Performance-Portable Sparse Tensor Decomposition Kernels on Emerging Parallel Architectures.

Abstract

—We leverage the Kokkos library to study perfor- mance portability of parallel sparse tensor decompositions on CPU and GPU architectures. Our result shows that with a single implementation Kokkos can deliver performance comparable to hand-tuned code for simple array operations that make up tensor decomposition kernels on a wide range of CPU and GPU systems, and superior performance for the MTTKRP kernel on CPUs.