Cute-Learning

Welcome to the Cute-Learning repository! This project showcases several example implementations using Cutlass CuTe, a powerful tool for high-performance computing.

Features

This repository includes implementations for:

• GEMM (General Matrix Multiply)
• GEMV (General Matrix-Vector Multiply)
• Flash-Decoding
• Data Copy
• LDSM (ldmatrix instruction)
• Tensor Dequant
• TODO... (More features to come!)

GEMM

The GEMM implementation is optimized for performance. Below is a performance graph showcasing its efficiency:

Refer to the following blog:

• 基于 CUTE 的 GEMM 优化【1】—— Baseline 实现
• 基于 CUTE 的 GEMM 优化【2】—— 高效 GEMM 实现，超越 Cublas 20%

LDSM

Refer to the following blog:

• 基于 CUTE 的 GEMM 优化【1】—— Baseline 实现

---

We hope you find this repository useful for your learning and development needs. Contributions and feedback are welcome!