Parallel sparse polynomial multiplication on modern hardware architectures
Francesco Biscani · 2012
We present a high performance algorithm for the parallel multiplication of sparse multivariate polynomials on modern computer architectures. The algorithm is built on three main concepts: a cache-friendly hash table implementation for the storage of polynomial terms in distributed form, a statistical method for the estimation of the size of the multiplication result, and the use of Kronecker substitution as a homomorphic hash function. The algorithm achieves high performance by promoting data access patterns that favour temporal and spatial locality of reference. We present benchmarks comparing our algorithm to routines of other computer algebra systems, both in sequential and parallel mode.