NVIDIA cuTile Python Tutorial: Building Tiled GPU Kernels for Vector Addition, Matrix Addition, and Matrix Multiplication in Colab
English summary
This tutorial walks through setting up NVIDIA cuTile Python in a Colab notebook, checking GPU/CUDA/driver compatibility, and implementing tiled kernels for vector addition, matrix addition, and matrix multiplication using direct load/store, gather/scatter, and matrix multiply accumulate. It provides wrapper functions that fall back to PyTorch when cuTile is unavailable, and validates outputs against PyTorch operations with correctness checks. The workflow includes benchmarking kernel performance against PyTorch equivalents and visualizing median runtimes, then suggests further experiments such as tile size tuning, precision comparison, and operation fusion. The notebook remains fully executable in Colab even without the required cuTile runtime.
Chinese summary
本教程逐步讲解如何在Colab笔记本中设置NVIDIA cuTile Python，检查GPU/CUDA/驱动兼容性，并使用直接加载/存储、gather/scatter和矩阵乘累加实现向量加法、矩阵加法和矩阵乘法的分块内核。它提供了当cuTile不可用时回退到PyTorch的包装函数，并通过正确性检查验证输出与PyTorch操作的一致性。工作流程还包括对标内核性能与等价PyTorch操作，并可视化中位数运行时间，然后建议进一步的实验，如分块大小调优、精度比较和操作融合。即使没有所需的cuTile运行时，该笔记本也完全可在Colab中执行。
Key points
The tutorial provides a complete cuTile Python workflow covering environment setup, kernel definition, execution, validation, and benchmarking.
教程提供了完整的cuTile Python工作流程，涵盖环境设置、内核定义、执行、验证和基准测试。
It demonstrates tiled GPU kernels for vector addition, matrix addition, and matrix multiplication using cuTile's tile-based programming model.
展示了使用cuTile基于分块的编程模型实现的向量加法、矩阵加法和矩阵乘法的分块GPU内核。
Correctness is verified against PyTorch outputs, and a PyTorch fallback ensures the notebook runs even without cuTile support.
通过与PyTorch输出对比验证正确性，并提供PyTorch回退，确保笔记本在没有cuTile支持时也能运行。
Benchmark results compare cuTile kernels with equivalent PyTorch operations, showing median runtimes in a bar chart.
基准测试结果比较了cuTile内核与等价PyTorch操作，并以柱状图展示中位数运行时间。
Next-step suggestions include tile-size sweep, precision comparison, operation fusion, and attention kernel study for further exploration.
后续实验建议包括分块大小扫描、精度比较、操作融合以及注意力内核研究，以供深入探索。