High-Performance Matrix Multiplication Algorithms for Architectures withHierarchical Memories
John A. Gunnels, Greg Henry, Robert A. Geijn · 2001
During the last half-decade, a number of research eorts have centered around the development of software for generating automatically tuned matrix multiplication kernels. These include the PHiPAC project and the ATLAS project. The software products of both projects employ brute force to search a parameter space for blockings that accommodate multiple levels of memory hierarchy. We take a dierent approach. Using a simple model of hierarchical memories we employ mathematics to determine a locally-optimal strategy for blocking matrices.