Communication-optimal algorithms for contracting distributed tensors
Tensor contractions, generalized matrix multiplications that are time-consuming to calculate, make them among the most compute-intensive operations in several ab initio computational quantum chemistry methods. In this work, scientists from Pacific Northwest National Laboratory and The Ohio State University developed a systematic framework that uses three fundamental communication operators—recursive broadcast, rotation, and reduction, or RRR,—to derive communication-efficient algorithms for distributed contraction of arbitrary dimensional tensors on the IBM Blue Gene/Q Mira supercomputer. The framework automatically models potential space-performance trade-offs to optimize the communication costs incurred in executing tensor contractions on supercomputers. The paper documenting this work, "Communication-optimal Framework for Contracting Distributed Tensors," is a SC14 Best Paper award finalist.