This thread is old, but for the sake of archives: BLIS actually tells you how to write a fast production large-matrix GEMM, and the papers linked from https://

This thread is old, but for the sake of archives: BLIS actually tells you how to write a fast production large-matrix GEMM, and the papers linked from https://github.com/flame/blis https://github.com/flame/blis would be a better reference than the Goto and van de Geijn. For small matrices see the references from https://github.com/hfp/libxsmm https://github.com/hfp/libxsmm but you're not likely to be re-implementing that unless you're serious about a version on non-x86_64. See also https://news.ycombinator.com/item?id=17061947 https://news.ycombinator.com/item?id=17061947