An Optimized Sparse Approximate Matrix Multiply

Authors:
  • NBNicolas Bock
    Nicolas Bock
  • MCMatt Challacombe
    Matt Challacombe

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 $\mathcal{O} (n \ln n)$ computational complexity with respect to matrix dimension $n$. We find that the max norm of the max norm achieved with a \SpAMM{} tolerance of below $2 \times 10^{-8}$ is lower than that of the single-precision {\tt SGEMM} for quantum chemical test matrices, while outperforming {\tt SGEMM} with a cross-over already for small matrices ($n \sim 1000$). Relative to naive implementations of \SpAMM{} using optimized versions of {\tt SGEMM}, such as those found in 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 with for quantum chemical matrices of RHF/STO-2G and RHF/6-31G${}^{**}$ water clusters.