GEneral Matrix Multiply (SGEMM)

Seen this term quite a lot.

BLAS vs. GEMM?

• BLAS is a broad collection of routines for linear algebra.
• GEMM is a single function within BLAS (Level 3) that is highly optimized because matrix-matrix multiplication is a core operation in computing.

Resources

GPU

• https://siboehm.com/articles/22/CUDA-MMM
  • https://github.com/siboehm/SGEMM_CUDA
  • His inspo: https://github.com/wangzyon/NVIDIA_SGEMM_PRACTICE

CPU

• https://siboehm.com/articles/22/Fast-MMM-on-CPU
• https://sahnimanas.github.io/post/anatomy-of-a-high-performance-convolution/
• https://github.com/flame/how-to-optimize-gemm

SGEMM

Related

• BLAS