TensorCraft::tensorcraft, so treat the repository root as its canonical build entrypoint.

Local Quality Checks

pre-commit run --all-files
npm run docs:build
cmake --list-presets

References