How to Make Linear Algebra Go Fast

The blas was an effort at separating out the parts of linear algebra that a vendor could optimize from the parts oflinear algebra that a mathematician suter should write. And part of that was an insight about how the importance of memory locality has only increased over time. Where arithmetic is practically free on a computer memory band, with cost you money, hardware, essentially. But memory, late s is physics. You can't get round it.