tilelang-cuda-examples-torch

What problem does it solve?

PyTorch workflows often require custom CUDA kernels for performance; this Skill provides complete TileLang CUDA examples to implement high-performance operations directly in PyTorch.

Core Features & Use Cases

• CUDA-accelerated TileLang kernels: demonstrates GEMM (matmul), elementwise, layernorm, GEMV, and more.
• Guided integration: shows how to compile and call TileLang kernels from PyTorch, with out_idx conventions and shared memory usage.
• Education and prototyping: ideal for learning how to write and benchmark custom kernels in CUDA via TileLang.

Quick Start

Run a sample TileLang CUDA kernel with PyTorch to build a high-performance matrix operation.