Development of efficient computational kernels and linear algebra routines for out-of-order superscalar processors

We present methods for developing high performance computational kernels and dense linear algebra routines. The microarchitecture of AMD processors is analyzed with the goal to achieve peak computational rates. Approaches for implementing matrix multiplication algorithms are suggested for hierarchical memory computers. Block versions of matrix multiplication and LU-decomposition algorithms are considered. The obtained performance results for AMD processors are discussed in comparison with other approaches.