An Optimized Sparse Approximate Matrix Multiply for Matrices with Decay

Abstract

We present an optimized single-precision implementation of the Sparse Approximate Matrix Multiply (\SpAMM{}) [M. Challacombe and N. Bock, arXiv {\bf 1011.3534} (2010)], a fast algorithm for matrix-matrix multiplication for matrices with decay that achieves an $\mathcal{O}(n\log n)$ computational complexity with respect to matrix dimension $n$. We find that the max norm of the error achieved with a \SpAMM{} tolerance below $2\times 10^{-8}$ is lower than that of the single-precision {\tt SGEMM} for dense quantum chemical matrices, while outperforming {\tt SGEMM} with a cross-over already for small matrices ($n\sim 1000$). Relative to naive implementations of \SpAMM{} using Intel's Math Kernel Library ({\tt MKL}) or AMD's Core Math Library ({\tt ACML}), our optimized version is found to be significantly faster. Detailed performance comparisons are made for quantum chemical matrices with differently structured sub-blocks. Finally, we discuss the potential of improved hardware prefetch to yield 2--3x speedups.